ESSAYS ON LABOUR AND HEALTH ECONOMICS By Asenka Asenova A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Economics - Doctor of Philosophy 2015 ABSTRACT ESSAYS ON LABOUR AND HEALTH ECONOMICS By Asenka Asenova Chapter 1 from the Russia Longitudinal Monitoring Survey in order to explore the extent to which the negative consequences of job loss of the household head extend to the health of the ch ildren in the family. We employ a linear model to study various health outcomes, and our estimation unemployment. In particular, we find support for an adverse the development of chronic conditions in children and occurrence of depression, but a beneficial evidence of a detrimental impact of paternal job loss on the probability the kid has low height for age, while there is possibility that children of unemployed parents are under - diagnosed in terms of chronic conditions. The latter has potential policy implications, pointing towards the need to make children health care and regular check - ups more broadly available. Chapter 2 the Elder Retirement in Europe to investigate the effect of retirement of the elderly on their psychological well - being and social inclusion. W e use an instrumental variable strategy based on plausibly exogenous variation in retirement probabilities induced by the country - level statutory and early retirement ages. The key findings of the study tell a consistent story: while labour force exit has no significant impact on the mental hea mental health . The results also suggest a heterogeneous effect of retirement on the social connectedness of the elderly: exiting the labour force decreases the size of social networks for men but n no effect for male retirees. This heterogeneity of the retirement effect has important policy implications, as it points out the possibility that the trends in the Euro pean Union towards increasing the pensionable ages could lead to a loss of welfare for women. Th e last Chapter uses data from the Programme for the International Assessment of Adult Competencies to re - examine the immigrant - non - immigrant earnings gap. We ex ploit the availability of cognitive skills measures in the data , such as numeracy and literacy scores, allowing us to minimize the presence of unobserved effects. Our analysis employs a modified Mincer earnings function and Oaxaca - Blinder mean log - wage dec omposition (Oaxaca (1973) and Blinder (1973)); we also make use of the decomposition technique by DiNardo, Fortin and Lemieux (1996) to examine the earnings gap across the entire earnings distribution. We find that immigrants have lower returns to educatio n than native workers, yet higher returns to literacy proficiency, which is conforming to the statistical discrimination literature. T he Oaxaca - Blinder decomposition results imply that a log - wage model specified without cognitive skill measures would overe stimate the unexplained part of the mean immigrant - non - immigrant gap nearly twice, while including numeracy and literacy test scores reveals lower role for discrimination. Lastly, the DiNardo - Fortin - Lemieux decomposition suggests that numeracy and literacy test scores matter almost equally throughout the entire log - wage distribution but cannot fully explain the observed immigrant - native gap, except for the bottom and the top decile. iv ACKNOWLEDGMENTS I would like to express my sincere gratitude to my advisor, Todd Elder , for his guidance, patience and support, and for always providing me with invaluable feedback and suggestions. His guidance was profoundly important not only for shaping this dissertation but , more importantly, for my formation as a researcher and teacher in the Economics field . I also wish to thank my other committee members , Stephen Woodbury, Leah Lakdawala and Peter Berg, whose comments helped me complete this dissertation , as well as enhance its quality . This dissertation also greatly benefitted from suggestions by other faculty members, including Steven Haider, Gary Solon and Jeffrey Wooldridge. Last but not least, I wish to thank my fellow graduate student Cristian Alcocer for his valuable feedback, as well as for his friends hip and support throughout my studies. All remaining errors are mine. v TABLE OF CONTENTS LIST OF TABLES .......... viii LIST OF FIGURES xiii CHAPTER 1 THE IMPACT OF HEALTH .. ................................................................................................. .............................. 1 2 1. 2 LITERATURE REVIEW .. ..... 4 1.2.1 Theoretical background ...... . 4 1.2. 2 Empirical evidence ....... ....... 6 1.3 DATA AND VARIABLES.............................................. ............................. ....... 9 1.3.1 Data 9 1.3.2 Variable definitions .. ... 10 1.3.2.1 Unemployment definition ................................................. ... 10 1.3.2.2 Health outcome variables .................................................. ... 12 1.4 DATA ANALYSIS 1.4.1 Child healthcare provisions in Russia 14 1.4.1 Child healthcare provisions in Russia ........... ... 14 1.4 .2 Sample statistics of the households ................................................. .... 16 1.4.3 Sample statistics of the children ... 18 1.4.4 Dynamics 21 1.5 ECONOMETRIC MODEL ... 1.5.1 Job Destruction in Russia 25 1.5.1 Job destruction in Russia 25 1.5.2 Econometric model 26 1.6 ESTIMATION RESULTS 31 1.6.1 Short run health indicators ... 31 1.6.2 Long run health indicators ... 35 1.6.2.1 Objective health measures ... 35 1.6.2.1.1 Chronic conditions....................... ..................... .... 35 1.6.2.1.2 Low height for age ............................................ .... 37 1.6.2.2 Subjective health measures ... 40 1.6.2.2.1 Child anxiety and depression ............................ .... 40 1.6.2.2.2 43 1.6.3 Tests for strict exogeneity of unemployment in the fixed effects estimation 45 1.7 CONCLUSIONS AND CAVEATS 49 APPENDICES . . 51 APPENDIX A : MAIN TABLES AND FIGURES ..................... .... .. ... ....... . 52 vi APPENDIX B : SUPPLEMENTARY TABLES AND FIGURES 98 BIBLIOGRAPHY .............. . . 108 CHAPTER 2 THE EFFECT OF RETIREMENT ON MENTAL HEALTH AND SOCIAL INCLUSION OF THE ELDERLY 111 112 2.2 LITERATURE .. 115 2.2.1 Theoretical background .. 115 115 2.2.2 Empirical evidence . 116 116 2.3 DATA AND VARIABLES........................................................................... .. 120 120 2.3.1 Data .. 120 120 2.3.2 Variable definitions .. 121 121 2.3.2 .1 Mental health measures . 121 12 1 1 2.3.2 . 2 Retirement definition .. 122 122 2.3.2.3 Social networks .. 123 123 2.3.3 Sample statistics .. 124 124 2.4 ECONOMETRIC MODEL 1.5.1 Job Destruction in Russia 127 127 2.5 ESTIMATION RESULTS 133 133 2.5.1 First stage 133 2.5.1.1 Statutory and early retirement ages, and actual retirement 133 2.5.1.2 Estimation results ... 135 135 2.5.2 Second stage 136 136 136 2.5.2 .1 Mental health by age distance to statutory and early retirement 136 138 138 2.5.2.2.1 Mental health ....................................................... 138 138 2.5.2.2.2 Social networks .................................................... 142 2.5.3 More on the gender heterogeneity and mechanism of the effect ....... 144 144 2.6 CONCLU DING REMARKS ........... 147 APPENDICES 149 149 APPENDIX C : MAIN TABLES AND FIGURES .. ...... ........ . 150 APPENDIX D : SUPPLEMENTARY TABLES AND FIGURES . 198 BIBLIOGRAPHY .. 235 CHAPTER 3 THE IMMIGRANT - NON - IMMIGRANT WAGE GAP REVISITED: EVIDENCE FROM THE PROGRAMME FOR THE INTERNATIONAL ASSESSMENT OF ADULT COMPETENCIES 240 3 241 3 . 2 LITERATURE REVIEW 244 3 .3 DATA AND DESCRIPTIVE STATISTICS .................. ................................. 249 249 vii 3 .3.1 Data 249 249 3 .3.2 Variable definitions .. 250 250 3.3.2.1 Wage and immigrant .. 250 3.3.2.2 Skill measures .. 250 3.3.2.3 Other covariates .. 252 3 .3.3 Sample statistics .. 252 3.4 IDENTIFICATION STRATEGY 1.5.1 Job Destruction in Russia 257 3.4.1 Mean wage gap decomposition 257 3.4.2 DiNardo - Fortin - Lemieux decomposition........................ .................. 260 3 .5 ESTIMATION RESULTS 265 3.5.1 Mean wage gap decomposition 265 3.5.1.1 OLS results 265 3.5.1.2 Oaxaca - Blinder decomposition .................... ...................... 268 3.5. 2 DiNardo - Fortin - Lemieux decomposition........................ .................. 272 3.6 CONCLUDING REMARKS 279 APPENDICES 281 APPENDIX E : MAIN TABLES AND FIGURES 282 APPENDIX F : SUPPLEMENTARY TABLES AND FIGURES . 307 BIBLIOGRAPHY .. . 345 viii LIST OF TABLES Table A.1: Self report and job holder definition vs. BLS definition of u nemployment.................................................................................................................. . 55 Table A.2: Child healthcare costs................................................................ ..................... 5 5 Table A. 3 : Sample statistics (households) ......... ......................................... ...................... 56 Table A. 4 : Sample statistics ( children ) .............. ......................................... ...................... 58 Table A.5: Variable dynamics around the time of unemployment (children).................. 70 Table A. 6 : Variable dynamics around the time of unemployment (spouse)..................... 72 Table A.7: Estimation results: (short run health indicators, 1994 - 1998 sample).............. 73 Table A. 8 : Estimation results: (short run health indicators, post - 1998 sample)............... 75 Table A. 9 : Estimation results: (short run health indicators, total sample)........................ 77 Table A.10: Estimation results: (number of chronic conditions, post - 1998 sample)........ 79 Table A.11: Estimation results (low height for age, 1994 - 1998 sample )......................... 81 Table A.12: Estimation results (low height for age, post - 1998 sample).. ......................... 83 Table A.13: Estimation results (low height for age, total sample)........... ......................... 85 Table A.14: Estimation results (anxiety or depression, 2003 - 2 005 sample).................... 87 - 1998 sample)............................................................................................................ .................. 89 - 1998 sample)............................................................................................................. ................. 9 1 Table A.17: Estimation results sample)...................................................................................................................... ........ 9 3 Table A.18: Child and household attrition.............................. .................... ...................... 9 5 ix Table A.19: Tests for strict exogeneity of unemployment in the fixed effects estimation.............................................................................................................. ............ 9 6 Table B.1: Sample statistics of the households (pre - 1998 vs. post - 1998)... ..................... 99 Table B.2: Sample statistics of the children (pre - 1998 vs. post - 1998)........ ..................... 10 1 Table B.3 .1 : Other robustness checks (number of chronic co nditions, post - 1998 sample). ............................................................................................................ ................. 10 4 Table B.3 .2 : Other robustness checks (anxiety an d depression, 2003 - 2005 sample)....... 10 6 Table C.1: Country representation ................................................................................... 151 Table C.2: Labour force status by gender ......................................................................... 152 Table C.3: Sample statistics ............................................................................................. 153 Table C.4 : Statutory, early and actual retirement ages by country and gender (wave 1 & 2) ................................................................................. .. ............ ...................... 156 Table C.5 : Statutory, early and actual retirement ages by country and gender (wave 4 ) ................................................................................. .. ............ .............................. 157 Table C.6: First stage estimation results (men) ...................... ........................................... 162 Table C. 7 : First stage estimation results ( wo men) ............................................................ 164 Table C.8: Second stage estimation results, Pooled OLS (death ideation, men).............. 174 Table C. 9 : Second stage estimation results , Pooled IV (statutory retirement age) (death ideation, men) .................................................. .. ............ ......................................... ... .................................. 17 5 Table C. 10 : Second stage estimation results , Pooled IV ( early retirement age) (death ideation, m en) .................................................. .. ............ ......................................... ... .................................. 176 Table C.11: Second stage estimation result s, Pooled 2SLS (death ideation, men)........... 177 Table C. 12 : Second stage estimation results, Pooled OLS (death ideation, women)....... 178 Table C. 13 : Second stage estimation results , Pooled IV (statutory retirement age) (death ideation, wo men) .................................................. .. ............ .................................... ... .................................. 179 x Table C. 14 Second stage estimation results , Pooled IV ( early retirement age) (death ideation, wo men) .................................................. .. ............ .................................... ... .................................. 180 Table C.1 5 : Second stage estimation result s, Po oled 2SLS (death ideation, women)...... 181 Table C.16: Second stage estimation results (mental health, men) ................................... 182 Table C.17: Second stage estimation results (mental health, women)........ ...................... 184 Table C.1 8 : Second stage estimation results ( social networks , men) ............................... 186 Table C.1 9 : Second stage estimation results ( social networks , wo men) .......................... 188 Table C.20: Tests for equality of the effect of retirement by gender................................ 190 Table C.21: Mechanism of the effect (men, death ideation)............... .............................. 192 Table C.22: Mechanism of the effect (men, demotivation index)............ ........................ 193 Table C.23: Mechanism of the effect (men, Euro - D scale)...................... ........................ 194 Table C.24: Mechanism of the effect (women, death ideation)................. ...................... 195 Table C.25: Mec hanism of the effect (women, demotivation index)................ ............... 196 Table C.26: Mechanism of the effect (women, Euro - D scale)........................ ................. 197 Table D.1: Age specification robustness checks ............................................................... 200 Table D.2: Standard error clustering robustness checks ................................................... 202 Table D.3: Fixed effects estimation results ....................................................................... 206 Table D . 4.1: Second stage estimation results (New EU - member states, men)................. 210 Table D . 4. 2 : Second stage estimation results (New EU - member states, women)............ 212 Table D . 215 Table D . 5. 2 : First stage estimation results (wo 217 Table D.5.3 : Second stage estimation results, Pooled OLS (death ideation, men)........... 219 xi Table D.5.4 : Second stage estimation results , Pooled IV (statutory retirement age) (death ideation, men) .................................................. .. ............ ......................................... ... ... ............................... 220 Table D.5.5 : Second stage estimation results , Pooled IV ( early retirement age) (death ideation, men) .................................................. .. ............ ......................................... ... .................................. 221 Table D.5.6 : Second stage estimation results, Pooled 2SLS (death ideation, men)......... 222 Table D.5.7 : Second stage estimation results, Pooled OLS (death ideation, women)...... 223 Table D.5.8 : Second stage estimation results , Pooled IV (statutory retirement age) (death ideation, wo men) ...................................... .. ............ ................................................ ... .................................. 224 Table D.5.9 : Second stage estimation results , Pooled IV ( early retirement age) (death ideation, wo men) .......................................... .. ............ ............................................ ... .................................. 225 Table D.5.10 : Second stage estimation results, Pooled 2SLS (death ideation, women)... 226 Table D.5.11: Second stage estimation results (mental health, men).......... ..................... 227 Table D.5.12: Second stage estimation results (mental health, women)..... ..................... 229 Table D.5.13: Second stage estimation results (social networks, men)..... ....................... 231 Table D.5.14: Second stage estimation results (social networks, women)....................... 235 Table E.1: Descriptive statistics (total sample)................................................................. 283 Table E.2: Descriptive statistics (immigrant/non - immigrant subsamples)....................... 286 Table E.3: OLS results (pooled total sample)................................................................... 296 Table E.4: OLS results (pooled restricted sample)........................................................... 298 Table E.5: Oaxaca - Blinder decomposition....................................................................... 300 Table E.6: DFL decomposition results for selected quantiles........................................... 305 Table F.1A: Country - level OL S results, test scores excluded (Canada)........................... 308 Table F.1B: Country - level OLS results, test scores included (Canada)........................... 309 Table F.2A: Country - level OLS results, test scores excluded (Ireland)........................... 310 xii Table F.2B: Country - level OLS results, test scores included (Ireland)............................ 311 Table F.3A: Country - level OLS results, test scores excluded (Austria)........................... 312 Table F.3B: Country - level OLS results, test scores included (Austria)............................ 313 Table F.4A: Country - level OLS results, test scores excluded (United States)................. 314 Table F.4B: Country - level OLS results, test scores included (United States................... 315 Table F.5A: Country - level OLS results, test scores excluded ( United Kingdom)............ 316 Table F.5B: Country - level OLS results, test scores included ( United Kingdom)............ 317 Table F.6A: Country - level OLS results, test scores excluded (Germany)........................ 318 Table F.6B: Country - level OLS results, test scores included (Germany)......................... 319 Table F.7A: Country - level OLS results, test scores excluded (Norway).......................... 320 Table F.7B: Country - level OLS results, test scores included (Norway)........................... 321 Table F.8A: Country - level OLS results, test scores excluded (Netherlands)................... 322 Table F.8B: Country - level OLS results, test scores included (Netherlands).................... 323 Table F.9A: Country - level OLS results, test scores excluded (Spain)............................. 324 Table F.9B: Country - level OLS results, test scores included (Spain ).............................. 325 Table F.10A: Country - level OLS results, test scores excluded (France).......................... 326 Table F.10B: Country - level OLS results, test scores included (France)........................... 327 Table F.11A: Country - level OLS results, test scores excluded (Estonia)......................... 328 Table F.11B: Country - level OLS results, test scores included (Estonia)......................... 329 xiii LIST OF FIGURES Figure A.1: Unemployment definit ions and resulting unemployment rate...................... 5 4 Figure A. 2: Probability the child has a regular physician ........................ ...... ................... 6 1 Figure A. 3: Probability the child was ever vaccinated ........ ................. ....... ..................... 6 2 Figure A. 4: Probability the child had a medical check up in the last 12 months.............. 6 3 Figure A. 5: Probability the child had a medical check up in the last 3 months................ 6 4 Figure A. 6: Probability the child was hospitalised in the last 3 months ..... ...................... 65 Figure A. 7: Probability the child has a diagnosed chronic condition .......... ..................... 66 Figure A. 8: Probability the child had any health problems in the last 30 days................ 67 Figure A. 9: Probabili ty the child has low height for age............................ ...................... 68 Figure A. 10: Probability the child suffers from anxiety or depression....... ...................... 69 ..................... 15 8 Figure C. 2: Retirement age histograms ( Sw itzerland ...................... 15 9 Figure C. 3: Retirement age histograms (Poland ........ ...................... 1 60 Figure C. 4: Retirement age histograms (Czech Republic ) ............. ....................... 16 1 Figure C.5: Mental health by age distance to statutory retirement age (death ideation) ................................................................................................................. 16 6 Figure C.6: Mental health by age distance to statutory retirement age (demotivation index)...................................................................................................... ... 16 7 Figure C. 7 : Mental health by age distance to statutory retirement age ( affective suffering index)............................................................................... .................. ................ 16 8 Figure C.8: Mental health by age distance to statutory retirement age (Euro - D scale)................................................................................................. .................. .............. 16 9 xiv Figure C. 9 : Mental health by age distance to early retirement age (death ideation)........ 1 70 Figure C. 10 : Mental health by age distance to early retirement age (demotivation index)...................................................................................................... ... ....................... 17 1 Figure C. 11 : Mental health by age distance to early retirement age ( affective suffering index)............................................................................... .................. ............................... 17 2 Figure C. 12 : Mental health by age distance to early retirement age (Euro - D scale)................................................................................................. ............................ .... 17 3 Figure E.1: Mean wage conditional on education .. .................................... . ...................... 28 9 Figure E.2: Mean wage conditional on labour market experience .............. ...................... 2 90 Figure E.3: Mean wage conditional on sector of employment .................... ..................... 2 91 Figure E.4: Mean wage conditional on skill - based occupational category....................... 29 2 Figure E.5: Mean wage conditional on numeracy test score ....................... ..................... 29 3 Figure E.6: Mean wage conditional on literacy test score ........................... ..................... 29 4 Figure E. 7 : Mean wage conditional on problem solving test score ............. ..................... 29 5 Figure E.8: Estimated densities of log - earnings (actual)....................... ...... ..................... 301 Figure E.9: Estimated densities of log - earnings (actual immigrants vs. counterfactual natives) ...................................................................................................... ........................ 30 2 Figure E.10: Estimated densities of log - earnings (actual natives vs. counterfactual natives) ...................................................................................................... ........................ 30 3 Figure E.11: Estimated log - earnings gap (actual vs. unexplained) ............. ...................... 30 4 Figure F.1: Total sample DFL results: specification (0) .............................. ..................... 330 Figure F.2: Total sample DFL results: specification (A) .................................. ................. 3 31 Figure F.3: Total sample DFL results: specification (B) ............................... .................... 3 32 Figure F.4: Total sample DFL results: specification (C) .................................. ................. 33 3 Figure F.5: Total sample DFL results: specification (D) ........... ...................... .................. 33 4 xv Figure F.6: Total sample DFL results: specification (E) .................................. ................. 33 5 Figure F.7: Restricted sample DFL results: estimated densities of log - earnings (actual).............................................................................................................. .. .............. 33 6 Figure F.8: Restricted sample DFL results: specification (0) ....................... .................... 33 7 Figure F.9: Restricted sample DFL results: specification (A) ....................... ................... 33 8 Figure F.10: Restricted sample DFL results: specification (B) ..................... .................... 33 9 Figure F.11: Restricted sample DFL r esults : specification (C ) ..................... ....................... 3 40 Figure F.12: Restricted sample DFL results : specification (D ) ..................... .................... 3 41 Figure F.13: Restricted sample DFL results : specification (E ) ..................... ..................... 34 2 Figure F.14: Restricted sample DFL results: specification (F) .................... ..................... 34 3 Figure F.15: Restricted sample DFL results: specification (G) ................... ..................... 34 4 1 CHAPTER 1 2 1.1 INTRODUCTION Up until the previous decade, the vast majority of economic literature focused on the - economic outcomes, with little or no attention being paid to families. Recent years have given rise to a number of studies examining the consequences of job loss for the entire family. For instance, Charles and Stephens (2004) and Ahituv and Robert (2005) show that spousal unemployment significantly increases the probability that the couple divorces. Moreover, Oreopou los (2008) demonstrates the existence of important intergenerational effects of job displacement and Oreopoulos et al. (2005) find strong support for the idea that parental employment and childhood poverty have causal effects on educational outcomes. Turni ng to transition economies, Kertesi and Kézdi (2007) estimate a substantial causal effect of unexpected long - term unemployment of the parents on the probability their kid drops out of secondary school. However, evidence on the effect of parental job loss o input for human capital formation (see e.g. Currie and Moretti (2007)); hence, a better understanding of the factors influencing it is of crucial importance. Indication of a significant long - term well - being of the children of displaced workers, and has potentially important policy implication s. Moreover, the answer to this research question may help shed light on the more The main question addressed in this paper is: is there evidence of a significant effect of ? The question of interest is examined using household - level longitudinal data on children and parents (working age household heads). All 3 data analysed comes from thirteen waves of the Russia Longitudin al Monitoring Survey of the University of North Carolina. The key findings of this study can be summarised as follows. First and foremost, in line responds diff main contribution of the study. To elaborate more on this, we find an adverse effect of male ren and child anxiety and depression. It is puzzling, however, that the opposite holds for short - term health indicators (measured as incidence of child health probl ems during the month before the impact the children. We also provide tentative evidence of a harmful impact of paternal job loss on the likelihood that a child is in g ood health, as well as on the probability the kid lags behind in the possibility that children of unemployed parents are under - diagnosed in terms of chronic di seases. The remainder of this paper is organised as follows. Section 1.2 discusses the theoretical framework behind this research question and reviews the relevant literature. Section 1.3 describes the data and variable definitions employed in the study. This is followed by detailed data analysis in section 1.4 . Section 1.5 outcomes and discussed the estimation strategies. Finally, section 1.6 presents the estimation 4 1.2 LITERATURE REVIEW 1.2 .1 Theoretical background modelled with the help of a s imple health production function approach. The first key factor his/her genetic endowment, and inherited from both of his/her parents. This stock establ ishes each child's susceptibility to certain diseases such as chronic heart disease, diabetes or learning - ion, and includes monetary and time resources, such as family income and available parental time both of which are affected by unemployment . It is important to note here that parental income and parental time have a direct ome (e.g. by offering better quality nutrition), as well as, an key to sustaining and improving his/her inherent health stock. In addition to this, various community - level facto production function, and may help maintain his/her health stock, or conversely lead to depleting it, even without any changes in the family income or time endowment. The most important example in this respect is t he ease to access public health care, as well as its quality. One other channel for the effect of parental unemployment to materialise is suggested by and causes them to engage in risky behaviours such as smoking and drinking. In the light of the production function itself, thus, making the production of health less effi cient even if all inputs 5 remained unchanged. Next, certain shocks, most evident of which accidents leading to trauma or injury, but also other events such as divorce or death in the family, may not only reduce the resources available to a child but may als o have a direct adverse effect on his/her health due to increased level of stress (see e.g. Mauldon (1990) discussing the effects of parental divorce). Lastly, it is worth mentioning an additional mechanism for the effect of parental job loss (1993)). In particular, the model ascertains that by optimally allocating their time between home and market activities, in addition to leisure, parents aim at maximizing the total utility of all household members (which is in turn determined by leisure, consumption and child outcomes). In this context, unemployment of one of the parents in a two - parent family might enforce a suboptimal allocation of labor in the household. To elaborate more on this, paternal job loss in a family with a working mother, for instance, may induce the father to take on a bigger share of home production, and if he is less efficient in providing care for the child than the mother, then this may poten Taking all this into consideration, the direction of the effect of parental job loss on their - priori. To elaborate more on this, economic theory suggests that a drop in household inco me resulting from parental job loss may make the family likely to decrease the quality and quantity of nutritional intake available to the children, as well as reduce the investment in their health. As a consequence, the children may suffer a depletion of their health stock. At the same time, however, job loss increases the amount of parental time spent particular, more time available to a parent may lead to an 6 production functio n is assumed to be increasing in both types of parental inputs income and time the theoretical prediction of the overall impact of unemployment in the family on of the two effects prevails. 1.2 .2 Empirical evidence is scarce, and draws mostly upon US data. Early work in the field indicated that poor health and infant mortalit y are more common among children of poor families (e.g. Mare (1982)); yet, - health may not be driven by poverty per se but by low parental education (Edwards and Grossman (1982)). More recently Dehejia and Lleras - Muney (2004) used U.S. nation - wide birth certificate data from year 1975 onwards, and reported that babies conceived in times of high unemployment have a reduced incidence of low and very low birth weight, fewer congenital malformations, and lower post neonatal mortality. At the same time, however, Lindo (2011) utilised detailed work and fe rtility histories from the Panel Study of Income Dynamics to explore the extent to which the health effects of job displacement extend to the children of displaced workers, and reached the opposite conclusion loss was found to have a signif icantly negative effect on infant health. The empirical evidence is even more perplexing when it comes to the effect of maternal maternal employment is related to the o utcomes of children, and reported that maternal labour 7 smoking and drinking. In addition, he observed that limited maternal market work during the child's fourt h through ninth year of age benefits children of low - income families. Yet, other studies found that maternity leave related non - health. One example in this respect is work by Baker and Milligan (2008), who examined an from work after birth, yet no effect on maternal or child health. In contrast, Liu et al. (2009) employed various parametric and non - parametric met hods to study data from the US National Longitudinal Survey of Youth 79 and reported that mother's full - time employment has an adverse effect on her children's body mass index and the likelihood of becoming overweight. There has not been much research on health outcomes outside the US, although the topic has gained interest in the past decade. Baten and Boehm (2010) investigated the effect of parental unemployment in the East Germany area on hropometric indicators, and reported that increasing unemployment is a major driving force for the decline in the average height of children. One of the few studies of the issue in developed economies was done by Liu and Zhao (2014), who utilised data from the China Health and Nutrition Survey and found that paternal job loss has a significant negative effect on appeared insignificant. Finally, Yasin et al. (2004) used data from Pakistan and concluded that Taken as a whole, several key findings emerge from the review of the body of recent literature on the impact of parental job loss on although some authors find insignificant or even significant beneficial effects. Secondly, for the 8 most part the literature focused on body weight and height indicators as the child health measure, behavioural outcomes, as well. Lastly, most of the empirical evidence dr aws upon studies of the effect of paternal unemployment, while relatively little attention has been paid to the issue of This paper adds to the body of literature by studying how unemployment of the family head in a sample of househ olds from the Russian Federation affects a number of child health outcomes, including both physical and psychological health indicators. Due to the rich data available, we are able to account for other parental and household - level indicators, as well as to include controls for community - level access to healthcare. In addition, we able to shed some health response to unemployment of the household head vary depending on whether the household head is male or female? 9 1.3 DATA AND VARIABLES 1.3 .1 Data A major challenge of the research question addressed in this paper is to find a source of plausibly exogenous variation in the family conditions under which children are being raised. This is why a preferred methodological strategy is to look at the effect of parental unemployment exogenous variation. Such a natural experiment is provided by the post - communist transition of many European countries that adopted large - scale subsidized employment during the communist system, but experienced fast and large job destruction once that system collapsed at the beginning of the 90s. During this period many workers (and amongst those, many parents), who had stable employment for most of their lives, lost their jobs for a prolonged period of time. In order to explore this phenomenon, we utilize annual micro - level data from the second round of the Russia Longitudinal Monitoring Survey 1 collected in the period 1994 to 2006. The RLM S - HSE is a series of nationally representative survey s in the Russian Federation aiming at studying the effects of the reforms in on the health and economic welfare of the households. An advantage of this dataset is that, in addition to data on household income and employment, it includes a wide variety of h ealth status variables of the household members over a period of nearly fifteen years. As the design of RLMS - HSE is based on dwelling - units (i.e. individuals are tracked if they remain in the same dwelling unit as in the baseline year), the sample is rest ricted to those households who remained in the same dwelling unit between years the years 1994 and 2006. The 1 The Russia Longitudinal Monitoring Survey, RLMS - HSE, has been organized and coordinated the Carolina Population Center at the University of North Carolina at Chapel Hill. Referred to as RLMS - HSE or RLMS henceforth. 10 analysis in the paper focuses on the employment status of the household head and attempts to infer on whether it has an effect on the health outco me of the children in that household. With this in mind, the sample is additionally restricted to families with working - age household heads, whose work status is either employed or unemployed. For the households in the remaining sample, head of household i s assigned according to the following demographic hierarchy: (1) the oldest working - aged male in the household, (2) if no working - aged males, then the oldest working - age female. The resulting sample consists of 1,637 households and 2,163 children (individu als under age of 14) in the base year 1994. 2 1.3 .2 Variable definitions 1.3.2.1 Unemployment definition reasons. A Bureau of Labour Statistics (BLS) definition of unemploym ent, which is available in RLMS - HSE, has the disadvantage that it only classifies as unemployed workers who have no job at present and have been actively searching for a job four weeks prior to being interviewed 3 , and this may not be well - suited to reflect the fact that Russia has been through a long period of transition and economic restructuring. Faced with massive job destruction, very few employment prospects, and long joblessness experience the unemployed individuals may have entirely altering their at titudes by becoming discouraged about their prospects of obtaining a job, and thus, lowering their search efforts. Hence, a BLS definition would not account for the 2 Due to high attrition the samples falling into the Moscow and St Petersburg regions were replaced with a new sample in year 1999. 3 BLS defines as active job search activities including, but not limited to: having a job interview, using an employment agen cy, sending out resumes, filling out applications, etc. In contrast, passive job - search methods include attending a job training program, or reading about job openings that are posted in the media. 11 discouraged workers, and presumably, these are the ones of major interest in view of the pa rticular research question in this paper as the effect of parental unemployment on their former communist countries, one could be working while not official ly employed (i.e. has no labour contract) or even while officially unemployed (i.e. registered at a state employment office as such), or could be searching less intensively for a job as s/he is engaged in unreported activities (see e.g. Grogan and van den Berg (1999)). Further, a registered unemployment definition is available in RLMS - HSE, as well; however, it has been reported that the unemployed in Russia often do not make use of the state employment agencies. For instance data from the Labor Force Survey showed that the true unemployment rate in the Russian Federation in the transition years was much higher than the official rate reported by the Federal Employment Service, which is limited to workers registered as unemployed at the local employment office s (Grogan and van den Berg (1999)). In this respect, employing a definition of unemployment status based on whether one is registered at an employment office is expected to severely underestimate the true unemployment rate and misclassify a significant fra ction of the unemployed. Since only a small fraction of the unemployed utilize the state employment offices, whether one is registered as unemployed or not, is not taken into consideration in the unemployment definition in the paper. Taking into account all these considerations and aiming at fully benefiting from the data available at RLMS - HSE, this paper utilises an unemployment definition based on a job holder indicator and a self report. In particular, it defines as unemployed an in dividual who has no job and considers him/herself unemployed, even if s/he had not been actively searching for a job in the past four weeks. Table A.1 shows a cross - tabulation of this definition and the BLS definition 12 of unemployment. As can be seen the tw o definitions are identical in defining the employed persons; however, they differ in classifying one as unemployed: while the BLS definition classifies 851 persons as not in the labour force, the definition used in this paper includes them in the category of unemployed. It is also worth noting that while the correlation between the two definitions is high (0.70) it is far from 1. 4 1.3.2.2 Health outcome variables con sensus in the relevant literature about the proper measurement of health, and this is even 5 As already mentioned the RLMS - HSE contains a wide range of health indicators such as parental ev aluation of their clinical data on the children, RLMS - HSE, like nearly all surveys has to rely on parental health 4 As Fig. 1 illustrates, the unemployment definition used in this analysis results in a pattern of the unemployment rate (among the household heads in the sample), which follows the pattern of the BLS definition and the definition based on one being registered at an employment and 9.13%); finally, the definition used in the paper is the highest (ranging from 4.93% to 16.97%). It is important to also note that all definitions show a peak of the unemployment rate in 1997, which is consistent with the findings of Earle and Brown, 2003 for a peak of job destruction ra tes and and 9.13%); finally, the definition used in the paper is the highest (ranging from 4.93% to 16.97%). It is important to also note that all definitions show a peak of the unemployment rate in 1997, which is consistent with the findings of Earle and Brown (2003) for a peak of job destruction rates and layoffs in Russia occurring in year 1997 (see later). 5 health outcome (see e.g. Mare, 1982). However, chil d mortality rates are relatively low in the Russian Federation; in addition, the under - 5 mortality rate in Russia has been consistently declining annually in the post - communist period (marking a drop from 27 children per 1,000 in 1990 to 12 in 2011 (UNICEF (2011)), contrary to the unemployment upheaval in the early years sample of about 3,000 children is likely to be very low even in a period of thirteen years. 13 reports, which raises the concern that these may be subject to bias. One issue with the use of parental reported health evaluation is that it might suffer from measurement error since perceptions of indivi even when employing reports of diagnosed health conditions: e.g. Bakes at al. (2004) compared e such errors in its reports are an issue of concern if the reporting error is in any way correlated to the key variable of interest. Based on this, we look at measurement error. The first group of indicators concerns child healthcare access and utilisation , and includes variables such as whether the child has a regular physician, whether it has had preventive medical care visits in the last 3 months and in the last 12 months, and vaccinations. Further, physical health status measures are analysed, such as p arental health evaluation of the the last thirty days, and anthropometric indicators (for children below seven years). In addition to l and mental health is also considered, based on a parental report of whether the child feels any anxiety or depression. Lastly, the paper looks at various community level health care access and quality indicators, as well as child healthcare expenses. 14 1.4 DATA ANALYSIS 1.4 .1 Child healthcare provisions in Russia The core provisions on child healthcare in Russia are stipulated in the Fundamentals of the existence of federal guarantees for free medical child care in all state and municipal health age). 6 In addition to the medical procedures prescribed at a federal le vel, the district authorities have the option to further extend the number of free medical procedures and treatments for children provided by the district hospitals and polyclinics, and to require the availability of qualified medical personnel in the nurs ing and child care facilities. Moreover, a Government Ordinance of June 21, 2003 prescribes that all children under the age of three have the right to free medication. Vaccination of infants and children against a number of diseases is also guaranteed by l aw and free of charge. Under such legislative framework, a drop in family in child health care. At the same time, however, there have been reports on breaches o f the federal legislation on children healthcare. One example in this respect comes from a series of inspections conducted by the Office of the Russian Prosecutor General in 2005, which found violations of the children rights to health care in numerous reg ions of the Russian Federation. 7 In particular, not only did the local authorities fail to fully implement the 1993 Act on child healthcare, but they often 6 2007, available at http://www.loc.gov/law/help/child - rights/pdfs/childrensrights - russia.pdf 7 Same as above. 15 intentionally decreased the number of free child health services provided in the district healthcare facilities below the federal guarantee. In addition to this, the inspection reported that no region in Russia had completely put into practice the Government Ordinance of 2003. The results from these inspections clearly show that at present the authoriti es in Russia fail to guarantee the implementation of the children healthcare rights stipulated in the legislation. The RLMS - HSE data also shows evidence suggestive of the lack of full compliance with the legislation regarding the federal child healthcare guarantees. In particular, over the period Table A.2 shows the sam ple mean responses: 14.16% of the RLMS - HSE households reported paying for the last medical appointment of their child, and 4.64% reported paying extra for medical tests and procedures. 8 It is important to note that these responses cannot be attributed to health clinics, and 11.79% were charged for additional medical tests in those clinics (v s. 50.22% and 69.28%, respectively for those who visited a private practice). Medicine also appears costly: medication, even though their child was less than three years old at the time. Finally, it is worth noting that 25.38% of the families incurred some travel costs to the medical facility. All the evidence presented in the above paragraphs indicates that regardless of the favourable legislation, child healthcare in R ussia involves certain costs. Therefore, households experiencing a drop in their income resulting from unemployment may be forced to reduce their healthcare investment in their children, which in turn may adversely affect their health. 8 The cited numbers are based on stratification - adjusted means. In addition, Table 2 reports the non - adjusted means as well. 16 1.4 .2 Sample statist ics of the households section of the paper compares the health outcomes of children of unemployed parents and those of employed parents. The key hypothesis of the s ubsequent analysis is that if parental job loss heath outcomes in the subsample of children of unemployed parents, and vice versa. Failure to observe significant differences in the health outcomes of the two groups of children might imply that the two opposing effects that of decresed parental income and of increased parental time available to the kids, are offsetting each other. Table A.3 presents the (stratification adjusted ) 9 statistics of the total sample of households, and separately for the two subsamples of employed and unemployed household heads. It is eviden t from here that households with unemployed heads and those with employed heads differ in several important ways. First and foremost, children of unemployed household heads are roughly 25% more likely to live in poor families, and this difference is signif icant at the 1% level both when looking at the all - Russia poverty line and at the regional poverty indicators. Secondly, these households tend to be larger they have more kids below six years of age, more kids in the age range seven to eighteen, and are more likely to live with a post - working age relative. Finally, such households are located in communities with significantly worse public 9 The RLMS - HSE employed a stratified sampling based on geographical factors and level of urbanization (stratum referred to as region). Oversampling was concentrated in large urban areas, where the highest non - response rate was e xpected. The post - stratification adjustment weights are based on the 1989 census and 1994 micro census for rounds 5 to 12 (years 1994 to 2002), and the 2002 census starting with round 13 (year 2003 onwards). The sampling weights are utilised in this paper in order to obtain consistent estimates of the population moments. In addition, standard errors are adjusted accordingly. 17 health services compared to households with employed heads they are about 15% less likely to have access to hospital and to paediatrician, and those with access to medical facilities live farther away from them. Further, an examination of the characteristics of the parents shows that household heads who are employed and unemployed do not appear different in terms of de mographics they are of similar age and gender structure. Even though the unemployed household heads are significantly less educated by one year, the sample mean of their years of education is still very high fifteen years, suggesting a large fraction o f university graduates in this group. Somewhat unexpectedly, however, the unemployed workers in the sample appear healthier than their employed counterparts: they are 10% less likely to suffer from a diagnosed chronic condition and the difference is highly significant (the difference being driven by spinal and gastrointestinal conditions). In addition, unemployed parents do not seem to show consistent signs of increased risky behaviours compared to their employed counterparts: the difference in the frequen cy of monthly alcohol consumption is significant but essentially zero in magnitude, and the two proxy for alcoholism). Cigarette smoking shows that unemployed par ents are considerably more likely to smoke, although both sample means being very high. non - household member 10 children from families where the head is unemployed are about 9% less likely to have been cared for by a non - household member, and the difference is significant at the 1% level. This gap remains high when restricting the attention to kids of pre - school age (age below seven years), although for both 10 - friends, workers teachers, or relatives who live separately. 18 subsamples parental care for small kids appears higher. The observed differences in parental time are in line with the idea that unemployment increases the available parental time, thus increasing the time spent with the child. It should be noted, however, that this gap may, at least in part, be driven by the fact that kids with an unemployed household head tend to live in larger families. 11 1.4 .3 Sample statistics of the children Turning to characteristics of the children of employed and unemployed household heads, the upper panel of Table A.4 suggests that children from both subsamples have similar demographics atistically significant. The table also suggests no significant differences between the two subsamples of kids in terms of immunization history 98% of the children in both samples obtained vaccines. In order to investigate the possibility that children o f unemployed parents have their immunisations delayed the analysis also presents the difference in vaccinations rates of babies aged one year or below (when most vaccinations are due). The results suggests that kids of unemployed parents might, indeed, be seeing some delay in their vaccinations they are nearly 5% less likely to be vaccinated by the age of one, but the difference is only significant at the 10% level. Further, kids from households where the household head is unemployed have significantly lo wer overall healthcare utilisation they are less likely to have had a routine medical check - up (i.e. not because of illness) in the last 3 or 12 months, and are less likely to have a regular physician, all differences being significant both in terms of m agnitude and statistically. Taken as 11 Table 1 in the Appendix presents the sample statistics of the households separately for the 1994 - 1998 sample and post - 1998 sample. The table illustrates that while employed and unemployed household heads, and t heir households, appear different in both periods, the sample means are closer for the period 1994 - 1998. 19 a whole, these observations give grounds to expect that kids of unemployed parents would have worse health outcomes as their parents seem to invest less in their health. At the same time, however, the lower panel of Ta ble A.4 presents a somewhat unexpected picture: on average, children from families with unemployed head appear less likely to have had health problems in the last 30 days, have better parent reported health evaluation 12 and appear less likely to suffer from anxiety or depression. 13 Perhaps even more surprising is the fact that these differences arise even when looking at particular diagnosed chronic conditions as there seems to be some evidence that children of unemployed parents are less likely to suffer fro m gastrointestinal and spinal conditions (significant at the 1% and 5% level, respectively). As can be expected, hospitalisation rates are very low in both children samples and the difference between them is not statistically significant. Yet, one health m arker on which children of unemployed parents seem to lag behind is the anthropometric indicator 14 height for age: the fraction of kids with low height for their age but normal weight for their height is 6% larger for kids of unemployed parents, while the o pposite is observed for the fraction of kids with both normal height for age and weight for height. This suggests that children of unemployed parents may possibly suffer from height impairment, possibly due to worse nutrition (as suggested by the significa ntly lower sample mean of daily caloric intake). 15 12 13 ears 2005, 2006 and 2007 of the survey; hence, the variable is only available in these years. Possible answers included into a single bin due to a very low fracti 14 Anthropometric indicators are only available for kids aged 7 or below. All are computed on the full RLMS - w eight. 15 Comparison of these statistics by time period suggests that gap between the fractions of children of low height for age has been increasing in time. The sample means for the period 20 Figure s A. 2 to A. 10 provide a closer look to the most important children health indicators Figure A.2 the probability the child has a regula r physician decreases linearly as the kids age for both groups of children. However, the fraction of kids with a registered doctor is always higher for kids of employed parents, and the gap seems to nearly double as the children get older than one year. Fi gures A. 4 and A. 5 illustrate a similar pattern in the probability that the child had a medical check - up in the last 12 and 3 months, respectively. It is evident from here that the fraction of kids who get to visit a doctor is very high (roughly 80%) for bo th groups of children in the first two years of their life, but considerably drops thereafter, the drop being much more pronounced for children of unemployed parents. A similar pattern can be seen in the medical check - up rates during the 3 months preceding the interview: children of employed parents are more likely to have had a As alr eady mentioned, no significant differences are observed in the likelihood of a child getting a vaccination between the two groups of kids, except for babies below 1 year of age. Figure A.3 reinforces this observation and, in addition, illustrates that virt ually all immunisations take place by the age of 2, regardless of parental work status. Further, Figure A.6 illustrates the child hospitalization rates in the last 3 months, confirming there are no important differences between the two groups of children: the two probabilities are not statistically different from each 1994 - 1998 are 0.090 and 0.125 (in household with employed/unemploy ed household head, respectively), and this difference is significant at the 5% level. However, for the post - 1998 sample the corresponding fractions are 0.075 and 0.195, or nearly 12pp (significant at the 1% level). This might suggest the effect of househol height (if any) takes longer time to materialise. 21 Next, Figure A.7 shows the probability that a child suffers from a chronic condition likely to have been diagnosed with a chronic illness for all ages above two. This may sugge st an effect of parental work status, but it might as well have genetic or environmental causes, as the fraction of employed parents suffering from chronic conditions is higher than that amongst unemployed parents. The pattern of the probabilities that a k id experienced a minor health problem in the last 30 days also reveals a gap favouring children of unemployed household heads. Lastly, Figure A.1 0 illustrates the probability that a child shows symptoms of anxiety or depression 16 , which suggests that childr en of unemployed parents are less likely to suffer from ages, which is in line with established facts in the medical and psychology literature typically repo rting signs of anxiety first shown as early as 7 9 months of age. 1.4 .4 Dynamics period right before and right after the household head loses his/her job, as well as in the period when s/he exited unemployment into employment (in case exit from unem ployment occurred). taking place around the time of unemployment. The results of this analysis are presented in Table s 5 . The left panel reports the sample means based on the full sample of household heads, who lost their job during the period 1994 - 2006. As can be seen from here, some of the child health characteristics show a considerable 16 The analysis uses larger age groups as the number of observations for this variable is smaller due to the fact that the question was only asked in years 2005 to 2007. 22 variation around the period of unemployment. In particular, there is a mental health indicator in the period when the household head becomes unemployed (the mean declines from 0.21 to 0.16), but anxiety and depression rise back to nearly the pre - unemployment level once the household head gets ree mployed (level of 0.18). In addition to this, the mean number of chronic conditions increases by nearly 0.03 (from 0.17 to 0.20) in the year when the household head gets unemployed as compared to the pre - unemployment level. The number of chronic illnesses remains at the increased level even after the family head transitions back into employment, likely due to the fact that once developed a certain chronic condition can only be treated but not cured. Equally important, the fraction of kids with low height fo r their age sees an increase by 4.5 percentage points in the year of parental unemployment (15.6% vs. 11.1%), and drops back once the household head gets reemployed (11.9% once reemployment occurs). At the same time, however, other variables show little o r no variation before and after the event of unemployment. For instance, the share of children who experienced some health problem in the 30 days preceding the interview marks a very slight drop when household head becomes unemployed, while the sample frac tion of kids who had a minor health issue seems to be trending up. The proportion of kids in bad health (based on a parental report) also increases to some extent in the time when unemployment occurs, and this is mirrored by a minor decline in the fraction of children in good health. In contrast, the percentage of children who got vaccinated also remains essentially unchanged, regardless of the unemployment transition of the household head. Further, it is interesting to note that some variation is noticeabl e in the health care - ups from 0.56 to 0.52 in the year the household head becomes unemployed, but those rise right after exit form 23 unemployment; no such pattern is visible in t he 3 - month check - ups. Lastly, the dynamics of the indicator of whether the child has a regular physician seems to be reflecting a downward trend. For completeness, the right panel of Table A.5 sample restricted to o nly those household heads for whom both entry and exit into unemployment were observed. To be more specific, Columns (4) and (5) now report the means can be see suggesting that unemployment might matter less in case transition into employment occurred. ound the time when the household head lost their job. It is important to note that the reviewed sample means may reflect a simple trend in the data and, while they do not necessarily imply that the change in possibility. Before concluding this section, it is also useful to look at the employment status of the spouse of the household head in the two - parent families, where the head experienced unemp loyment. These are shown in Table A.6 . As can be seen from here, spousal employment seems to drop in periods when the household head becomes unemployed, possibly reflecting the situation at the local labour market. At the same time, the fraction of wives out of the workf orce considerably declines in the period when the main earner loses his job (0.09 vs. 0.19 in the earners enter the labour force as a response to this adverse event. Yet, it seems from Column (2) that a large percentage of the secondary earners enter the labour force only to join the category of unemployed, rather than find a job the fraction of spouses who are unemployed noticeably 24 increases in the period when the house hold head gets unemployed, while the fraction of employed wives marks a decline. Virtually, the opposite spousal work status force transitions occur when the household head transits back into employment. The lack of an added worker effect points at the pos sibility that households whose head experiences unemployment fail to make up for the loss of income. The same conclusions prevail when looking at the restricted sample conditional on household head exiting unemployment. Taken as a whole, the evidence pres ented in this Chapter illustrates one notable pattern: while children from families where the household head is unemployed tend to have significantly worse preventive healthcare, this does not materialise in having worse health on the contrary, those chi ldren seem to have better health indicators. At first glance, this suggests that a parental However, it also raises an important question: could the observed di fference be due to children of unemployed parents being under diagnosed, as these kids are less likely to have visited a physician? This possibility is also supported by the analysis of the child heath indicators around the time of unemployment. Verifying any of these possibilities would require accounting for other parental and household indicators, capturing community - level effects, as well as allowing for the effect of parental unemployment to vary based on other characteristics of the parent. The next C hapter shall attempt to model this. 25 1.5 ECONOMETRIC MODEL 1 . 5.1 Job destruction in Russia Most studies, which analysed the job reallocation rates in the Russian Federation following the fall of the communist system, reach an agreement in identifying the time period marked by utmost job destruction. For instance, according to Earle and Brown (20 03), who used survey data from a sample of industrial enterprises in the Russian Federation during the period 1990 to 1999, the end - year job destruction rate started escalating in 1994 (reaching a level of 11.79 from 7.85 during the previous year); maintai ned high levels through 1998, before dropping to 5.94 in 1999. An analogous pattern was observed in the worker separation rates (layoffs): they sharply increased in 1994 compared to the previous year (level of 9.10 in 1994 vs. 6.69 in 1993), continued risi ng in the years to follow, and peaked at 14.71 in 1997. Similarly to the job destruction rates, the worker layoff rates saw a considerable decline in 1999, while the rehiring rates remained stable throughout the entire period of data, suggesting no recalls occurred. Finally, turning to employment growth rates, those appeared negative in all analyzed years except 1999, and their pattern mimicked the pattern of involuntary job separations; in particular, the employment turndown nearly doubled in 1994 reaching a level of - 9.6, remained high in the following years, before positive employment growth was marked in 1999. The general picture considerably increased in magnitude durin g the years 1994 to 1998, and this was especially pronounced for job destruction and involuntary separation rates. The pattern in job destruction rates described above suggests that earlier waves of RLMS - HSE could be utilized to explore the influx of mass ive layoffs and plant closure in Russia 26 those comprise a convincing case of an exogenous shock in household income. At the same time, however, the years post - 19 98 were marked by overall employment growth and job creation, suggesting that a different approach is required. In the latter case setting, solving the potential endogeneity of job loss becomes a major issue of concern. The main problem stems from the fact that individuals who get unemployed (either by being selected for a lay - off, or by voluntarily outcomes. Recent literature focused on a narrow category of job l osses job displacements, 17 arguing that those provide an exogenous shock to household income (see e.g. Lindo (2010)); however, data on job displacements is not available in RLMS - HSE making the approach implausible in this paper. In a natural experiment outcome variable and a binary treatment for parental unemployment as the major variable of estimati on of the causal effect of unexpected long - health. In the absence of a natural experiment, the central identifying assumption is that job loss provides an exogenous variation in family income, once unobserved time - invariant heterogeneity is accounted for. 1 . 5.2 Econometric model Consider the following hierarchical linear model: Health ijt 0 1 ParentalJobLoss jt + 2 + 3 + 4 + D t j + c ij + u ijt , (1) 17 According to the definition of the U.S. Bureau of Labor Statistics for a worker to qualify as a 27 where the left - hand - side variable, Health, represents the health outcome of child i in family j at time period t, ParentalJobLoss is a binary variable for whether the household head in family j experienced unemployment resulting from job destruction (in th e natural experiment setting of the early RLMS - HSE rounds) or any type of job loss (in the later rounds of data) at period t . X C is a vector of characteristics of the child (age category). Vector X H consists of parental and family - level controls including gender, education and, in some specification, health conditions of the household head, work force status of the spouse of the household head 18 , as well as a household wealth indicator and number of kids in the family aged below 7 and between 7 and 18 years. Finally, X R incorporates community - level indicators for the availability of health care (access to paediatrician), and controls for region of residence. Including regional dummies in the model is important for several reasons: first, they account for diff erential employment opportunities across region; secondly, they account for various environmental factors which play a role in certain health conditions; lastly, they are needed to correct for the fact the RLMS - HSE sample is stratified rather than random. 19 18 force status controls include several mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour for ce, single female head, single male head, and living with spouse but spousal labour force status missing. The definition of spouse unemployed uses the same definition of unemployed as for the household head: the spouse of the household head is considered u nemployed if they report holding no job and consider themselves unemployed. 19 Since the stratification in RLMS - HSE is based on an observable characteristic (region), the unweighted estimation including controls for the strata is consistent, as well as the weighted estimation. Unweighted estimation with controls for the strata is preferred in this paper as weighting makes cluster - robust inference a challenging issue. Eight main region categories are available in RLMS - HSE (those are: metropolitan areas Mos cow and St. Petersburg, Northern and North Western, Central and Central Black - Earth, Volga - Vaytski and Volga Basin, North Caucasian, Ural, Western Siberian, Eastern Siberian and Far Eastern), together with sub - region location categories for every region. 28 Further, D t represents time effects (both individual and household invariant). In particular, D t includes year effects to account for changes in the overall economic and environmental conditions, as well as developments in the overall state of the health care system; in addition, month effects are included to allow for seasonal patterns in certain health conditions. Error component j represents time - invariant unobserved household - level heterogeneity that stance, parental child abuse, nutritional habits of the parents or parental innate ability, all of which might be correlated with parental employment status). Error component c ij thought of as the certain health problems like heart disease, diabetes, or allergies. Finally, u ijt is an idiosyncratic error component which is assumed to be uncorre lated with all the right - hand - side variables. natural experiment setting relies on the assumption that the assignment to treatment mechanism (i.e. ParentalJobLos s jt ) is practically random from the viewpoint of the outcome variable Health ijt or formally: cov(ParentalJobLoss ijt, j )=0,cov(ParentalJobLoss ijt, c ij )=0 and cov(ParentalJobLoss ijt, u ijs )=0, t, s. If the key identifying assumptions of a natural experiment setting hold, both pooled OLS and fixed - effects estimators would consistently estimate the effect of interest. In the absence of a natural experiment, identification hinges on the assumption that cov(ParentalJobLoss ijt, u ijs )=0, t, s . This essentially means that parental job loss is uncorrelated with the time varying unobservable characteristics of the children and their families, which could alidity in the absence of a natural 29 experiment is failure of cov(ParentalJobLoss ijt, j )=0 , as household heads who lose their jobs (e.g. as a result of being fired) might be worse in their permanent unobservable charactestistics than those who remain emplo yed. One such example arises if household heads with lower innate abilities are also the ones, who are more likely to become unemployed, i.e. cov(ParentalJobLoss ijt, j )<0. Since parents with better abilities are likely to be better caregivers and make hi cov(Health ijt, j )>0. The last two conditions taken together would result in a downward inconsistency of the pooled OLS estimate of the parameter on interest 1 . In addition to this, condition c ov(ParentalJobLoss ijt, c ij )=0 may fail to hold as well if, for example, employers are more or less likely to fire parents of children with lower inherent health stock, or if parents of such kids are more likely to voluntarily remain unemployed longer in or der to provide care for a frail child (reverse causality). The latter may not be an issue of major concern as this paper looks at household heads, who are the main earner in their family, and their employment status is less likely to be affected by the hea lth of their children; yet, it does call for proper treatment. Lastly, cov(ParentalJobLoss ijt, c ij )=0 is violated if children respond in a heterogeneous way to the unemployment of the household head, and if this response varies by unobservable characterist ics of the kids. In cases as the ones described above, where unemployment of the parents is likely to be correlated with both the unobserved child effect c ij and the family effect j , eliminating c ij along with j becomes an attractive estimation strategy . 20 If the key identifying assumption 20 It is important to note that one benefit of using a linear model and fixed effects estimation is that the relationship between the unobserved heterogeneity j , c ij and the covariates X i it is unrestricted. Alternative nonlinear models, such as correlated random effects, impose additional restrictions on the distribution of the time constant unobserved heterogeneity conditional on the covariates (see Chamberlain (1980)). 30 cov(ParentalJobLoss ijt, u ijs )=0, t, s holds, and the unobserved child and family effects are, indeed, time invariant, the standard fixed - effect estimation at the individual level would produce a consistent estimate of the key parameter of interest in model (1). Finally, it is important to not e that since RLMS - HSE is an unbalanced panel, using fixed effects estimation leads to eliminating all observations which appear only once. This does not lead to selection bias under the assumption that selection into being observed only once is exogenous ( i.e. uncorrelated with u ijs ), and in cases where selection into being observed only once is correlated with the unobserved heterogeneity c ij or j , a fixed effects estimation would eliminate this source of selection bias. The estimation results of the mode l described above are presented in the next section. Another point is that, because of the nested structure of model (1), there are differen t fixed effects estimators available. Employing household - level fixed effects estimation eliminates the group effect j ; however, unit specific effects c ij are still part of the composite error term. Since there are reasons to suspect failure of both condi tion cov(ParentalJobLoss ijt, c i )=0 and condition cov(ParentalJobLoss ijt, j )=0 , this paper utilised individual level fixed effects estimation. 31 1.6 ESTIMATION RESULTS 1.6 . 1 Short run health indicators Tables 5A to 5C show the estimation results from model (1) with a binary variable of whether the child had any health problems in the last 30 days as the health outcome. Due to the differential trend in the job destruction rates in Russia in the period analysed, all model specifications are estimated separately on the 1994 - 1998 sample, post - 1998 sample and the total sample of data. The analysis shall comment on these in turn. Column (1a) of Table A.7 presents the estimation results from the 1994 - 1998 sample when employing a pooled OLS estimator, and without accounting for any child or household - specific characteristics that might have a ffected the outcome variable. As can be seen from here, children of unemployed parents appear 5.2pp less likely to have suffered from an illness in the past 30 days, ceteris paribus , even after accounting for regional differences, yearly time trends and po ssible seasonal patterns in certain health conditions. age (a binary indicator for age below 7), 21 household composition and parent characteristics, as well as acces s to paediatrician in the community. In addition, this specification allows for the impact of parental unemployment to differ depending on whether the household head is male or female in order to capture the fact that parents of different gender may differ in their abilities to provide childcare. 21 As shown in the data analysis section, the probability that a child experience certain health conditions considerably increases as the kid starts school, all model specifications control for age below 7. This is preferred to estimating model (1) separately on the below and above 7 years of age subsamples due to a low number of observations for particular heal th outcomes. Since we from zero at the conventional levels i n all specifications); as gender is time constant its inclusion in model when employing fixed estimation is of no relevance. 32 The coefficient on unemployment shows the partial effect of job loss for male household heads, and implies that compared to children in families with an employed male head, kids in households with an unemployed fa ther are about 5pp less likely to have experienced health on unemploymen t and the interaction term between unemployment status and female household head. The magnitude of this effect is also negative; however, it is not statistically different from zero at the conventional levels. Next, the specification reported in Column (1 c) adds an additional covariate to the regression total income of the household. It is important to note, that since the main channels available parental time , the model does not control for the time proxy (whether a non - household member cared for the child in the last 7 days ). However, looking at a specification with total income available to the household is meaningful as it consists of labour earnings and in come from other sources (e.g. property rents). As can be seen from here, the model is robust to adding the log of family income (the latter appearing insignificant at the conventional levels once labour force status of the household head and his spouse are accounted for). The results from the fixed effects estimation are presented in Columns (3a) to (3c) of Table A.7 . As can be seen from here, once unobserved time - invariant child and household heterogeneity has been eliminated, the parameter estimate on par ental unemployment drops somewhat in magnitude, and appears significant only at the 10% level 22 in the specification with 22 Note that the decline in significance is also due to the fact that the fixed effects estimation is based on a lower number of obs ervations: due to the unbalanced nature of the panel, all units which are only observed for one time period are not used in the fixed effects estimation. 33 no covariates. Moreover, the estimation results of the specifications with controls shown in Columns (3b) and (3c), suggest that neithe health problems in the last month. Finally, Panel POLS 2 of Table A.7 estimates all model specifications on a restrict ed sample of children, who were observed at least twice in the period 1994 - 2005. These results are reported for comparison purposes as they are obtained from the same sample as the sample used in the fixed effects estimation. It is evident from here that t he parameter estimates on the unemployment dummy are consistent with the results obtained by pooled OLS on the full sample; however, the coefficient on the interaction term between female household head and unemployment status now appears positive in magni tude and statistically significant. Turning briefly to the post - 1998 sample, the estimation results are presented in Table A.8 . The implications of these results differ than the ones obtained on the 1994 - 2005 RLMS - HSE sample in an interesting way. In part icular, all specifications and estimation methods point to the conclusion that, children of unemployed male household heads do not significantly differ in the probability to have suffered from a health problem in the 30 days prior to the interview. At the same time, the coefficient on the interaction term appears highly significant, positive and large in magnitude between 0.16 and 0.18 in the different specifications and estimation methods. This implies that children in families with an unemployed female head are nearly 18pp more likely to have undergone a medical problem in the previous month and the effect is 34 significant at the 10% level 23 (based on the results from the fixed effects estimation of specification (3c). Finally, the model is estimated on t he full sample of children in order to take advantage of the larger sample size, and the results are reported in Table A.9 . In contrast to the models estimated on the two subsamples of data, the estimation results based on the total sample overall reveal a n economically and statistically significant effect of parental unemployment for both that when looking at children in families with a male household head, those whose father is unemployed are about 3.5pp less likely to have suffered a health issue in the month prior to the interview, ceteris paribus, although the effect is only significant at the 10% level. Further, other factors equal, children in families with a single mother are more likely to experience health problems if the mother is unemployed, and the effect is both economically and statistically significant (magnitude of 0.958, significant at the 10% level in the fixed effects estimation from specificatio n 3c). Taken as a whole, the estimation results presented in this section tend to suggest that outcomes if the household head is male (suggesting a two - parent househ old 24 ). At the same time, on the likelihood the child has experienced a health problem recently, if she is the household head (i.e. if the child lives in a singl e parent household). 23 The p - value of the Wald test for significance of the sum of coefficients on parental unemployment an d female household head is 0.0654. 24 Only 1.15% of the sample of households with a male head report no presence of a working - age female in the family. In contrast, all of the households with a female head report no working age male. 35 1.6 . 2 Long run health indicators 1.6 . 2.1 Objective health measures 1.6.2.1 .1 Chronic conditions available in the RLMS - HSE prior to year 1998, which is why the paper only estimates model (1) - 1998 sample of data. The estimation results are shown in Table A.10 . As before, Column (1a) reports the pooled OLS estimation results from the model with no individual and household controls. As can be seen from here, after accounting for time trends and regional heterogeneity the difference between the predicted number of chronic conditions between children of unemployed and employed parents drops to zero. This is also supported by the pooled OLS estimation on the sample restricted to units observed at le ast twice, shown in Column (2a). At first glance, this seems at odds with the implications from the simple data analysis which suggested that kids of unemployed parents are healthier in terms of chronic conditions. However, the regression analysis looks at this difference after accounting for regional environmental factors, as well as for the availability of medical care in the community. The fact that once health care access has been controlled, children of unemployed parents no longer appear to have less predicted chronic conditions, might imply that the lower number of observed chronic illnesses for those kids is merely because they have not been diagnosed due to worse health care access, and not because they are healthier. At the same time, once other c ovariates are included and the effect of parental unemployment is allowed to vary by the gender of the parent, Column (1b) depicts a somewhat different story. The parameter estimate on the unemployment dummy in this specification 36 appears positive in sign a nd significant at the conventional 5% level, suggesting that having an conditions, ceteris paribus . It is important to note that since we are looking at male house hold heads, it is unlikely that the results are driven by reverse causality: the idea that the main earner in the family might prolong his unemployment spell in order to provide care for a frail child does not seem plausible. In contrast, the coefficient o n the interaction term between female household head and unemployment status is highly insignificant, indicating that maternal unemployment are validated when est imating the model on the restricted sample (shown in Panel POLS 2). Turning briefly to the results from the fixed effects estimation, the implications are significant at th e 5% level in specification (3b) and (3c) lending strong support for the idea that parameter estimate on paternal unemployment is 0.052, implying that kids in house holds with male unemployed heads are likely to develop about 0.05 m ore chronic conditions, on average, holding other factors fixed. It is important to note that even though this effect seems small in magnitude, it is of high economic significance as the sample mean number of chronic conditions for the children in RLMS - HSE is 0.19. In comparison, once unobserved child and household - insignificantly different from z ero in all specifications. Finally, although omitted from the estimation results reported in Table A.10 , it is worth chronic conditions (shown in Table A.3 in th e Appendix). This is an important consideration if 37 parental chronic conditions are correlated with their employment status (as suggested by the sample statistics) as medical theory predicts that children might inherit some chronic conditions from their par ents. The latter is supported by the estimation results the coefficient on household 0.064 in the fixed effects estimation). What is more relevant the estimat es on the key parameters remain essentially unchanged. unemployment increases the number of chronic conditions a child develops, ceteris paribus , while there is important, the estimation results indicate that once health care access has been controlled, children of unemployed parents no longer appear to have a lower number of predicte d chronic illnesses, suggesting their parents might be unaware of such conditions. This fits well with the increase the chance that a chronic disease is diagn osed and treated early, lessening the overall 1.6.2.1.2 Low height for age This subsection comments on the estimation result for the probability that a chi ld has low height for his/her age group. 25 Data availability allows estimating the model separately on the 25 Since anthropometric indicators are only available for children below 7, the sample is only for age z - score. Child in low height for age is defined as a child having low height fo r his/her age, regardless of whether s/he has low or normal weight for age (as the two subsamples of kids for age). The complementary (and mutually exclusive catego ry) is defined as normal height for age (again, regardless of weight). 38 pre - 1998 sample and on a sample from 1999 to 2003 ( Table A.11 and 7B, respectively), as well as on the total sample of children ( Table A.13 ). Apart from the potential endogeneity of unemployment due to unobserved ability of the household head, there is another reason why the pooled OLS estimates may suffer from omitted variable bias: parental height is not controlled in the model, while at the s ame time it is positively unemployment status, as suggested by s ome authors (see e.g. Cable and Judge (2004)). In addition, RLMS - yet, the Russian Federation is home to sizeable ethnic minorities, some of whom non - Caucasian. Since height of the household head is clearly time - invariant during the sample period, it is part of the unobserved heterogeneity j c ij ; for this reason the interpretation of the results only focuses on those obtained by fixed effects. These results for the sample 1994 - 1998 are shown in the rightmost panel of Table A.11 . As can be seen from here, once time constant unobserved child and h ousehold - level heterogeneity have been accounted for, unemployment status of the household head has no predictive power for the probability the child has low height, and this concussion is robust to allowing for heterogeneity of the effect by the gender of the household head. It is interesting to is negative 0.019 and significant at the 10% level, even after controlling for the employment status of the household head and his spouse, suggesting it might not unemployment per - se which 39 These observations are supported by the estimation results obtained from the 1999 - 2003 sample: the estimate on the log of household income is even larger in magnitude negative 0.036 , and significant at the 5% level, implying that a 10 - percent raise in family income would decrease the probability that a small child in the family lags behind in terms of height by 3.6pp, ceteris paribus . It is also worth mentioning that the coefficient on the interaction term between female head and unemployment status appears significant at the 10% level in the specification without an income control, lending support to the idea that children of unemployed single mothers are significantly less likely to have low height for age, compared to kids of employed male household heads, holding other factors fixed. However, once income of the household is accounted for, maternal u nemployment status appears irrelevant. Lastly, Table A.13 reports the results from model (1) estimated on the full sample of data. As before, parental job loss is insignificant at conventional levels in the specification with only time and regional control s. However, the results form Column (3b) differ from those obtained separately on the two subsamples: paternal unemployment now appears a significant predictor of the probability a child has low height. In particular, kids in families headed by an unemploy ed male are roughly 4.5pp more likely to lag behind their peers in terms of height, ceteris paribus , and there is no corresponding effect for female heads. 26 Yet, once household income is controlled, the estimate on paternal unemployment drops in significan ce, although its magnitude remaining unchanged. Moreover, family income is statistically different from zero at the 5% level and its magnitude is nearly 2pp; compared to the RLMS - HSE sample fraction of children in low height (0.09), this is a very large ef fect. 26 The p - value of the Wald test for the effect of maternal unemployment in specification (3b) estimated on the full sample of data is 0.3231. The corresponding p - values of the test from the pre - 1998 and post - 1998 sample estimation are 0.7128 and 0.4529, respectively. 40 The above paragraphs depict a somewhat unclear picture of the effect of parental job loss - robustness of the model to inclusion of a control for household income the estimation results lend only tentative support t evidence unambiguously points out that the income available to the family plays an important role in the likelihood the children in the househ old remain of low height compared to their peers. 1.6 . 2.2 Subjective health measures 1.6.2.2.1 Child anxiety and depression In years 2003 to 2005 t he child RLMS - d questionnaire, which makes it possible to study this metal health outcome in addition to studying the physical health of the children. One shortcoming of looking at this indicator is, however, the fact that the available sample size is relatively small, especially when employing fixed effects estimation. The estimation results of the binary response model for child anxiety and depression are presented in Table A.14 . The pooled OLS parameter estimate on unemployment in specification (1a) is - 0.05, implying that after accounting for seasonal effects and environmental factors (often considered as possible causes for depression), children of unemployed parents are nearly 5pp less likely to suffer from anxiety and depression, ceteris paribus , and the effect is statistically significant at the 10% level. Adding individual and group - specific controls in Column (1b) and allowing the effect on paternal job lo ss becoming insignificant at the conventional levels. At the same time, however, 41 the effect of maternal unemployment (obtained as the sum of the parameter estimates on the unemployment dummy and the interaction term) is negative, large in magnitude and sta tistically significant at the 5% level. 27 In particular, it implies that compared to children of working female household heads, kids of single mothers who are unemployed are 19.91 pp less likely to suffer from anxiety or depression, ceteris paribus . The mo del appears robust to inclusion of a household income - level indicator reported in Column (1c). As before, all model specifications are also estimated on the restricted sample of children, who were observed at least twice in the period 2003 - 2005, and the r esults are reported in Panel POLS 2 of Table A.14 . Taken as a whole, the parameter estimates somewhat drop, both in terms of magnitude and in significance, compared to the pooled OLS estimates on the entire sample. Most notably, the coefficient on the interaction term between female household head and her being unemployed is no longer significantly different from zero at the conventional levels in both specifications (2b) and (2c). Lastly, Columns (3a) to (3c) report the estimation results when employing individual level fixed effects estimation, which ha s the advantage of eliminating the time invariant unobserved heterogeneity related to child depression (such as parental inclination to morbidity and child abuse). As can be seen from here, the estimate on parental job loss in the model with no covariates is similar in magnitude to the one estimated by pooled OLS, although less precisely estimated. At the same time, however, the coefficient on male household head being unemployed unemployment increases the likelihood the child suffers from anxiety or depression by about 10pp (obtained from the specification with a household income control). Further, the interaction term between 27 P - value of the Wald test 0.0298. 42 household head being unemployed and female in Columns (3b) and (3c) also changes considerably in magnitude compared to the pooled OLS estimation, which is somewhat puzzling. Taken at face value, it implies that among families with a single mother, her unemployment reduces the probability that a child suffers from anxiety and depression by roughly 42.2pp, ceteris paribus , and this effect is significant at the 10% level (p - value of 0.0982). anxiety and depression does not se unemployment in the family is supported by recent empirical evidence. For instance, Rege at al. (2011) used Norwegian data on plant closure to investigate the effect of parental unemployment erse effect, while Similarly, Lindo (2013) studied the impact of economic downturns on child abuse, and established that male layoffs increase the rates of child abu se, whereas female layoffs reduce them. Moreover, a disparate effect of unemployment of fathers and mothers is in line with the psychology literature from the last decade documenting that the mental distress caused by unemployment is more severe for men th an women (see e.g. McKee - Ryan et al. (2005)). As a last remark, since chronic illness in children is often reported as a leading cause for depressive symptoms in children (see e.g. Bennett (1994)), model (1) is also estimated after accounting for presence of child chronic condition. In addition, as hospital stay and certain depression ( Miller et al. (2008) through Pinquart (2010)), a binary indicator for hospital st ay in 43 the past 3 months is included in this robustness check. The results are presented in Table A.4 in the Appendix and, overall, confirm the observations from the psychology literature hospital stay and presence of chronic illness are significant predi ctors of child anxiety and depression, although the later being not statistically significant in the fixed effects estimation. Most importantly, the estimates of the parameters on the unemployment dummy and the interaction term appear robust to inclusion o f these additional controls . 1.6.2.2.2 This section presents the estimation results of model (1) with the outcome variable - hand side variable is defined as a binary outcome for the probability that a child is in very good or good health versus average, bad or very bad health. 28 Table A.15 presents the estimation results from the 1994 - 1998 sample of data. As can be seen from here, the effect on parental unemployment is not significantly different from zero in all model specifications, regardless of estimation methods employed. This holds even when allowing for the effect to vary depending on the g ender of the household head. Hence, the estimation results give grounds to conclude that the unemployment status of mothers and fathers plays no role in determining the probability that their child is in good health, measured as the luation. It is worth noting, however, that when looking at two - parent families, wife of the household head being out of the labour force considerably increases the probability that the child is in good health, ceteris paribus , and the effect is significant at the 1% 28 This binning is preferred in order to address the issue that category very bad health (and to some extent category very good heath) ha s a low number of observations; moreover, this makes the distinction between good and bad health straightforward, and might possibly correct for reporting biases. 44 level in all specifications and large in magnitude (0.0613 in the pooled OLS and 0.0890 in the fixed effects estimation of specification (c). Further, Table A.16 reports the estimation results for the 1999 - 2006 RLMS - HSE sample. These results app ear similar with the implications of the pre - 1998 sample estimation, with one marginally significant at the 10% level in the model specifications with covariates acros s all estimation methods, lending some support to the idea that paternal unemployment might have a that children in families where the household head is unem ployed are about 4pp less likely to be in good health measured as a parental evaluation, other factors being equal. In contrast, the coefficient on maternal unemployment appears positive in all specifications, although not statistically different from zero at low levels. Lastly, Table A.17 estimates the response probability model for a child being in good health based on all years of data in order to take advantage of the extended sample size. As can be seen from here, the effect of parental unemployment is essentially zero in the specification specifications allowing for mothe different way, with the exception of specification (2c) where paternal unemployment appears negative but only marginally significant. As a final remark, it is interesting to note that the poo led OLS estimate on the interaction term between female head and unemployment status is large and significant; moreover, the sum of this coefficient and the coefficient on the 45 unemployment dummy is positive 0.07 and significant at the 10% level 29 ; yet, once fixed effects might be harmful for the children in the family. 1.6 . 3 Tests for strict exogeneity of unemployment in the fixed effects estimation The prec eding analysis showed some evidence suggesting that unemployment of the differs based on the gender of the household head. However, a major issue of concern remains even when fixed effects estimation is employed, namely: household heads, who become unemployed at a certain point of time might differ in their time - varying unobservables from those who are employed, and that it is this difference which is driving the results rather than job loss itself. In addition to this, as RLMS - HSE is a very unbalanced panel, attrition is worrisome even in the fixed effects estimation. One main source of attrition stems from the fact that, by design, RLMS - HSE is restricted to households which remain in the same dwelling unit. This implies that once a family or a family member change their dwelling unit, they leave the sample a complexity resulting in attrition rates in RLMS - HSE considerably higher than those in other longitud inal datasets. Attrition becomes a particular source of concern if families tend to change location as a response to unemployment of the household head. In this respect, Table A.18 29 P - value of the Wald test 0.0765. 46 presents the number of times each child and household was observed in the s ample, separately for the subsamples of employed and unemployed household heads, and it does seem from here that attrition is higher in the subsample of unemployed household heads. Moreover, there is an extra complication when studying children as childr en get older than 14 years they leave the children sample, in which case their health outcomes are no longer observed. 30 Lastly, an issue closely related to attrition is the fact that households might select themselves into being observed only once, and thi s selection may be correlated not only with the unobserved heterogeneity c ij and/or j (in which case a fixed effects estimator would eliminate the selection bias), but also correlated with the time - varying idiosyncratic error. In order to address these issues, this section performs a test for strict exogeneity of parental unemployment in the fixed effects estimation. Following Wooldridge (2010) a test for strict exogeneity using fixed effects consists of estimating an expanded v ersion of model (1) from section (IV): Health ijt = 0 1 ParentalJobLoss jt jt+1 + 2 + 3 + 4 + D t + j + c ij + u ijt , (1') where ParentalJobLoss jt+1 is the one - period lead of the d ummy for unemployment status of the household head. Under strict exogeneity, =0; hence, a test for strict exogeneity is a tests of the null hypothesis H 0 : =0 in equation (1') when employing a fixed effects estimation. The main idea is that if the estimat ion results are the capturing the effect of unobserved household or child - level heterogeneity rather than the causal effect of unemployment, then it is likely that future 30 The reason for this is that the adult questionnaire and the child questionnaires differ considerably in the outcomes they study, including the health outcomes. It should also be noted that there is yet another source of attrition in RLMS - HSE, namely the fact that the samples falling into the Moscow and St Petersburg regions were replaced with a new sample 1999. However, this particular type of attrition is not an issue of concern as it is random. 47 unemployment of the household head will have predictive power for the current health outcome of the child, even after controlling for current unemployment status (as future unemployment is correlated in a similar way to the unobserved effects). The results from the test are reported in Table A.19 . As can be seen from here, the coefficient on future unemployment status of the household head is not significantly different from zero at low levels for all health outcomes, with one notable exception. In particular, when modelling child depression/anxiety the lead of unemployment is significant at the 5% level in the specifications with controls, suggesting that that unemployment and/or household attrition is endogenous. In addition, when looking at the probability the child had any health problems in the last 30 days, the p - values of the test ar e comparatively low (although higher than 0.10) when the model is estimated on the pre - 1998 and total samples, providing some evidence against exogeneity of parental job loss. In contrast, for the rest of the analysed health outcomes the p - values of the te st are large, both when looking at the 1994 - 1998 and post - 1998 subsamples separately, and when estimating the model on the total sample of data. For instance, when the child health outcome is number of chronic conditions, the p - value of the test ranges fro m 0.299 to 0.669 in the different specifications, suggesting that there is no evidence against the assumption of strict exogeneity of unemployment in the of fixed effects estimation. It is hard to argue why the strict exogeneity assumption fails in the mo del for child anxiety and depression. However, one extra complication when examining this particular outcome may partly explain this, namely: since it is a subjective measure, it is possible that presence of a parent at home (due to him/her being unemploye d) makes them more likely to notice symptoms of anxiety/depression their kids might experience. This might mean that child anxiety and depression in cases where both parents are employed are underreported, leading to 48 non - classical measurement error and inc onsistency of the fixed effects results. This is not likely for example, as they are based on far more objective health indicators. Overall, while the results presented in this section are no firm proof that the estimation of the model of parental unemployment suffers from no potential issues, they do provide somewhat of a reassurance that the fixed effects estimation results are not driven by reverse causality and potential endogeneity of parental unemployment status, at least when examining health outcomes 49 1.7 CONCLUSIONS AND CAVEATS This paper used longitudinal from the Russia Longitudinal Monitoring Survey to study the consequences of unemployment of the household head on the health of the children in the family. Several important conclusions can be drawn from the presented analysis. Most importantly, in line with the findings of Rege et al. (2011) and Lindo (2013), our results indicate and this is considered the main contribution of the stu dy. In terms of specific health outcomes, employment status. Furt her, in line with the results of Ruhm (2008) we observes some beneficial Somewhat unexpectedly, however, the opposite is observed for the incidence of child health unemployment appears to negatively impact the children. We also find som e tentative evidence in good health based on a parental report, as well as on the likelihood the child has low height for age, and there are no corresponding i The implications of these findings are twofold. First, in terms of methodology, the results health needs to be investigated only when accounting for the interaction of this effect with the Secondly, some of the findings in the study have potentially important policy implications. For 50 instance , the fact that once access to healthcare has been accounted for, children of unemployed parents no longer appear to have a lower number of predicted chronic conditions, suggests that their parents might be unaware of such health issues and the kids might be under - diagnosed. This may be an issue of particular concern for kids aged 3 years and above as medical check - ups considerably decline after this age. In this respect, one policy suggestion might be to make children health care and regular check - ups more broadly available, e.g. by having medical offices in public schools and kindergartens converse to the trends in the post - communist states in the past decades. The results presented in this paper should be interpreted with care due to several important limitations. A major issue of concern remains the suspected endogeneity of parental unemployment (particularly when studying child anxiety and depression), which would invalidate the pooled OLS results and lead to questionable validity of the fixed effects estimation if the unobserved child and household - level heterogeneity is not time invariant. Another notable limitation of the analysis is the excessively high attrition rate in the RLMS - HSE sample, which is tcomes. However, correcting for attrition in model) assume that all right - hand side variables are always observed an assumption which in most cases fails t o holds as there are missing values in virtually all explanatory variables. For this reason this paper has made no attempt at addressing the issue of attrition bias. One would his type of research, which would eventually be able to account for the various econometric challenges and 51 APPENDICES 52 APPENDIX A MAIN TABLES AND FIGURES 53 Table A.1 : Self report and job holder definition vs. BLS definition of unemployment Definition Bureau of Labor Supply definition Self report and job holder definition Labour force status Employed Unemployed Not in the Labour Force Employed 16,187 0 0 Unemployed 0 892 851 Not in LF 0 0 0 Note: the numbers reflect total person - year observations. 54 Fig ure A. 1: Unemployment definitions and resulting unemployment rate 0% 2% 4% 6% 8% 10% 12% 14% 16% 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 Self report & Job holder definition BLS definition Registered at Employement Office definition 55 Table A.2 : Child healthcare costs Variable Stratified sample mean Non - stratified sample mean Paid for medical visit 0.046 0.046*** (0.003) 0.14 2 0.139*** (0.0 10 ) Visited a private medical facility 0.03 1 0.030*** (0.00 4 ) Paid for medical visit (district/city/village medical facility) 0.03 1 0.030 *** (0.00 4 ) (district/city/village medical facility) 0.11 8 0.11 6 *** (0.01 2 ) Paid for medical visit (private/commercial medical facility) 0.502 0.500 *** (0.087) (private/commercial medical facility) 0.69 3 0.696*** (0.056 ) Paid for medicine 0.72 6 0.726*** (0.01 2 ) Paid for medicine (child below 3 years of age) 0.525 0.526 *** (0.0 40 ) Any travel cost to medical facility 0.254 0.25 1 *** (0.008) Number of observations (children) 3,881 3,881 Number of clusters (households) 2,129 2,129 Notes: Variables are available for the period 1994 - 2005. 2) Missing stratified standard error because of stratum with a single sampling unit. Number of strata (regions) 144. 3) Test for zero population mean reported for the non - stratified sample means. *** denot es significance at the 1% level; ** denotes significance at the 5% level, and * denotes significance at the 10% level. 56 Table A. 3: Sample statistics (households) Characteristics Total sample Restricted samples Employed household head Unemployed household head Difference (Employed Unemployed) Household level Age of household head 37.554 (0.059) 37.122 (0.061) 37.334 (0.197) 0.21 2 (0.206) Years of education of household head 16.182 (0.02 7 ) 16.3 60 (0.028) 15.007 (0.091) 1. 292 *** (0.096) Number of children aged 0 - 6 in the household 0.676 (0.006 ) 0.657 (0.006) 0.84 9 (0.02 3 ) - 0.19 2 *** (0.023) Number of children aged 7 - 18 in the household 1.14 4 (0.007) 1.12 7 (0.007) 1.316 (0.02 6 ) - 0.1 90 *** (0.027) Number of post - working age females in the household 0.216 (0.003) 0.197 (0.003) 0.333 (0.01 2 ) - 0.136*** (0.121) Number of post - working age males in the household 0.07 1 (0.00 2 ) 0.06 5 (0.00 2 ) 0.122 (0.00 8 ) - 0.05 8 *** (0.00 8 ) Fraction of female household heads 0.13 1 (0.00 3 ) 0.12 8 (0.00 3 ) 0.125 (0.008) 0.003 (0.00 9 ) Total household monthly income (real, in rubbles) 10506.58 (210.574) 10807.12 (230.4 90 ) 7257.917 (295.31) 3549.20*** (374.61 2 ) Notes: 1) Means corrected for stratification; linearised standard errors. Number of strata (regions) 144. 2) Test for equality of means reported for the samples of employed and unemployed household heads. *** denotes significance at the 1% level; ** denotes significance at the 5% level, and * denotes significance at the 10% level. 3) Variables paediatrician/hospital; the variables are not available for year 2006. Chronic conditions are only available after 1998. 4) The total number of observations exceeds the sum of employed and unemployed as some households/children are observed in both states. 57 Household below the all Russia poverty line 0.4 20 (0.00 4 ) 0. 390 (0.00 9 ) 0.662 (0.015) - 0.2 72 *** (0.012) Household below the regional poverty line 0.270 (0.003) 0.242 (0.003) 0.502 (0.012) - 0.2 60 *** (0.01 3 ) Frequency of monthly alcohol use (household head) 4.28 4 (0.01 1 ) 4.275 (0.01 2 ) 4.375 (0.037) - 0.099*** (0.039) Drinks without eating (household head) 0.18 8 (0.003) 0.18 7 (0.00 4 ) 0.200 (0.012) - 0.014 (0.01 3 ) Household head a smoker 0.63 2 (0.004 ) 0.627 (0.004 ) 0.692 (0.01 2 ) - 0.065*** (0.012) Number of chronic conditions (household head) 0.5 30 (0.008) 0.525 (0.009) 0.42 6 (0.028) 0.09 9 *** (0.029) Non - household member cared for the child in the last 7 days 0.28 5 (0.003) 0.295 (0.00 4 ) 0.21 1 (0.010) 0.08 5 *** (0.01 1 ) Non - household member cared for the child in the last 7 days (kids below seven) 0.17 6 (0 .004 ) 0.18 0 (0.050 ) 0.138 (0.128) 0. 042 *** (0.019) Community level Fraction with access to paediatrician 0.8 48 (0.00 3 ) 0.858 (0.003 ) 0.71 3 (0.01 1) 0.14 6*** (0.011) Distance to paediatrician (in kilometres) 23.169 (0.405) 22.61 5 (0.499) 25.53 5 (0.668) - 2.920 *** (0.83 4 ) Hospital in the community 0.816 (0.00 3 ) 0.82 9 (0.003) 0.669 (0.01 2 ) 0. 159 *** (0.012) Distance to hospital (in kilometres) 19.7 10 (0.23 5 ) 18.622 (0.25 3 ) 24.207 (0.671) - 5.58 5 *** (0. 717 ) Distance to hospital (in hours) 0.20 3 (0.009) 0.175 (0.0 10 ) 0.30 9 (0.0267) - 0.13 4 *** (0.02 8 ) Number of observations (children) 6,509 6,203 1,054 Number of clusters (households) 4,214 4,049 678 58 Table A.4 : Sample statistics (children) Characteristics Total sample Restricted samples Employed household head Unemployed household head Difference (Employed Unemployed) Demographic Age 7.6 80 (0.029) 7.729 (0.032) 7.61 1 (0.09 8 ) 0.118 (0.10 3 ) Fraction of boys 0.50 8 (0.00 4 ) 0.508 (0.00 4 ) 0.5 10 (0.012) - 0.00 2 (0.01 3 ) Child healthcare access and utilisation Has a regular physician 0.58 1 (0.007) 0.591 (0.008) 0.45 7 (0.02 7 ) 0.13 4 *** (0.028) Had a medical check - up in the last 3 months 0.394 (0.00 4 ) 0.39 5 (0.004 ) 0.336 (0.01 3 ) 0.0 06 *** (0 .013) Had a medical check - up in the last 12 months 0.646 (0.00 6) 0.647 (0.006) 0.488 (0.02 2 ) 0.15 9 *** (0.022) Notes: 1) Means corrected for stratification; linearised standard errors. Number of strata (regions) 144. 2) Test for equality of means reported for the samples of employed and unemployed household heads. *** denotes significance at the 1% level; ** denotes significance at the 5% level, and * denotes significance at the 10% level. 3) Variable - up in the last 12 available for years 2003 to 2005. Anthropometric indicators only available for years 1994 to 2003. 4) The total number of observations exceeds the sum of employed and unemployed as some households/children are observed in both states. 59 T Ever vaccinated (all children) 0.98 1 (0.001) 0.980 (0.001) 0.98 3 (0.003) - 0.00 2 (0.00 3) Ever vaccinated (children aged 1 year or below) 0.933 (0.007) 0.93 8 (0.008 ) (0.89 1 ) (0.02 9 ) 0.04 9* (0.024 ) Health conditions Any health problems in the last 30 days 0.38 8 (0.004 ) 0.39 4 (0.00 4 ) 0.31 1 (0.011) 0.00 3 *** (0.012) Minor health problems in the last 30 days 0.202 (0.005) 0.208 (0.00 6 ) 0.12 6 (0.01 5 ) 0.08 3 *** (0.01 6 ) Hospitalised in the last 3 months 0.04 2 (0.001) 0.042 (0.00 2 ) 0.039 (0.00 5 ) 0.00 3 (0.005) Feels anxiety or depression 0.251 (0.007) 0.26 3 (0.008) 0.128 (0.020 7 ) 0.13 5 *** (0.022) Health evaluation Good 0.61 7 (0.00 4 ) 0.60 8 (0.0039) 0.695 (0.011) - 0.087*** (0.01 2 ) Average 0.357 (0.003) 0.366 (0.00 4 ) 0.278 (0.011) 0.08 8 *** (0.01 2 ) Bad 0.026 (0.001) 0.026 (0.001) 0.027 (0.00 4 ) - 0.00 1 (0.004) Chronic heart condition 0.02 5 (0.001) 0.026 (0.00 2 ) 0.0 20 (0.005) 0.006 ( 0.00 5 ) Chronic lung condition 0.021 (0.001) 0.020 (0.00 2 ) 0.01 4 (0.00 4 ) 0.00 9 (0.004) Chronic liver condition 0.011 (0.001 ) 0.01 2 (0.001 ) 0.00 7 (0.00 3 ) 0.00 5 (0.003) Chronic kidney condition 0.02 1 (0.001) 0.021 (0.00 2 ) 0.019 (0.005 ) 0.00 2 (0.00 5 ) Chronic gastrointestinal condition 0.045 (0.002) 0.048 (0.002) 0.02 9 (0.00 6 ) 0.0 19 *** (0.006) Chronic spinal condition 0.0 30 (0.00 2 ) 0.030 (0.00 2 ) 0.0 20 (0.00 5 ) 0.01 1 ** (0.005) Other chronic condition 0.09 0 (0.00 3 ) 0.091 (0.003) 0.0721 (0.0086) 0.01 9 ** (0.009) Diabetes 0.00 3 (0.000) 0.00 3 (0.000 ) 0.0028 (0.0012) - 0.000 (0.001) Any chronic condition 0.18 8 (0.00 4 ) 0.19 2 (0.004) 0.1327 (0.0114) 0.059 *** (0.012) 60 T able Anthropometric indicators (children aged 7 years or below) Height for age z - score - 0.130 (0.018) - 0.09 5 (0.0 120 ) - 0.3373 (0.0673) 0.242 *** (0.06 9 ) Weight for height z - score 0.277 (0.019 ) 0.259 (0.021) 0.3486 (0.0683) - 0.089 (0.07 2 ) Fraction with normal height for age and normal weight for height 0.851 (0.00 5 ) 0.860 (0.005 ) 0.8006 (0.0170) 0.06 0 *** (0.016) Fraction with normal height for age and low weight for height 0.058 (0.003) 0.05 7 (0.003) 0.0573 ( 0.0100) - 0.000 ( 0.01 1 ) Fraction with low height for age and normal weight for height 0.088 (0.00 4 ) 0.0 8 0 (0.00 4 ) 0.1371 (0.0146) - 0.057*** (0 .015) Fraction with low height for age and low weight for height 0.003 (0.00 1 ) 0.002 (0.00 1 ) 0.0049 (0.0028) - 0.00 3 (0.00 3 ) Nutrition Total daily caloric intake 1593.821 (5.58 2 ) 1602.54 (5.907) 1519.48 (16.932) 83.05*** (18.19 6 ) Percent calories from fat 31.359 (0.07 8 ) 31.536 (0.081) 29.852 (0.255) 1.68 4 *** (0.25 3 ) Percent calories from proteins 11.998 (0.025) 11.99 8 (0.02 7 ) 11.992 (0.077) 0.005 (0.08 6 ) Number of observations (children) 6,509 6,203 1,054 Number of clusters (households) 4,214 4,049 678 61 Figure A.2 : Probability the child has a regular physician 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% <1 year [1;2) years [2; 3) years [3; 4) years [4; 6) years [6; 9) years [9; 12) years > 12 years Household head unemployed Household head employed 62 Figure A.3 : Probability the child was ever vaccinated 50% 60% 70% 80% 90% 100% 110% <1 year [1;2) years [2; 3) years [3; 4) years [4; 6) years [6; 9) years [9; 12) years > 12 years Household head unemployed Household head employed 63 Figure A.4 : Probability the child had a medical check up in the last 12 months 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% <1 year [1;2) years [2; 3) years [3; 4) years [4; 6) years [6; 9) years [9; 12) years > 12 years Household head unemployed Household head employed 64 Figure A.5 : Probability the child had a medical check up in the last 3 months 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% <1 year [1;2) years [2; 3) years [3; 4) years [4; 6) years [6; 9) years [9; 12) years > 12 years Household head unemployed Household head employed 65 Figure A.6 : Probability the child was hospitalised in the last 3 months 0% 5% 10% 15% 20% 25% <1 year [1;2) years [2; 3) years [3; 4) years [4; 6) years [6; 9) years [9; 12) years > 12 years Household head unemployed Household head employed 66 Figure A.7 : Probability the child has a diagnosed chronic condition 0% 5% 10% 15% 20% 25% 30% <1 year [1;2) years [2; 3) years [3; 4) years [4; 6) years [6; 9) years [9; 12) years > 12 years Household head unemployed Household head employed 67 Figure A.8 : Probability the child had any health problems in the last 30 days 0% 10% 20% 30% 40% 50% 60% <1 year [1;2) years [2; 3) years [3; 4) years [4; 6) years [6; 9) years [9; 12) years > 12 years Household head unemployed Household head employed 68 Figure A.9 : Probability the child has low height for age to 7 years as child height and weight indicators in RLMS - HSE are only available for kids aged 7 or below. 0% 5% 10% 15% 20% 25% <2 years [2; 4) years [4; 6) years >= 6 years Household head unemployed Household head employed 69 Figure A.1 0 : Probability the child suffers from anxiety or depression Note: For child anxiety/depression age is grouped in larger intervals compared to graphs A to G, in order to avoid having categories with less than 50 observati ons, as the variable is only a vailable for years 2003 - 2005. 0% 5% 10% 15% 20% 25% 30% 35% 40% <5 years [5; 9) years [9; 12) years >= 12 years Household head unemployed Household head employed 70 Table A.5 : Variable dynamics around the time of unemployment (children) Characteristics Full sample Restricted sample (both entry and exit into unemployment observed) (1) Period right before becoming unemploy ed (2) Period right after becoming unemployed (3) Period right after exiting unemployment (4) Period right before becoming unemployed conditional on exiting (5) Period right after becoming unemployed conditional on exiting (6) Period right after exiting unemployment Child of the household head Has a regular physician 0.481 0.473 0.416 0.436 0.447 0.338 Had a medical check - up in the last 3 months 0.33 3 0.33 6 0.350 0.341 0.333 0.358 Had a medical check - up in the last 12 months 0.55 6 0.516 0.542 0.527 0.501 0.514 Note s : 1) Means corrected for stratification. Number of strata (regions) 144. 2) During the entire sampling period, there are 1,743 observations of unemployed household heads. The table reports lower number s oyed when first observed or there was a gap in the panel) and/or exit from unemployment is not observed (i.e. household head was s till unemployed when last observed or there was a gap in the panel). The number of such cases is 508 and 499, respectively. 71 Table A.5 Ever vaccinated (all children) 0.98 6 0.987 0.988 0.985 0.987 0.994 Any health problems in the last 30 days 0.323 0.315 0.327 0.305 0.325 0.318 Minor health problems in the last 30 days 0.14 7 0.151 0.164 0.134 0.148 0.180 Ever vaccinated (all children) 0.98 6 0.987 0.988 0.985 0.987 0.994 Any health problems in the last 30 days 0.323 0.315 0.327 0.305 0.325 0.318 Minor health problems in the last 30 days 0.14 7 0.151 0.164 0.134 0.148 0.180 Hospitalised in the last 3 months 0.034 0.02 7 0.032 0.036 0.030 0.033 Feels anxiety or depression 0.20 8 0.15 6 0.184 0.138 0.159 0.202 Health evaluation Good health 0.687 0.66 8 0.674 0.685 0.683 0.675 Average health 0.29 7 0.30 5 0.306 0.299 0.298 0.314 Bad health 0.016 0.027 0.020 0.016 0.018 0.011 Number of chronic conditions 0.172 0.203 0.201 0.101 0.121 0.175 Child has low height for age 0.112 0.156 0.119 0.114 0.151 0.126 Number of observations 661 664 633 437 437 437 72 Table A.6 : Variable dynamics around the time of unemployment (spouse) Characteristics Full sample Restricted sample (both entry and exit into unemployment observed) (1) Period right before becoming unemploy ed (2) Period right after becoming unemployed (3) Period right after exiting unemployment (4) Period right before becoming unemployed conditional on exiting (5) Period right after becoming unemployed conditional on exiting (6) Period right after exiting unemployment Spouse of the household head Spouse employed 0.55 7 0.2 66 0.532 0.543 0.267 0.563 Spouse unemployed 0.065 0.465 0.078 0.070 0.462 0.082 Spouse not in the labour force 0.1921 0.0905 0.192 0.209 0.098 0. 186 Number of observations 661 664 633 437 437 437 73 Table A.7 : Estimation results: (short run health indicators, 1994 - 1998 sample) Controls Dependent variable: child had any health problems in the last 30 days 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1994 - 1998 Estimation method POLS 1 POLS 2 FE Household head unemployed - 0.051*** (0.018) - 0.053** (0.022) - 0.053** (0.023) - 0.040* (0.023) - 0.044* (0.026) - 0.065*** (0.019) - 0.045* (0.026) - 0.043 (0.030) - 0.044 (0.031) Household head female 0.016 (0.020) 0.016 (0.021) - 0.019 (0.018) - 0.007 (0.018) 0.056 (0.045) 0.054 (0.045) Household head unemployed & female 0.038 (0.060) 0.024 (0.060) 0.126** (0.054) 0.150*** (0.052) 0.073 (0.085) 0.055 (0.088) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denote s significance at the 5% level, * denote s significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the fo llowing mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour force, single female head (same category as household head female), single male head, living with spouse but spousal labour force st atus missing; number of children in the household aged below 7; numb er of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and month of interview. 74 Table A.7 Child's age below 7 years - 0.020 (0.023) - 0.025 (0.023) 0.063*** (0.020) 0.083*** (0.021) 0.001 (0.040) 0.005 (0.040) years of education - 0.015 (0.010) - 0.016 (0.010) - 0.013 (0.009) - 0.014 (0.010) - 0.032* (0.0193) - 0.034* (0.020) years of education squared 0.001* (0.000) 0.001* (0.0003) 0.001 (0.000) 0.001 (0.000) 0.001 (0.001) 0.001* (0.001) Log of real total household income (in Rubles) 0.004 (0.006) 0.002 (0.006) 0.004 (0.008) Region (time invariant) yes yes yes yes yes yes no no no Number of children 4,567 4,412 4,379 2,240 2,185 2,166 2,240 2,185 2,166 Number of clusters (households) 3,040 2,928 2,911 1,530 1,488 1,479 1,530 1,488 1,479 Time periods (unbalanced) 5 5 5 5 5 5 5 5 5 75 Table A.8 : Estimation results (short run health indicators, post - 1998 sample) Controls Dependent variable: child had any health problems in the last 30 days 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1999 - 2006 Estimation method POLS 1 POLS 2 FE Household head unemployed 0.024 (0.018) 0.004 (0.023) 0.010 (0.0237) 0.033 (0.021) 0.013 (0.027) 0.022 (0.028) 0.014 (0.021) - 0.006 (0.027) - 0.003 (0.0279) Household head female - 0.013 (0.021) - 0.010 (0.022) - 0.027 (0.026) - 0.019 (0.027) - 0.016 (0.041) - 0.019 (0.0418) Household head unemployed & female 0.177*** (0.064) 0.173*** (0.065) 0.171** (0.080) 0.162** (0.081) 0.176** (0.082) 0.181** (0.083) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denote s significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour force, single female head (same category as household head female), single male head, living with spouse but spousal labour force st atus missing; number of children in the household aged below 7; number of children in the household aged 7 to 18 ; an indicator of whether the household has access to paediatrician; year and month of interview. 76 Table A.8 Child's age below 7 years 0.0 50 ** (0.02 1 ) 0.05 1 ** (0.021 ) 0.071 *** (0.02 5 ) 0.074*** (0.025) 0.02 2 (0.034 ) 0.024 (0.034) Household years of education 0.005 (0.01 1 ) 0.006 (0.01 1 ) 0.01 1 (0.01 4 ) 0.01 2 (0.01 4 ) - 0.00 4 (0.019) 0.00 1 (0.0 20 ) years of education squared 0.000 (0.000) 0.000 (0.000) 0.000 (0.00 1 ) - 0.000 (0.000) 0.000 (0.001 ) 0.000 (0.00 1 ) Log of real total household income (in Rubles) 0.009 (0.008) 0.016 (0.01 1 ) - 0.000 (0.012 ) Region (time invariant) yes yes yes yes yes yes no no no Number of children 3,054 2,555 2,547 2,152 1,813 1,809 2,152 1,813 1,809 Number of clusters (households) 2,190 1,817 1,813 1,598 1,338 1,335 1,598 1,338 1,335 Time periods (unbalanced) 8 8 8 8 8 8 8 8 8 77 Table A.9 : Estimation results (short run health indicators, total sample) Controls Dependent variable: child had any health problems in the last 30 days 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1994 - 2006 Estimation method POLS 1 POLS 2 FE Household head unemployed - 0.020 (0.013) - 0.033** (0.016) - 0.030* (0.016) - 0.003 (0.015) - 0.017 (0.019) - 0.013 (0.019) - 0.019 (0.015) - 0.038* (0.019) - 0.035* (0.019) Household head female 0.002 (0.015) 0.002 (0.015) - 0.008 (0.018) - 0.006 (0.019) 0.010 (0.028) 0.008 (0.028) Household head unemployed & female 0.102** (0.044) 0.091** (0.045) 0.154*** (0.053) 0.141*** (0.053) 0.136** (0.054) 0.128** (0.055) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denote s significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour force, single female head (same category as household head female), single male head, living with spouse but spousal labour force st atus missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and month of interview. 78 Table A.9 Child's age below 7 years 0.009 (0.015) 0.007 (0.015) 0.048** (0.022) 0.0500** (0.022) 0.006 (0.026) 0.008 (0.025) Household years of education - 0.008 (0.008) - 0.008 (0.008) - 0.006 (0.010) - 0.005 (0.010) - 0.019 (0.013) - 0.018 (0.013) years of education squared 0.001* (0.000) 0.000* (0.000) 0.000 (0.000) 0.000 (0.000) 0.001 (0.001) 0.001 (0.001) Log of real total household income (in Rubles) 0.005 (0.005) 0.002 (0.006) 0.009 (0.006) Region (time invariant) yes yes yes yes yes yes no no no Number of children 6,506 5,867 5,830 3,807 3,422 3,406 3,807 3,422 3,406 Number of clusters (households) 4,213 3,739 3,720 2,526 2,230 2,222 2,526 2,230 2,222 Time periods (unbalanced) 13 13 13 13 13 13 13 13 13 79 Table A.10 : Estimation results (number of chronic conditions, post - 1998 sample) Controls Dependent variable: number of chronic conditions 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1999 - 2006 Estimation method POLS 1 POLS 2 FE Household head unemployed 0.017 (0.018) 0.050** (0.021) 0.052** (0.021) 0.035 (0.022) 0.059** (0.025) 0.062** (0.025) 0.026 (0.019) 0.052** (0.021) 0.052** (0.022) Household head female 0.071** (0.029) 0.075** (0.030) 0.097** (0.039) 0.102*** (0.039) 0.041 (0.036) 0.041 (0.037) Household head unemployed & female - 0.062 ( 0.077) - 0.058 (0.078) 0.015 (0.109) 0.017 (0.113) - 0.025 (0.090) - 0.018 (0.092) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force sta tus of the spouse/cohabiting partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour forc e, single female head (same category as househ old head female), single male head, living with spouse but spousal labour force status missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatr ician; year and month of interview. 80 Table A.10 Child's age below 7 years - 0.095*** (0.025) - 0.095*** (0.026) - 0.085*** (0.028) - 0.084*** (0.028) - 0.028 (0.031) - 0.029 (0.031) years of education - 0.029** (0.012) - 0.029** (0.012) - 0.044*** (0.015) - 0.044*** (0.015) - 0.035** (0.015) - 0.035** (0.015) years of education squared 0.001** (0.000) 0.001** (0.000) 0.001*** (0.001) 0.001*** (0.001) 0.001* (0.001) 0.001* (0.001) Log of real total household income (in Rubles) 0.007 (0.009) 0.009 (0.009) - 0.000 (0.011) Region (time invariant) yes yes yes yes yes yes no no no Number of children 3,039 2,542 2,534 2,147 1,808 1,804 2,147 1,808 1,804 Number of clusters (households) 2,184 1,809 1,805 1,596 1,334 1,331 1,596 1,334 1,331 Time periods (unbalanced) 8 8 8 8 8 8 8 8 8 81 Table A.11 : Estimation results (low height for age, 1994 - 1998 sample) Controls Dependent variable: child has low height for age 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1994 - 1998 Estimation method POLS 1 POLS 2 FE Household head unemployed 0.021 (0.019) 0.035 (0.023) 0.027 (0.023) 0.047* (0.028) 0.072** (0.034) 0.068** (0.035) 0.026 (0.027) 0.052 (0.034) 0.050 (0.035) Household head female - 0.003 (0.020) - 0.007 (0.021) 0.009 (0.031) 0.009 (0.032) 0.024 (0.038) 0.028 (0.036) Household head unemployed & female - 0.053 (0.043) - 0.051 (0.045) - 0.137*** (0.052) - 0.148*** (0.053) - 0.009 (0.047) - 0.006 (0.048) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force status of the spouse/c ohabiting partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour forc e, single female head (same category as household head female), s ingle male head, living with spouse but spousal labour force status missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and mon th of interview. 82 Table A.11 Child's age below 7 years years of education 0.008 (0.010) 0.006 (0.010) 0.020 (0.015) 0.020 (0.016) - 0.017 (0.019) - 0.019 (0.019) years of education squared - 0.000 (0.000) - 0.000 (0.000) - 0.001 (0.001) - 0.001 (0.001) 0.001 (0.001) 0.001 (0.001) Log of real total household income (in Rubles) - 0.014* (0.007) - 0.008 (0.009) - 0.019* (0.010) Region (time invariant) yes yes yes yes yes no no no Number of children 1,955 1,913 1,891 872 860 847 872 860 847 Number of clusters (households) 1,566 1,527 1,509 704 692 683 704 692 683 Time periods (unbalanced) 5 5 5 5 5 5 5 5 5 83 Table A.12 : Estimation results (low height for age, post - 1998 sample) Controls Dependent variable: child has low height for age 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1999 - 2003 Estimation method POLS 1 POLS 2 FE Household head unemployed 0.030 (0.029) 0.061 (0.040) 0.047 (0.041) 0.048 (0.038) 0.101* (0.054) 0.095* (0.053) 0.020 (0.036) 0.052 (0.048) 0.037 (0.048) Household head female 0.070** (0.035) 0.055 (0.035) 0.086* (0.048) 0.073 (0.048) 0.072 (0.059) 0.056 (0.059) Household head unemployed & female - 0.189*** (0.065) - 0.183*** (0.068) - 0.229** (0.093) - 0.231** (0.097) - 0.143* (0.084) - 0.136 (0.091) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denote s significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the follow ing mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour forc e, single female head (same category as household head female), single male head, living with spouse but spousal labour force st atus missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and month of interview. 84 Table A.12 Child's age below 7 years years of education - 0.030** (0.015) - 0.034** (0.014) - 0.023 (0.018) - 0.027 (0.018) - 0.011 (0.025) - 0.015 (0.024) years of education squared 0.001* (0.001) 0.001** (0.001) 0.001 (0.001) 0.001 (0.001) 0.000 (0.001) 0.001 (0.001) Log of real total household income (in Rubles) - 0.028** (0.013) - 0.024 (0.015) - 0.036** (0.016) Region (time invariant) yes yes yes yes yes yes no no no Number of children 947 817 813 603 531 531 603 531 531 Number of clusters (households) 798 691 688 532 469 469 532 469 469 Time periods (unbalanced) 5 5 5 5 5 5 5 5 5 85 Table A.13 : Estimation results (low height for age, total sample) Controls Dependent variable: child has low height for age 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1994 - 2003 Estimation method POLS 1 POLS 2 FE Household head unemployed 0.026* (0.016) 0.048** (0.020) 0.037* (0.020) 0.050** (0.021) 0.081*** (0.027) 0.072*** (0.027) 0.018 (0.021) 0.045* (0.027) 0.039 (0.027) Household head female 0.014 (0.017) 0.008 (0.017) 0.020 (0.023) 0.027 (0.013) 0.022 (0.027) 0.021 (0.026) Household head unemployed & female - 0.098*** (0.032) - 0.094*** (0.033) - 0.158*** (0.041) - 0.159*** (0.041) - 0.067 (0.045) - 0.064 (0.048) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denote s significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the fo llowing mutually exclusive categories: spouse une mployed, spouse employed (omitted in the regressions), spouse not in the labour force, single female head (same category as household head female), single male head, living with spouse but spousal labour force st atus missing; number of children in the hous ehold aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and month of interview. 86 Table A.13 Child's age below 7 years years of education - 0.009 (0.009) - 0.012 (0.009) - 0.001 (0.017) - 0.013 (0.012) - 0.020 (0.013) - 0.024* (0.014) years of education squared 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.000 (0.000) 0.001 (0.001) 0.001* (0.001) Log of real total household income (in Rubles) - 0.018*** (0.001) - 0.018** (0.008) - 0.020** (0.008) Region (time invariant) yes yes yes yes yes yes no no no Number of children 2,593 2,429 2,406 1,406 1,324 1,312 1,406 1,324 1,312 Number of clusters (households) 2,006 1,865 1,847 1,108 1,032 1,025 1,108 1,032 1,025 Time periods (unbalanced) 10 10 10 10 10 10 10 10 10 87 Table A.14 : Estimation results (anxiety or depression, 2003 - 2005 sample) Controls Dependent variable: child suffers from anxiety or depression 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 2003 - 2005 Estimation method POLS 1 POLS 2 FE Household head unemployed - 0.046* (0.025) - 0.036 (0.032) - 0.034 (0.033) - 0.027 (0.031) - 0.031 (0.033) - 0.031 (0.034) - 0.047 (0.034) 0.082** (0.042) 0.101** (0.046) Household head female 0.065* (0.037) 0.071* (0.039) 0.007 (0.042) 0.011 (0.044) 0.112 (0.168) 0.132 (0.167) Household head unemployed & female - 0.264*** (0.096) - 0.262*** (0.096) - 0.169 (0.114) - 0.169 (0.115) - 0.557*** (0.192) - 0.555*** (0.187) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denote s significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour forc e, single female head (same category as household head female), single male head, living with spouse but spousal labour force status missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the hou sehold has access to paediatrician; year and month of interview. 88 Table A.14 Child's age below 7 years - 0.088*** (0.026) - 0.087*** (0.026) - 0.071** (0.030) - 0.070** (0.030) - 0.058 (0.074) - 0.061 (0.076) years of education 0.011 (0.018) 0.010 (0.018) - 0.006 (0.021) - 0.008 (0.021) - 0.044 (0.040) - 0.044 (0.042) years of education squared - 0.000 (0.001) - 0.000 (0.001) 0.000 (0.001) 0.000 (0.001) 0.002 (0.001) 0.002 (0.002) Log of real total household income (in Rubles) 0.007 (0.017) 0.003 (0.019) 0.038 (0.025) Region (time invariant) yes yes yes yes yes yes no no no Number of children 1,785 1,313 1,309 1,095 1,009 1,006 1,095 1,009 1,006 Number of clusters (households) 1,357 1,017 1,014 868 796 794 868 796 794 Time periods (unbalanced) 3 3 3 3 3 3 3 3 3 89 Table A.15 - 1998 sample) Controls Dependent variable: child in good health (parental report) 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1994 - 1998 Estimation method POLS 1 POLS 2 FE Household head unemployed 0.011 (0.018) 0.004 (0.022) 0.008 (0.022) - 0.009 (0.023) - 0.015 (0.025) - 0.014 (0.025) - 0.005 (0.023) 0.013 (0.028) 0.011 (0.029) Household head female 0.012 (0.021) 0.020 (0.022) 0.019 (0.028) 0.025 (0.029) - 0.003 (0.045) 0.001 (0.047) Household head unemployed & female 0.040 (0.052) 0.046 (0.053) - 0.043 (0.066) - 0.044 (0.067) - 0.057 (0.072) - 0.052 (0.075) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force sta tus of the spouse/cohabiting partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour forc e, single female head (same category as househ old head female), single male head, living with spouse but spousal labour force status missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatr ician; year and month of interview. 90 Table A.15 Child's age below 7 years 0.072*** (0.022) 0.073*** (0.022) 0.025 (0.039) 0.022 (0.039) 0.024 (0.038) 0.017 (0.038) years of education - 0.003 (0.010) - 0.003 (0.010) 0.010 (0.012) 0.010 (0.013) 0.013 (0.019) 0.010 (0.020) years of education squared 0.000 (0.000) 0.000 (0.000) - 0.000 (0.000) - 0.000 (0.0004) - 0.001 (0.001) - 0.000 (0.001) Log of real total household income (in Rubles) 0.006 (0.006) 0.005 (0.007) - 0.008 (0.008) Region (time invariant) yes yes yes yes yes yes no no no Number of children 4,571 4,415 4,382 2,243 2,188 2,169 2,243 2,188 2,169 Number of clusters (households) 3,042 2,929 2,912 1,531 1,489 1,480 1,531 1,489 1,480 Time periods (unbalanced) 5 5 5 5 5 5 5 5 5 91 Table A.16 - 1998 sample) Controls Dependent variable: child in good health (parental report) 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1999 - 2006 Estimation method POLS 1 POLS 2 FE Household head unemployed - 0.011 (0.017) - 0.039* (0.021) - 0.038* (0.022) - 0.025 (0.020) - 0.049** (0.025) - 0.050** (0.025) - 0.014 (0.019) - 0.038* (0.022) - 0.042* (0.023) Household head female - 0.054** (0.022) - 0.050** (0.023) - 0.040 (0.026) - 0.040 (0.027) - 0.017 (0.035) - 0.018 (0.036) Household head unemployed & female 0.088 (0.061) 0.090 (0.062) 0.052 (0.075) 0.062 (0.077) 0.024 (0.076) 0.031 (0.078) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour forc e, single female head (same category as household head female), single male head, living with spouse but spousal l abour force status missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and month of interview. 92 Table A.16 Child's age below 7 years 0.003 (0.0215) 0.002 (0.022) - 0.009 (0.025) - 0.011 (0.025) - 0.078** (0.031) - 0.079*** (0.031) years of education - 0.002 (0.011) - 0.004 (0.011) 0.006 (0.013) 0.003 (0.013) 0.003 (0.017) - 0.001 (0.017) years of education squared 0.000 (0.000) 0.000 (0.000) - 0.000 (0.000) - 0.000 (0.000) - 0.000 (0.001) 0.000 (0.001) Log of real total household income (in Rubles) 0.007 (0.008) 0.001 (0.011) - 0.001 (0.010) Region (time invariant) yes yes yes yes yes yes no no no Number of children 3,054 2,555 2,547 2,153 1,814 1,810 2,153 1,814 1,810 Number of clusters (households) 2,190 1,817 1,813 1,599 1,339 1,336 1,599 1,339 1,336 Time periods (unbalanced) 8 8 8 8 8 8 8 8 8 93 Table A.17 : Estimation Controls Dependent variable: child in good health (parental report) 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1994 - 2006 Estimation method POLS 1 POLS 2 FE Household head unemployed 0.003 (0.012) - 0.010 (0.015) - 0.009 (0.015) - 0.017 (0.014) - 0.026 (0.017) - 0.031* (0.017) - 0.001 (0.014) - 0.009 (0.017) - 0.013 (0.017) Household head female - 0.018 (0.016) - 0.013 (0.016) - 0.012 (0.025) - 0.010 (0.026) - 0.022 (0.019) - 0.022 (0.019) Household head unemployed & female 0.075* (0.039) 0.083** (0.040) 0.009 (0.049) 0.025 (0.049) 0.020 (0.050) 0.033 (0.051) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the fo llowing mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour force, single female head (same category as household head female), single male head, living with spouse but spousal labour force st atus missing; number of children in the household ag ed below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and month of interview. 94 Table A.17 Child's age below 7 years 0.044*** (0.015) 0.044*** (0.015) 0.015 (0.022) 0 .011 (0.022) - 0.040 (0.023) - 0.041* (0.024) years of education - 0.005 (0.007) - 0.006 (0.008) 0.000 (0.010) - 0.001 (0.010) 0.001 (0.012) - 0.001 (0.012) years of education squared 0.000 (0.000) 0.000 (0.000) - 0.000 (0.000) 0.000 (0.000) - 0.000 (0.000) 0.000 (0.000) Log of real total household income (in Rubles) 0.005 (0.005) - 0.004 (0.006) - 0.006 (0.006) Region (time invariant) yes yes yes yes yes yes no no no Number of children 6,509 5,869 5,832 3,809 3,425 3,409 3,809 3,425 3,409 Number of clusters (households) 4,214 3,740 3,721 2,527 2,232 2,224 2,527 2,232 2,224 Time periods (unbalanced) 13 13 13 13 13 13 13 13 13 95 Table A.18 : Child and household attrition Number of rounds observed Employed household head Unemployed household head Children Households Children Households 1994 - 1998 sample 1 53.95% 36.20% 73.96% 47.61% 2 24.43% 22.79% 19.55% 23.91% 3 13.44% 13.51% 5.21% 10.20% 4 6.58% 9.47% 1.28% 6.48% 5 1.59% 5.76% 0.00% 3.93% Post - 1998 sample 1 37.90% 27.77% 68.10% 45.40% 2 24.65% 20.29% 20.92% 22.55% 3 16.27% 14.17% 7.86% 10.83% 4 9.27% 10.07% 3.12% 6.38% 5 6.28% 7.58% 0% 4.01% 6 3.76% 5.93% 0% 3.12% 7 1.86% 4.11% 0% 2.08% Notes : 1) Due to high attrition the Moscow and St. Petersburg samples were completely changed in year 1998. 2) Cases where children are observed only once do not necessarily imply attrition; these could also mean the child was born during the last round of data or that s/he was aged 13 in the first year when the household entered the survey (the latter is 1994 for all households, except for the newly - sampled Moscow and St Petersburg households in 1998). 96 Table A.19 : Tests for strict exogeneity of unemployment in the fixed effects estimation Dependent variable Model specification P - value of the test Conclusion (5% significance level) Child had any health problems in the last 30 days, sample 1994 - 1998 3a 0.221 Fail to reject the null for strict exogeneity 3b 0.145 Fail to reject the null for strict exogeneity 3c 0.182 Fail to reject the null for strict exogeneity Child had any health problems in the last 30 days, sample 1999 - 2006 3a 0.369 Fail to reject the null for strict exogeneity 3b 0.682 Fail to reject the null for strict exogeneity 3c 0.748 Fail to reject the null for strict exogeneity Child had any health problems in the last 30 days, sample 1994 - 2006 3a 0.167 Fail to reject the null for strict exogeneity 3b 0.106 Fail to reject the null for strict exogeneity 3c 0.133 Fail to reject the null for strict exogeneity Number of chronic conditions, sample 1999 - 2006 3a 0.669 Fail to reject the null for strict exogeneity 3b 0.324 Fail to reject the null for strict exogeneity 3c 0.299 Fail to reject the null for strict exogeneity Child has depression/anxiety, sample 2003 - 2005 3a 0.109 Fail to reject the null for strict exogeneity 3b 0.032 Reject the null for strict exogeneity 3c 0.036 Reject the null for strict exogeneity Child in good health, sample 1994 - 1998 3a 0.427 Fail to reject the null for strict exogeneity 3b 0.392 Fail to reject the null for strict exogeneity 3c 0.351 Fail to reject the null for strict exogeneity 97 Table A. 1 9 Child in good health, sample 1999 - 2006 3a 0.681 Fail to reject the null for strict exogeneity 3b 0.280 Fail to reject the null for strict exogeneity 3c 0.296 Fail to reject the null for strict exogeneity Child in good health, sample 1994 - 2006 3a 0.625 Fail to reject the null for strict exogeneity 3b 0.404 Fail to reject the null for strict exogeneity 3c 0.437 Fail to reject the null for strict exogeneity Child has low height for age, sample 1994 - 1998 3a 0.980 Fail to reject the null for strict exogeneity 3b 0.962 Fail to reject the null for strict exogeneity 3c 0.920 Fail to reject the null for strict exogeneity Child has low height for age, sample 1999 - 2003 3a 0.706 Fail to reject the null for strict exogeneity 3b 0.308 Fail to reject the null for strict exogeneity 3c 0.306 Fail to reject the null for strict exogeneity Child has low height for age, sample 1994 - 2003 3a 0.346 Fail to reject the null for strict exogeneity 3b 0.722 Fail to reject the null for strict exogeneity 3c 0.843 Fail to reject the null for strict exogeneity 98 APPENDIX B SUPPLEMENTARY TABLES AND FIGURES 99 Table B. 1: Sample statistics of the households (pre - 1998 vs. post - 1998) Household characteristics 1994 - 1998 Post - 1998 Employed Unemployed Diff erence Employed Unemployed Diff erence Age of household head 37.015 36.579 0.436 37.85 0 38.224 - 0.374 Years of education of household head 16.248 15.054 1.193*** 16.4 60 15.080 1.379 *** Number of children aged 0 - 6 in the household 0.650 0 .850 - 0. 200 *** 0.663 0 .847 - 0.183 *** Number of children aged 7 - 18 in the household 1.2 10 1.34 4 - 0.134 *** 1.050 1.285 - 0.235 *** Number of post - working age females in the household 0.175 0.286 - 0.112 *** 0.218 0 .387 - 0.169 *** Notes: 1) Means corrected for stratification; linearised standard errors. Number of strata (regions): 144. 2) Test for equality of means reported for the samples of employed and unemployed household heads. *** denotes significance a t the 1% level; ** denotes significance at the 5% level, and * denotes significance at the 10% level. not available for year 2006. Chronic conditions are only available after 1998. 4) Total number of observations in the 1994 - 1998 sample: (childre n): 4,571; number of clusters (households): 3,042; 1999 - 2006 sample: (children): 3,054; number of clusters (households): 2,190. The total number of observations is below the sum of employed and unemployed as some households/children are observed in both s tates. 100 Table B. Number of post - working age males in the household 0.058 0.101 0.042 *** 0.07 1 0.146 - 0.075 *** Fraction of female household heads 0 .116 0.129 - 0.0134 0.138 0.119 0.019 Total household monthly income (real, in rubbles) 7815.49 5933.54 1881.95*** 13553.02 8766.25 4786.76*** Household below the all Russia poverty line 0.575 0.795 - 0.219 *** 0.22 1 0 .512 - 0 .291 *** Household below the regional poverty line 0.36 2 0.618 - 0.25 7 *** 0.13 3 0 .370 - 0 .238 *** Frequency of monthly alcohol use (household head) 4.397 4.386 0.01 2 4.164 4.364 - 0.200 *** Drinks without eating (household head) 0.142 0.170 - 0.02 8 0 .244 0.239 0.00 5 Household head a smoker 0.61 6 0.693 - 0.077 *** 0 .638 0.69 1 - 0.052*** Number of chronic conditions (household head) 0.536 0.555 - 0.019 0 .523 0 .400 0.123 *** Non - household member cared for the child in the last 7 days 0.28 2 0.236 0.04 6 *** 0 .308 0.18 2 0.126 *** Non - household member cared for the child in the last 7 days (kids below seven) 0.186 0.14 7 0.0 40 *** 0 .211 0 .124 0.087 *** Community level Fraction with access to paediatrician 0 .871 0 .782 0 .089 *** 0.84 5 0 .629 0 .216 *** Distance to paediatrician (in kilometres) 22.213 22.929 - 0.71 6 23.006 27.478 - 4.472 *** Hospital in the community 0 .835 0 .721 0 .114 *** 0 .821 0 .602 0 .220 *** Distance to hospital (in kilometres) 17.507 22.554 - 5.047*** 19.790 25.651 - 5.86 1 *** Distance to hospital (in hours) 0 .172 0 .326 - 0 .154 *** 0 .178 0 .294 - 0.115 *** Number of observations (children) 4,306 696 2,905 479 Number of clusters (households) 2,889 448 2,107 316 101 Table B. 2: Sample statistics of the children (pre - 1998 vs. post - 1998) Characteristics 1994 - 1998 Post - 1998 Employed Unemployed Diff erence Employed Unemployed Diff erence Demographic Age 7.904 7.492 0.41 2 *** 7.567 7.7 50 - 0 .183 Fraction of boys 0.51 3 0 .514 - 0.00 2 0 .504 0.504 - 0 .000 Child healthcare access and utilisation Has a regular physician NA NA NA 0 .591 0 .457 0 .134 *** Had a medical check - up in the last 3 m onths 0.243 0 .226 0.01 8 0 .582 0.52 9 0 .053 *** Notes: 1) Means corrected for stratification; linearised standard errors. Number of strata (regions): 144. 2) Test for equality of means reported for the samples of employed and unemployed household heads. *** denotes significance a t the 1% level; ** denotes significance at the 5% level, and * denotes significance at the 10% level. not available for year 2006. Chronic conditions are only available after 1998. 4) Total number of observations in the 1994 - 1998 sample: (childre n): 4,571; number of clusters (households): 3,042; 1999 - 2006 sample: (children): 3,054; number of clusters (households): 2,190. The total number of observations is below the sum of employed and unemployed as some households/children are observed in both s tates. 1 02 Table B. 2 Had a medical check - up in the last 12 months NA NA NA 0.64 7 0.488 0.15 7 *** Ever vaccinated (all children) 0.973 0.97 8 - 0.00 5 0.98 7 0 .988 - 0.001 Ever vaccinated (children aged 1 year or below) 0.918 0.87 2 0.046 0.959 0.946 0.01 3 Health conditions Any health problems in the last 30 days 0 .387 0.304 0.08 3 *** 0 .400 0 .319 0.08 1 *** Minor health problems in the last 30 days NA NA NA 0.208 0.12 6 0.08 3 *** Hospitalised in the last 3 months 0.03 6 0.044 - 0.00 9 0 .048 0.034 0 .014 ** Feels anxiety or depression NA NA NA 0 .263 0 .128 0.13 5 *** Health evaluation Good 0.57 8 0 .664 - 0 .087 *** 0 .636 0 .729 - 0 .094 *** Average 0.3 90 0. 297 - 0.09 3 *** 0.3 45 0.258 0.0871*** Bad 0 .033 0 .039 - 0 .005 0 .020 0 .013 0 .007 Chronic heart condition 0 .023 0 .019 0 .005 0 .026 0 .020 0 .006 Chronic lung condition 0 .023 0 .008 0 .016 * 0 .020 0 .015 0 .005 Chronic liver condition 0 .020 0 .015 0.005 0 .010 0 .005 0 .005 * Chronic kidney condition 0 .019 0 .040 - 0 .021 0.02 1 0 .015 0 .006 Chronic gastrointestinal condition 0.045 0.034 0.011 0.048 0 .027 0.021 *** Chronic spinal condition 0.027 0 .033 - 0 .006 0 .031 0 .017 0.014 *** Other chronic condition 0 .084 0 .099 - 0 .015 0 .092 0 .067 0.025 *** 103 Table B. Diabetes 0 .002 0.003 - 0 .001 0 .003 0 .002 0.000 Any chronic condition 0 .186 0 .160 0 .026 0 .193 0 .127 0 .065 *** Anthropometric indicators (children aged 7 years or below) Height for age z - score - 0 .186 - 0 .305 0.1 20 0.03 5 - 0 .390 0 .425 *** Weight for height z - score 0 .213 0 .292 - 0 .080 0 .326 0 .458 - 0 .132 Fraction with normal height for age and normal weight for height 0 .842 0 .821 0 .022 0 .886 0. 763 0 .123 *** Fraction with normal height for age and low weight for height 0 .069 0.0 57 0 .012 0 .040 0 .058 - 0.018 Fraction with low height for age and normal weight for height 0 .086 0 .122 - 0 .036 *** 0 .071 0 .165 - 0 .094 *** Fraction with low height for age and low weight for height 0 .003 0.000 0 .003 *** 0 .002 0 .014 - 0 .012 Number of observations (children) 4,306 696 2,905 479 Number of clusters (households) 2,889 448 2,107 316 104 Table B . 3.1 : Other robustness checks (number of chronic conditions, post - 1998 sample) Controls Dependent variable: number of chronic conditions 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 1999 - 2006 Estimation method POLS 1 POLS 2 FE Household head unemployed 0.017 (0.018) 0.052* (0.021) 0.055** (0.022) 0.035 (0.022) 0.062** (0.025) 0.066** (0.025) 0.026 (0.019) 0.052** (0.022) 0.057** (0.022) Household head female 0.075** (0.030) 0.050 (0.030) 0.102*** (0.039) 0.072* (0.0378) 0.041 (0.037) 0.022 (0.037) Household head unemployed & female - 0.057 (0.078) - 0.051 (0.079) 0.017 (0.113) 0.020 (0.112) - 0.018 (0.092) - 0.013 (0.093) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiti ng partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour forc e, single female head (same category as household head female), single m ale head, living with spouse but spousal labour force status missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and month of i nterview. 105 Table B . 3.1 years - 0.095*** (0.026) - 0.093*** (0.025) - 0.084*** (0.028) - 0.084*** (0.027) - 0.029 (0.031) - 0.031 (0.031) years of education - 0.029** (0.012) - 0.026** (0.012) - 0.044*** (0.015) - 0.035*** (0.016) - 0.035** (0.015) - 0.031** (0.016) years of education squared 0.001** (0.000) 0.001* (0.000) 0.001*** (0.001) 0.001*** (0.001) 0.001* (0.001) 0.001* (0.001) Log of real total household income (in Rubles) 0.007 (0.0091) 0.005 (0.009) 0.009 (0.009) 0.007 (0.009) - 0.000 (0.011) - 0.002 (0.011) Number of chronic conditions 0.085*** (0.012) 0.103*** (0.014) 0.064*** (0.012) Region (time invariant) yes yes yes yes yes yes no no no Number of children 3,039 2,534 2,522 2,147 1,804 1,797 2,147 1,804 1,797 Number of clusters (households) 2,184 1,805 1,797 1,596 1,331 1,326 1,596 1,331 1,326 Time periods (unbalanced) 8 8 8 8 8 8 8 8 8 106 Table B.3.2 : Other robustness checks (anxiety and depression, 2003 - 2005 sample) Controls Dependent variable: child suffers from anxiety or depression 1a 1b 1c 2a 2b 2c 3a 3b 3c Sample 2003 - 2005 Estimation method POLS 1 POLS 2 FE Household head unemployed - 0.046* (0.025) - 0.034 (0.033) - 0.043 (0.032) - 0.027 (0.031) - 0.031 (0.034) - 0.054 (0.035) - 0.047 (0.034) 0.101** (0.046) 0.098** (0.045) Household head female 0.071* (0.039) 0.062 (0.039) 0.011 (0.044) - 0.002 (0.043) 0.132 (0.167) 0.093 (0.167) Household head unemployed & female - 0.262*** (0.096) - 0.228** (0.096) - 0.170 (0.115) - 0.209* (0.108) - 0.555*** (0.187) - 0.543*** (0.186) Notes: 1) Standard errors clustered at household level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 2) All specifications control for: labour force status of the spouse/cohabiting partner of the household head, defined as the following mutually exclusive categories: spouse unemployed, spouse employed (omitted in the regressions), spouse not in the labour forc e, single female head (same category as household head female), single male head, living with spouse but spousal l abour force status missing; number of children in the household aged below 7; number of children in the household aged 7 to 18; an indicator of whether the household has access to paediatrician; year and month of interview. 107 Table B . 3.2 below 7 years - 0.087*** (0.026) - 0.079*** (0.026) - 0.070** (0.030) - 0.069** (0.029) - 0.061 (0.075) - 0.051 (0.076) years of education 0.010 (0.018) 0.013 (0.018) - 0.008 (0.021) 0.001 (0.020) - 0.044 (0.042) - 0.038 (0.041) years of education squared - 0.000 (0.001) - 0.000 (0.001) 0.000 (0.001) - 0.000 (0.001) 0.002 (0.002) 0.002 (0.001) Log of real total household income (in Rubles) 0.007 (0.017) 0.009 (0.016) 0.003 (0.019) 0.004 (0.018) 0.038 (0.025) 0.041 (0.024) Child has a chronic condition 0.134*** (0.029) 0.135*** (0.036) 0.066 (0.062) Child stayed in hospital in the last 3 months 0.110** (0.046) 0.110 (0.072) 0.158*** (0.060) Region (time invariant) yes yes yes yes yes yes no no no Number of children 1,785 1,309 1,309 1,095 1,006 1,006 1,095 1,006 1,006 Number of clusters (households) 1,357 1,014 1,014 868 794 794 868 794 794 Time periods (unbalanced) 3 3 3 3 3 3 3 3 3 108 BIBLIOGRAPHY 109 BIBLIOGRAPHY Ahituv, Avner, and Robert I. Lerman. "Job Turnover, Wage Rates, and Marital Stability: How Are They Related?." (2005). Baker, Michael, and Kevin Milligan . "Maternal employment, breastfeeding, and health: Evidence from maternity leave mandates." Journal of health economics 27, no. 4 (2008): 871 - 887. Baten, Jörg, and Andreas Böhm. "Children's Height and Parental Unemployment: A Large Scale Anthropometric Stud y on Eastern Germany, 1994 2006." German Economic Review 11, no. 1 (2010): 1 - 24. Becker, Gary S. "A Theory of the Allocation of Time." The economic journal (1965): 493 - 517. Bennett, David S. "Depression among children with chronic medical problems: a meta - analysis." Journal of Pediatric Psychology 19, no. 2 (1994): 149 - 169. Charles, Kerwin Kofi, and Melvin Stephens. "Disability, job displacement and divorce." Journal of Labor Economics 22, no. 2 (2004): 489 - 522. Dehejia, Rajeev Harsha, and Adriana Lleras - Muney. Booms, busts, and babies' health . National Bureau of Economics Research, 2004. David Brown, J., and John S. Earle. "The reallocation of workers and jobs in Russian industry." Economics of Transition 11, no. 2 (2003): 221 - 252. Edwards, Linda N., and Michael Grossman. "Income and race differences in children's health in the mid - 1960s." Medical care (1982): 915 - 930. Grossman, Michael. "The demand for health: a theoretical and empi rical investigation." NBER Books (1972). Kertesi, Gabor, and Gabor Kezdi. "Children of the post - communist transition: Age at the time of the parents' job loss and dropping out of secondary school." The BE Journal of Economic Analysis & Policy 7, no. 2 (2007). Lindo, Jason M. "Are children really inferior goods? Evidence from displacement - driven income shocks." Journal of Human Resources 45, no. 2 (2010): 301 - 327. Lindo, Jason M. "Parental job loss and infant health." Journal of Health Economics 30, no. 5 (2011): 869 - 879. Lindo, Jason M., Jessamyn Schaller, and Benjamin Hansen. Economic conditions and child abuse . No. w18994. National Bureau of Economic Research, 2013. 110 Liu, Hong, and Zhong Zhao. "Parental job loss and children's health: Ten years after the massive layoff of the SOEs' workers in China." China Economic Review 31 (2014): 303 - 319. Mare, Robert D. "Socioeconomic effects on child mortality in the United States." American Journal of Pub lic Health 72, no. 6 (1982): 539 - 547. Mauldon, Jane. Demography 27, no. 3 (1990): 431 - 446. McKee - Ryan, Frances, Zhaoli Song, Connie R. Wanberg, and Angelo J. Kinicki. "Psychological and physical well - being during unemployment: a meta - analytic study." Journal of applied psychology 90, no. 1 (2005): 53. Montgomery, Scott M., Derek G. Cook, Mel J. Bartley, and Michael EJ Wadsworth . "Unemployment, cigarette smoking, alcohol consumption and body weight in young British men." The European Journal of Public Health 8, no. 1 (1998): 21 - 27. Rege, Mari, Kjetil Telle, and Mark Votruba. "Parental job loss and children's school performance." T he Review of Economic Studies (2011): rdr002. Ruhm, Christopher J. "Maternal employment and adolescent development." Labour Economics 15, no. 5 (2008): 958 - 983. Oreopoulos, Philip, Marianne Page, and Ann Huff Stevens. The intergenerational effect of worker displacement . No. w11587. National Bureau of Economic Research, 2005. Pinquart, Martin, and Yuhui Shen. "Depressive symptoms in children and adolescents with chronic physical illness: an updated meta - analysis." Journal of pediatric psychology 36, no. 4 (2 011): 375 - 384. Stevens, Ann Huff . "Persistent effects of job displacement: The importance of multiple job losses." Journal of Labor Economics (1997): 165 - 188. Torpy, Janet M., Annie Campbell, and Richard M. Glass. "Chronic diseases of children." JAMA 303, no. 7 (2010): 682 - 682. Wooldridge, Jeffrey M. Econometric analysis of cross section and panel data . MIT press, 2010. Russia Longitudinal Monitoring Survey , RLMS - HSE, conducted by Higher School of ion Center, University of North Carolina at Chapel Hill and the Institute of Sociology RAS. (RLMS - HSE: http://www.hse.ru/org/hse/rlms ) 111 CHAPTER 2 THE EFFECT OF RETIREMENT ON MENTAL HEALTH AND SOCIAL INCLUSION OF THE ELDERLY 112 2.1 INTRODUCTION Faced with the challenge of population ageing and the need to ensure the sustainability of the public health and pension systems, most countries in Europe have taken steps towards increasing reti rement eligibility ages. This, in turn, makes understanding the consequences of an - being of considerable importance. Until the last decade, the mainstream literature in the field focused on studyin g the retirement of male longevity outpaces that for men, have given rise to a number of studies examining the effect of Fawaz (2009)). Revealing the mental health effects of retirement has important implications for the well - being o f the elderly and may have significance for policy - making. To elaborate more on this, evidence of high psychic costs of labour force exit would imply that increasing the retirement ages would work towards preserving the emotional well - being of the workers. In contrast, indications of a beneficial impact of retirement might highlight a potential detrimental aspect of the present policies of encouraging continued employment of the older adults. This paper utilizes the empirical methodology developed in a rec ent study by Coe and Zamarro (2011) to investigate the effect of retirement on the mental health of the elderly, and extends their analysis in several ways. First, in contrast to Coe and Zamarro (2011) who study exclusively the labour force exit of men and how it interacts with their physical and mental health, this paper examines the heterogeneity of the impact of retirement for male and female workers while restricting its attention to psychological well - being as the outcome of interest. 113 Secondly, while C oe and Zamarro (2011) are able to look at 11 developed economies from the first wave of the Survey of Health, Ageing and Retirement in Europe (SHARE), this study makes use of an extended version of SHARE including three waves of data on 17 countries, among which 5 post - transition economies. 31 Finally, since the last wave of SHARE enquired about the social and family networks, the analysis presented here is able to shed some light on a secondary question of interest: does retirement cause social isolation of the elderly? The key findings of this paper can be summarised as follows. In line with the conclusions of Coe and Zamarro (2011), the analysis in this study indicates that retirement has no significant - bei ng. At the same time, however, the paper provides strong her depr ession score measured as a composite demotivation index and as the Euro - D depression scale. In addition, the paper finds some evidence of a heterogeneous effect of exiting work on the social contacts of the elderly while retirement decreases the size of the social network for men, it has no effect for women; moreover, retirement appears to enhance contact with parents for female workers, but not for males. These findings on the effect of retirement on the psychological well - being and social networks of th e elderly have important policy implications. 31 This paper uses data from SHARE wave 4 release 1.1.1, as of March 28th 2013 or SHARE wave 1 and 2 release 2.5.0, as of May 24th 2011 or SHARELIFE release 1, as of November 24th 2010. The SHARE data collection has been primarily funded by the European Commission through the 5th Framework Programme (project QLK6 - C T - 2001 - 00360 in the thematic programme Quality of Life), through the 6th Framework Programme (projects SHARE - I3, RII - CT - 2006 - 062193, COMPARE, CIT5 - CT - 2005 - 028857, and SHARELIFE, CIT4 - CT - 2006 - 028812) and through the 7th Framework Programme (SHARE - PREP, N° 2 11909, SHARE - LEAP, N° 227822 and SHARE M4, N° 261982). Additional funding from the U.S. National Institute on Aging (U01 AG09740 - 13S2, P01 AG005842, P01 AG08291, P30 AG12815, R21 AG025169, Y1 - AG - 4553 - 01, IAG BSR06 - 11 and OGHA 04 - 064) and the German Ministr y of Education and Research as well as from various national sources is gratefully acknowledged (see www.share - project.org for a full list of funding institutions). 114 The remainder of this paper is organised as follows. Section 2.2 discusses the theoretical framework behind the main research question and reviews the relevant literature. Section 2.3 describes the data and variable definitions employed in the study, followed by detailed data discussed the identification strategy. Finally, Section 2.5 presents the estimation result s, followed by concluding remarks. 115 2.2 LITERATURE REVIEW 2.2.1 Theoretical background hand, retirement is an event involving a major lifestyle change, and the mainstream psychology literature views it as potentially stressful for the retirees (see, e.g., O. Salami (2010)). Since research suggests the existence of a causal relationship between stress and depressive episodes (Hammen (2005)), this i - being. In addition, a strand of the sociology literature the so - (1934)) maintains the idea that work provides a sense of identity, worth and fulfilmen t for the individual; hence, retirement may lead to loss of a social role, and emotional distress. Further, exiting employment often results in a drop in the income available to an individual or a family, and several studies have shown that insufficient fi nancial resources are related to lower life satisfaction and subjective well - being (Diener et al. (1992)). Finally, some authors argue that retirement may cause a decrease in social contact and disruption of social networks, thus leading to perceived lonel iness and isolation. One example in this respect is a study by Sugisawa et al. (1997), who studied retirement of male workers in Japan and found that early retirement tends to decrease the frequency of social interactions. Recently Börsch - Supan and Schuth (2014) examined data from SHARE and concluded that retirement in general, and early retirement in - family members and immediate colleagues. At the same time, however, others believe that withdrawal from work is a beneficial life change. Retirement dramatically increases the leisure time available to the retiree, which may offset the loss of income to cause a net favourable effect on psychological well - being. In 116 addition, a jo b may be stressful, dissatisfying and strenuous to the individual; hence, retirement would work towards preserving emotional health. Further, a competing theory to the social role theory the continuity theory (e.g., Atchley (1999)) hypothesizes that th e elderly will typically maintain their earlier lifestyle activities, relationships, and identity, even after exiting their jobs; hence, they need not experience any loss of self worth after retirement. Lastly, retirees often get engaged in volunteer ing and charity work, which has been linked to various psychological outcomes. In particular, a number of studies report that volunteering increases life satisfaction (see e.g. Meier and Stutzer (2004)), reduces depression rates (Musick and Wilson, (2003); Lum and Lightfood (2005)), and has a positive impact on subjective well - being (Morrow - Howell et al. (2003)). 2.2.2 Empirical evidence The empirical evidence on the effect of retirement on the occurrence of depressive symptoms is largely mixed: while several studies have found support for a beneficial effect of retirement, a number of other publications reported no significant impact of w orkforce exit, or a detrimental effect. One seminal paper by Charles (2004) utilised data from the Health and Retirement Study (HRS) and the National Longitudinal Survey of Mature Men (NLS - MM) to examine the effect of d reported that permanent exit from employment improves psychological well being. Similarly, using data from the Wisconsin Longitudinal Study, Coursolle et al. (1994) provided support for the idea that retirement is associated with fewer depressive symptom s. More recently, Bound and Waidmann (2007) examined data on morbidity 117 from the English Longitudinal Study of Ageing and concluded that retirement has a positive, albeit small, effect on mental health for men. Yet, the mainstream relevant literature repor emotional well being. Early work (see, e.g., Portnoi (1983)) used cross - sectional data and concluded that retirement is strongly associated with depression; however, those results typically do not have a causal i nterpretation as they did not address the potential endogeneity of workforce exit. More recently, Dave et al. (2008) analysed data from the HRS and documented that full retirement caused a 6 to 9% decline in mental health. Another contemporary study by Bo nsang - being measures from the German Socio - Economic Panel and indicated no significant effect of voluntary retirement, but an adverse effect when it is involuntary (i.e. resulting from employment constraints). F inally, a number of authors have reported that retirement plays no significant role in NLS - MM to study the effect of retirement on life satisfaction and found no impact. Clar k and Fawaz (2009) used two European panels SHARE and the British Household Panel Study and showed that psychological well - being remains largely unchanged following labour force exit. Lastly, Coe and Zamarro (2011) utilised cross - country data on 11 Eur opean states in SHARE and found that, once endogeneity of retirement is accounted for, it appears to have no effect on occurrence of - All this research typically focused on studying the mental health labour force exit . At the same time, however, the rising labour force participation rate of females in the developed economies in general, and in the EU in particular, has tremendously increased the scope of this research 118 force participation rate in the EU reached its highest value over the past two decades, 64.70% (compare e.g. to 56.41% as of 1990). 32 Moreover, the ratio of female - to - male labour participation in the EU has been constantly increasing as well, reaching a re cord high of 77.68% as of 2011. to concerns of sample selection and cohort effects. An important point should be made here regarding the first concern: given the re search question addressed in this paper, sample selection is not an issue as the SHARE sample is representative of the population of interest women who are in the labour force are studied as they subsequently transit into retirement, and this is the exac t population one would like to study (put differently, selection is exogenous, not endogenous). The second issue, however, is potentially problematic: cohort effects are present in the EU and are particularly relevant for women, as females born in the 60s and 70s are more likely to participate in the labour force over their life - cycle. Additionally, these effects vary by country (see e.g. Balleer et al. (2009)). However, given the identification strategy employed in this study, cohort effects are a problem correlated to the statutory and early retirement ages a much stronger statement. Another reason to study t he retirement of female workers is the potential presence of heterogeneous effects, and there are several reasons why labour force exit may, indeed, have a differential impact across gender. First and foremost, a consistent long - standing observation in the social epidemiology literature is the gender gap in depression, namely that depression is more prevalent amongst females than amongst males. For instance, Van de Velde et al. (2010) used 32 Calculation based on the female population aged 15 to 64. Source: International Labour Organization, Key Indicators of the Labour Market database. 119 data on 23 countries from the European Social Survey, and found high er levels of depression for women in all countries, although the gender gap exhibited a considerable cross - national variation. Moreover, some authors hypothesise the gender gap in psychological well - being is due to fact that women combine paid employment w ith engaging in a disproportionately larger share of the housework (see e.g. Mirowski (1996) and Lennon and Rosenfield (1992)). This direction of thought implies that exiting employment into retirement may provide an additional channel for a beneficial eff ect of retirement on mental health for women, but not for men. In addition to this, a number of studies suggest that women and men who retire experience a loss in social role to a different extent. To elaborate more on this, women typically have more frag mented work histories and lower attachment to the labour market and to a particular occupation than men, while at the same time strong workplace attachment has been associated with more a painful transition into retirement (see e.g. Tibbitts (1954)). Simi larly, a contemporary study conducted in the United Kingdom by Barnes and Parry (2004) found that status upon retirement, compared to women. Lastly, a number of European states still maintain different pension eligibility age for men and women, resulting in lower replacement rates for women. 33 Because economic factors - being, this may result in a differential effe ct of retirement for both genders. Taking all this into consideration, the analysis in this paper studies both men and women aiming at shedding some light on the potentially heterogeneous effect of retirement by gender. 33 The most notable gender differential in replacement rates is observed in Italy, Poland, and Slovenia (European Commission, 2012). 120 2.3 DATA AND VARIABLES 2.3.1 Data and sample The analysis in this paper utilises data from the Survey of Health, Ageing and Retirement in Europe (SHARE). SHARE is a cross - national European survey, containing micro - level data on persons aged 50 and older at the time of the first interview, and their spouses. The survey is based on probability samples in all participating countries; following the individuals from the baseline wave in 2004, subsequent interviews were conducted, on average, once in two years. 34 Since wave 3 in SHARE was entire ly retrospective, the paper uses data from waves 1, 2 and 4 only. This results in a sample containing data on 17 European countries: Austria, Belgium, Czech Republic, Denmark, Estonia, France, Germany, Greece, Hungary, Italy, Netherlands, Poland, Portugal, Slovenia, Spain, Sweden, Switzerland. 35 A detailed country representation for each wave in SHARE is shown in Table C.1 . Since SHARE collects data on the elderly for a large multinational sample over a relatively long period of time, it is particularly well suited for studying the link between retirement and health outcomes. In addition to the basic demographic and socio - eco nomic variables, SHARE provides detailed information for the purposes of this paper labour supply outcomes and psychological health. An important strength of the dataset is the quality of the mental health information collected: the respondents were aske d series of questions on their 34 For wave 1 interviews were conducted in year 2004 (80.8% of the sample) and 2005 (19.2%), for wave 2 in year 2007 (75.4% of the sample) and 2006 (24.6% of the sample), and for wave 4 predominantly in year 2011 (93.7% of the sample), and a small fraction of the respondents were interviewed i n 2010 (2.8%) and 2012 (3.5%). Due to attrition, samples were drawn in later waves in most first - wave countries, aiming at keeping the national samples representative of the elderly population. 35 Data on Ireland is also available in wave 2; how ever, since it does not contain key variables such as a household identifier, and income imputations, the country is excluded from the analysis. 121 overall emotional condition, as well as whether they experienced certain depression symptoms. Further, SHARE contains comprehensive information on variables considered key determinants of depression, such as physical health, hospital stay, and household income. Finally, wave 4 enquired about the social and family networks, which allows inferring upon the effect of retirement on the social inclusion of the elderly. Since the central research question of this pap er focuses on the effect of being retired on the mental health of the elderly, attention is restricted to individuals who were aged 50 or over at the time of the first interview, and were either employed or retired at that time. 36 Persons who consider them selves unemployed, disabled or a homemaker are excluded from further analysis. In addition, individuals who never worked for pay or have not worked for pay since the age of 50, are considered out of the labour force and dropped from the sample. 2 .3.2 Vari able definitions 2 .3.2 .1 Mental health measures This paper focuses on several measures of mental health. First, the Euro - D depression scale is an instrument developed by a number of European countries for screening the mental health of the elderly, and is available in SHARE. The scale is largely based on the Geriatric Mental State examination (Copeland et al. (1986)) and includes the following self - reported symptoms: indicators of being sad or depressed during the last month, pessimism, suicidal thoughts, f eelings of guilt, trouble sleeping, loss of interest, loss of appetite, irritability, fatigue, poor concentration, enjoyment, and tearfulness. Each item is coded as a binary indicator, and the 36 SHARE was designed for persons aged 50 or over at the time of the first interview, and their spouses; since s ome spouses are aged below 50, those are excluded from the sample. 122 Euro - D index is then composed as the summation of all indicator s (on a 0 to 12 scale, where 0 stands for no depression indication and 12 for severe depression). Further, since a number of European psychometric studies report two types of major components of mental health of the elderly affective suffering and demot ivation symptoms (see, e.g., Prince et al. (2004) and Castro - Costa et al. (2007)) this paper defines two separate indices measuring those components. Following Castro - Costa et al. (2007) the index of demotivation symptoms is composed of dummies for pessi mism, loss of interest, poor concentration and lack of enjoyment (0 to 4 scale), while the measure associated with affective suffering symptoms includes all the remaining items from the Euro - D index (0 to 8 scale). Next, in view of the fact that death ideation is often associated with severe depression and increased suicide risk (see O'Riley et al. (2013)), the analysis examines the effect of retirement on this particular indicator. Lastly, the paper also looks at th e self - report of feeling sad or depressed in the month before the interview. 2 .3.2 . 2 Retirement definition This paper employs the following definition of retirement. An individual is considered retired if: 1) s/he considers him/herself retir ed and reports supplying no work; or 2) s/he considers him/herself retired but may supply some part time work (i.e. works no more than 20 hours a week), and, in addition, 3) is not unemployed, disabled or a homemaker. 37 This definition is preferred as it captures the idea of retirement as a state of mind (i.e., one considers themselves retired although s/he might still be supplying some part - time work), while at the same time 37 Individuals in SHARE, who report themselves a homemaker, are 97% female, and since they do not consider themselves either employed, unemployed or retired, this paper classifies them as not in labour force. Hence, they are excluded from further analysis. 123 reflecting the notion of retirement as a comp lete withdrawal from the labour force or a withdrawal from active work into the state of being retired. alternative state of remaining employed. The latter category i s composed of individuals who report themselves employed or self - employed; in addition, persons who consider themselves retired but continue supplying more than 20 hours of labour per week are also classified as working. These definitions of the retirement and employment states allow capturing the key aspect of the research question addressed in this paper, namely that work may be either draining or rewarding for the individual; thus, withdrawal from active labour versus continuation of active work may be e ither beneficial or harmful for their mental health. The final sample after restrictions consists of 81,823 observations, of which 53.04% are males. The resulting retirement rate is roughly 62% in the total sample, and when looking at males and females se parately (see Table C.2 ). 2.3.2.3 Social networks As a secondary question of interest, the analysis looks at several measures of the social defined as t 38 Since individuals who exit work are likely to have less contact with their former co - workers, we examine the number of persons in the social network with daily contact. Next, the respondents satisfaction with their network is based on a self - rated measure on a 0 to 10 scale (where 0 stands for completely dissatisfied and 10 for completely satisfied). In addition, the paper studies the 38 124 effect of retirement on child - parent bonds by foc using on two binary indicators for presence of is investigated . 2 .3.3 Sample statistics Table C.3 presents the descriptive statistics for the full sample of observations, as well as for the male and female subsamples separately. 39 The mean age of the persons in the final sample is 65.8 years, with women being older by 0.5 years (significant at the 1% level). While the percentage of males and females who hav e reached early retirement age is roughly the same (67%), the fraction of women who have reached statutory retirement age is higher by 4.2 percentage points (pp); significant at the 1% level. Women are also less educated by 0.4 years and considerably more likely to be widowed the difference in means equals 0.15; significant at low levels. Further, the mean number of children of the elderly is 2.1; the gender difference being statistically different from zero but small in magnitude. To complete the demogra phic representation, 9.3% of the females and 8.0% of the males report being born in a country other than their country of residence. Next, examining the labour force outcomes of the elderly in SHARE shows that a somewhat higher fraction of females is reti red, but the 1.2pp difference in means is not significant at the conventional levels. Amongst those who are still working, the highest fraction reports holding a job in the private sector (roughly 50% of the total sample of employed), followed by the publi c sector (30.0%) and self employed individuals (20.4%). Women are more 39 Means and standard errors corrected for inverse probability weighted sampling; t - test for equali ty of means with equal variances reported. 125 likely to work in the public sector and significantly less likely to be self employed, and this pattern holds when looking at the last job history of the retirees, as well. The lower pa nel of Table C.3 mental health outcomes depicts a vivid illustration of the gender gap in depression in Europe. To elaborate more on this, women are more likely to experience both the affective suffering symptoms and the demotivation symptoms of depres sion (the difference in means being significant at the 1% level for all indicators), resulting in a considerable gap in the composite Euro - D index of 0.95. The mental health measures exhibiting highest difference in means (relative to the female sample mea n) are tearfulness (0.66), death ideation (0.47), and trouble sleeping (0.43). It is also worth noting that for both genders feeling sad or depressed in the last month appears to be the most common affective suffering symptom, while lack of hopes for the f uture is the most frequently reported demotivation symptom. The two groups also differ in their physical health. In particular, women report more limitations to activities of daily living (0.23 vs. 018 for men) and mobility difficulties (1.79 vs. 1.09), a s well as a higher number of chronic conditions (1.63 vs. 1.46); all differences are statistically significant at the 5% level. In addition, females in the sample are 4.0 pp more likely to evaluate their health as fair (vs. excellent or very good). Somewha t surprisingly, however, men appear worse when examining the indicator for hospital stays in the last year, although the difference is small in magnitude (0.7pp) and significant only at the 10% level. Turning briefly to the social outcomes of the elderly s uggests that, on average, women have better social and family connectedness they have broader social networks and are more likely to keep a close relationship with children and parents. Yet, both groups seem equally satisfied with their social network the difference in means while statistically significant is low 126 in magnitude. Lastly, it is interesting to note men take higher participation in volunteering and charity work, although the difference in means is small in magnitude. In sum, while the fractio ns of retired women and men in the sample are comparatively close, women appear older, more likely to be widowed and to suffer from ill health and, ultimately, in noticeably poorer psychological condition. This raises an interesting question: could retirem ent have a heterogeneous effect on the mental health of the two groups possibly adversely affecting females and having a beneficial or no effect for males, or it is the unfavourable socio - demographic factors (such as loss of a spouse) which induce depres sive suffering for women? Verifying either of the two possibilities requires that the effect of labour household characteristics and country - level indicators. We tur n to this analysis in the next section. 127 2.4 ECONOMETRIC MODEL - being: Y i c t 0 1 Retired i c t + 2 + 3 + 4 + d t + ( + ) , (1) where Y represents the mental health outcome of individual i in country c at time t, and Retired is a binary indicator for whether the person is retired or still employed. X OWN consist of individual characteristics considered to be important determinants of depression, such as old age, poor education and immigrant status, which have all typically been reported as drivers of mental suffering (Buber and Engelhardt (2006)). 40 In ad dition, since a number of studies found a protective effect of living with a partner and having children against depressive occurrences (Buber and Engelhardt (2006)), X OWN has any kids. Controls for physical health are also incorporated, as declining physical health is often thought of as a key factor for emotional distress (e.g. Beekman et al. (1997)). Lastly, X OWN includes sector of employment at the current/last job as a measur cteristics. F urther, X HHD consists of a household - level control for aggregate annual income (converted to EUR, PPP - adjusted, and where missing, imputed), and incorporates a set of 40 Based on the mainstream literature that supports the idea of a U - shaped relationship between age and depression (see e.g. Stone et al. (2010)), we specify age in quadratics in the model. However, since age is a key determinant of mental health we perform several robustness checks to the age functional form specification in Appendix A1. Education is one of the most wide - ranging variables in Europe. Wave 1 and 2 in SHARE used the 1997 International Standard Class ification of Education ISCED - 97 to group the education variables into standardised levels of attained education. The latter are, however, not available in wave 4. For this reason, the analysis utilises number of years of schooling as a measure of education . Since these are only available in waves 2 and 4, the paper imputes years of education in wave 1 the following way: 1) for observations which appear both in waves 1 and 2, years of education in wave 1 is set to the report from wave 2; 2) for those appeari ng only in wave 1, years of education in wave 1 is set to the sample mean years of education in each ISCED - 97 category (based on wave 2). 128 country dummies accounting for country - level heterogeneity, such as the cross - country differences in depression prevalence (Van de Velde (2010)). Next, d t denotes year effects to control for the overall economic, public health and environmental conditions that play a role in al. (2000)), as well as month - of - interview dummies as certain depressive symptoms exhibit a seasonal pattern. Next, the error component a i represents time - invariant unobserved individual - level factors that could affect mental health outcomes. One such exam ple is genetic predisposition, as recent research reported an association between certain genes and various anxiety and depression disorders (Donner et al. (2008)). Finally, u i c t health, such as stressful life events; e.g., illness or death in the family. mental heal th is the reverse causality between the two while being retired might possibly force (Conti et al. (2008)). Following the identification strategy developed by Co e and Zamarro (2011) this paper uses the exogenous variation in the early and statutory retirement ages as instruments for the state of being retired. Since there are two potential instrumental variables available, two estimation methods could be employed: pooled instrumental variable (IV) estimator using either the statutory retirement age or the early retirement age as a single excluded instrument, and pooled two - stage least squares (2SLS) using both instruments. The later has been shown to be the most ef ficient IV estimator under certain assumptions (Wooldridge (2010)). Formally, the first stage regression in the two - stage least squares (2SLS) estimation has the following form: Retired i c t = 0 + Z i c t 1 + 2 + 3 + 4 + d t + i c t , (2) 129 where Z i c t = (Z 1i c t , Z 2i c t ) is the vector of excluded instruments. In particular, Z 1i c t denotes a binary variable for whether person i in country c has reached the statutory retirement age as of time t , and Z 2i ct whether s/he has reached the early retirement age at that time. It is worth noting that both instruments vary across countries (as the pension eligibility ages vary between states in the across time (as some of the countries in SHARE changed the retirement eligibility ages during the period of the survey). There are several identifying assumptions for consistency of the IV/2SLS estimator. Since the paper involves the use of a binary instr ument and a binary instrumented variable, adopting the notation in the seminal work by Imbens and Angrist (1994) is convenient. Let Y i denote a vector of all actual mental health outcomes of individual i , and D i denote their actual retirement outcome (rega rded as the treatment). Next, define Y i0 and Y i1 as the potential values of the outcome of interest when the binary treatment takes on values 0 and 1, respectively, and D i0 and D i1 as the level of the treatment received if the instrument takes on values 0 and 1. In this way e.g., when the instrument is the statutory retirement age Y ict 0 stands for the potential mental health outcome of person i in country c at time t has s/he not reached full retirement age, while Y ict 1 stands for the potential mental health of the individual has s/he reached that age. Likewise, D ict 0 and D ict 1 denote the potential retirement outcomes conditional on the value of the instrument in that country and time period. Under this framework, the first key identifying assumption refers to the relevance of the instrument(s) and states that conditional on the observable covariates the probability of being retired should be a non - trivial function of the instrument: ( D i Z i - trivial function of k, (A1) where k 130 In other words, reaching early or statutory retirement age should have an effect on the retirement propensity. The second assumption is often referred to as independence of all potential outcomes of the instrument , or formally: { Y itc 0 , Y itc1 , D itc0 , D itc 1 } Z ict . (A2) Statement (A2) incorporates two properties of the instrument: exogeneity and excludability. The first refers to the requirement that the instrument is essentially randomly assigned with respect to the composite error in that time period (put differently, t his requires )=0 t and contemporaneous exogeneity of the instruments )=0 t ). 41 Since the early and full retirement ages are decided at country level, there are no reasons to believe that they are related to the unobserved heterogeneity at individual level or the idiosyncratic error at that time. The second part of assumption (A2) captures the restriction that there is no direct link between the instrument and the outcome of interest. Put differently, the pension eligibility ages - being other than through the state of being retired. S ince the compulsory health insurance scheme in the EU covers major and 41 For the countries in SHARE observed at least once, an alternative estimation strategy is available fixed effects I V (FEIV). In contrast to pooled 2SLS, which assumes ( )=0 t and contemporaneous exogeneity of the instruments ( )=0 t , FEIV allows ( ) 0 but imposes the stronger restriction ( ) = 0, r, t (strict exogeneity, see e.g. Wooldridge (2010)). Since the statutory and early retirement ages are decided at country level, the value of the instruments in each time period only depends on the pensionable ages in a given country and age at that time period; hence, there is no reason to believe that ( )=0 would fail to hold as a i only varies at individual level. It is more worrisome, however, to assume that ( ) = 0, r, t as it would rule out the possibility that the retirement ages were changed at country level as a response to shocks in the past, which may have also affected the the paper. In addition, Appendix 3 reports the main results when model (1) is estimated under less restrictive assumptions than the ones imposed by FEIV, namely, fixed effects estimation (see Appendix A3). 131 minor risks for all employees and retirees, and this coverage does not discontinuously change when reaching a certain age, be that early or full retirement age, 42 one would not expect th e 43 The last assumption requires that the retirement probability is monotonic in the instruments: Either D i1 D i0 i , or D i0 D i 1 i . (A3) In other words, while reaching early or statutory retirement age may have no effect on the instrument should (A3) is likely to hold; in particular, it is credible that D i1 D i0 for all i , as there is no reason to believe any person would be more likely to retire while being below pensionable age but less likely thereafter. Under assumptions (A1) through (A3), the IV estimand captures the local average treatment effect (LATE), i.e., it ide ntifies the average treatment effect of retirement on mental health for the subpopulation of retirees whose retirement was induced by the instrument. Several important notes can be made here. First, the effect of retirement need not be the same when employ ing the early and statutory retirement age as an instrument since the groups affected by 42 Source: Healthcare Systems in the EU: a Comparative Study, European Parliament (2010) 43 equacy Report 2010 - 2050), the statutory and early retirement ages are expected to be linked to the country - average physical health of the elderly. One might worry, then, that this implies a country pensionable ages as SHARE is a nationally representative survey and the national - average physical health depends alth, it is not likely to be the case when studying the effect of retirement on depression of outpatients (mental illness has been shown to lower life expectancy for inpatients due to the detrimental physical health effect of antipsychotic medication; see e.g. Crystal et al. (2009)). 132 each of these instruments are different. Secondly, given that the retirement eligibility ages would most likely affect planned voluntary retirement rather than involu ntary retirement, the implications of the analysis are most relevant for the former. Lastly, it is worth menti oning that a usual criticism of LATE is that it often identifies an effect which is not important from a policy perspective; however, in our paper , LATE is of particular interest as it identifies an effect caused by the exact variables that could be targeted by policy - makers the early and statutory retirement ages. 133 2.5 ESTIMATION RESULTS 2.5.1 First stage 2.5.1.1 Statutory and early retirement ages, and actual retirement ages in SHARE Table C.4 shows the statutory, early and actual mean retirement ages in SHARE for each country in the sample, separately for waves 1 - 2 and wave 4. 44 Several observations are worth noting at this point. First, even though there has been some convergence of the statutory retirement ages towards age 65 and the early retirement ages towards age 60, some cross - country variation in those ages still exists. Secondly, on average, the post transition economies provide access to early and full retirement considerably earlier than the EU - 15 and Switzerland; in addition, the new EU members are more likely to maintain different pensionable ages for women and men. Furthermore, although not a perfect predictor of the actual ages of retirement, statutory retirement age in Europe is Sweden, setting the full retirement age at 67 a s of 2010, and it is also the country with highest actual retirement ages for men and one of the highest for women. Next, an increase of the statutory retirement age appears to result in an increase of the actual age of retirement: e.g., Italy increased t his age for women from 60 to 65 years following wave 2, and considerably higher than the increase for men (0.5 years). Finally, while women tend to retire earlier in all countries, the gender differential in the mean retirement ages is lower for countries 44 The question about year of retirement was asked in waves 2 and 4 only. Year of retirement imputed for the retired individuals in wave 1 based on the report from wave 2. Retirement age derived as the difference between year o f retirement and year of birth. Waves 1 and 2 are grouped together as the main sources of information for the early and statutory retirement ages in years 2004 and 2007 report no changes in those in the period. 134 with equal treatment of women and men; for instance, for wave 4, this differential was 2.7 years in Poland but only 0.1 years in Sweden. To further illustrate the li nk between the legislative provisions for retirement and the actual ages of retirement, it is also useful to examine the entire histogram of the ages of labour force exit. Figure C.1 to C.4 show these histograms for four of the countries in the SHARE sampl e Sweden and Switzerland selected amongst the states with high statutory and early retirement ages, and the Czech Republic and Poland amongst those with low eligibility ages. It is apparent that the largest fractions of women and men in Sweden, which has equal treatment for both genders, retire at the pre - 2010 statutory retirement age, and the two histograms exhibit very similar patterns (Figure C.1) . Males in Switzerland also appear most likely to exit from labour when reaching full retirement age (65 ye ars), while the largest fraction of females stops working at the early retirement age (62 years), followed by relatively equal shares of retirees at age 63 and the full retirement age, 64 (Figure C.2) . Turning to the post - transition countries, Figure C.3 s hows the retirement probabilities in Poland display a clear peak at the (pre - 2009) early retirement age for both genders, followed by a secondary peak at the respective statutory retirement ages. Lastly, while most men in the Czech sample exit the labour f orce at the single early retirement age, 60, the retirement probabilities for females are high, albeit declining, for all ages 55 through 59, likely due to the linkage of retirement eligibility to number of children (Figure C.4) . Overall, these examples st rongly confirm that the early and statutory retirement ages strongly influence the distribution of retirement ages. 135 2.5.1.2 Estimation results Tables C.6 and C.7 report the first stage estimation results separately for the male and female subsamples. Co lumn (1a) reports estimates from Model (2) using the statutory retirement age as a single excluded instrument, column (1b) uses the early retirement age only, and column (1c) uses both instruments. Columns (2a - 2c) repeat the specifications but add a binar y indicator for being in bad health; this measure is potentially endogenous to mental health, but we include it to assess its effect on the estimated treatment effect of interest. As can be seen from the table, the statutory and the early retirement ages are strong predictors of retirement for both genders. For instance, column (1a) of Table C.6 implies that having reached statutory retirement age on average increases the probability that a male has retired by 23.8 pp, ceteris paribus , and the effect is highly significant. The corresponding specification from Table C. 7 states that reaching full retirement age would make a female 28.8 pp more likely to exit the labour force, other factors being equal. The early retirement age is also es timated to induce retirement with a high probability for both men and women (magnitudes of 0.25 and 0.28, respectively), and the effects are statistically different than zero at low levels. Essentially the same conclusions prevail when both instruments are employed, and the models are robust to the exclusion of the bad health indicator. It is also interesting to note that age appears a significant predictor of exiting work, even after accounting for reaching full and early retirement age. Lastly, it is use ful to examine the first stage F - statistic and the F - statistic on the excluded instruments as suggested by several studies (see, e.g., Stock and Yogo (2005)). In particular, a number of authors reported a correspondence between the first stage F - statistic and the bias of the IV estimator relative to the bias of the OLS estimator, and some proposed rules of thumb for 136 evaluating the relevance of the instruments. For instance, Staiger and Stock (1997) suggested an F - statistic on the excluded instruments of at least 10. The lower panel of Tables C6 and C7 reports the non - robust and the cluster - robust F - statistic of the regression they are considerably higher than 10 in all specifications. 45 2.5.2 Second stage 2.5.2 .1 Mental health by age distance to statutory and early retirement Figure s C. 5 to C.8 illustrates the pattern of the mental health indicators 46 for men and women by age distance to statutory retirement age; in addition, a histogram of the distance between the actual and statutory retirement a ges in each age group are presented. The graphs for women (depicted on the left - hand - retirement occurs five years prior to reaching statutory retirement age (nearly 30 percent of the actual reti rement ages are in this group). In addition, a large fraction of women exit from work the year when reaching full retirement age. For men (right - hand side graphs), the majority of retirements occur at statutory age (30% of the actual retirement ages), foll owed by a peak five years prior this age. Panels A and B of Figure C. 5 show the mean death ideation by age category for women and men, respectively. Focusing on the changes that occur around the statutory retirement age 45 The critical values and rules of thumb fo r the F - statistic on the excluded instruments are based on the assumption of i.i.d. errors. Since SHARE collects household - level country data, heteroskedasticity and serial correlation are likely to be present; for this reason, the tables also report the c luster - robust F - statistic on the excluded instruments. The related theoretical results do not extend to proposing critical values for the robust F - statistic but a recent study by Bun and De Haan (2010) used simulations and showed that a decrease in the rob ust F - statistic is enough to offset the increase in the IV bias relative to OLS; in other words, even values lower than 10 would suffice. 46 The graphs do not look at the indicator for feeling sad or depressed during the month preceding the interview as this measure is particularly likely to exhibit seasonal patterns. 137 reveals a large improvement in this indicator for women in the years before reaching statutory declining two years before the cut - off, and its mean remains at a lower level two years after full retireme nt age, before gradually increasing thereafter. For men, suicidal wishing is characterised by a large jump at the cut - off and no drop prior to it; in addition, the decline in this measure following full retirement age is mirrored by an almost equally sized increase the year after. Next, Figure C.6 illustrate s the patterns of the demotivation measure: while this index sees a sizeable drop for females a year before reaching statutory retirement, the index for men shows a sharp increase at the cut - off. Turning to the affective suffering index ( Figure C.7 ) shows a large decline in this measure in the years around the cut - off for women, while the improvement for men is not as striking. Lastly, the patterns in the Euro - D scale (Panels G and H on Figure C.8 ), point towards a substantial decline for women around the statutory retirement age, while suggesting only a minor favourable development for men following the year after reaching statutory retirement age. It is also worth noting that for women very few retiremen ts occur two years past statutory retirement age, after which virtually all mental health indicators increase in a nearly linear fashion. The later suggests that in the absence of retirement, the drop of all depressive symptoms around statutory retirement age may not have occurred, and that, instead, mental health would have deteriorated linearly with age. A similar conclusion can be drawn for men, as well, although the improvement in emotional well - being for males being concentrated only the year after rea ching statutory retirement age. Figure s C. 9 to C.12 illustrates the analogous graphs for both genders by distance to early retirement age (restricting attention to 12 years around that age). The largest proportions of women and men tend to retire when they are first eligible for early retirement (the retir ement age 138 histograms reaching nearly 0.2 for women at the cut - off, and about 0.25 for men). The pattern of the mean death ideation for women (Panel A , Figure C.9 ) depicts an improvement when early retirement age is reached and the year after, and only a mi nor increase during the following six years. There is a parallel drop in this mental health measure for men, as well (Panel B , Figure C.9 ), occurring at early retirement age the age when most male workers retire; however, this is followed by sharp increa se thereafter. Examining the demotivation index (Panels C and D , , Figure C.10 ) also suggests an improvement for both genders at the time when the elderly are most likely to retire, with this improvement being more pronounced for females. The patterns of t he affective suffering index and the Euro - D scale are essentially the identical at the cut - off: a large and sustained decline for women and only a temporary drop for men; the male indices also improve two years after reaching early retirement. Overall, for females all mental health measures exhibit a nearly linear upward trend starting about five years after early retirement eligibility age, after which very few retirements are observed. This implies that, in the absence of retirement, the sizeable improvem - off may possibly not have occurred. A similar conclusion can be drawn for men, as well, although the mental health patterns being not as clear for this subsample. 2.5.2 .2 Estimation results 2.5.2.2.1 Mental health As shown in the previous section, both instruments the statutory and the early retirement age are strong predictors of retirement. In the absence of weak instrument concerns the 2SLS estimator combining both IVs provides efficiency gains; for th is reason, the main part 139 of the subsequent analysis focuses on the estimation results when using both instruments, but we will also examine second stage results based on each instrument separately. Table C.8 to C.15 report these results for men and women, respectively, when the mental health outcome of interest is whether the person had suicidal thoughts. Table C.8 shows the pooled OLS estimates for the male sample when model (1) includes controls for age, time and country dummies only (specification (1a), as well as when employing all covariates (specification (1b). Due to suspected endogeneity of the binary indicat or for being in bad health, column (1c) reports the estimation results when omitting this variable. As can be seen from here, retirement on death ideation, ceteris paribus . In contrast, panels (2) to (4) of Tables C.9 to C. 11 report the parameter estimates when employing an instrumental variable estimation on the same specifications as in Panel (1). The key implication from this set of results is that once endogenei role in the occurrence of suicidal thoughts the coefficient on retirement appears negative in sign but insignificantly different from zero in all but one specificati on. The only exception is the 2SLS estimate from column (4a) when both instruments are employed it is negative 0.018 and marginally significant, but it drops in magnitude and significance once covariates are included in specifications (4b) and (4c). Ta ble s C.12 to C.15 reports the corresponding estimation results for the female sample. As before, the pooled OLS estimates on retirement exhibit an upward bias, although the coefficients are somewhat smaller in magnitude and significance than the ones obtai ned on the male sample (refer to Table C.12) . Columns (2a) to (2c) of Table C.13 report the results when employing the statutory retirement age as an instrument for retirement, and the parameters on 140 retirement have the interpretation of an average treatmen t effect for the female subpopulation of compliers with the full retirement age. These results imply that for women whose labour force exit is induced by the statutory retirement age, retirement reduces the occurrence of suicidal thoughts by nearly 4pp, ce teris paribus , and the effect is statistically significant at the 5% level. Next, specifications (3a) to (3c) of Table C.14 report the estimation results on the female subsample when the early retirement age is employed as the single excluded instrument. I n these specifications, the parameters on retirement are still negative in sign but lower in magnitude and less precisely estimated, implying no important effect of labour force exit on death ideation for the women whose retirement is induced by the early retirement age. Further, Table C. 15 reports the 2SLS results when both instruments are used; the average treatment effect of retirement for both groups of compliers is roughly negative 0.03, implying a beneficial effect of labour force exit on suicidal tho ughts for these women (also note the considerably lower standard errors on the estimates illustrating the efficiency gain of pooled 2SLS compared to pooled IV). Lastly, it is worth noting that compared to the female sample mean of the death ideation indica tor, 0.087, the estimated magnitude of the effect of retirement of 0.03 is very large. Taken as a whole, the estimation results examined so far suggest a large and significant beneficial effect of retirement on death ideation for women but no corresponding effect for men. In addition, although this impact is significant when looking at both groups of compliers as shown by the 2SLS results, it does seem stronger for the compliers with the statutory retirement age. The next sections shall focus on the 2SLS es timation results in order to make use of the efficiency gain when employing both instruments and aiming at reporting an average treatment effect for both groups of compliers. 141 Table s C. 16 and C.17 show the estimation results on the parameter of interest fo r all the remaining mental health measures. When the outcome is the composite demotivation index (the later ranges from 0 to 4, where higher values imply worse psychological well - being), the pooled OLS estimates on retirement for men (reported in Panel A o f Table C.16 ) are positive and highly significant in all specifications. However, once retirement is instrumented by the statutory and early retirement ages, the key parameter of interest appears negative in sign and not statistically different from zero i n all specifications. Turning to panel B of Table C.17 , the pooled OLS estimates overall imply a detrimental effect of retirement on the composite demotivation index for women; yet, once a 2SLS estimator is employed, the impact of retirement for the female s complying with the instruments appears negative in sign and statistically significant at low levels across all specifications. For instance, the estimate from column (2b), obtained when including all covariates, is - 0.187, implies that, other factors equ al, labour force exit has a beneficial effect on the demotivation index for women (interpreted as a local average treatment effect). Moreover, the magnitude of this effect is very large roughly one - third of the female sample mean for this mental health i ndicator. The next section of Table s C. 16 and C.17 reports the second stage results for the effect of retirement on the affective suffering index (scale ranging from 0 to 8). As can be seen from here, both sets of estimation results tell a similar story while the pooled OLS estimator implies that retirement increases the occurrence of affective suffering symptoms both for women and for men, the pooled 2SLS estimator suggests that labour force exit has no important impact on this composite mental health i ndex for either gender. Table s C. 16 and C.17 also illustrates the estimation results when the outcome of interest is the Euro - D index. As before, the pooled OLS estimates on retirement are positive and 142 significant for both genders, meaning that exiting th well - being, ceteris paribus . 2SLS leads to entirely different conclusions, namely that retirement plays no significant role in determining the Euro - D index for men, but it has a favourable effect for women. The magni tude of this effect is non - negligible (0.24, based on the specification with covariates), compared to the female sample mean of the Euro - D scale, 2.78. Lastly, we examine the results when the mental health outcome variable is a dummy for feeling sad or d epressed in the month preceding the interview (reported in the lowest section of each Table C. 16 and C.17 ). In short, while the pooled OLS estimator suggests a detrimental effect of retirement, the 2SLS estimates imply that retirement is not a significant predictor of the occurrence of sadness and depression episodes, either for women or for men. 47 2.5.2.2.2 Social networks This subsection of the paper uses the last wave in SHARE to estimate model (1) when the dependent variable represents a social outcome of interest rather than mental health. In particular, Y it number of persons in the network with daily contact; a binary indicator for children and parents present in the network, as well as participation in voluntary work. All covariates are the same as before, except that vector X OWN includes number of childre n rather than a dummy for having kids in all 47 The same identification strategy could be employed to investigate the presence of spousal retirement cross - effects amongst the couple households in SHARE (21,528 couple observations). Treating spousal retirement as endogenous and instrumenting both own an d spousal retirement (the later by whether spouse has reached full and early retirement age, and controlling for spousal age) reveals that, conditional on own retirement, spousal retirement has no significant mportant to note that the gender heterogeneity in the effect of retirement holds when restricting the attention to couple household only retirement specification w ith covariates) and demotivation index (magnitude negative 0.135 in the specification with covariates), while having no effect for men. 143 specifications but the one for volunteering, and an additional control for number of living parents model is estimated using the identification strategy described in section III in order to account for the reverse causality between retirement and social outcomes. To elaborate more on this, prior studies report that labour force exit reduces social contacts and induces soci al isolation (Sugisawa et al. (1997)), while at the same time social networks and interactions have been Since both the statutory and the early retirement ages are likely exogenous in a model of social outcomes and affect those outcomes only through retirement, employing them as instruments for retirement becomes an attractive estimation strategy. The top section of Table C. 1 8 and C.19 (panels A and B for men a nd women, respectively) report the estimation results when the outcome of interest is the number of persons on to exit the labour force is accounted (compared to a sample mean of 2.28), while there is no analogous effect for females. A somewhat similar suggestion of an ad verse effect of retirement on social contacts for men can be drawn based on the next section of Table C. 1 8 . Specifically, the 2SLS results from column (2a) net work, ceteris paribus , although, this effect is not different from zero at low significance levels once other covariates are included. Again, there is no corresponding effect for women (panel B , Table C.19 ). At the same time, Table s C. 1 8 and C.19 also sugg est that retirement has no 144 significant impact on the overall satisfaction of the elderly with their social network the parameter on retirement is low in magnitude and significance for both genders. The next two sets of regressions look at the effect of l abour force exit on child - parent bonding. In particular, Table s C. 1 8 and C.19 reports the results from estimating model (1) on a restricted sample of elderly with at least one living child, when the outcome of interest is a a retired pa women or for men. Further, Tables C.18 and C.19 show the estimation results when the dependent variable is a dummy for presence of a parent in the social network (base d on the subsample of respondents with at least one living parent). Those results reveal that retirement significantly increases the probability that a female keeps contact with a parent by roughly 19pp, ceteris paribus . There is also some tentative eviden ce that men are more likely to have a parent in their social network once they exit work (column (2a) of Table C.18 ), but this effect drops in magnitude and significance once controls are included. Lastly, we examine the effect of retirement on an important social activity of the elderly volunteering. The central implication from Table s C. 1 8 and C.19 is that labour force exit significantly increases the probability of involvement with voluntary or charitable work for both genders, ceteris paribus . The magnitude of this effect is 0.08 for males and 0.11 for females (based on the specifications will all covariates). This comprises large effects compared to the sample means of voluntary work (0.18 for men and 0.16 for women, respectively). 2.5.3 More on the gender heterogeneity and mechanism of the effect In order to complete the discussion on the gender heterogeneity of the effect of retirement 145 Tables C.20 presents a test for equality of the paramete r on retirement in model (1) by gender; in other words, they report the results from testing the hypothesis H 0 : = As different for men and women: the bootstrap estimates of this difference are large in magnitude and statistically significant (significance at the 1% level in t he model with no controls, and at the 5% level in the specifications with covariates). At the same time, however, the test cannot reject the null that the coefficient on retirement is the equal for both genders when the outcome of interest is any other psy chological well - being measure, or a measure of the social connectedness of the elderly (in the later case the comparison is further complicated by the reduced sample size and lower estimation precision). Overall, this provides further support for the idea of gender heterogeneity of the effect of labour force exit on mental health measured by the composite demotivation index. Before concluding, the paper addresses an issue which has been largely overlooked by previous research: does the mechanism of the eff through their social network? This may be the case as retirement was shown to affect the social connectedness of the elderly effect for females, which may potentially explain why labour force exit appears beneficial for connectedness overall, and better relations with children in particular, both of which have been hypothesised to lower depression rates. Lastly, exiting work was revealed to increase volunteering of the elderly, which has in turn been linked to lower depression rates (Lum and Lightfood (2005)). We proceed by estimating model (1) from sectio n III on the last wave of data and include a number of social inclusion measures, such as size of the social network, children in 146 the network and volunteering. We then examine the resulting change in the estimated effect of retirement, compared to the mod el with no controls for social networks. The results are reported in Tables C.20 to C.26 . It is evident that the model is robust to inclusion of social network size and presence of children in the network for both genders. s retirement drops both in magnitude and in significance when volunteering is included in the regressions for the demotivation and Euro - D measures (columns (2d) of Table C.22 column and (3d) of Table C.23 ) , but not in the model for death ideation. For the male sample, the effect of labour force exit on the death ideation and demotivation index also changes in significance once volunteering is controlled (column (2d) of Table C. 22 ); yet, the magnitude of these changes is essentially zero. Based on the above results, this paper fails to find any evidence that the effect of the analysis suggests that, at least in part, the beneficial effect of retirement on the comp osite Euro - D and demotivation indices for women is explained by the increase in volunteering following their labour force exit. 147 2.6 CONCLUDING REMARKS This study utilised household - level multinational data from 17 countries in Europe to explore the effects of labour force exit on the mental health of the elderly. Following the identification strategy developed by Coe and Zamarro (2011) the paper explored the exogenous variation in the retirement propensity of the older workers, induced by the national statutory and early retirement ages. Consistent with the findings of Coe and Zamarro (2011) the analysis presented here provides support for the idea that r psychological well - being, other factors being equal. At the same time, however, this study finds important contribution to the literature. In particular, exiting the workforce is predicted to decrease the likelihood that a female has suicidal thoughts by about 3pp, ceteris paribus , and to improve her mental health as measured by the composite demotivation and Euro - D depression scores. The magnitude of this effect is large for the death ideation and demotivation indicators, while relatively low for the Euro - D index. Lastly, there is no evidence that retirement plays an important role on the occurrence of a recent depressive epis ode and on the composite affective depression measure for either gender. The central estimates also uncover a role for retirement on the social contacts of older adults. In particular, the analysis presented evidence that exiting work decreases the size o f the immediate social network for male retirees (in agreement with the findings of Sugisawa et al. (1997)) with no corresponding effect for women. Moreover, retirement significantly increases the probability of a parent present in the social network for f emales, but not for males. Lastly, the paper found no evidence that retirement induces self - perceived social isolation exit from work 148 beneficial effect on volunt eering for both genders. The implications of these findings are twofold. First, the finding retirement affects the mental health of men and women differently, is in line with contemporary theories in the psychology literature suggesting a differential i emotional well - being. Secondly, the results in this paper have potentially important policy psychological health implie s that the recent trends in the EU towards increasing the statutory and early ages of retirement would lead to no detrimental consequences for their mental health, and may have a favourable impact on their social connectedness. At the same time, however, t he - being and relationship with parents, cannot rule out the possibility that increasing the pensionable ages as well as equalizing those ages across gender would lead to a los s of social welfare for women. 149 APPENDICES 150 APPENDIX C MAIN TABLES AND FIGURES 151 Table C.1 : Country representation Country Fraction of total sample wave 1 wave 2 wave 4 Austria 0.068 0.037 0.092 Germany 0.117 0.082 0.026 Sweden 0.141 0.106 0.037 Netherlands 0.088 0.068 0.039 Spain 0.062 0.049 0.041 Italy 0.081 0.079 0.052 France 0.117 0.093 0.100 Denmark 0.071 0.093 0.044 Greece 0.090 0.083 NA Switzerland 0.040 0.048 0.070 Belgium 0.124 0.084 0.080 Czech Republic NA 0.108 0.125 Poland NA 0.070 0.024 Hungary NA NA 0.053 Portugal NA NA 0.034 Slovenia NA NA 0.047 Estonia NA NA 0.136 Total 18,632 22,257 40,934 152 Table C.2 : Labour force status by gender Note: sample restricted to individuals for whom the main variables of interest are not missing. Labour force status Number of observations Employment / retirement rate Males Females Total Males Females Total Employed 16,372 14,610 30,982 37.7% 38.0% 37.9% Retired 27,028 23,813 50,841 62.3% 62.0% 62.1% Total 43,400 38,423 81,823 100.00% 100.0% 100.0% 153 Table C.3 : Sample statistics Characteristic Total sample Male subsample Female subsample Demographic Age 65.790 (0.070) 65.574 (0 .092) 66.053 (0.108) Male 0.530 (0.499) 1.000 0.000 Has reached statutory retirement age 0.540 (0.003) 0.521 (0.004) 0.563 (0.005) Has reached early retirement age 0.672 (0.003) 0.671 (0.004) 0.673 (0.005) Education (in years) 10.832 (0.027) 11.021 (0 .039) 10.603 (0.039) Marital status Married /partnered 0.719 (0.449) 0.819 (0.384) 0.605 (0.488) Divorced / separated 0.092 (0.289) 0.068 (0.251) 0.119 (0.324) Widowed 0.133 (0.339) 0.063 (0.244) 0.212 (0.408) Never married 0.056 (0.229) 0.049 (0.216) 0.063 (0.243) Number of children 2.075 (0.010) 2.128 (0.013) 2.010 (0.015) Foreign country of birth 0.087 (0.001) 0.080 (0.002) 0.093 (0.003) Labour force status and employment history Retired (vs. still employed) 0.618 (0.003) 0.611 (0.004) 0.627 (0.004) Current job in the public sector (conditional on being employed) 0.299 (0.005) 0.256 (0.006) 0.353 (0.007) Current job in the private sector (conditional on being employed) 0.497 (0.005) 0.503 (0.008) 0.489 (0.008) Notes: 1) Means corrected for inverse probability weighed sampling; linearised standard errors reported in parentheses. 2) Social networks available for wave 4 only. Number of observations: 21,394 men and 21,416 wome n. 154 Table C.3 Current job as self employed (conditional on being employed) 0.204 (0.004) 0.240 (0.006) 0.157 (0.006) Last job in the public sector (conditional on being retired) 0.314 (0.003) 0.290 (0.004) 0.343 (0.005) Last job in the private sector (conditional on being retired) 0.519 (0.004) 0.545 (0.005) 0.489 (0.006) Last job as self employed (conditional on being retired) 0.166 (0.002) 0.165 (0.003) 0.168 (0.004) Mental health Affective suffering symptoms Felt sad or depressed last month 0.378 (0.003) 0.287 (0.004) 0.489 (0.004) Felt would rather be dead 0.065 (0.001) 0.046 (0.001) 0.087 (0.002) Tearfulness 0.221 (0.002) 0.117 (0.003) 0.348 (0.004) Feelings of guilt 0.211 (0.002) 0.173 (0.003) 0.258 (0.004) Trouble sleeping 0.314 (0.003) 0.234 (0.003) 0.411 (0.004) Loss of appetite 0.074 (0.001) 0.060 (0.002) 0.091 (0.002) Irritability 0.277 (0.003) 0.263 (0.004) 0.294 (0.004) Fatigue 0.316 (0.003) 0.266 (0.004) 0.377 (0.004) Affective suffering symptoms index (0 to 8) 1.731 (0.012) 1.343 (0.013) 2.202 (0.019) Demotivation symptoms Pessimism (no hopes for the future) 0.157 (0.002) 0.148 (0.003) 0.169 (0.003) Loss of interest 0.076 (0.001) 0.065 (0.002) 0.090 (0.002) Poor concentration (reading) 0.143 (0.002) 0.131 (0.002) 0.158 (0.003) Feels no enjoyment 0.132 (0.002) 0.125 (0.002) 0.142 (0.003) Demotivation symptoms index (0 to 4) 0.488 (0.005) 0.451 (0.007) 0.534 (0.008) 155 Table C.3 Euro - D depression index (0 to 12) 2.262 (0.014) 1.832 (0.017) 2.784 (0.023) Physical health Number of limitations to activities of daily living (0 to 6) 0.205 (0.005) 0.183 (0.006) 0.231 (0.008) Number of chronic conditions (0 to 12) 1.539 (0.009) 1.462 (0.012) 1.633 (0.014) Bad health (self report of less than very good health) 0.755 (0.003) 0.734 (0.004) 0.774 (0.004) Mobility, arm function and fine motor limitations (0 to 10) 1.407 (0.014) 1.091 (0.016) 1.791 (0.024) Hospital stay in the last 12 months 0.144 (0.002) 0.147 (0.003) 0.140 (0.003) Social networks Size of the immediate social network (number of persons) 2.506 (0.019) 2.346 (0.027) 2.688 (0.027) Number of persons in the social network with daily contact 1.208 (0.012) 1.231 (0.016) 1.182 (0.018 Social network satisfaction (0 to 10) 8.757 (0.018) 8.723 (0.024) 8.796 (0.027) Children in the social network (conditional on having a living child) 0.597 (0.006) 0.532 (0.009) 0.670 (0.008) Parents in the social network (conditional on having a living parent) 0.321 (0.016) 0.309 (0.026) 0.332 (0.019) Done voluntary or charity work (last year) 0.173 (0.004) 0.184 (0.007) 0.161 (0.006) No. observations 81,823 43,400 38,423 156 Table C.4 : Statutory, early and actual retirement ages by country and gender (wave 1 & 2) Country Wave 1 & 2 (interview year 2004 & 2007) Early retirement age Statutory retirement age Actual mean retirement age Male Female Male Female Male Female Austria 48 60 57 65 60 58.4 56.7 Belgium 60 60 65 64 60.1 58.7 Czech Rep 49 60 59 61y 10m 60 59.4 55.9 Denmark 50 65 65 65 65 62.8 62.5 Estonia NA NA NA NA NA NA France 51 56 56 60 60 59.2 59.4 Germany 52 63 63 65 65 61.0 60.0 Greece 53 55 55 65 60 60.3 60.4 Hungary 54 NA NA NA NA NA NA Italy 57 57 65 60 58.1 57.0 Netherlands 60 60 65 65 61.1 60.4 Poland 55 60 55 65 60 59.6 57.3 Portugal NA NA NA NA NA NA Slovenia NA NA NA NA NA NA Sweden 61 61 65 65 62.3 61.6 Switzerland 63 62 65 64 63.1 61.7 Spain 60 60 65 65 61.3 61.4 No. observations (retired individuals) 13,207 9,984 48 Statutory retirement age 61.5 years in the public sector; values 65 and 60 are assigned to all men/women in the samp le regardless of sector. 49 Statutory and early retirement age reduced by one year for women for each child up to the 4 th ; value of 60 and 59 for the statutory/early retirement age assigned to all women in the sample. 50 No option for early retirement provid ed in Denmark; value of the early retirement age set to equal the statutory retirement age. 51 Early retirement age linked to the number of years of contribution; value of 56 assigned to the entire sample. 52 Statutory retirement age increased to 65 years, 1 month as of Jan 1, 2012; gradual increase by one month every year planned until reaching age 67. 53 Early retirement age linked to the number of years of contribution; value of 56 assigned to the entire sample. 54 Early retirement age reduced by one year for each additional five - year period (men) or four - year period (women) of hazardous or unhealthy work. Age 60 assigned to the entire sample. 55 Access to early retirement abolished after 2008; value of the early retirement age in wave 4 set to equal the sta tutory retirement age. 157 Table C.5 : Statutory, early and actual retirement ages by country and gender (wave 4) Country Wave 4 (interview year 2011) Early retirement age Statutory retirement age Actual mean retirement age Male Female Male Female Male Female Austria 62 60 65 60 59.1 57.7 Belgium 60 60 65 65 60.2 59.2 Czech Rep 60 59 61y 10m 60 59.9 56.2 Denmark 65 65 65 65 62.8 62.3 Estonia 60 57y 6 m 63 60y 6m 62.5 60.0 France 56 56 62 62 59.1 59.5 Germany 63 63 65 65 61.2 60.7 Greece NA NA NA NA NA NA Hungary 60 60 62 62 58.2 56.2 Italy 57 57 65 65 58.7 58.1 Netherlands 60 60 65 65 61.5 61.1 Poland 65 60 65 60 59.5 56.8 Portugal 55 55 65 65 60.4 60.4 Slovenia 58 58 63 61 58.5 55.5 Sweden 61 61 67 67 62.6 62.5 Switzerland 63 62 65 64 63.2 61.8 Spain 60 60 65 65 61.9 61.5 No. observations (retired individuals) 13,821 13,829 158 Figure C.1 : Retirement age histograms (Sweden) Men Women 159 Figure C.2 : Retirement age histograms (Switzerland) Men Women 160 Figure C.3 : Retirement age histograms ( Poland ) Men Women 161 Figure C.4 : Retirement age histograms ( Czech Republic ) Men Women 162 Table C.6 : First stage estimation results (men) Outcome: retired (vs. still employed) Sample restricted to men (1a) (1b) (1c) (2a) (2b) (2c) Has reached statutory retirement age 0.238*** 0.215*** 0.238*** 0.215*** (0.016) (0.015) (0.016) (0.015) Has reached early retirement age 0.249*** 0.224*** 0.249*** 0.224*** (0.019) (0.017) (0.019) (0.018) Age (in years) 0.178*** 0.155*** 0.119*** 0.178*** 0.155*** 0.120*** (0.005) (0.006) (0.006) (0.005) (0.006) (0.006) Age (in years), squared - 0.001*** - 0.001*** - 0.001*** - 0.001*** - 0.001*** - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Education (in years) 0.003** 0.000 0.001 0.002* 0.000 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Education (in years), squared - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Notes: 1) All specifications control for: year, month and country dummies, and aggregate household income. Models estimated by poole d OLS. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 3) Omitted category for variable marital status: s eparated/divorced; omitted category for variable current/last sector of employment: self employed. 163 Table C.6 (con t Married/partnered 0.023*** 0.022*** 0.023*** 0.023*** 0.022*** 0.023*** (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Never married 0.011 0.010 0.009 0.012 0.011 0.010 (0.009) (0.009) (0.009) (0.009) (0.009) (0.009) Widowed 0.020*** 0.012 0.017** 0.020*** 0.012 0.017** (0.007) (0.008) (0.007) (0.007) (0.008) (0.007) Has bad health 0.029*** 0.029*** 0.029*** (0.004) (0.004) (0.004) Has any kids 0.004 0.002 0.001 0.004 0.002 0.001 (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Hospital stay (last 12 months) 0.017*** 0.018*** 0.017*** 0.021*** 0.022*** 0.021*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Foreign born 0.001 0.003 0.000 0.002 0.004 0.001 (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Public sector of employment (last / current job) 0.128*** 0.125*** 0.125*** 0.129*** 0.126*** 0.126*** (0.006) (0.006) (0.005) (0.006) (0.006) (0.005) Private sector of employment (last / current job) 0.107*** 0.106*** 0.105*** 0.108*** 0.107*** 0.106*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Intercept - 6.275*** - 5.650*** - 4.296*** - 6.279*** - 5.652*** - 4.301*** (0.192) (0.213) (0.205) (0.191) (0.213) (0.205) First stage F statistic (cluster - robust) 515.84 539.92 800.36 525.81 548.18 808.58 F statistic on the excluded instruments (cluster - robust) 211.09 177.71 279.95 210.39 177.51 278.17 F statistic on the excluded instruments (non - robust) 1,963.76 1,861.05 1,787.45 1,951.89 1,856.55 1,779.37 No. observations 43,291 43,291 43,291 43,315 43,315 43,315 R - squared 0.65 0.65 0.66 0.65 0.65 0.66 164 Table C.7 : First stage estimation results ( wo men) Outcome: retired (vs. still employed) Sample restricted to wo men (1a) (1b) (1c) (2a) (2b) (2c) Has reached statutory retirement age 0.288*** 0.231*** 0.289*** 0.231*** (0.021) (0.022) (0.021) (0.022) Has reached early retirement age 0.278*** 0.208*** 0.277*** 0.207*** (0.022) (0.023) (0.022) (0.023) Age (in years) 0.162*** 0.150*** 0.118*** 0.162*** 0.151*** 0.119*** (0.006) (0.007) (0.007) (0.006) (0.007) (0.007) Age (in years), squared - 0.001*** - 0.001*** - 0.001*** - 0.001*** - 0.001*** - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Education (in years) 0.004*** 0.002 0.003* 0.004*** 0.002 0.002* (0.002) (0.002) (0.001) (0.002) (0.002) (0.001) Education (in years), squared - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Notes: 1) All specifications control for: year, month and country dummies, and aggregate household income. Models estimated by poole d OLS. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 3) Omitted category for variable marital status: s eparated/divorced; omitted category for variable current/last sector of employment: self employed. 165 Table C.7 (con t Married/partnered 0.032*** 0.031*** 0.031*** 0.032*** 0.031*** 0.031*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Never married 0.008 0.007 0.006 0.007 0.007 0.006 (0.008) (0.009) (0.008) (0.008) (0.009) (0.008) Widowed 0.010* 0.010* 0.011** 0.010* 0.009* 0.011* (0.005) (0.006) (0.005) (0.005) (0.006) (0.006) Has bad health 0.031*** 0.032*** 0.031*** (0.004) (0.004) (0.004) Has any kids - 0.005 - 0.007 - 0.007 - 0.005 - 0.007 - 0.007 (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Hospital stay (last 12 months) 0.014*** 0.014*** 0.015*** 0.018*** 0.018*** 0.019*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Foreign born - 0.006 - 0.005 - 0.005 - 0.005 - 0.004 - 0.004 (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Married/partnered 0.081*** 0.076*** 0.078*** 0.081*** 0.076*** 0.078*** (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Never married 0.071*** 0.069*** 0.068*** 0.072*** 0.069*** 0.069*** (0.005) (0.006) (0.005) (0.005) (0.006) (0.005) Intercept - 5.582*** - 5.274*** - 4.082*** - 5.750*** - 5.460*** - 4.263*** (0.218) (0.233) (0.227) (0.229) (0.238) (0.230) First stage F statistic (cluster - robust) 467.93 435.22 598.15 477.91 444.90 609.45 F statistic on the excluded instruments (cluster - robust) 188.91 153.34 186.46 189.76 153.21 187.40 F statistic on the excluded instruments (cluster - robust) 2,639.01 2,215.51 1,953.00 2,639.40 2,207.36 1,950.08 No. observations 38,085 38,085 38,085 38,105 38,105 38,105 R - squared 0.68 0.68 0.69 0.68 0.68 0.69 166 Figure C.5 : Mental health by age distance to statutory retirement age (death ideation) Note: Distance to his/her country of residence, and rounded to integer. 167 Figure C.6 : Mental health by age distance to statutory retirement age (demotivation in dex) Note: his/her country of residence, and rounded to integer. 168 Figure C.7 : Mental health by age distance to statutory retirement age ( affective suffering index) Note: his/her country of residence, and rounded to int eger. 169 Figure C.8 : Mental health by age distance to statutory retirement age ( Euro - D scale ) Note: his/her country of residence, and rounded to integer. 170 Figure C.9 : Mental health by age distance to early retirement age (death ideation) early retirement age in his/her country of residence, and rounded to integer. 171 Figure C.10 : Mental health by age distance to early retirement age (demotivation index) Note: Distance to early retirement age computed as the difference country of residence, and rounded to integer. 172 Figure C.11 : Mental health by age distance to early retirement age (affective suffering index) Note: Distance to early country of residence, and rounded to integer. 173 Figure C.12 : Mental health by age distance to early retirement age (Euro - D scale) s/her country of re sidence, and rounded to integer. 174 Table C.8 : Second stage estimation results , Pooled OLS (death ideation, men) Outcome: death ideation Pooled OLS (1a) (1b) (1c) Retired 0.020*** 0.016*** 0.017*** (0.003) (0.003) (0.003) Age (in years) - 0.015*** - 0.012*** - 0.012*** (0.002) (0.002) (0.002) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.002*** - 0.003*** (0.001) (0.001) Education (in years), squared 0.000** 0.000* (0.000) (0.000) Married/ partnered - 0.016*** - 0.016*** (0.004) (0.004) Never married - 0.019*** - 0.019*** (0.007) (0.007) Widowed 0.029*** 0.029*** (0.007) (0.007) Has bad health 0.023*** (0.002) Has any kids - 0.013*** - 0.013*** (0.005) (0.005) Hospital stay ( last 12 months) 0.033*** 0.037*** (0.004) (0.004) Foreign born 0.003 0.004 (0.004) (0.004) No. observations 44,104 42,680 42,691 R - squared 0.02 0.03 0.03 Note: All specifications control for year, month and country dummies, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorc ed. 175 Table C.9 : Second stage estimation results , Pooled IV (statutory retirement age) (death ideation, men) Outcome: death ideation Pooled IV (statutory retirement age) (2a) (2b) (2c) Retired - 0.013 - 0.009 - 0.010 (0.013) (0.014) (0.014) Age (in years) - 0.007** - 0.007** - 0.006* (0.003) (0.003) (0.003) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.002*** - 0.003*** (0.001) (0.001) Education (in years), squared 0.000* 0.000 (0.000) (0.000) Married/ partnered - 0.016*** - 0.016*** (0.004) (0.004) Never married - 0.019*** - 0.018*** (0.007) (0.007) Widowed 0.029*** 0.029*** (0.007) (0.007) Has bad health 0.024*** (0.002) Has any kids - 0.013*** - 0.013*** (0.005) (0.005) Hospital stay ( last 12 months) 0.033*** 0.037*** (0.004) (0.004) Foreign born 0.003 0.004 (0.004) (0.004) No. observations 44,104 42,680 42,691 R - squared 0.02 0.03 0.03 Note: All specifications control for year, month and country dummies, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorced. 176 Table C.10 : Second stage estimation results , Pooled IV ( early ret irement age) (death ideation, men) Outcome: death ideation Pooled IV (early retirement age) (3a) (3b) (3c) Retired - 0.023 - 0.022 - 0.023 (0.015) (0.015) (0.015) Age (in years) - 0.005 - 0.004 - 0.003 (0.004) (0.004) (0.004) Age (in years), squared 0.000** 0.000* 0.000 (0.000) (0.000) (0.000) Education (in years) - 0.002*** - 0.002*** (0.001) (0.001) Education (in years), squared 0.000 0.000 (0.000) (0.000) Married/ partnered - 0.015*** - 0.015*** (0.004) (0.004) Never married - 0.019*** - 0.018*** (0.007) (0.007) Widowed 0.029*** 0.029*** (0.007) (0.007) Has bad health 0.024*** (0.002) Has any kids - 0.013*** - 0.013*** (0.005) (0.005) Hospital stay ( last 12 months) 0.034*** 0.038*** (0.004) (0.004) Foreign born 0.003 0.004 (0.004) (0.004) No. observations 44,104 42,680 42,691 R - squared 0.02 0.03 0.02 Note: All specifications control for year, month and country dummies, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separate d/divorced. 177 Table C.1 1 : Second stage estimation results , Pooled 2SLS (death ideation, men) Outcome: death ideation Pooled 2SLS (both instruments) (4a) (4b) (4c) Retired - 0.018* - 0.015 - 0.016 (0.010) (0.011) (0.011) Age (in years) - 0.006** - 0.005* - 0.005* (0.003) (0.003) (0.003) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.002*** - 0.002*** (0.001) (0.001) Education (in years), squared 0.000* 0.000 (0.000) (0.000) Married/ partnered - 0.016*** - 0.015*** (0.004) (0.004) Never married - 0.019*** - 0.018*** (0.007) (0.007) Widowed 0.029*** 0.029*** (0.007) (0.007) Has bad health 0.024*** (0.002) Has any kids - 0.013*** - 0.013*** (0.005) (0.005) Hospital stay ( last 12 months) 0.034*** 0.037*** (0.004) (0.004) Foreign born 0.003 0.004 (0.004) (0.004) No. observations 44,104 42,680 42,691 R - squared 0.02 0.03 0.03 Note: All specifications control for year, month and country dummies, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significan ce at the 10% level. Omitted category for variable marital status: separated/divorced. 178 Table C.12 : Second stage estimation results , Pooled OLS (death ideation, wo men) Outcome: death ideation Pooled OLS (1a) (1b) (1c) Retired 0.010** 0.005 0.008* (0.004) (0.004) (0.004) Age (in years) - 0.015*** - 0.015*** - 0.015*** (0.002) (0.002) (0.002) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.008*** - 0.008*** (0.001) (0.001) Education (in years), squared 0.000*** 0.000*** (0.000) (0.000) Married/ partnered - 0.040*** - 0.039*** (0.005) (0.005) Never married - 0.038*** - 0.038*** (0.008) (0.008) Widowed 0.006 0.006 (0.006) (0.006) Has bad health 0.040*** (0.003) Has any kids - 0.015*** - 0.015*** (0.006) (0.006) Hospital stay ( last 12 months) 0.039*** 0.043*** (0.005) (0.005) Foreign born 0.023*** 0.024*** (0.005) (0.005) No. observations 39,138 37,760 37,769 R - squared 0.03 0.05 0.04 Note: All specifications control for year, month and country dummies, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorced. 179 Table C.13 : Second stage estimation results , Pooled IV (statuto ry retirement age) (death ideation, wo men) Outcome: death ideation Pooled IV (statutory retirement age) (2a) (2b) (2c) Retired - 0.038** - 0.039** - 0.039** (0.016) (0.017) (0.017) Age (in years) - 0.004 - 0.005 - 0.004 (0.004) (0.004) (0.004) Age (in years), squared 0.000** 0.000** 0.000* (0.000) (0.000) (0.000) Education (in years) - 0.007*** - 0.008*** (0.001) (0.001) Education (in years), squared 0.000*** 0.000** (0.000) (0.000) Married/ partnered - 0.038*** - 0.038*** (0.005) (0.005) Never married - 0.037*** - 0.037*** (0.008) (0.008) Widowed 0.006 0.006 (0.006) (0.006) Has bad health 0.042*** (0.003) Has any kids - 0.015*** - 0.015*** (0.006) (0.006) Hospital stay ( last 12 months) 0.039*** 0.044*** (0.005) (0.005) Foreign born 0.023*** 0.024*** (0.005) (0.005) No. observations 39,138 37,760 37,769 R - squared 0.03 0.05 0.04 Note: All specifications control for year, month and country dummies, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category fo r variable marital status: separated/divorced. 180 Table C.14 : Second stage estimation results , Pooled IV ( early retirement age) (death ideation, wo men) Outcome: death ideation Pooled IV (early retirement age) (3a) (3b) (3c) Retired - 0.029 - 0.020 - 0.020 (0.019) (0.019) (0.019) Age (in years) - 0.006 - 0.009* - 0.008* (0.005) (0.005) (0.005) Age (in years), squared 0.000** 0.000*** 0.000** (0.000) (0.000) (0.000) Education (in years) - 0.008*** - 0.008*** (0.001) (0.001) Education (in years), squared 0.000*** 0.000*** (0.000) (0.000) Married/ partnered - 0.039*** - 0.038*** (0.005) (0.005) Never married - 0.037*** - 0.037*** (0.008) (0.008) Widowed 0.006 0.006 (0.006) (0.006) Has bad health 0.041*** (0.003) Has any kids - 0.015*** - 0.015*** (0.006) (0.006) Hospital stay ( last 12 months) 0.039*** 0.044*** (0.005) (0.005) Foreign born 0.023*** 0.024*** (0.005) (0.005) No. observations 39,138 37,760 37,769 R - squared 0.03 0.05 0.04 Note: All specifications control for year, month and country dummies, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorced. 181 Table C.15 : Second stage estimation results , Pooled 2SLS (death ideation, wo men) Outcome: death ideation Pooled 2SLS (both instruments) (4a) (4b) (4c) Retired - 0.034** - 0.031** - 0.031** (0.014) (0.014) (0.014) Age (in years) - 0.005 - 0.007* - 0.006 (0.004) (0.004) (0.004) Age (in years), squared 0.000*** 0.000*** 0.000** (0.000) (0.000) (0.000) Education (in years) - 0.007*** - 0.008*** (0.001) (0.001) Education (in years), squared 0.000*** 0.000*** (0.000) (0.000) Married/ partnered - 0.038*** - 0.038*** (0.005) (0.005) Never married - 0.037*** - 0.037*** (0.008) (0.008) Widowed 0.006 0.006 (0.006) (0.006) Has bad health 0.041*** (0.003) Has any kids - 0.015*** - 0.015*** (0.006) (0.006) Hospital stay ( last 12 months) 0.039*** 0.044*** (0.005) (0.005) Foreign born 0.023*** 0.024*** (0.005) (0.005) No. observations 39,138 37,760 37,769 R - squared 0.03 0.05 0.04 Note: All specifications control for year, month and country dummies, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category fo r variable marital status: separated/divorced. 182 Table C.16 : Second stage estimation results (mental health, men) Pooled OLS Pooled 2SLS (1a) (1b) (1c) (2a) (2b) (2c) Panel A: men Outcome: demotivation index (0 to 4) Retired 0.088*** 0.051*** 0.059*** - 0.059 - 0.059 - 0.065 (0.011) (0.011) (0.011) (0.042) (0.041) (0.042) No. observations 42,454 41,101 41,110 42,454 41,101 41,110 R - squared 0.08 0.11 0.11 0.08 0.11 0.10 Outcome: affective suffering index (0 to 8) Retired 0.128*** 0.063*** 0.094*** - 0.047 - 0.057 - 0.080 (0.024) (0.023) (0.023) (0.090) (0.087) (0.090) No. observations 42,280 40,940 40,947 42,280 40,940 40,947 R - squared 0.04 0.10 0.07 0.04 0.10 0.07 Outcome: Euro - D index (0 to 12) Retired 0.225*** 0.122*** 0.164*** - 0.116 - 0.128 - 0.153 (0.030) (0.029) (0.029) (0.109) (0.104) (0.107) No. observations 43,695 42,302 42,309 43,695 42,302 42,309 R - squared 0.08 0.14 0.11 0.07 0.14 0.11 Notes: 1) Specifications (1a) and (2a) control for year, month and country dummies. Specifications (1b) and (2b) control for control for year, month and country dummies; age and education (in quadratics); marital status, dummy for having children; bad health and hos pital stay in the last 12 months; foreign born dummy; sector of employment, and aggregate household income. Specifications (1c) and (2c) are the same as (1b) and (2b) with the exception of dropping the bad heath indicator. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. (Refer to Appendix 2 reporting a robustness check to different levels of clusterin g). 183 Table C.16 ) Outcome: sad or depressed last month Retired 0.016** 0.009 0.016** 0.011 0.011 0.007 (0.007) (0.007) (0.007) (0.026) (0.026) (0.027) No. observations 44,197 42,765 42,776 44,197 42,765 42,776 R - squared 0.02 0.05 0.04 0.02 0.05 0.04 184 Table C.17 : Second stage estimation results (mental health, wo men) Pooled OLS Pooled 2SLS (1a) (1b) (1c) (2a) (2b) (2c) Panel B: women Outcome: demotivation index (0 to 4) Retired 0.049*** 0.020 0.030** - 0.208*** - 0.187*** - 0.188*** (0.013) (0.013) (0.013) (0.045) (0.046) (0.045) No. observations 37,703 36,382 36,391 37,703 36,382 36,391 R - squared 0.11 0.14 0.13 0.10 0.13 0.13 Outcome: affective suffering index (0 to 8) Retired 0.173*** 0.077** 0.129*** - 0.032 0.003 - 0.002 (0.034) (0.033) (0.034) (0.106) (0.104) (0.108) No. observations 37,539 36,230 36,239 37,539 36,230 36,239 R - squared 0.05 0.11 0.08 0.05 0.11 0.08 Outcome: Euro - D index (0 to 12) Retired 0.211*** 0.090** 0.152*** - 0.304** - 0.244* - 0.245* (0.042) (0.040) (0.042) (0.129) (0.127) (0.130) No. observations 38,795 37,430 37,439 38,795 37,430 37,439 R - squared 0.09 0.15 0.12 0.08 0.15 0.11 Notes: 1) Specifications (1a) and (2a) control for year, month and country dummies. Specifications (1b) and (2b) control for control for year, month and country dummies; age and education (in quadratics); marital status, dummy for having children; bad health and hos pital stay in the last 12 months; foreign born dummy; sector of employment, and aggregate household income. Specifications (1c) and (2c) are the same as (1b) and (2b) with the exception of dropping the bad heath indicator. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 185 Table C.17 ) Outcome: sad or depressed last month Retired 0.027*** 0.016* 0.025*** - 0.005 0.007 0.006 (0.009) (0.009) (0.009) (0.029) (0.030) (0.030) No. observations 39,197 37,794 37,803 39,197 37,794 37,803 R - squared 0.03 0.06 0.04 0.03 0.06 0.04 186 Table C.18 : Second stage estimation results (men, social networks) Pooled OLS Pooled 2SLS (1a) (1b) (1c) (2a) (2b) (2c) Panel A: men Outcome: size of the social network (number of persons) Retired - 0.035 0.009 0.007 - 0.157 - 0.201 - 0.205* (0.035) (0.037) (0.037) (0.126) (0.123) (0.123) No. observations 21,394 20,306 20,322 21,394 20,306 20,322 R - squared 0.04 0.05 0.05 0.04 0.05 0.05 Notes: 1) Samples restricted to individuals with at least one living child in the model with outcome restricted to individuals with at least one 2) Specifications (1a) and (2a) of control for year, month and country dummies. Specifications (1b) and (2b) control for control f or year, month and country dummies; age and education (in quadratics); marital status, number of children (dummy for having children in Table C.16 ); bad health and hospital stay in the last 12 months; foreign born dummy; sector of employment, aggregate hou sehold income, and number of living parents in the model with outcome variable the exception of dropping the bad heath indicator. 3) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 18 7 Table C.18 ) Outcome: number of persons in social network with daily contact Retired - 0.044** - 0.029 - 0.028 - 0.119* - 0.111 - 0.112 (0.019) (0.020) (0.020) (0.070) (0.071) (0.071) No. observations 20,809 19,755 19,770 20,809 19,755 19,770 R - squared 0.07 0.13 0.13 0.07 0.13 0.13 Outcome: social network satisfaction (0 to 10) Retired - 0.040 - 0.019 - 0.032 0.096 0.051 0.061 (0.031) (0.032) (0.032) (0.116) (0.115) (0.116) No. observations 20,901 19,889 19,894 20,901 19,889 19,894 R - squared 0.02 0.04 0.03 0.02 0.04 0.03 Outcome: children in the social network Retired 0.001 0.005 0.004 - 0.069 - 0.073 - 0.074 (0.012) (0.013) (0.013) (0.046) (0.047) (0.047) No. observations 18,515 17,601 17,606 18,515 17,601 17,606 R - squared 0.05 0.06 0.06 0.05 0.06 0.06 Outcome: parents in the social network Retired - 0.007 - 0.015 - 0.013 0.161* 0.129 0.131 (0.025) (0.025) (0.025) (0.094) (0.097) (0.096) No. observations 3,395 3,218 3,218 3,395 3,218 3,218 R - squared 0.08 0.17 0.17 0.06 0.16 0.16 Outcome: done voluntary/charity work in the last 12 months Retired 0.021** 0.044*** 0.041*** 0.069** 0.076** 0.076** (0.010) (0.010) (0.010) (0.033) (0.032) (0.032) No. observations 21,269 20,208 20,220 21,269 20,208 20,220 R - squared 0.09 0.10 0.10 0.08 0.10 0.10 188 Table C.19 : Second stage estimation results ( wo men, social networks) Pooled OLS Pooled 2SLS (1a) (1b) (1c) (2a) (2b) (2c) Panel B : wo men Outcome: size of the social network (number of persons) Retired - 0.083* - 0.039 - 0.040 - 0.148 - 0.113 - 0.112 (0.042) (0.043) (0.043) (0.148) (0.147) (0.147) No. observations 21,416 20,399 20,418 21,416 20,399 20,418 R - squared 0.07 0.09 0.09 0.07 0.09 0.09 Notes: 1) Samples restricted to individuals with at least one living child in the model with outcome Samples restricted to individuals with at least one 2) Specifications (1a) and (2a) of control for year, month and country dummies. Specifications (1b) and (2b) control for c ontrol for year, month and country dummies; age and education (in quadratics); marital status, number of children (dummy for having children in Table C.16 ); bad health and hospital stay in the last 12 months; foreign born dummy; sector of employment, aggre gate household income, and number of living parents in the model with outcome variable the exception of dropping the bad heath indicator. 3) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 189 Table C.19 ) Outcome: number of persons in social network with daily contact Retired - 0.037 - 0.066*** - 0.064*** 0.022 - 0.001 - 0.001 (0.023) (0.023) (0.023) (0.082) (0.080) (0.080) No. observations 20,803 19,818 19,837 20,803 19,818 19,837 R - squared 0.10 0.15 0.15 0.10 0.15 0.15 Outcome: social network satisfaction (0 to 10) Retired - 0.016 - 0.010 - 0.024 - 0.004 - 0.032 - 0.037 (0.032) (0.033) (0.033) (0.112) (0.113) (0.113) No. observations 21,122 20,168 20,178 21,122 20,168 20,178 R - squared 0.02 0.03 0.03 0.02 0.03 0.03 Outcome: children in the social network Retired - 0.010 - 0.006 - 0.006 - 0.024 - 0.034 - 0.034 (0.012) (0.013) (0.013) (0.042) (0.044) (0.044) No. observations 18,748 17,889 17,897 18,748 17,889 17,897 R - squared 0.04 0.06 0.06 0.04 0.06 0.06 Outcome: parents in the social network Retired 0.021 0.017 0.016 0.231*** 0.191** 0.190** (0.025) (0.025) (0.025) (0.088) (0.086) (0.086) No. observations 4,134 3,892 3,893 4,134 3,892 3,893 R - squared 0.08 0.17 0.17 0.06 0.16 0.16 Outcome: done voluntary/charity work in the last 12 months Retired 0.033*** 0.052*** 0.050*** 0.087*** 0.112*** 0.112*** (0.010) (0.010) (0.010) (0.033) (0.035) (0.035) No. observations 21,324 20,339 20,353 21,324 20,339 20,353 R - squared 0.07 0.09 0.08 0.07 0.08 0.08 190 Table C.20 : Tests for equality of the effect of retirement by gender Dependent variable Instrument list Model specification Difference 1 Female 1 Male P - value for the test H 0 : = Mental health Death ideation Statutory retirement age No controls - 0.025 0.214 All controls - 0.030 0.145 All controls, bad health omitted - 0.029 0.148 Early retirement age No controls - 0.005 0.821 All controls 0.002 0.933 All controls, bad health omitted 0.003 0.920 Statutory & early retirement age No controls - 0.015 0.355 All controls 0.016 0.326 All controls, bad health omitted - 0.014 0.420 Demotivation index Statutory & early retirement age No controls - 0.154*** 0.005 All controls - 0.141** 0.013 All controls, bad health omitted - 0.123** 0.018 Affective suffering index Statutory & early retirement age No controls 0.015 0.906 All controls - 0.052 0.749 All controls, bad health omitted 0.078 0.565 Euro - D index Statutory & early retirement age No controls - 0.188 0.200 All controls - 0.116 0.509 All controls, bad health omitted - 0.092 0.566 Sad or depressed last month Statutory & early retirement age No controls - 0.016 0.683 All controls 0.003 0.974 All controls, bad health omitted - 0.001 0.988 Note: Test for parameter equality based on bootstrap estimates, 500 replications. 191 Table C.20 Dependent variable Instrument list Model specification Difference 1 Female 1 Male P - value for the test H 0 : = Social networks Size of the social network Statutory & early retirement age No controls 0.009 0.959 All controls 0.045 0.798 All controls, bad health omitted 0.093 0.655 Social network satisfaction Statutory & early retirement age No controls - 0.100 0.554 All controls - 0.114 0.515 All controls, bad health omitted - 0.101 0.570 Children in the social network Statutory & early retirement age No controls 0.044 0.500 All controls 0.046 0.465 All controls, bad health omitted - 0.005 0.918 Parents in the social network Statutory & early retirement age No controls 0.070 0.542 All controls 0.037 0.805 All controls, bad health omitted 0.058 0.593 # persons in social network with daily contact Statutory & early retirement age No controls 0.141 0.132 All controls 0.168 0.111 All controls, bad health omitted 0.111 0.277 Volunteering Statutory & early retirement age No controls 0.018 0.659 All controls 0.008 0.872 All controls, bad health omitted 0.036 0.399 192 Table C.21 : Mechanism of the effect (men, death ideation) Characteristics Death ideation (1a) (1b) (1c) (1d) Retired - 0.006 - 0.007 - 0.005 - 0.006 (0.015) (0.015) (0.016) (0.015) Other covariates (including bad health) yes yes yes yes Size of the social network - 0.004*** (0.001) Children in the social network - 0.008*** (0.003) Volunteering - 0.010** (0.003) Test for equality with 1 from specification (a): difference & p - value NA 0.001 0.001 - 0.000** 0.336 0.111 0.045 No. observations 19,944 19,944 17,468 19,930 R - squared 0.03 0.03 0.03 0.03 Notes: 1) All models estimated on data from wave 4 only. Samples in specification (1c) restricted to individuals with at least one living child. 2) All specifications control for year, month and country dummies; age and education (in quadratics); marital status, dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy, and aggregate household income. Omitted category for variable marital status: separated/divorced. 3) Test for parameter equality based on bootstrap estimates, 500 replications. 193 Table C.2 2: Mechanism of the effect (men, demotivation index ) Characteristics Demotivation index (2a) (2b) (2c) (2d) Retired - 0.068 - 0.074 - 0.070 - 0.065 (0.062) (0.062) (0.062) (0.062) Other covariates (including bad health) yes yes yes yes Size of the social network - 0.041*** (0.003) Children in the social network - 0.053*** (0.012) Volunteering - 0.080*** (0.012) Test for equality with 1 from specification (a): difference & p - value NA 0.009 0.005 - 0.006** 0.106 0.190 0.020 No. observations 19,208 19,208 16,837 19,197 R - squared 0.11 0.12 0.11 0.11 Notes: 1) All models estimated on data from wave 4 only. Samples in specification (1c) restricted to individuals with at least one living child. 2) All specifications control for year, month and country dummies; age and education (in quadratics); marital status , dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy, and aggregate household income. Omitted category for variable marital status: separated/divorced. 3) Test for parameter equality based on bootstrap estimates, 500 replications. 194 Table C.2 3: Mechanism of the effect (men, Euro - D scale ) Characteristics Euro - D scale (3a) (3b) (3c) (3d) Retired - 0.232 - 0.235 - 0.228 - 0.230 (0.150) (0.151) (0.155) (0.151) Other covariates (including bad health) yes yes yes yes Size of the social network - 0.012 (0.009) Children in the social network - 0.060*** (0.029) Volunteering - 0.070*** (0.037) Test for equality with 1 from specification (a): difference & p - value NA 0.001 0.005 - 0.006 0.824 0.170 0.208 No. observations 19,749 19,749 17,316 19,737 R - squared 0.13 0.13 0.13 0.13 Notes: 1) All models estimated on data from wave 4 only. Samples in specification (1c) restricted to individuals with at least one living child. 2) All specifications control for year, month and country dummies; age and education (in quadratics); marital status, dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy, and aggregate household income. Omitted category for variable marital status: separated/divorced. 3) Test for parameter equality based on bootstrap estimates, 500 replications. 195 Table C.2 4 : Mechanism of the effect ( wo men, death ideation) Characteristics Death ideation (1a) (1b) (1c) (1d) Retired - 0.054** - 0.054** - 0.055** - 0.053** (0.022) (0.022) (0.024) (0.022) Other covariates (including bad health) yes yes yes yes Size of the social network - 0.003** (0.001) Children in the social network - 0.012** (0.005) Volunteering - 0.001 (0.005) Test for equality with 1 from specification (a): difference & p - value NA 0.000 0.000 0.000 0.484 0.554 0.709 No. observations 20,234 20,234 17,817 20,216 R - squared 0.05 0.05 0.04 0.05 Notes: 1) All models estimated on data from wave 4 only. Samples in specification (1c) restricted to individuals with at least one living child. 2) All specifications control for year, month and country dummies; age and education (in quadratics); marital status , dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy, and aggregate household income. Omitted category for variable marital status: separated/divorced. 3) Test for parameter equality based on bootstrap estima tes, 500 replications. 196 Table C.2 5 : Mechanism of the effect ( wo men, demotivation index ) Characteristics Demotivation index (2a) (2 b ) (2 c ) (2 d ) Retired - 0.239*** - 0.244*** - 0.277*** - 0.227*** (0.067) (0.067) (0.073) (0.068) Other covariates (including bad health) yes yes yes yes Size of the social network - 0.048*** (0.004) Children in the social network - 0.073*** (0.014) Volunteering - 0.099*** (0.014) Test for equality with 1 from specification (a): difference & p - value NA 0.001 0.003 - 0.011*** 0.928 0.460 0.001 No. observations 19,507 19,507 17,212 19,492 R - squared 0.12 0.13 0.12 0.12 Notes: 1) All models estimated on data from wave 4 only. Samples in specification (1c) restricted to individuals with at least one living child. 2) All specifications control for year, month and country dummies; age and education (in quadratics); marital status , dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy, and aggregate household income. Omitted category for variable marital status: separated/divorced. 3) Test for parameter equality based on bootstrap estima tes, 500 replications. 197 Table C.2 6: Mechanism of the effect ( wo men, Euro - D scale ) Characteristics Euro - D scale (3a) (3b) (3c) (3d) Retired - 0.494** - 0.493** - 0.486** - 0.477** (0.184) (0.184) (0.195) (0.187) Other covariates (including bad health) yes yes yes yes Size of the social network 0.005 (0.011) Children in the social network - 0.039 (0.038) Volunteering - 0.131*** (0.041) Test for equality with 1 from specification (a): difference & p - value NA - 0.000 - 0.001 - 0.014** 0.925 0.275 0.044 No. observations 20,050 20,050 17,660 20,034 R - squared 0.14 0.14 0.13 0.19 Notes: 1) All models estimated on data from wave 4 only. Samples in specification (1c) restricted to individuals with at least one living child. 2) All specifications control for year, month and country dummies; age and education (in quadratics); marital status , dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy, and aggregate household income. Omitted category for variable marital status: separated/divorced. 3) Test for parameter equality based on bootstrap estimates, 500 replications. 198 APPENDIX D SUPPLEMENTARY TABLES AND FIGURES 199 Appendix D . 1: Allowing for a more flexible age specification The mainstream literature agrees that the relationship between age and psychological well - being is U - shaped once covariates have been accounted for (see e.g. Stone et al. (2010)). In addition, most authors who studied this relationship for the elderly adul ts reported a linear relationship between age and mental health after a certain age (e.g. Wu et al. (2014) observed a linear increase in depressive symptoms after age 65) a finding largely supported by this paper (refer to section 4B.1 in the main text). However, since age is a key driver of depression, we explicitly address the concern that a quadratic in age might not be flexible enough, and double - check that the key results from the paper still hold and are not due to misspecification. In this section , we show the results of estimating model (1) from the main text when allowing for a cubic specification in age. As can be seen from Table A1 on the next page, the model is robust to including a cubic in age, and the key implications for a beneficial effec the parameter estimates for females are unchanged up to 3 decimals places. Moreover, in the female models with death ideation and demotivation index as the mental health outcome, the linear, quad ratic and cubic terms age are not jointly significant (see columns (1b) and (2b)). Since the quadratic terms are always jointly significant across all outcomes and since all models are robust to changing the age specification, our preferred specification i s a quadratic in age. It is also worth noting the results are robust to including a 4th order polynomial in age, and that higher order polynomials or age dummies (for every year of age) show signs of severe multicollinearity. 200 Table D. 1: Age specifi cation robustness checks ariable Outcome: death ideation Outcome: demotivation index Outcome: Euro - D index (1a) (1b) (2a) (2b) (3a) (3b) Panel A: men Retired - 0.015 (0.011) - 0.017 (0.011) - 0.063 (0.040) - 0.049 (0.042) - 0.134 (0.045) - 0.172 (0.105) Age - 0.005* (0.003) - 0.014 (0.014) - 0.012 (0.012) 0.032 (0.052) - 0.128*** (0.027) - 0.385*** (0.120) Age squared 0.000*** (0.000) 0.000 (0.000) 0.000*** (0.000) - 0.000 (0.000) 0.001*** (0.000) 0.005*** (0.02) Age cubic - 0.000 (0.000) 0.000 (0.000) - 0.000** (0.000) Test for joint significance H 0 : age = age 2 = =0 P - value 0.0000 H 0 : age = age 2 = age 3 =0 P - value 0.2475 H 0 : age = age 2 =0 P - value .0000 H 0 : age = age 2 = age 3 =0 P - value 0.0295 H 0 : age = age 2 =0 P - value 0 .0000 H 0 : age = age 2 = age 3 =0 P - value 0.0000 No. observations 42,680 41,101 42,302 Notes: 1) All specifications control for year, month and country dummies; marital status, a dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy; sector of employmen t, aggregate household income. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 201 Table D. 1 Panel B: women Retired - 0.031** (0.014) - 0.031** (0.014) - 0.187*** (0.046) - 0.186*** (0.046) - 0.244*** (0127) - 0.244** (0.126) Age - 0.007** (0.004) 0.022 (0.017) - 0.014 (0.012) - 0.021 (0.054) - 0.077** (0.033) - 0.375*** (0.141) Age squared 0.000*** (0.000) - 0.000 (0.000) 0.000*** (0.000) 0.000 (0.000) 0.001*** (0.000) 0.005** (0.002) Age cubic 0.000 (0.000) - 0.000 (0.000) - 0.000** (0.000) Test for joint significance H 0 : age = age 2 =0 P - value 0.0497 H 0 : age = age 2 = age 3 =0 P - value 0.2658 H 0 : age = age 2 =0 P - value 0.0000 H 0 : age = age 2 = age 3 =0 P - value 0.9002 H 0 : age = age 2 =0 P - value 0.0000 H 0 : age = age 2 = age 3 =0 P - value 0.0184 No. observations 37,760 36,382 37,430 202 Appendix D.2: Robustness to different levels of standard error clustering Table D. 2 : Standard error clustering robustness checks Men Women (1a) (1b) (1c) (2a) (2b) (2c) Outcome: death ideation Retired - 0.018 - 0.015 - 0.016 - 0.034 - 0.031 - 0.031 (0.010)* (0.011) (0.011) (0.014)** (0.014)** (0.014)** [0.010] [0.011] [0.011] [0.017]* [0.017]* [0.017]* {0.013} {0.014} {0.014} {0.017}* {0.018}* {0.017}* No. observations 44,104 42,680 42,691 39,138 37,760 37,769 R - squared 0.02 0.03 0.03 0.03 0.05 0.04 Outcome: demotivation index (0 to 4) Retired - 0.059 - 0.059 - 0.065 - 0.208 - 0.187 - 0.188 (0.042) (0.041) (0.042) (0.045)*** (0.046)*** (0.045)*** [0.036] [0.039] [0.036] [0.046]*** [0.043]*** [0.041]*** {0.050} {0.050} {0.059} {0.050}*** {0.048}*** {0.044}*** Notes: 1) Standard errors clustered at age - country - 2) Standard errors clustered at country - 4) Specifications (1a) and (2a) control for year, month and country dummies. Specifications (1b) and (2b) control for control for year, month and country dummies; age and education ( in quadratics); marital status, dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy; sector of employment, and aggregate household income. Specifications (1c) and (2c) are the same as (1b) and (2b) with the ex ception of dropping the bad heath indicator. 203 Table D. 2 No. observations 42,454 41,101 41,110 37,703 36,382 36,391 R - squared 0.08 0.11 0.10 0.10 0.13 0.13 Outcome: affective suffering index (0 to 8) Retired - 0.047 - 0.057 - 0.080 - 0.032 0.003 - 0.002 (0.090) (0.087) (0.090) (0.106) (0.104) (0.108) [0.090] [0.087] [0.090] [0.138] [0.125] [0.121] {0.108} {0.101} {0.105} {0.172} {0.165} {0.158} No. observations 42,280 40,940 40,947 37,539 36,230 36,239 R - squared 0.04 0.10 0.07 0.05 0.11 0.08 Outcome: Euro - D index (0 to 12) Retired - 0.116 - 0.128 - 0.153 - 0.304 - 0.244 - 0.245 (0.109) (0.104) (0.107) (0.129)** (0.127)* (0.130)* [0.100] [0.099] [0.103] [0.158]* [0.141]* [0.135]* {0.118} {0.106} {0.112} {0.206} {0.198} {0.185} No. observations 43,695 42,302 42,309 38,795 37,430 37,439 R - squared 0.07 0.14 0.11 0.08 0.15 0.11 Outcome: sad or depressed last month Retired 0.011 0.011 0.007 - 0.005 0.007 0.006 (0.026) (0.026) (0.027) (0.029) (0.030) (0.030) [0.029] [0.030] [0.031] [0.043] [0.042] [0.042] {0.035} {0.035} {0.036} {0.056} {0.057} {0.056} No. observations 44,197 42,765 42,776 39,197 37,794 37,803 R - squared 0.02 0.05 0.04 0.03 0.06 0.04 204 Appendix D . 3: Fixed Effects Estimation Consider again model (1) in the text: Y ict 0 1 Retired ict + 2 + 3 + 4 + d t + ( + ) . (1) For the countries in the sample observed for at least two waves, an alternative identification strategy is available, namely: fixed effects estimation at individual level. 56 Fixed effects (FE) estimation allows for identifying the parameters on the time varying repressors only; for this reason the model is estimat ed with the following controls: vector includes age (in quadratics), marital status, a binary indicator for being in bad health, and a dummy for hospital stay; vector includes household income; as before, d t stands for year and month dummies. Identification of the causal effect of interest by FE relies on the assumption that retirement is uncorrelated with the time varying unobservable characteristics of the elderly, which could affect their mental health outcome Y ict , or formally: cov ( Retired ict, u ics )=0, t, s. This condition rules out the possibility that the elderly exit the labour force as a response to shocks affecting their mental health. In addition, since SHARE is unbalanced panel, FE estimation leads to elimin ating all observations which appear in one wave only. 57 This does not lead to attrition bias under the assumption that selection into being observed only once is exogenous (i.e. uncorrelated with u ics ). Under the assumptions stated above FE estimation cons We check whether the key findings of the paper continue to hold when a FE estimator is employed, and present the FE estimation results for death ideation, motivat ion index and Euro - D scale in the rightmost panel of Table D3. The leftmost panel of the table reports the pooled 2SLS 56 Countries in SHARE observed once are: Hungary, Portugal, Slovenia and Estonia (all observed in wave 4 only). 57 This, together with the above res triction, results in dropping 54.06% of the total male sample and 51.01% of the total female sample in the FE estimation. 205 estimation results based on the full sample of countries; in contrast, the centre panel reports the 2SLS estimation results based on the same sample as the one used in the FE estimation (i.e. a sample restricted to countries observed at least twice , and persons in those countries observed at least twice) . As can be seen from here, the FE estimation results suggest no significant effect of l interest; in addition, the parameter on retirement is not significantly different from zero for women when the outcome of interest is death ideation (see Table A2, panel B). At the same time, however, the FE estimates reported in Tables A2 point to a beneficial effect of exiting work on the motivation index and Euro - D scale for women, although this effect is lower both in terms of magnitude and in significance compa red to the 2SLS estimates on the full sample. In particular, the FE estimate of the retirement effect on the motivation index from specification (3b) of is of magnitude - 0.042 compared to - 0.156 based on the 2SLS estimation; likewise, the FE estimate of th e effect on the Euro - D scale is of magnitude negative 0.106 compared to 0.247 based on the 2SLS estimation (see column (1b)). One reason for this may be the fact that the FE estimation identifies an ATE for all retirees, while the 2SLS estimation identifie s a LATE for the two groups of women complying with the statutory and early retirement ages, and the effect for the later group may be stronger. Another potential explanation is that the FE estimates are obtained on a different set of countries in SHARE, n amely the countries that participated in the survey for at least two waves. To address the later, it is worth this leads to estimates generally lower in magni tude and in significance compared to the full sample of observations in SHARE, suggesting the effect of beneficial effect of retirement on 206 T able D. 3: Fixed effects estimation results Pooled 2SLS (full sample) Pooled 2SLS (restricted sample) Fixed effects (restricted sample) (1a) (1b) (1c) (2a) (2b) (2c) (3b) (3b) (3c) Panel A: men Outcome: death ideation Retired - 0.018* - 0.010 - 0.016 - 0.020 - 0.010 - 0.017 0.006 0.006 0.006 (0.010) (0.010) (0.011) (0.014) (0.014) (0.014) (0.005) (0.005) (0.005) No. observations 44,104 42,680 42,691 23,881 23,800 23,819 23,881 23,800 23,819 R - squared 0.02 0.04 0.03 0.02 0.03 0.02 0.00 0.01 0.01 Notes: 1) Specifications (1a - c) and (2a - c) are analogous to specifications (2a - c) in Table 7. Specification (3a) controls for year, month and country dummies, and age (in quadratics). Specification (3b) controls for year, month and country dummies, age (in quadratic s), marital status, a binary indicator for being in bad health, and a dummy for hospital stay. Specification (3c) is the same as (3b) with the exception of dropping the bad heath indicator. 2) Standard errors clustered at age - country - year level in the po oled 2SLS estimation and at household level in the FE estimation. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 3) Restricted sample includes only observations from countries ob served at least twice. 207 Table D. Outcome: demotivation index (0 to 4) Retired - 0.059 - 0.034 - 0.065 - 0.050 - 0.010 - 0.045 - 0.012 - 0.013 - 0.015 (0.041) (0.040) (0.041) (0.052) (0.051) (0.052) (0.020) (0.020) (0.020) No. observations 42,454 41,101 41,110 22,706 22,633 22,633 22,706 22,633 22,633 R - squared 0.08 0.14 0.10 0.07 0.12 0.09 0.01 0.02 0.02 Outcome: Euro - D index (0 to12) Retired - 0.116 - 0.027 - 0.153 - 0.137 - 0.006 - 0.156 - 0.055 - 0.051 - 0.060 (0.109) (0.099) (0.107) (0.139) (0.128) (0.139) (0.046) (0.046) (0.046) No. observations 43,695 42,302 42,309 23,601 23,528 23,541 23,601 23,528 23,541 R - squared 0.07 0.19 0.11 0.06 0.17 0.09 0.02 0.06 0.03 Panel B: women Outcome: death ideation Retired - 0.033** - 0.023* - 0.031** - 0.020 - 0.007 - 0.016 - 0.008 - 0.008 - 0.008 (0.014) (0.014) (0.014) (0.016) (0.017) (0.017) 0.008 0.008 0.008 No. observations 39,150 37,760 37,769 18,801 18,765 18,776 18,801 18,765 18,776 R - squared 0.03 0.06 0.04 0.03 0.06 0.04 0.00 0.01 0.00 Outcome: demotivation index (0 to 4) Retired - 0.208*** - 0.156*** - 0.188*** - 0.137** - 0.093* - 0.099* - 0.042* - 0.042* - 0.041* (0.045) (0.044) (0.045) (0.055) (0.056) (0.055) (0.024) (0.024) (0.024) No. observations 37,703 36,382 36,391 17,840 17,804 17,815 17,840 17,804 17,815 R - squared 0.10 0.15 0.13 0.10 0.15 0.12 0.01 0.02 0.01 Outcome: Euro - D index (0 to 12) Retired - 0.304** - 0.247* - 0.245* - 0.143 - 0.082 - 0.104 - 0.115* - 0.106* - 0.114* (0.129) (0.127) (0.130) (0.158) (0.154) (0.158) (0.066) (0.064) (0.065) No. observations 38,795 37,430 37,439 18,605 18,569 18,580 18,605 18,569 18,580 R - squared 0.08 0.19 0.11 0.08 0.18 0.10 0.01 0.05 0.02 208 Appendix D.4: New EU member states In this section we assess whether heterogeneity across country exists by estimating model (1) separately on the sample of new EU member - states in SHARE (Czech Republic, Poland, Hungary, Slovenia and Estonia). The main motivation for this is that previous r esearch has generally not studied the effect of retirement on mental health in the post - communist states, while at the same time there may be reasons why the effect differs in those countries. Tables D4. 1 and D4. 2 show the pooled - IV estimation results obt ained from a sample of 8,561 men and 11,145 women in those countries. The model is robust to the inclusion of a dummy for being in bad health; hence, the paper omits reporting the specifications when this dummy is not controlled. Since for both genders all mental health measures have considerably higher sample means in the post - communist countries than in all countries in SHARE, the estimated magnitudes are not directly comparable to the estimates obtained on the full sample. In order to allow inference on the parameter magnitudes, Tables D4. 1 and D4. 2 also report the sample means for the post - communist economies. As can be seen from Table D4. ideation, demotivation index and the probability of feelin g sad or depressed, ceteris paribus. However, in contrast to the results obtained on the full sample of countries, the results reported in panels (3) and (4) suggest a statistically significant beneficial effect of retirement on the affective suffering ind ex and Euro - D scale for men. The effect is of economic importance, as well its magnitude is roughly a third of the mean for the affective suffering measure, and 25% of the mean for the Euro - D scale. Turing briefly to the results for women, Table D4. 2 imp lies a - being in the new EU members: the parameter on retirement is negative and highly significant for all mental health 209 measures, except for the demotivation index in specification (2b) . The magnitude of the effect is also very large: ranging from a third of the mean for the Euro - D index to just above half of the mean for suicide wishing. Taken together, these results suggest a somewhat stronger favourable effect of retirement on women - being in the post - communist states than in the entire female SHARE sample, and a favourable effect on some depression measures for men in those countries. It is worth noting here that all the new EU member - states in SHARE are reasonab ly ratios similar to the mean EU - 27, and with the exception of Poland and Slovenia the at - risk - of - poverty rate (at 60% of median income) for retirement age pers ons in those countries is lower that the mean EU - 27 (European Commission (2012)). However, in contrast to the old EU member - states where most retirement transitions occur around statutory retirement age, the vast majority of women and men in the post - commu nist economies retire when first eligible at the early retirement age (see e.g. Figure E.1 in the body of the paper), and it may be that this difference in retirement patterns is driving the results. 210 Table D . 4. 1 : Second stage estimation results (New EU - member states, men) Characteristics Death ideation Demotivation index Affective suffering index (1a) (1b) (2a) (2b) (3a) (3b) Retired - 0.033 - 0.037 - 0.019 - 0.060 - 0.593*** - 0.550*** (0.024) (0.024) (0.098) (0.097) (0.203) (0.195) Other c ovariates (including bad health) no yes no yes no yes Sample mean outcome (weighted) 0.063 (0.005) 0.650 (0.020) 1.577 (0.037) No. obs ervations 8,544 8,399 8,177 8,043 8,122 7,990 R - sq uared 0.01 0.02 0.09 0.13 0.04 0.11 Notes: 1) Sample restricted to Czech Republic, Poland, Hungary, Slovenia and Estonia. 2) Sample means corrected for inverse probability weighed sampling; linearised standard errors reported in parentheses. 3) Specifications (a) control for year, month and country dummies, and age (in quadratics). Specifications (b) control for year, month and country dummies; age and education (in quadratics); marital statu s, dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy, and aggregate household income. Omitted category for variable marital status: separated/divorced. 4) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 211 Table D . 4. 1 Characteristics Euro - D scale Sad/depressed last month (4a) (4b) (5a) (5b) Retired - 0.550** - 0.567** - 0.087 - 0.075 (0.226) (0.226) (0.062) (0.060) Other c ovariates (including bad health) no yes no yes Sample mean outcome (weighted) 2.293 (0.048) 0.368 (0.011) No. obs ervations 8,434 8,294 8,561 8,415 R - sq uared 0.09 0.15 0.03 0.06 212 Table D . 4.2 : Second stage estimation results (New EU - member states, wo men) Characteristics Death ideation Demotivation index Affective suffering index (1a) (1b) (2a) (2b) (3a) (3b) Retired - 0.084** - 0.068* - 0.243** - 0.148 - 0.799*** - 0.648*** (0.041) (0.040) (0.123) (0.111) (0.243) (0.240) Other c ovariates (including bad health) no yes no yes no yes Sample mean outcome (weighted) 0.124 (0.007) 0.792 (0.021) 2.543 (0.042) No. obs ervations 11,139 10,968 10,636 10,478 10,567 10,410 R - sq uared 0.02 0.05 0.10 0.15 0.05 0.12 Notes: 1) Sample restricted to Czech Republic, Poland, Hungary, Slovenia and Estonia. 2) Sample means corrected for inverse probability weighed sampling; linearised standard errors reported in parentheses. 3) Specifications (a) control for year, month and country dummies, and age (in quadratics). Specifications (b) control for year, month and country dummies; age and education (in quadratics); marital status, dummy for having childre n; bad health and hospital stay in the last 12 months; foreign born dummy, and aggregate household income. Omitted category for variable marital status: separated/divorced. 4) Standard errors clustered at age - country - year level and shown in parentheses. * ** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 213 Table D . 4.2 Characteristics Euro - D scale Sad/depressed last month (4a) (4b) (5a) (5b) Retired - 1.145*** - 0.881*** - 0.218*** - 0.202*** (0.320) (0.308) (0.068) (0.067) Other c ovariates (including bad health) no yes no yes Sample mean outcome (weighted) 3.392 (0.051) 0.543 (0.010) No. obs ervations 11,000 10,835 11,145 10,973 R - sq uared 0.08 0.17 0.02 0.06 214 Appendix D .5 : Allowing for country - specific trends The subsequent section presents the estimation results when adding country - specific trends in model (1) allowing for the trends in psychological well - being and social networks to vary by country: Y ict 0 1 Retired ict + 2 + 3 + 4 + d t + d t 5 + ( + ) , where d t denotes the interaction terms between year and country dummies. As can be seen from Tables D4. 1 through D4. 6, both the first and the second stage of the model are robust to inclusion of country - specific trends and the key implications from the estimation results remain unchanged. Given this, the specification without c ountry - specific trends is preferred in order to avoid introducing high collinearity in the model. 215 Table D . 5. 1: First stage estimation results (men) Outcome: retired (vs. still employed) Sample restricted to men (1a) (1b) (1c) (2a) (2b) (2c) Has reached statutory retirement age 0.238*** 0.215*** 0.238*** 0.215*** (0.016) (0.015) (0.016) (0.015) Has reached early retirement age 0.249*** 0.224*** 0.249*** 0.224*** (0.019) (0.017) (0.019) (0.017) Age (in years) 0.177*** 0.155*** 0.119*** 0.178*** 0.155*** 0.120*** (0.005) (0.006) (0.006) (0.005) (0.006) (0.006) Age (in years), squared - 0.001*** - 0.001*** - 0.001*** - 0.001*** - 0.001*** - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Education (in years) 0.003** 0.001 0.001 0.003** 0.001 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Education (in years), squared - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Notes: 1) All specifications control for: year, month and country dummies, country specific trends, and aggregate household income. Mod els estimated by pooled OLS. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 3) Omitted category for variable marital status: sepa rated/divorced; omitted category for variable current/last sector of employment: self employed. 216 Table D . 5. Married/partnered 0.025*** 0.025*** 0.025*** 0.025*** 0.025*** 0.025*** (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Never married 0.012 0.011 0.009 0.013 0.012 0.01 (0.009) (0.009) (0.009) (0.009) (0.009) (0.009) Widowed 0.022*** 0.013* 0.019** 0.022*** 0.014* 0.019*** (0.007) (0.008) (0.007) (0.007) (0.008) (0.007) Has bad health 0.028*** 0.029*** 0.029*** (0.004) (0.004) (0.004) Has any kids 0.004 0.002 0.001 0.004 0.002 0.001 (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Hospital stay (last 12 months) 0.017*** 0.018*** 0.017*** 0.021*** 0.022*** 0.021*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Foreign born 0.001 0.003 0.000 0.002 0.004 0.000 (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Public sector of employment (last / current job) 0.127*** 0.125*** 0.125*** 0.128*** 0.126*** 0.126*** (0.005) (0.006) (0.005) (0.006) (0.006) (0.005) Private sector of employment (last / current job) 0.107*** 0.106*** 0.105*** 0.108*** 0.107*** 0.106*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) First stage F statistic (robust) 625.80 644.11 868.06 640.93 658.96 972.51 F statistic on the excluded instruments (robust) 215.46 180.92 293.67 215.64 180.16 292.41 F statistic on the excluded instruments (non - robust) 1,958.44 1,867.45 1,786.61 1,953.12 1,859.39 1,780.11 No. observations 43,291 43,291 43,291 43,315 43,315 43,315 R - squared 0.65 0.65 0.66 0.65 0.65 0.66 217 Tabl e D . 5. 2 : First stage estimation results ( wo men) Outcome: retired (vs. still employed) Sample restricted to wo men (1a) (1b) (1c) (2a) (2b) (2c) Has reached statutory retirement age 0.290*** 0.232*** 0.290*** 0.232*** (0.021) (0.022) (0.021) (0.022) Has reached early retirement age 0.278*** 0.208*** 0.278*** 0.208*** (0.022) (0.023) (0.022) (0.022) Age (in years) 0.161*** 0.150*** 0.118*** 0.162*** 0.151*** 0.119*** (0.006) (0.007) (0.006) (0.006) (0.007) (0.006) Age (in years), squared - 0.001*** - 0.001*** - 0.001*** - 0.001*** - 0.001*** - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Education (in years) 0.004*** 0.002 0.002* 0.004*** 0.002 0.002* (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Education (in years), squared - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Notes: 1) All specifications control for: year, month and country dummies, country specific trends, and aggregate household income. Mod els estimated by pooled OLS. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes signifi cance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 3) Omitted category for variable marital status: separated/divorced; omitted category for variable current/last sector of employ ment: self employed. 218 Table D . 5. 2 (con t Married/partnered 0.034*** 0.032*** 0.032*** 0.034*** 0.033*** 0.033*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Never married 0.008 0.008 0.006 0.008 0.008 0.006 (0.008) (0.009) (0.008) (0.008) (0.009) (0.008) Widowed 0.010* 0.010* 0.011** 0.010* 0.010* 0.011* (0.005) (0.006) (0.006) (0.006) (0.006) (0.006) Has bad health 0.031*** 0.031*** 0.031*** (0.004) (0.004) (0.004) Has any kids - 0.005 - 0.007 - 0.007 - 0.005 - 0.007 - 0.007 (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) Hospital stay (last 12 months) 0.014*** 0.015*** 0.016*** 0.018*** 0.019*** 0.019*** (0.004) (0.004) (0.004) (0.004) (0.004) (0.004) Foreign born - 0.005 - 0.005 - 0.004 - 0.004 - 0.004 - 0.003 (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Married/partnered 0.080*** 0.075*** 0.078*** 0.081*** 0.076*** 0.078*** (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Never married 0.071*** 0.069*** 0.069*** 0.071*** 0.069*** 0.069*** (0.005) (0.006) (0.005) (0.005) (0.006) (0.005) First stage F statistic (robust) 505.09 515.95 660.93 479.58 503.04 709.23 F statistic on the excluded instruments (robust) 198.44 154.35 192.08 199.20 154.76 193.00 F statistic on the excluded instruments (non - robust) 2,654.12 2,227.93 1,959.65 1,740.41 2,221.65 1,955.28 No. observations 38,085 38,085 38,085 38,105 38,105 38,105 R - squared 0.68 0.68 0.69 0.68 0.68 0.69 219 Table D . 5. 3: Second stage estimation results, Pooled OLS (death ideation, men) Outcome: death ideation Pooled OLS (1a) (1b) (1c) Retired 0.019*** 0.015*** 0.017*** (0.003) (0.003) (0.003) Age (in years) - 0.015*** - 0.012*** - 0.012*** (0.002) (0.002) (0.002) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.003*** - 0.003*** (0.001) (0.001) Education (in years), squared 0.000** 0.000** (0.000) (0.000) Married/ partnered - 0.016*** - 0.016*** (0.004) (0.004) Never married - 0.020*** - 0.019*** (0.007) (0.007) Widowed 0.028*** 0.029*** (0.007) (0.007) Has bad health 0.023*** (0.002) Has any kids - 0.013*** - 0.013*** (0.005) (0.005) Hospital stay ( last 12 months) 0.033*** 0.036*** (0.004) (0.004) Foreign born 0.003 0.004 (0.004) (0.004) No. observations 44,104 42,680 42,691 R - squared 0.02 0.03 0.03 Note: All specifications control for year, month and country dummies, country specific trends, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorce d. 220 Table D . 5. 4: Second stage estimation results, Pooled IV (statutory retirement age) (death ideation, men) Outcome: death ideation Pooled IV (statutory retirement age) (2a) (2b) (2c) Retired - 0.003 0.004 0.004 (0.013) (0.014) (0.014) Age (in years) - 0.009*** - 0.010*** - 0.009*** (0.003) (0.003) (0.003) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.002*** - 0.002*** (0.001) (0.001) Education (in years), squared 0.000* 0.000 (0.000) (0.000) Married/ partnered - 0.016*** - 0.016*** (0.004) (0.004) Never married - 0.019*** - 0.018*** (0.007) (0.007) Widowed 0.029*** 0.029*** (0.007) (0.007) Has bad health 0.024*** (0.002) Has any kids - 0.013*** - 0.013*** (0.005) (0.005) Hospital stay ( last 12 months) 0.033*** 0.037*** (0.004) (0.004) Foreign born 0.003 0.003 (0.004) (0.004) No. observations 44,104 42,680 42,691 R - squared 0.02 0.03 0.03 Note: All specifications control for year, month and country dummies, country specific trends, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at t he 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorced. 221 Table D . 5. 5: Second stage estimation results, Pooled IV (early retirement age) (death ideation, men) Outcome: death ideation Pooled IV (early retirement age) (3a) (3b) (3c) Retired - 0.016 - 0.011 - 0.011 (0.015) (0.015) (0.015) Age (in years) - 0.006 - 0.006* - 0.006 (0.004) (0.004) (0.004) Age (in years), squared 0.000*** 0.000** 0.000** (0.000) (0.000) (0.000) Education (in years) - 0.002*** - 0.002*** (0.001) (0.001) Education (in years), squared 0.000* 0.000 (0.000) (0.000) Married/ partnered - 0.016*** - 0.016*** (0.004) (0.004) Never married - 0.019*** - 0.018*** (0.007) (0.007) Widowed 0.029*** 0.029*** (0.007) (0.007) Has bad health 0.024*** (0.002) Has any kids - 0.013*** - 0.013*** (0.005) (0.005) Hospital stay ( last 12 months) 0.033*** 0.037*** (0.004) (0.004) Foreign born 0.003 0.003 (0.004) (0.004) No. observations 44,104 42,680 42,691 R - squared 0.02 0.03 0.03 Note: All specifications control for year, month and country dummies, country specific trends, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at t he 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorced. 222 Table D . 5.6: Second stage estimation results, Pooled 2SLS (death ideation, men) Outcome: death ideation Pooled 2SLS (both instruments) (4a) (4b) (4c) Retired - 0.015 - 0.008 - 0.009 (0.011) (0.011) (0.011) Age (in years) - 0.007** - 0.007** - 0.006** (0.003) (0.003) (0.003) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.002*** - 0.002*** (0.001) (0.001) Education (in years), squared 0.000* 0.000 (0.000) (0.000) Married/ partnered - 0.016*** - 0.016*** (0.004) (0.004) Never married - 0.019*** - 0.018*** (0.007) (0.007) Widowed 0.029*** 0.029*** (0.007) (0.007) Has bad health 0.024*** (0.002) Has any kids - 0.013*** - 0.013*** (0.005) (0.005) Hospital stay ( last 12 months) 0.033*** 0.037*** (0.004) (0.004) Foreign born 0.003 0.003 (0.004) (0.004) No. observations 44,104 42,680 42,691 R - squared 0.02 0.03 0.03 Note: All specifications control for year, month and country dummies, country specific trends, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorce d. 223 Table D . 5.7 : Second stage estimation results, Pooled OLS (death ideation, wo men) Outcome: death ideation Pooled OLS (1a) (1b) (1c) Retired 0.010** 0.005 0.008* (0.004) (0.004) (0.004) Age (in years) - 0.015*** - 0.015*** - 0.015*** (0.002) (0.002) (0.002) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.008*** - 0.008*** (0.001) (0.001) Education (in years), squared 0.000*** 0.000*** (0.000) (0.000) Married/ partnered - 0.040*** - 0.039*** (0.005) (0.005) Never married - 0.038*** - 0.038*** (0.008) (0.008) Widowed 0.006 0.006 (0.006) (0.006) Has bad health 0.040*** (0.003) Has any kids - 0.015*** - 0.015*** (0.006) (0.006) Hospital stay ( last 12 months) 0.039*** 0.043*** (0.005) (0.005) Foreign born 0.023*** 0.024*** (0.005) (0.005) No. observations 39,138 37,760 37,769 R - squared 0.03 0.05 0.04 Note: All specifications control for year, month and country dummies, country specific trends, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at t he 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorced. 224 Table D . 5.8 : Second stage estimation results, Pooled IV (statutory retirement age) (death ideation, wo men) Outcome: death ideation Pooled IV (statutory retirement age) (2a) (2b) (2c) Retired - 0.038** - 0.036** - 0.036** (0.016) (0.016) (0.016) Age (in years) - 0.004 - 0.006 - 0.005 (0.004) (0.004) (0.004) Age (in years), squared 0.000** 0.000** 0.000** (0.000) (0.000) (0.000) Education (in years) - 0.008*** - 0.008*** (0.001) (0.001) Education (in years), squared 0.000*** 0.000** (0.000) (0.000) Married/ partnered - 0.038*** - 0.038*** (0.005) (0.005) Never married - 0.037*** - 0.037*** (0.008) (0.008) Widowed 0.006 0.006 (0.006) (0.006) Has bad health 0.042*** (0.003) Has any kids - 0.015*** - 0.015*** (0.006) (0.006) Hospital stay ( last 12 months) 0.039*** 0.044*** (0.005) (0.005) Foreign born 0.023*** 0.024*** (0.005) (0.005) No. observations 39,138 37,760 37,769 R - squared 0.03 0.05 0.04 Note: All specifications control for year, month and country dummies, country specific trends, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at t he 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorced. 225 Table D . 5.9 : Second stage estimation results, Pooled IV (early retirement age) (death ideation, wo men) Outcome: death ideation Pooled IV (early retirement age) (3a) (3b) (3c) Retired - 0.023 - 0.011 - 0.014 (0.018) (0.020) (0.019) Age (in years) - 0.008 - 0.012** - 0.010** (0.005) (0.005) (0.005) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.008*** - 0.008*** (0.001) (0.001) Education (in years), squared 0.000*** 0.000*** (0.000) (0.000) Married/ partnered - 0.039*** - 0.039*** (0.005) (0.005) Never married - 0.037*** - 0.038*** (0.008) (0.008) Widowed 0.006 0.006 (0.006) (0.006) Has bad health 0.041*** (0.003) Has any kids - 0.015*** - 0.015*** (0.006) (0.006) Hospital stay ( last 12 months) 0.039*** 0.044*** (0.005) (0.005) Foreign born 0.023*** 0.024*** (0.005) (0.005) No. observations 39,138 37,760 37,769 R - squared 0.03 0.05 0.04 Note: All specifications control for year, month and country dummies, country specific trends, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at t he 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separated/divorced. 226 Table D . 5.10: Second stage estimation results, Pooled 2SLS (death ideation, women) Outcome: death ideation Pooled 2SLS (both instruments) (4a) (4b) (4c) Retired - 0.033** - 0.030** - 0.029** (0.014) (0.014) (0.014) Age (in years) - 0.005 - 0.007* - 0.006* (0.004) (0.004) (0.004) Age (in years), squared 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Education (in years) - 0.007*** - 0.008*** (0.001) (0.001) Education (in years), squared 0.000*** 0.000*** (0.000) (0.000) Married/ partnered - 0.039*** - 0.038*** (0.005) (0.005) Never married - 0.037*** - 0.037*** (0.008) (0.008) Widowed 0.006 0.006 (0.006) (0.006) Has bad health 0.041*** (0.003) Has any kids - 0.015*** - 0.015*** (0.006) (0.006) Hospital stay ( last 12 months) 0.039*** 0.044*** (0.005) (0.005) Foreign born 0.023*** 0.024*** (0.005) (0.005) No. observations 39,138 37,760 37,769 R - squared 0.03 0.05 0.04 Note: All specifications control for year, month and country dummies, country specific trends, aggregate household income and sector of employment. Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. Omitted category for variable marital status: separate d/divorced. 227 Table D . 5. 11: Second stage estimation results (mental health, men) Pooled OLS Pooled 2SLS (1a) (1b) (1c) (2a) (2b) (2c) Panel A: men Outcome: demotivation index (0 to 4) Retired 0.089*** 0.052*** 0.060*** - 0.053 - 0.063 - 0.069* (0.011) (0.011) (0.011) (0.041) (0.040) (0.041) No. observations 42,454 41,101 41,110 42,454 41,101 41,110 R - squared 0.09 0.11 0.11 0.08 0.11 0.10 Outcome: affective suffering index (0 to 8) Retired 0.125*** 0.060*** 0.092*** 0.005 - 0.063 - 0.086 (0.024) (0.023) (0.023) (0.089) (0.086) (0.088) No. observations 42,280 40,940 40,947 42,280 40,940 40,947 R - squared 0.05 0.10 0.07 0.04 0.10 0.07 Outcome: Euro - D index (0 to 12) Retired 0.222*** 0.120*** 0.163*** - 0.120 - 0.134 - 0.162 (0.030) (0.029) (0.029) (0.106) (0.103) (0.106) No. observations 43,695 42,302 42,309 43,695 42,302 42,309 R - squared 0.08 0.14 0.11 0.07 0.14 0.11 Notes: 1) Specifications (1a) and (2a) control for year, month and country dummies, and country specific trends. Specifications (1b) and (2b) control for control for year, month and country dummies; age and education (in quadratics); marital status, dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy; sector of employment, and aggregate household income. Specifications (1c) and (2c) are the same as (1b) and (2b) with the exception of dropping the bad heath indicator. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 228 T able D . 5. 11 ) Outcome: sad or depressed last month Retired 0.015** 0.008 0.015** 0.010 0.009 0.005 (0.007) (0.007) (0.007) (0.026) (0.026) (0.027) No. observations 44,197 42,765 42,776 44,197 42,765 42,776 R - squared 0.02 0.05 0.04 0.02 0.05 0.04 229 Table D . 5.12 : Second stage estimation results (mental health, wo men) Pooled OLS Pooled 2SLS (1a) (1b) (1c) (2a) (2b) (2c) Panel B: women Outcome: demotivation index (0 to 4) Retired 0.048*** 0.020 0.030** - 0.211*** - 0.183*** - 0.183*** (0.013) (0.013) (0.013) (0.045) (0.045) (0.044) No. observations 37,703 36,382 36,391 37,703 36,382 36,391 R - squared 0.11 0.14 0.13 0.11 0.13 0.13 Outcome: affective suffering index (0 to 8) Retired 0.174*** 0.078** 0.129*** - 0.041 0.002 - 0.005 (0.034) (0.033) (0.034) (0.105) (0.104) (0.107) No. observations 37,539 36,230 36,239 37,539 36,230 36,239 R - squared 0.05 0.11 0.08 0.05 0.11 0.08 Outcome: Euro - D index (0 to 12) Retired 0.213*** 0.091** 0.153*** - 0.312** - 0.252** - 0.256** (0.042) (0.040) (0.042) (0.129) (0.126) (0.129) No. observations 38,795 37,430 37,439 38,795 37,430 37,439 R - squared 0.09 0.15 0.12 0.08 0.15 0.11 Notes: 1) Specifications (1a) and (2a) control for year, month and country dummies, and country specific trends. Specifications (1b) and (2b) control for control for year, month and country dummies; age and education (in quadratics); marital status, dummy for having children; bad health and hospital stay in the last 12 months; foreign born dummy; sector of employment, and aggregate household income. Specifications (1c) and (2c) are the same as (1b) and (2b) with the exception of dropping the bad heath indicator. 2) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 230 Table D . 5.12 ) Outcome: sad or depressed last month Retired 0.028*** 0.016* 0.025*** - 0.007 0.003 0.003 (0.009) (0.009) (0.009) (0.029) (0.030) (0.030) No. observations 39,197 37,794 37,803 39,197 37,794 37,803 R - squared 0.03 0.06 0.05 0.03 0.06 0.04 231 Table D . 5.13 : Second stage estimation results (men, social networks) Pooled OLS Pooled 2SLS (1a) (1b) (1c) (2a) (2b) (2c) Panel A: men Outcome: size of the social network (number of persons) Retired - 0.037 0.009 0.004 - 0.147 - 0.240* - 0.194 (0.035) (0.037) (0.037) (0.126) (0.124) (0.123) No. observations 21,394 20,306 20,322 21,394 20,306 20,322 R - squared 0.04 0.05 0.05 0.04 0.05 0.05 Note s: 1) Samples restricted to individuals with at least one living child in the model with outcome 2) Specifications (1a) and (2a) of control for year, month and country dummies, and country - specif ic trends. Specifications (1b) and (2b) control for control for year, month and country dummies; age and education (in quadratics); marital status, number of children (dummy for having children in Table E.1 6); bad health and hospital stay in the last 12 mo nths; foreign born dummy; sector of employment, aggregate household income, and number of living parents in the same as (1b) and (2b) with the exception of dr opping the bad heath indicator. 3) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 232 Table D . 5. 13 Outcome: number of persons in social network with daily contact Retired - 0.041 - 0.019 - 0.033 0.096 0.051 0.062 (0.031) (0.032) (0.032) (0.116) (0.115) (0.117) No. observations 20,809 19,755 19,770 20,809 19,755 19,770 R - squared 0.02 0.04 0.03 0.02 0.04 0.03 Outcome: social network satisfaction (0 to 10) Retired - 0.045** - 0.029 - 0.030 - 0.116* - 0.113 - 0.109 (0.019) (0.020) (0.020) (0.070) (0.071) (0.071) No. observations 20,901 19,889 19,894 20,901 19,889 19,894 R - squared 0.07 0.13 0.13 0.07 0.13 0.13 Outcome: children in the social network Retired 0.001 0.005 0.004 - 0.069 - 0.074 - 0.073 (0.012) (0.013) (0.013) (0.046) (0.047) (0.047) No. observations 18,515 17,601 17,606 18,515 17,601 17,606 R - squared 0.05 0.06 0.06 0.05 0.06 0.06 Outcome: parents in the social network Retired - 0.007 - 0.015 - 0.013 0.161* 0.129 0.131 (0.025) (0.025) (0.025) (0.094) (0.097) (0.096) No. observations 3,395 3,218 3,218 3,395 3,218 3,218 R - squared 0.08 0.17 0.17 0.06 0.16 0.16 Outcome: done voluntary/charity work in the last 12 months Retired 0.021** 0.044*** 0.041*** 0.073** 0.076** 0.078** (0.010) (0.010) (0.010) (0.033) (0.032) (0.032) No. observations 21,269 20,208 20,220 21,269 20,208 20,220 R - squared 0.09 0.10 0.10 0.08 0.10 0.10 233 Table D . 5.14 : Second stage estimation results ( wo men, social networks) Pooled OLS Pooled 2SLS (1a) (1b) (1c) (2a) (2b) (2c) Panel B : wo men Outcome: size of the social network (number of persons) Retired - 0.084** - 0.039 - 0.042 - 0.139 - 0.110 - 0.111 (0.043) (0.043) (0.043) (0.148) (0.147) (0.147) No. observations 21,416 20,399 20,418 21,416 20,399 20,418 R - squared 0.07 0.09 0.09 0.07 0.09 0.09 Note s: 1) Samples restricted to individuals with at least one living child in the model with outcome 2) Specifications (1a) and (2a) of control for year, month and country dummies, and country - specific trends. Specifications (1b) and (2b) control for control for year, month and country dummies; age and education (in quadratics); marital status, number of children (dummy for having children in Table E.1 6); bad health and hosp ital stay in the last 12 months; foreign born dummy; sector of employment, aggregate household income, and number of living parents in the same as (1b) and (2 b) with the exception of dropping the bad heath indicator. 3) Standard errors clustered at age - country - year level and shown in parentheses. *** denotes significance at the 1% level, ** denotes significance at the 5% level, * denotes significance at the 10% level. 234 T able D . 5.14 ) Outcome: number of persons in social network with daily contact Retired - 0.016 - 0.01 - 0.024 - 0.008 - 0.034 - 0.040 (0.032) (0.033) (0.033) (0.112) (0.113) (0.113) No. observations 20,803 19,818 19,837 20,803 19,818 19,837 R - squared 0.02 0.03 0.03 0.02 0.03 0.03 Outcome: social network satisfaction (0 to 10) Retired - 0.037* - 0.066*** - 0.054*** 0.025 0.029 0.030 (0.023) (0.023) (0.020) (0.082) (0.073) (0.073) No. observations 21,122 20,168 20,178 21,122 20,168 20,178 R - squared 0.10 0.15 0.31 0.10 0.15 0.31 Outcome: children in the social network Retired - 0.010 - 0.006 - 0.005 - 0.022 - 0.034 - 0.031 (0.012) (0.013) (0.013) (0.042) (0.044) (0.043) No. observations 18,748 17,889 17,897 18,748 17,889 17,897 R - squared 0.04 0.06 0.06 0.04 0.06 0.06 Outcome: parents in the social network Retired 0.022 0.017 0.016 0.231*** 0.188** 0.200** (0.025) (0.025) (0.025) (0.088) (0.086) (0.087) No. observations 4,134 3,892 3,893 4,134 3,892 3,893 R - squared 0.08 0.17 0.17 0.06 0.16 0.16 Outcome: done voluntary/charity work in the last 12 months Retired 0.032*** 0.053*** 0.050*** 0.086** 0.112*** 0.112*** (0.010) (0.010) (0.010) (0.033) (0.035) (0.035) No. observations 21,324 20,339 20,353 21,324 20,339 20,353 R - squared 0.07 0.09 0.08 0.07 0.08 0.08 235 BIBLIOGRAPHY 236 BIBLIOGRAPHY Atchley, Robert C. "Continuity theory, self, and social structure." The self and society in aging processes 94 (1999): 121. Balleer, Almut, Ramón Gómez - Salvador, and Jarkko Turunen. Labour force participation in the euro area: a cohort based analysis . No. 1049. European Central Bank, 2009. Barnes, Helen , and Jane Parry. "Renegotiating identity and relationships: Men and women's adjustments to retirement." Ageing and Society 24, no. 02 (2004): 213 - 233. Beck, Scott H. " Adjustment to and satisfaction with retirement." Journal of Gerontology 37, no. 5 (1982): 616 - 624. Beekman, A. T. F., B. W. J. H. Penninx, D. J. H. Deeg, J. Ormel, A. W. Braam, and W. Van Tilburg. "Depression and physical health in later life: results fro m the Longitudinal Aging Study Amsterdam (LASA)." Journal of affective disorders 46, no. 3 (1997): 219 - 231. Börsch - Supan, Axel and Morten Schuth. Discoveries in the Economics of Aging. University o f Chicago Press. 2014, 225 250. Bonsang, Eric, and Tobias Klein. "Retirement and subjective well - being." Journal of Economic Behavior & Organization (2012). Bound, John, and Timothy Waidmann. "Estimating the health effects of retirement." Michigan Retirement Research Center Research Paper No. UM WP 168 (2007). Buber, Isabella, older mothers and fathers: findings from the Survey of Health, Ageing and Retirement in Europe." Eur opean Journal of Ageing 5, no. 1 (2008): 31 - 45. Castro - Costa, Erico, Michael Dewey, Robert Stewart, Sube Banerjee, Felicia Huppert, Carlos Mendonca - Lima, Christophe Bula et al. "Prevalence of depressive symptoms and syndromes in later life in ten European countries The SHARE study." The British Journal of Psychiatry 191, no. 5 (2007): 393 - 401. Charles, Kerwin Kofi. "Is retirement depressing?: Labor force inactivity and psychological well - being in later life." Research in Labor Economics 23 (2004): 269 - 299 . Clark, Andrew E., and Yarine Fawaz. "Valuing jobs via retirement: European evidence." National Institute Economic Review 209, no. 1 (2009): 88 - 103. Coe, Norma B., and Gema Zamarro. "Retirement effects on health in Europe." Journal of health economics 30, no. 1 (2011): 77 - 86. 237 Conti, Rena M. , Ernst R. Berndt, and Richard G. Frank. Early retirement and public disability insurance applications: Exploring the impact of depression . No. w12237. National Bureau of Economic Research, 2006. Copeland, J. R. M., Michael E. Dewey, and H. M. Griffiths - Jones. "A computerized psychiatric diagnostic system and case nomenclature for elderly subjects: GMS and AGECAT." Psychological medicine 16, no. 01 (1986): 89 - 99. Coursolle, Kathryn M. , Megan M. Sweeney, James M. Raymo , and Jeong - Hwa Ho. "The Association Between Retirement and Emotional Well - being: Does Prior Work Family Conflict Matter?." The Journals of Gerontology Series B: Psychological Sciences and Social Sciences 65, no. 5 (2010): 609 - 620. Crystal, Stephen , Mark O lfson, Cecilia Huang, Harold Pincus, and Tobias Gerhard. "Broadened use of atypical antipsychotics: safety, effectiveness, and policy challenges." Health Affairs 28, no. 5 (2009): w770 - w781. Dave, Dhaval , Inas Rashad, and Jasmina Spasojevic. The effects of retirement on physical and mental health outcomes . No. w12123. National Bureau of Economic Research, 2006. DG - ECFIN, E. C. "The 2009 Ageing Report: economic and budgetary projections for the EU - 27 Member States (2008 2060)." Joint Report prepared by the European Commission (DG ECFIN) and the Economic Policy Committee (AWG). Luxembourg: Office for Official Publications of the European Communities (2009). Diener, Ed , Ed Sandvik, Larry Seidlitz, and Marissa Diener. "The relationship between income and subjec tive well - being: Relative or absolute?." Social Indicators Research 28, no. 3 (1993): 195 - 223. Donner, Jonas , Sami Pirkola, Kaisa Silander, Laura Kananen, Joseph D. Terwilliger, Jouko Lönnqvist, Leena Peltonen, and Iiris Hovatta. "An association analysis o f murine anxiety genes in humans implicates novel candidate genes for anxiety disorders." Biological Psychiatry 64, no. 8 (2008): 672 - 680. Duflo, Esther, and Emmanuel Saez. Implications of information and social interactions for retirement saving decisions . Pension Research Council Working Paper 2003 - 13, Philadelphia, 2003. European Commission . "Pension adequacy in the European Union 2010 - 2050." Report prepared jointly by the Directorate - General for Employment, Social Affairs and Inclusion of the European C ommission and the Social Protection Committee (2012). European Parliament. "Healthcare Systems in the EU: a Comparative Study.", Directorate General for Research, (2010). Hammen, Constance . "Stress and depression." Annu. Rev. Clin. Psychol. 1 (2005): 293 - 319. 238 Imbens, Guido W., and Joshua D. Angrist. "Identification and estimation of local average treatment effects." Econometrica 62, no. 2 (1994): 467 - 475. Lavikainen, Juha , Eero Lahtinen, and Ville Lehtinen, eds. Public health approach on men tal health in Europe . Stakes, 2000. Lennon, Mary Clare, and Sarah Rosenfield. "Women and mental health: the interaction of job and family conditions." Journal of health and social behavior (1992): 316 - 327. Lum, Terry Y., and Elizabeth Lightfoot. "The effec ts of volunteering on the physical and mental health of older people." Research on aging 27, no. 1 (2005): 31 - 55. Mead, George Herbert. Mind, self, and society: From the standpoint of a social behaviorist . University of Chicago press, 2009. Meier, Stephan, and Alois Stutzer. "Is volunteering rewarding in itself?." Economica 75, no. 297 (2008): 39 - 59. Mirowsky, John. "Age and the gender gap in depression." Journal of Health and Social Behavior (1996): 362 - 380. Morrow - Howell, Nancy, Jim Hinterlong, Philip A. Rozario, and Fengyan Tang. "Effects of Volunteering on the Well - Being of Older Adults." Journals of Gerontology Series B: Psychological Sciences & Social Sciences 3 (2003). M usick, Marc A., and John Wilson. "Volunteering and depression: The role of psychological and social resources in different age groups." Social science & medicine 56, no. 2 (2003): 259 - 269. O'Riley, Alisa A., Kimberly A. Van Orden, Hua He, Thomas M. Richardson, Carol Podgorski, and Yeates Conwell. "Suicide and death ideation in older adults obtaining aging services." The American Journal of Geriatric Psychiatry (2013). Portnoi, V. A. "Postretirement depression: myth or reality." Comprehensive therapy 9, no. 7 (1983): 31. Prince, Martin J. , A. T. Beekman, D. J. Deeg, R. Fuhrer, S. L. Kivela, B. A. Lawlor, A. Lobo et al. "Depression symptoms in late life assessed using the EURO - D scale. Effect of age, gender and marital status in 14 European centres." The British Journal of Psychiatry 174, no. 4 (1999): 339 - 345. Salami, Samuel O. "Retirement context and psychological factors as predictors of well - being among retired teachers." 6, no. 2 (2010): 47 - 64. Staiger, Douglas , and James H. Stock. "Instrumental variables re gression with weak instruments." Econometrica 65, no. 3 (1997): 557 - 586. 239 Stock, James, and Motohiro Yogo. Asymptotic distributions of instrumental variables statistics with many instruments . Vol. 6. Chapter, 2005. Stone, Arthur A., Joseph E. Schwartz, Joan E. Broderick, and Angus Deaton. "A snapshot of the age distribution of psychological well - being in the United States." Proceedings of the National Academy of Sciences 107, no. 22 (2010): 9985 - 9990. Sugisawa, Atsudo, Hidehiro Sugis awa, Yomei Nakatani, and Hiroshi Shibata. "Effect of retirement on mental health and social well - being among elderly Japanese" [Nihon koshu eisei zasshi] Japanese journal of public health 44, no. 2 (1997): 123 - 130. Tibbitts, Clark. "Retirement problems in American society." American Journal of Sociology (1954): 301 - 308. Van de Velde, Sarah, Piet Bracke, and Katia Levecque. "Gender differences in depression in 23 European countries. Cross - national variation in the gender gap in depression." Social Science & Medicine 71, no. 2 (2010): 305 - 313. Wooldridge, Jeffrey M. Econometric analysis of cross section and panel data . The MIT press, 2002. Wu, Zheng, Christoph M. Schimmele, and Neena L. Chappell. "Aging and late - life depression." Journal of aging and health 2 4, no. 1 (2012): 3 - 28. 240 CHAPTER 3 THE IMMIGRANT - NON - IMMIGRANT WAGE GAP REVISITED: EVIDENCE FROM THE PROGRAMME FOR THE INTERNATIONAL ASSESSMENT OF ADULT COMPETENCIES 241 3.1 INTRODUCTION Immigrants comprise a large and growing fraction of the population worldwide. Over the past two decades the international migrant stock nearly doubled from 154.2 million persons in 1990 to 231.5 in 2013. 58 A large number of those immigrants reside in countries in the Organisation for Economic Co - operation and Development (OECD) and comprise a significant proportion of the total population in these states. For instance, in 2013 the U.S. hosted nearly 46 million immigrants more than any other country in the world with immigrants accounting fo r about 20% of its total population. The number of foreign - born residents of Germany is at present close to 10 million, while the number of immigrants in the United Kingdom, France and Canada overpasses 7 million. 59 Even countries like Japan and Korea, whic h were traditionally highly homogeneous societies, have seen a steady increase in their immigrant inflows since year 2000. At the same time, there is ample evidence in the economic literature that immigrants in all countries earn less than the native - born workers, and are typically hit harder by a worsening of the economic conditions (see e.g. Chiswick (1978), Borjas (1985), and OECD and DESA - UN Report (2013)). Understanding the source of this immigrant - non - immigrant earnings gap has potentially important policy implications. To elaborate more on this, evidence of presence of labour market discrimination might put forward the need for stronger anti - discrimination legislative provisions, and for the implementation of government financial assistance programs targeting the immigrant population. Conversely, evidence that the gap originates from lower workplace skills of the immigrant workers might call for the development of education and , hence, speeding up the convergence of their earnings to those of native - born workers. 58 Source: DESA, UN. "Trends in international migrant stock: The 2013 revision." (2013). 59 Source: OECD and DESA - UN. " World Migration in Figures. " (2013). 242 The primary objective of this study is to examine the immigrant - native earnings gap using a cross - section of 21 countries from the 2011/2012 Programme for the Internat ional Assessment of Adult Competencies, 60 and to make inferences on the extent of labour market discrimination against the immigrant workers. To this end, we employ a modified Mincerian earnings function and a standard Oaxaca - Blinder mean wage decompositio n (Oaxaca (1973) and Blinder (1973)). In addition to this, we extend our analysis to include the decomposition technique by DiNardo, Fortin and Lemieux (1996), aiming at capturing the differences in the earnings of immigrant and natives across the entire w age distribution. Last but not least, we seek to establish the particular way in which the labour market returns for immigrants and natives differ, and to evaluate these in the light of the labour market discrimination theories of the past decades. Our c ontribution to the literature in the field is twofold. First, in contrast to previous studies which typically focused on a single country, we examine the immigrant - non - immigrant wage gap in a large pool of countries from PIAAC. As a consequence, we are abl e to make cross - country comparisons due to the same data structure and methodology applied. Secondly, we make use of the advantages of PIAAC for examining our main question of interest, in particular the fact that the dataset contains a number of measure skills and competences (e.g. numeracy and literacy test scores). This allows us to minimize the presence of unobserved effects such as ability bias, and ultimately to have more credible results. The key findings of the paper can be summarized as follows. First, we find evidence that immigrants in PIAAC have lower returns to education than native workers but enjoy higher 60 PIAAC henceforth 243 returns to cognitive skills, and the latter is especially pronounced for literacy test score. This observati on is line with the statistical discrimination theory suggesting that employers may view educational attainment as a less reliable productivity signal for immigrants, and that in the absence of other reliable productivity signals they place higher weight o proficiency. Secondly, the Oaxaca - Blinder decomposition of the immigrant - native mean earnings gap in the PIAAC sample suggests that a log - wage model specified with the controls usually employed by previous literature would overestima te the unexplained part of the gap nearly two times. In contrast, including measures for numeracy and literacy proficiency reveals a much lower role of the labour market discrimination component just below 7 percent, while the composition effect is twice more important. Lastly, the DiNardo, Fortin and Lemieux (1996) results reveal that numeracy and literacy proficiency matter in the same way throughout the entire distribution of log - wages. Yet, even after controlling for these test scores, the differences in observables cannot fully account for the differences in the log - earning distributions between native and immigrant workers, except for the bottom 10 and the top 10 percentile earners. Hence, much like the Oaxaca - Blinder results, the DiNardo, Fortin and Lemieux results imply presence of labour market discrimination against the non - native workers, but suggest the magnitude of this discrimination is lower than that implied without controls for numeracy and literacy proficiency. The remainder of this stud y is organised as follows. Section 3.2 reviews the main findings of the literature in the field. Section 3.3 discusses the data and defines the key variable employed in the study; this is followed by detailed data analysis. Section 3.4 proceeds by describi ng the methodology employed in the study for the immigrant - non - immigrant gap decomposition. Section 3.5 presents the key results, followed by concluding remarks. 244 3.2 LITERATURE REVIEW Most of the literature on the immigrant - native wage gap draws upon data from the U.S. and Canada (see e.g. Borjas (1994)). One of the earliest works in the field was done by Chiswick (1978), who analyzed the earnings of foreign - born and native men in the U.S., and observed that immigrants earned less than natives at the beginning of their working lives but enjoyed higher wage gains with the increase of their working experi ence and accumulation of skills, so that 10 to 15 years later their earnings surpassed those of non - immigrant workers. This study gave rise to what is and at what speed do immigrant earnings converge to those of native workers. Another early study employing the assimilation approach was done by Tandon (1978), who used data from the 1971 Canadian census to study the immigrant - native wage gap, and reached a sim ilar yet different conclusion for Canada, as compared to the findings by Chiswick (1978). In particular, the author reported that immigrants in Canada earn less than natives when they enter the labour market and have steeper wage - experience profiles than n atives; however, in contrast to the U.S., the wage gap between immigrant and native workers in Canada only narrowed over time but remained substantial. During the last decade several authors have taken advantage of the availability of long panels to re - eva luate the early findings in the field. One example in this respect is a study by Lubotsky (2000), who used longitudinal data from the U.S. Social Security records and reported that the immigrant - non - immigrant wage gap closed by 10 15 percent during the fir st 20 years of immigration, or nearly twice slower compared to the typical estimates based on cross - sectional data. 245 A number of important studies in the field focused on analyzing the role of the changing composition of immigrant flows to the U.S. on the ir labour market performance and on the speed of convergence of their earnings to those of native workers. For instance, Borjas (1984) used the 1970 and 1980 censuses to study the earnings growth of various immigrant cohorts during that period. The key fin ding of that study was that for most immigrant groups the within - cohort growth was considerably smaller than the one predicted by cross - section regressions, which the and Topel (1991) used the 1970 and 1980 censuses, as well, but found that Asian and Mexican immigrants saw an earnings increase of roughly 20 percent in the first 10 years in the U.S., which did not lend strong support for the idea of declining immigrant quality. The issue of Aydemir and Skuterud (2005) attempted to explore it. In particular, the authors analyzed data from five Canadian Censuses between 198 1 and 2001 to explore the reasons for the declining entry earnings of immigrant men and women, and reported that nearly a third of this decline is explained by compositional shifts in the language proficiency of the immigrants. At the same time, the author s found evidence of a decline in the returns to foreign labour market experience but no evidence of a decline in the returns to foreign education. The past decades gave rise to several European studies in the field. A large fraction of these studies employ ed the assimilation approach, yet several authors used a methodology approach allows for decomposing the mean immigrant - non - immigrant earnings gap to a part du e to differences in observables and a part due to discrimination. An early assimilation study by Pischke (1992) analyzed data from the German Socioeconomic Panel from the 1980s and 246 reported a sizeable native - immigrant earnings gap of roughly 20 to 25 perce nt; in addition, the author found little evidence that the earnings of foreign - born workers catch up with those of Germans a finding he attributed to the fact immigrants were concentrated in unskilled and low - skilled jobs. Another study by Kee (1993) exa mined the employment likelihood of native Dutch men in the 80s, on the one hand, and four groups of immigrants, on the other: Antilleans, Surinamese, Turks and Moroccans. The key findings of this study suggested that while Moroccans and Antilleans would ha ve enjoyed the same employment probabilities had they had the same characteristics as native workers, for Surinamese and Turk employees at least part of the gap was due to discrimination (25% and 60%, respectively). Further, Le Grand and Szulkin (2002) emp loyed the decomposition approach to analyze a large sample of Swedish workers in 1995 and found that immigrants from countries other than those in Western Europe earn considerably less than native workers 5.5 percent for male workers and 2.8 percent for females, and that a large part of the observed gap could be attributed to discrimination. A recent study by Coppola et al. (2013) used a nationally - representative data from Italy to examine the labour market outcomes of immigrants and non - immigrants in the country, and found a considerable wage differential between immigrants and natives which increases along the wage distribution. Most of the empirical studies in the field focused on a single country rather than on a broader cross - country analysis. One of the few comprehensive analyses was one by Adsera and Chiswick (2004), who used the 1994 - 2000 waves of the European Community Household Panel to analyze of the earnings of native and immigrant workers, and reported a significant negative effect of immigran countries. In particular, immigrants in Germany and Portugal enjoyed highest earnings relative to those of native workers, while immigrants in countries such as Sweden, Denmark, Sp ain and 247 Luxembourg were paid lowest relative to native workers. The authors also reported some gender and country - of - origin heterogeneity in the immigrant wage levels with Asian, Latin - American and Eastern European men, and Latin - American and Eastern Euro pean women being at the bottom of the male and female wage distributions. To sum up the review of the relevant literature, most research on the immigrant - native earnings gap is country - specific with only few cross - country studies. While this has the benef it of large country samples and the potential to develop models that are better fitted to a specific labour market, it also has the disadvantage that clear cross - country comparisons are unfeasible due to the fact that the data structure as well as the appr oach to measure the native - immigrant earnings gap generally differ between studies. This paper aims at extending the current decomposition literature in that it intends to revisit the previous findings on the immigrant - native wage gap based on a large cros s - country snapshot from the 2011/2012 PIAAC. Even though some PIAAC country - immigrant samples are small making estimation for that particular country unfeasible, using a single pooled dataset has the advantage that it allows for the application of a unifie d research approach in the 20 members - states of OECD, and Russia, and makes it possible to draw country - level conclusions, as well as conclusions for the entire area. In addition to this, we aim at adding to the previous findings by employing better measu res of the worker skills. To elaborate more on this, prior research typically estimated Mincer - type earnings functions with years of formal schooling and labour market experience as the key human capital measures. In addition, some authors included other o bservable characteristics such as gender and ethnic background; type of education; skill - based occupational category and industry of employment (see e.g. Reitz (2001)), and only few studies had access to ency (see e.g. Aydemir and Skuterud (2005) 248 and Coppola et al. (2013)). All this has the drawback that presence of unobserved effects, and most notably ability bias, cannot be ruled out as a potential threat to the validity of the results. We take advantage of a unique feature of PIAAC, which makes the dataset particularly well - suited for our key question of interest, namely: the fact PIAAC contains various measures of the ive workers returns on education and experience, as the vast body of literature did, but to also examine the returns of the two groups of workers to workplace competences, and ultimately to make better inference on the presence of labour market discrimin ation against immigrant workers, and on its importance. 249 3.3 DATA AND DESCRIPTIVE STATISTICS 3.3.1 Data and sample This paper uses data from the Programme for the International Assessment of Adult Competencies. PIAAC is an international survey, implemented in 24 countries and targeted at adults aged between 16 and 65 years. Data was collected in 2011 and 2012. Data on Australia not yet available; in addition, we exclude Sweden due to the fact wage data in this country is restricted in PIA AC and could not be imputed from other publicly available sources (see below). The final sample consists of data from 21 countries: Austria, Belgium (Flanders only), Canada, Czech Republic, Denmark, Estonia, Finland, France, Germany, Ireland, Italy, Japan , Korea, Netherlands, Norway, Poland, Russian Federation, Slovak Republic, Spain, United Kingdom, and the United States. In order to obtain a sample of workers with comparable labour force attachment, we focus our attention to all prime - age individuals in PIAAC, who are full - time employed at the time of the survey. We employ the OECD definition of prime age, i.e. persons aged 25 to 54 years. In turn, full - time employment is defined as working at least 30 hours a week. 61 In addition to this, following the mainstream literature, we drop the lowest and the highest wage percentile in each country in order to limit the influence of outliers and observations with implausible hourly wage values. All persons with missing wage ar e also dropped from the sample. The final sample consists of 45,697 observations on full - time employed prime age individuals. 61 There is a considerable variation in the full - time working week duration by country. E.g. the Affordable Care Act in the U.S. defines a full - time week 30 hours or more; countries in Europe have typically a definition between 30 and 40 hours a week (e.g. 37 hours in Denmark; 38 in Belgium; 35 - 40 h ours in the Netherlands and Germany; 40 hours in Poland, etc). For this reason this paper adopts a definition of a minimum of 30 hours a week. 250 3.3.2 Variables definitions 3.3.2 .1 Wage and immigrant status The wage measure employed is gross hourly earnings in USD and PPP - adjusted. For Austria, Canada, Germany, Sweden, and the United States the public use PIAAC files only contain wage data as deciles in the hourly wage distribution. For this reason, for all these countries except Sweden we impute the hourly wage as the mean wage for full - time employed workers in each decile in the corresponding country. 62 We focus our analysis on measuring the wage gap between first generation immigrants and non - immigrants. In turn, immigrants are defined as persons who are foreign - born, an d have at least one foreign - born parent. 3.3 .2. 2 Skill measures The unique feature of PIAAC is that apart from providing demographic and socio - economic information, the respondents answered a series of questions aimed at measuring their cognitive and wor kplace skills and competences. The main skill measures in PIAAC are divided into three domains: 1. Numeracy skills: PIAAC defines numeracy as the ability to use and interpret mathematical information, and it encompasses solving a problem in mathematical content. 62 All calculations are based on OECD data on the mean wages by deciles and country for full - time employed persons in 2009 (in USD and PPP - adjusted). OECD values are reported annually for Austria, weekly for Canada and the U.S., and monthly for Germany. All values were converted to hourly equivalents assuming the following work - interval durations: 8 hours per day, 40hours per week, 173 hours per month and 2,080 hours per year. The same week/month/year durations are used by PIAAC. In turn, all the resulting we ekly values were CPI adjusted to 2011 values based on the 2011 - 2009 CPI (source : http://www.inflation.eu/inflation - rates/historic - cpi - inflation.aspx). Data for Sweden is only reported for deciles 10, 50 and 90, making the wage imputation implausible; for t his reason the country is excluded from the analysis. 251 2. Literacy proficiency: literacy is defined as the ability to understand written text, and it includes skills such as text decoding, as well as comprehension and interpretation. In each countries in PIAAC the literacy score reflects the responden of that country. This implies that for foreign - language immigrants the literacy test serves as a measure of their proficiency in the language of the receiving country (rather than their native language), making the test of particular importance in our analysis. 3. Problem solving in technology - rich environment: this domain is defined as the ability to use information technology to obtain, assess and communicate information. The design of the skill testing in PIAAC was b ased on matrix sampling where each respondent answered only a subset of questions from the total question pool. Item response theory scaling was then applied in order to obtain cognitive test scores for the entire PIAAC sample in terms of a common scale. I n order to increase the accuracy of these measurements, score. For each skill domain, ten plausible score values were computed; we use the first plausible value in each domain. 63 All test scores are measured on a scale ranging from 0 to 500. Various working environment skill measures, such as cooperation, communication, and time organizing are also available. 63 No plausible value is better than either of the others; in principle, we could have used all 10 plausible values. However, this has the disadvantage that computing standard errors with 10 plausible va lues and 80 replicates for each country would require computing 800 statistics (80 replicates times 10 plausible vales), and 10 more statistics using the whole sample and final sample weight, i.e. a total of 810 computations of statistic of interests. For this reason we focus on using the first plausible value only, i.e. the first test score imputation for every respondent in the sample. 252 3.3.2.3 Other covariates Demographic characteristics such as age, gender, education, marital status, number of children and parental education level are also available in PIAAC. In addition, education and labour force characteristics, such as actual working experience (years of pa id work during lifetime), job tenure, and skill - based occupational category are also available. However, the public data files do not contain the raw variables in some countries. For instance, age in years is not available for Austria, Canada, Germany and the US. For this reason, we employ a categorical age definition based on 5 - year age intervals, which are available for all countries. Likewise, working experience in PIAAC was top - coded at 47 years for Austria, Canada, Germany and the US. Given that we foc us on workers age between 25 and 54, the top coding could not lead to censoring in our sample, so we opt for using experience as a continuous variable. Lastly, various education measures are available in PIAAC: highest level of education grouped in ISCED c ategories, years of education (not available for Germany, but imputed), and education categorized as below high school, high school and above high school (not available for Austria, but imputed). 64 3.3.3 Sample statistics Table E.1 presents the key sample statistics in the total pooled sample, and by country. Several things are worth noting here. First and foremost, countries with higher than the pooled 64 respondent based on their education by ISCED category, and the typical duration for each ISCED educational category in Germany. Education categorized as below high school, high school and above high school is not available for Austria. We impute this vari able based on the available education by ISCED categories for Austria. Source: Classifying Educational Programmes: Manual for ISCED - 97 Implementation in OECD Countries. OECD 1999, available a t http://www.oecd.org/edu/skills - beyond - school/1962350.pdf 253 sample mean wages also have a higher fraction of immigrants, on average. E.g. Ca nada and the U.S. are the countries with highest mean wages in the PIAAC sample, and the share of immigrant population in these countries is much above the OECD average. This is in line with the theoretical literature which lists the economic conditions in the country of destination amongst the key pull factors of migration. Turning briefly to demographics, the prime age full - time employed in Ireland and Poland appear significantly younger than average, with the fraction of workers in the 25 - 34 age range of above 40 percent. Lastly, there is a large country variation in the share of women in the full - time prime age employed population: the Netherlands, Germany, Italy, Korea and Japan have particularly low fraction of females, while Finland, Denmark, post - com munist Europe, and the US have relatively high fractions of females amongst the full - time employed prime age workers. Turning to the education and skill - levels in the country samples, it appears from Table E.1 that the full - time employed prime age workers in Ireland and Norway are particularly well educated (sample mean years of schooling of roughly 16 and 15 years, respectively). Four other European countries Italy, France, Spain and Austria are on the other end of the spectrum with mean years of educ ation of just between 11.5 and 12.5 years. The mean years of actual labour market experience by country naturally follows the age structure and educational attainment of the employees. Last but not least, workers in Japan, Finland, Netherlands and Norway a ppear to score highest across all cognitive skill measures; those countries also have below average to average fraction of immigrant population. In contrast, Italy, Spain, the U.S, Poland and France have lowest across in the PIAAC sample across most skill domains; all countries in this group, except Poland, have a large immigrant population amongst the full - time employed prime age individuals. 254 A glance at the labour market characteristics of the two subsamples reveals that non - immigrants appear significant ly better qualified in terms of education both measured as years of schooling, and as highest degree obtained (Table E.2) . It is also worth mentioning that nearly 19% of the respondents in the immigrant sample are high - school drop - outs versus only 9% in the non - immigrant sample. In addition to this, immigrants have nearly 2.5 years less of working experience during their lifetime. Further, non - immigrants are employed in better occupations: they have higher number years of formal education required to obta in the current job they hold (13.1 years vs. 12.3 for immigrants), and are nearly 10% more likely to be employed at skilled occupations. Related to this, non - immigrants are about 10% more likely to have undergone on - the - job training during the year precedi ng the interview. The two groups also differ by sector of employment with non - immigrants being considerably more likely to be employed in the public sector (28% of all versus 18% for non - immigrants). Turning to cognitive competences, non - immigrants perfor m significantly better in all skill domains. The gap between the two groups is particularly pronounced for the numeracy test scores nearly 40 points, or roughly 0.8 standard deviations. Immigrants also score lower on the literacy test (difference in mean s of just over 30 points, or 0.65 standard deviations); this is expected as nearly 75% of all immigrants have a foreign - language background. In addition, those workers also demonstrate worse problem solving skills (gap of about 20 points, or 0.4 standard d eviations). Moreover, native workers appear more likely to use numeracy, reading and writing skills at work, and have a higher index of learning and readiness to learn at work. All the above observations strongly point towards immigrants in the PIAAC samp le having, on average, worse labour market characteristics and cognitive abilities. It is somewhat puzzling then, at first, that immigrants earn slightly higher wages, on average ($18.62 vs. 255 $18.05); however, this is merely a reflection of the fact that co untries with higher wages have larger immigrant - populations. This suggests that in order to get the complete picture of the earnings differences between the two groups, we need to examine those outcomes when controlling for country - specific heterogeneity. We present this analysis in the next paragraphs. Figures E. 1 to E. 7 depict the mean wage gap between native and immigrant workers conditional on observable covariates, and after accounting for country - level heterogeneity. Figures E. 1 to E. 4 present the di fferences in mean wages conditional on educational attainment, labour market experience, sector of employment and occupation. These graphs reveal an interesting pattern: a sizeable and statistically significant earnings gap, even within each category. For instance, the mean wages of native workers are considerably larger than those of immigrants conditional on educational attainment, and this gap remains stable across all education categories. The existence of such a gap might imply presence of labour marke t discrimination against immigrant workers, but there is an alternative plausible explanation, as well, namely: there might be considerable heterogeneity in education quality between immigrant and native workers within each educational attainment group. Vi rtually the same observation prevails when examining the immigrant and non - immigrant mean earnings by labour market experience and sector of employment (Figures E. 2 and E. 3), although the gap in the public sector appears somewhat smaller than the gap obser ved in the private and non - profit sectors. Figure E.4 also reveals earnings differences within a skill - based occupational category, although those differences are smaller than the ones observed within each education and experience group; moreover, the gap is lower for high skill occupations than for elementary, semi - skilled white collar and semi - skilled blue collar ones. 256 All these observations lend support for idea that at least some of the immigrant - non - immigrant wage gap observed within each educational, experience, sector of employment and occupational groups, might be due to differences in skills. This explanation is further reinforced when examining the mean wages of native and immigrant workers conditional on numeracy, literacy and problem solving sco res (standardized to have a mean of zero and a standard deviation of 1), presented on Figures E. 5, E. 6 and E. 7. As can be seen from here, in all skill domains the mean wage gap within a given bin is considerably lower than the gap conditional on the conventional observable characteristics, such as education and working experience. In addition to this, the mean ea rnings gap between native and foreign born workers decreases with the increase of test scores, and is not statistically different from zero in the top score ranges. This is particularly well pronounced for numeracy and literacy proficiency, and to a lesser extent for increase with problem solving test score. This is consistent with the findings of a recent paper by Hanushek et al., 2013 who used data from PIAAC and reported considerably larger labour market returns to numeracy and literacy test skills than for problem solving skills. 257 3.4 IDENTIFICATION STRATEG Y 3.4.1 Mean wage gap decomposition Following the mainstream literature, we start with a standard linear model written explicitly with an intercept, relating log - wages and covariates, defined for two mutually exclusive groups of interest: Y gi = + , { I, N }, (1) where I stands for immigrants, and N for non - immigrants; Y denotes log - earnings, X is a vector of observable covariates and u ig the idiosyncratic error term. It is well established in the economic literature that the raw wage gap (defined as the difference in the mean outcomes for the two groups, = ) can be represented as: = + + ( ) where and , { I, N }, are the estimates of the intercept and slope parameters from model (1), for group I and N , respectively. This representation is known as the Oaxaca - Blinder decomposition (Oaxaca (1973) and Blinder (1973)). The first term in the decomposition, + , is often ( ) , is the so covariates. It is worth noting that is the returns to skills effect for the baseline group, and it will generally depend on which group is chosen as a baseline (see e.g. Oaxaca and Ransom (1999)). In order to illustrate the decomposition more formally, it is convenient to adopt the framework developed by Fortin, Lemieux and Firpo (2011), accord ing to which the wage 258 structure effect can be interpreted as a treatment effect. To elaborate more on this, immigrant status can be viewed as a binary treatment D i , taking a value of 1 if the person is an immigrant, and 0 otherwise; in this way, the treat ment identifies two distinct and mutually exclusive groups, I and N . Next, define Yi N and Yi I as the potential values of the outcome of interest for worker i when the binary treatment takes on values 0 and 1, respectively, and D iN and D iI as the potential treatment, i.e. moving from group I to group N , and vice versa . Of course, the main difficulty of the analysis stems from the fact that the counterfactual outcomes are not observed; we only observe the actual earnings Y i of worker i , whic h can be expresses as Y i = D gi Y gi , { I, N }. Hence, we could only compare the actual mean wages of the two groups, and . However, using a program evaluation approach we can represent the mean wage gap the following way: = = + 259 (where we have applied the Law of Iterated Expectations, and used the assumption that ). Here the first term equals , and represents the average treatment effect of the treated (ATT), or more intuitively the difference between the actual mean wages of immigrants, and the potential wages of immigrants had they been rewarded according to the wage structure of natives. 65 The second term in the mean wage gap, , represents the difference between the potential earnings of non - immigrants if they had the same observable characteristi cs as immigrants and the actual mean non - immigrant wages. Given this set - up, the wage decomposition problem could be restated the following way: what would the wages of non - immigrants be if they had the same returns to skills as immigrant workers, or, in other words consistently estimating ATT (or ATUT, respectively). This representation is useful as it allows applying well - known results from the program evaluation literature consistent estimation of ATT requires that two key assumptions are satisfied : A1) overlapping support Stated in simple terms, the overlap assumption requires that X in its support , (see e.g. Wooldridge (2010)). Intuitively, this rules out cases where the factors affecting the log - earnings may differ across the two groups of interest. In the literature of wage decomposition, this assumption often fails to hold (see e.g. Fortin, Lemieux and Firpo (2011)). Studies o f the immigrant - non - 65 The choice of a reference group is arbitrary; an alternative representation would be: = . In this case, the first term represents the average treatment effect of the untreated (ATUT). 260 immigrant wage gap also face a potential fail of the overlap assumption, if e.g. factors such as age and country of immigration are important determinants of immigrant wages as these factors do not affect wages of native workers and wou ld, hence, serve to unambiguously identify group I . In order to ensure holds, we impose the same dimension of the two vectors of covariates and . This assumption states that a sufficient condition for consistent estimation of ATT, is that conditional on the observable covariates , the distribution of the unobservables is the same in the two groups: D ( | , D=1) = D ( | , D=0) (Wooldridge (2010)). As a corollary, the consistent estimation of ATT is possible, even though e.g. unobserved innate ability and educational attainment are correlated, as long as once the observable characteristics (such as age, labour market exper ience, cognitive skills test scores, etc.) in vector X have been accounted for, the dependence structure between ability and education is the same in groups I and N . 3.4.2 DiNardo - Fortin - Lemieux decomposition Numerous authors have proposed further extensions of the Oaxaca - Blinder decomposition framework to allow for decomposing the differences in the entire log - wage distribution, rather than merely focusing on the mean - wage gap. We follow one commonly used approach in the literature, proposed by DiN ardo, Fortin and Lemieux (1996) (referred to as to construct counterfactual wage distributions, and can be summarized as follows. 261 Adapting the notation from DiN ardo, Fortin and Lemieux (1996) and DiNardo (2002) to the notation of this paper, let be the actually observed distribution of log - wages for natives, and the actual distribution of log - wages for immigrants. DFL represent the actual dist ribution for the native and immigrant workers, and , respectively as a conditional distribution of log - wages on individual attributes, 66 integrated over the distribution of individual attributes, or formally: where is the distribution of covariates is each group. Further, consider the counterfactual density for native workers , i.e. the density that would have been observed if native workers had the distribution of individual covariates of The counterfactual density for immigrants is composed analogously as: Then difference between the actual density for native workers and the counterfactual density for those workers can be expressed as: 66 This follows from representing as a joint distribution of wages , integrated over the distribution of individual attributes, and then app lying Law of Iterated Expectation. 262 and reveals the part of the wage gap, which is due to differences in the distribution of covariates between native and immigrant workers. The part due to discrimination is given by the difference between the counterfactual density for natives and the actual d ensity for immigrants: However, since generally includes several explanatory variables, integrating over multiple dimensions of covariates may be unfeasible. In order to avoid a potentially unsolvable problem, DFL suggest representing the counterfactual density for natives as a reweighted actual densit y for natives: = where is the weight applied to the actual density of natives. 67 covariates for natives and immigrants as: 67 Similarly, the counterfactual density for immigrants can be represented as: = . 263 where is unconditional distribution of covariates in the total population. Plugging in the expressions for and in the definition of yields: The advantage of this representation of is that it replaces the conditional distribution of covariates with the conditional probabilities and , which are analogous to standard binary response models. and ar e often worker belongs to group { I , N } respectively, given the set of his/her observable characteristics. Alternatively, employing the treatment approa ch from the previous section and viewing immigrant status as the treatment, is the probability a given worker would have been exposed to treatment conditional on the set of his/her observable covariates. The unconditional probabilities and are simply the fractions of immigrants and natives in the total population. Given this set - up, estimating the reweighting function boils down to estimating the propensity score; the later can be done by the following steps: 1. Define a binary treatment variable D , such that D i =1 if worker i is an immigrant, and 0 otherwise. 2. Run a logit or probit of the treatment indicator D on the set of observable covariates X . 3. Obtain the predicted probability . The unconditional probabilities in expression above, and , can be estimated simply as the fraction of immigrants and the fraction of non - immigrants in the pooled sample, and do not vary by observation. In practice, these terms 264 can be ignored i n the estimation of since the statistical packages apply a subsequent normalization to the weighting variable such that it sums up to 1 (see e.g. DiNardo (2002)). It is important to note an additional step for data with sampling weights, such as PIAAC: the final re - weighting factor, in such cases is composed as the product of and the final sampling weight (see DiNardo (2002)). Once the weighting factor has been obtained, the counterfactual density for native workers can be obtained via a non - parametric kernel density estimation, 68 where their actual log - earnings density is reweighted by . 68 A general kernel density estimator has the form: , where h > 0 is the so - called bandwidth and k ( · ) is the kernel function. It can be shown that under weak conditions, if and , (see Wooldridge (2010)). We use a commonly used kernel Epanechnikov which has the form k ( u ) = (1 - , - 1< u - of - bandwidth incorporated in the quantitative software package STATA. 265 3.5 ESTIMATION RESULTS 3.5.1 Mean wage gap decomposition 3.5.1.1 OLS results Before we get to the mean wage gap computation and decomposition, it is useful to examine the OLS estimation results from model (1) from the previous section. This model is estimated separately on the pooled PIAAC immigrant and non - immigrant subsamples, and with a different set of controls (see Tables 3 and 4). In particular, specification (A) of Table E.3 report s the results from estimating a Mincer - type equation including education, labour market experience and experience squared in the vector of covariates X ; in addition, the model is specified with country dummies in order to account for country - level heterogeneity. Specification (B) adds gender dummies and controls for sector of employment (parameterized as public, private and non - governmental sector) in the list of covariates. 69 Next, columns (C) to (G) in both tables take advantage of the novelty of PIAAC in that it contains a wide range of measures of cognitive skills and competences. To be more specific, specifications (C) to (E) include separate skill measures for the numerac y, literacy and problem solving in technology - r ich environment domains; specification (F) controls for numeracy and literacy test scores, and specification (G) controls for all these measures jointly. 70 As can be seen from column (A) of Table E.3 , foreign - born workers appear to have lower return to education than natives (difference of 1.5 percentage points; significant at the 1% level), 69 Other relevant variables, such as job tenure, on - the - job training, or skill - based occupational category are available in PIAAC. However, these have the disadvantage that they have a high non - response rate (e.g. on - the - job training), or are not available in the public data - files for part of the countries in PIAAC. We opt for omitting them in the regression in ord er to avoid further lowering the sample size. 70 Note that France, Italy and Spain did not take part in the survey of the problem solving in technology - rich environment skills, which results in a lower number of observations in the specification where this test score is controlled. 266 although slightly higher return to experience. The results from specification (B) confirm this observation. Once cogniti ve skill measures have been accounted for in columns (C) to (E), the gap between the return to education for immigrants and natives further deepens. At the same time, however, immigrants appear to have higher returns to cognitive abilities, although the di fference between the two groups is not always statistically significant. The later holds across all skill domains, and is particularly well pronounced in the literacy and problem solving test domain both in terms of magnitude and in terms of statistical significance. For instance, the estimation results in specification (D) imply that a one standard deviation increase in literacy solving test score increases immigrants earnings by 9.2% vs. 8.0% for non - immigrants, ceteris paribus , while specification (E) of Table E.4 implies that the wage gain associated with a one standard deviation increase in problem solving test score is 9.2% for foreign - born workers vs. 7.6% for non - immigrants, ceteris paribus . Lastly, when all test scores are included in the regressi on (column (G) of Table E.4 nearly three times lower than those of immigrants, and the difference is statistically significant at the 5% level. It is also worth noting that the difference in the return to education between the two groups is statistically significant at the conventional levels across all specifications. Tables F. 1A to F. 11B in Appendix F present the analogous results estimated at a country - level for the countries in PIAAC with av erage and above - average share of immigrants (as compared to the total sample mean). 71 Since the country - level immigrant subsamples are small, we omit reporting the results for the specification jointly controlling for all test scores due to 71 The country results are reported in decreasing order of the fraction of immigrant population. 267 high collinearity between these scores. 72 Overall, the country - level results are in line with the implication from the pooled analysis immigrants tend to have lower returns to education but considerably higher returns to numeracy, literacy and problem so lving proficiency, although the differences are not always statistically significant (the latter is expected due to the small country samples). The immigrant - non - immigrant gap in returns to cognitive skills is particularly marked in Canada, Norway, Belgium , the UK, and Estonia. At the same time, however, it is worth noting that several of the PIAAC countries foreign - born workers have worse returns to both education and cognitive skills; such examples are France, Spain, the U.S., and Austria (although, in th e last two countries the returns to education of both groups are virtually the same, once cognitive skills have been accounted for). Taken as a whole, the estimation results of model (1) suggest that immigrants have lower returns to schooling than non - imm igrants, but enjoy higher returns to cognitive skills. This is line with the theory of statistical discrimination suggesting that employers are likely to view educational attainment as a less reliable signal of productivity for an immigrant than for a non - immigrant worker. In addition to this, it is interesting that in some model specifications the relative earnings gain associated with literacy test score is two - to - three times larger for immigrants than for non - immigrants. This might suggest that in the ab sence of other reliable weight on language proficiency than they do for a native - born worker. The later makes sense also because nearly 100% of the natives report the primary language of their country of residence as their mother tongue vs. only a quarter of the immigrants; hence, for a foreign - born worker to 72 The sample correlation between numeracy and literacy test score is 0.83; 0.79 between problem solving test and literacy test score , and 0.74 between numeracy and problem solving test score. 268 have acquired a certain level of proficiency in the language of the receiving country would reveal more about t heir productivity than it would for a native speaker. 3.5.1.2 Oaxaca - Blinder decomposition Table E.5 presents the results from the immigrant - non - immigrant wage gap estimation and the Oaxaca - Blinder decomposition of this gap based on model (1). In turn, model (1) includes the same set of controls as specifications (A) to (G) from the previous subsection, and adds a baseline specification, denoted as (0) . Specifications (0) , and (A) to (E) illustrate the results based on the entire PIAAC sample, while specifications (F) and (G) are only estimated on the sample of countries which administered the problem sol ving test score. Column (1) reports the estimated raw wage gap between the immigrant and native workers, and columns (2) and (3) report the Oaxaca - Blinder decomposition for each specification. In particular, Column (2) illustrates the difference in earning s between immigrants and natives due to differences in their observable characteristics (as already noted previously, we shall refer to this part as the differ The baseline specification (0) reports the results when no observable covariates are included in the log - wage regressions for both groups, other than country dummies. Column (1) states that the raw mean wage gap between the immigrant and native workers estimated on the total sample of 45,697 prime age ful l time workers in OECD and Russia is 0.065, and this gap is statistically significant at the 5% level. While the sign of the mean gap may appear surprising at 269 first, it is merely a reflection of the fact that, on average, countries with higher wages are al so countries with higher shares of immigrants. The later can also be illustrated by looking at the composition effect reported in Column (2): this effect is positive in sign, implying that on average, immigrants have better observables, i.e. they are more concentrated in countries with higher wages. Or, more intuitively, the potential earnings of the natives if they had the same country distribution as immigrants would have surpassed their actual earnings. The magnitude of the explained part of the gap is 0.25 meaning that immigrants enjoy a 25 percent earnings gain compared to native workers due differences the in country - distribution. At the same time, however, a large fraction of the mean wage gap remains unexplained negative 0.185 (Column (3)), imply ing that immigrants earn, on average, roughly 19 percent less than non - immigrant workers after accounting for country - level heterogeneity. Alternatively, a program evaluation interpretation would imply that the gap between the actual mean wages of immigran t workers and the potential mean wages those workers had they been awarded according to the wage structure of natives, is 19 percent. We proceed by examining the Oaxaca - Blinder decomposition results from the specifications that include both country indica tors, as well as various individual characteristics. Specification (A) depicts these results when including education and labour market experience as 73 Column (2) reports the wage gap that would have arisen due to differences in covariates; this part is still positive in sign but lower in magnitude compared to the baseline specification (0.193 vs. 0.25) the drop being a reflection of the fact that, on a verage, immigrants have lower educational attainment and working experience 73 The raw mean gap now appears significant only at the 10% level, likely due to the slight drop in sample size because of missing values for variables education and/or experience for 426 observati ons (1% of the sample). 270 compared to native workers. The part due to discrimination drops in magnitude, as well (Column (3)) suggesting better average returns to schooling and experience for immigrants, ye t remains negative and substantial, - 0.127. As expected, with the inclusion of individual covariates the relative importance of the composition effect in the total observed wage gap increases. Including controls for gender and sector of employment in speci fication (B) leaves the results essentially unchanged. Next, we turn to specifications (C) to (E), which include numeracy and literacy cognitive skill measures. Once numeracy test score is controlled in specification (C) shows that, the results look notic eably different; in particular, the composition effects drops further to 0.137 reflecting the fact that, on average, immigrant workers score lower on the numeracy test. Equally important, the unexplained part of the gap is now nearly twice lower in magnitu de compared to specification (B) just - 0.071, further lending support to the idea suggested in the previous section that immigrants enjoy higher returns to test scores, and numeracy skills in particular. As expected, the relative importance of the observ able characteristics in the total mean wage gap additionally increases the explained part of the gap is now nearly twice as important as the unexplained. The implications from controlling for literacy test score (specification (D)) are virtually the same . To complete the analysis specification (E) includes controls for both literacy and numeracy skills. This results in a further drop of the magnitude of the unexplained gap to negative 0.067, implying that immigrants enjoy better joint returns to both sets of cognitive skills, even though the discrimination component remains statistically significant at the conventional 5% level and large in magnitude. The remaining specifications (F) and (G) add problem solving test score in the vector of covariates for a subsample of countries excluding France, Italy and Spain, which did not 271 administer this test. For comparison purposes we also report the results from specification (B) estimated on this smaller set of countries. The estimated raw wage gap is 0.118, of whi ch 0.221 is explained and - 0.103 is unexplained (the relative importance of composition being roughly twice larger than that of discrimination). Controlling for the measure of problem solving skills in specification (F), results in the unexplained part of the gap decreasing in magnitude by nearly 40% to - 0.06, while the explained part correspondingly shifts to 0.178. Put differently, the results now imply a much higher relative importance of endowments than the part due to discrimination. Lastly, controllin g for all test scores jointly (specification (G)) further reinforces the idea that the composition effect plays a key role in the observed immigrant - non - immigrant gap roughly 17 percent, while discrimination plays a much smaller part only 5 percent. T o sum up, the results obtained by model (1) specified with the observable covariates typically included by the mainstream literature would have lead us to incorrectly conclude that the unexplained part of the mean immigrant - non - immigrant wage gap obtained from the entire PIAAC sample is roughly 13 percent. In contrast, including controls for numeracy and literacy skills reveals that the role of discrimination is nearly twice lower 6.7 percent; yet, the wage structure effects remains statistically signific ant at the conventional 5% level and considerable in magnitude, implying that the actual mean wages of immigrants are nearly 7 percent lower than what they would have been, had immigrants enjoyed the same returns as natives. Lastly, our analysis suggests t hat the key reason behind the observed immigrant - native gap in the analyzed countries is the fact that immigrants have, on average, worse workplace skills and competences the part of the wage gap due to endowments is roughly twice larger than the part du e to discrimination. 272 3.5.2 DiNardo - Fortin - Lemieux decomposition Figures E. 8 through E. 10 depict the estimated actual log - earnings kernel density for both groups of workers, and the counterfactual kernel density for non - immigrants. The latter comprises t he hypothetical earnings distribution of native workers that would have prevailed if those workers had the same observable characteristics distribution as immigrants, and enjoyed returns according to the native wage structure. Figure E.11 shows the estimat ed actual difference between the densities of the two groups plotted against the unexplained part of this difference, obtained when employing a different set of controls. Figure E.8 shows the actual log - wage densities for immigrant and natives (the solid b lue and dark red line, respectively), together with the corresponding 95% confidence intervals. As can be seen from here, the density for native workers appears somewhat skewed to the right, while that of immigrants is more symmetric. In addition to this, 1st to 20th percentile), implying a sizeable wage gap in favour of immigrant workers in this range. This is mirrored by a correspondingly lower share of native earners in the range between the 20th and 50th percentile. At the same time, however, natives are more densely concentrated in the above - mean range (65th - 90th percentile), although the raw gap between the two densiti es being relatively small. Lastly, there is a noticeably higher share of immigrants above the 90th percentile of the log - earnings range, implying a larger fraction of top - earners amongst immigrants. The solid red line in Figure E.11 depicts the difference in the actual densities between immigrant and non - immigrant workers and confirms the implication from Figure E.8 , namely: immigrants enjoy higher earnings a reflection of the fact they are more prevalent in countries with higher wages. 273 Next, Figure E.9 plots the actual wage density for immigrants (solid blue line) and the counterfactual densities for natives, together with the corresponding 95% confidence intervals (shown by the dashed lines). In turn, the counterfactual densities are obtained when using three different sets of propensity score weights: 1) only accounting for the differences in the country distribution between immigrants and natives (solid black line); 2) accounting for the differences in the country distribution and the distribution of observable covariates typically included in the log - earnings function regressions, as listed in specification B from the previous section (solid green line); 3) accounting for the differences in literacy and numeracy proficiency, in addition to the control s included in 2 (solid orange line). 74 Comparison between the actual density for immigrant workers and the counterfactual densities for native workers would reveal the contribution of each of these sets of observable covariates to the raw wage gap across the entire wage distribution. The diffe rence between these densities is depicted on Figure E.11 using the same colour coding as the one used on Figure E.9 ; in addition, Figure E.11 illustrates the gap between the actual immigrant and native log - wage densities (red line). We examine the counterf actual densities in turn. The black line on Figure E.9 displays the counterfactual density of log - wages that would have been observed if native workers had the same country distribution as immigrants but were paid according to their own wage structure. As can be seen from here, this line is shifted considerably to the right of the blue line for most of the earnings distribution range, and the 95% confidence intervals of the two densities show virtually no overlap. This is particularly 74 We do not plot the results for all specifications (0) to (G) as described in Table E. 5 from the previous section on the same graph for the purposes of better visibility; those are presented on separate graphs in Appendix F , Figure s F. 1 to F. 6. In addition, we opt not to include a control for problem solving test score due to the fact this score is not available f or the entire sample. Appendix F , Figures F. 7 to F. 1 5 presents the complete analysis on the set of countries that administered this test. 274 pronounced for the ab log - wage of roughly 3.7 and above, or the 95th percentile. The difference between the tw o densities is depicted by the black line on Figure E.11 , which confirms these findings; in addition, it provides a vivid illustration of the fact that a large fraction of the gap between the wage densities of immigrants and natives remains unexplained whe n only accounting for country of residence. Taken as whole, this comparison indicates that native workers would have enjoyed higher earnings than immigrants at nearly the entire range of the wage distribution if they had the advantage of being concentrated in higher - earning countries. Turning to the green line on Figure E.9 would have been if they had the same country, education, labour market experience, gender and employment sector distribution as immigr ant workers, but retained their own returns. Compared in the entire earnings range, and the 95% confidence intervals of the two counterfactual densities do disadvantages in terms of the labour market characteristics listed above play an important role in the distribution of the observed earnings gap. At the same time, however, t he distance between the green and the blue lines remains substantial and would imply presence of considerable discrimination virtually in the entire log - earnings range. One other important observation is that the distance between the black and the green li nes is roughly the same between the 5th and the 95th percentile suggesting that the included observable characteristics matter equally across most the earnings distribution. This interpretation is further reinforced by examining the green line of Figure E. 11 . 275 Lastly, the orange line on Figure E.9 distribution obtained after accounting for differences in the country distribution, and the distribution of all other observable covariates, including literacy a nd numeracy test scores. Several important observations can be made here. First, the orange line is much closer to the blue line than the green line is, and this holds over the entire range of the wage distribution, although the 95% confidence intervals sh ow some overlap below the 65 th percentile. This suggests that literacy and numeracy proficiency play an important role in the observed wage distribution in the entire earnings range. What is more, the density depicted by the orange line appears much like a leftward shift of the density depicted by the green line, suggesting that numeracy and literacy test scores matter practically equally throughout the whole distribution of log - earnings. It is also worth examining particular ranges of the wage distributio n in more detail. At the very bottom of the earnings distribution (log - wages or roughly below 2, corresponding to 1st to 15th percentile) the orange and blue lines nearly coincide and the 95% confidence interval of the suggesting that differences in observables almost entirely explain the wage gap between natives and immigrants in this range. This can also be seen from Figure E.11 , showing that the unexplained part of t he gap is essentially zero below value of log - earnings below roughly 2, corresponding to the 15 th percentile. A similar observation can be made for the top 5 percentile earners, as well. At the same time, however, for all earners in the range of roughly be tween the 15th and the 95th percentile the counterfactual density for natives is shifted to the right of the actual density for immigrants, and their confidence intervals show no overlap. This suggests that even if native workers had the observable charact eristics of immigrants, but were paid according 276 elaborate more on this, focusing on the 15th - 65th percentile range reveals that the non - immigrant workers would have had a lower concentration in this earnings range than immigrants, even if actual immigrant density and the counterfactual native density based of the fullest set of controls in earning percentiles 65th to 95th. These observations are of considerable importance as they reveal that differences in the observable covariates cannot fully account for the differences in the log - earning distributions between natives and immigrants, even after controlling for numeracy and literacy proficiency. This, in turn, implies presence of considerable labour market discrimination against the non - native workers especially in the above - mean earnings range. Yet, the magnitude of thi s discrimination is considerably lower than the magnitude implied based on the counterfactual density constructed without controlling for numeracy and literacy proficiency. Turning briefly to Figure E.10 , it is interesting to compare the actual log - wage d ensity of native workers (dark red line) and the counterfactual log - wage density of these workers obtained with the fullest set of controls (orange line). Perhaps the most interesting observation here is that for values of log - earnings of roughly above the 75th percentile, these lines nearly coincide and the 95% confidence interval of the counterfactual density includes the actual density. Taken together, these imply that for this earnings range, even if native workers had the characteristics of immigrants they would have still earned the same as their actual wages as long as they were log - earnings percentile, the explained component of the observed wage gap is very small and that the main part of the raw gap is due to discrimination. The same conclusion prevails after examining the orange line on Figure E.11 , depicting the difference between the actual wage 277 distribution for immigrant workers and the counterfa ctual distribution for native workers obtained after controlling for all observable covariates. In particular, it is evident that beyond values of log - wage of about 3.15, corresponding to the 75th percentile, the orange line overlaps with the red line, sug gesting that in this earnings range practically the entire raw earnings gap remains unexplained. Table E.6 allows examining the DFL results from a somewhat different perspective, namely, it reports the decomposition into an explained part and wage structur e effect for selected quanitles in the log - wage distribution for all specifications (0) to (E) as described in the previous section. Even though the implications from this representation of the results are largely the same as the ones that follow from the DFL graphical analysis, several interesting observations can be made. First, as a fuller set of controls is added, the contribution of the observable characteristics in the raw gap increases, overall, and this is particularly well - pronounced in the low pe rcentiles. For instance, specification (B) reports the DFL decomposition results obtained when accounting for differences in the country distribution, as well as education, labour market experience, gender and employment sector. Focusing on the 10th percen tile, the results from this specification indicate that the explained part of the wage gap for the low earners in PIAAC is 0.437, implying surpass their actual earnings by roughly 44 percent. The part of the 10th percentile gap due to discrimination is negative 0.150, meaning that a 15 percent wage gap between immigrants and natives in this quantile originates from differences in the returns to skills. In contras t, the DFL decomposition results from specification (E), adding controls for numeracy and literacy test scores, indicate an explained effect of 0.331 and only - 0.044 unexplained, leading to the 278 conclusion that for low earners composition plays considerably larger role in the observed gap than discrimination does. A similar conclusion prevails for all earners below the 50th percentile. Turning to the above - median earners, including literacy and numeracy test scores causes the unexplained part of the wage gap in each percentile to drop by roughly half, suggesting a noticeably lower role for discrimination than what is implied by specific ation (B). At the same time, however, the explained part of the gap drops correspondingly, as well (as expected since immigrants score lower on those tests), implying a lower role for the composition effect relative to the unexplained. To elaborate more on this, when looking at e.g. the 80th percentile of the earnings distribution, specification (E) suggests that immigrants earn 7.2 percent lower wages due to differences in returns to skills, while specification (B) would have attributed a nearly twice larg er share to discrimination 13.6 percent. Yet, the relative importance of composition is lower in the model with test scores, suggesting that labour market discrimination may be stronger for earners close to the right tail of the log - wage distribution. Ne vertheless, the magnitude of this discrimination appears nearly uniform across the entire range of the wage distribution, except for the bottom 10 percent and the top 10 percent of the workers. These are essentially the same observation that prevailed from the graphical analysis. 279 3.6 CONCLUDING REMARKS This study utilised a large cross - section of 21 countries from the 2011/2012 Programme for the International Assessment of Adult Competencies in order to evaluate the earnings gap between immigrant and native workers in these countries. Following the mainstream literature in the field, we employ a decomposition methodology, in particular a modified Mincer regression and Oaxaca - Blinder mean lo g - wage decomposition (Oaxaca (1973) and Blinder (1973)), as well as the decomposition technique developed by DiNardo, Fortin and Lemieux (1996). Consistent with the findings of several authors for the U.S. (e.g. Bratsberg and Terrell (2002), Betts and Lo fstrom (2000)), we find that immigrants have lower returns to education than native workers, yet higher returns to cognitive abilities, especially in the literacy domain. Moreover, we observe this same pattern in most countries in PIAAC, as well as in the pool of data. This is consistent with the main idea of the statistical discrimination theory that employers may view educational attainment as a less reliable productivity signal for immigrants, and that in the absence of other reliable productivity signal s, they place substantial weight on language proficiency for foreign - born workers. Next, a central finding of the Oaxaca - Blinder decomposition is that a log - wage regression specified without controls for cognitive skills would greatly overestimate the unex plained part of the mean immigrant - native wage gap, while including numeracy and literacy test scores reveals a much lower role of discrimination. Finally, the DiNardo - Fortin - Lemieux results show that numeracy and literacy test scores matter considerably a nd almost equally throughout the entire log - wage distribution, yet differences in numeracy and literacy proficiency between natives and immigrants cannot fully explain the observed earnings gap, except for the bottom and the top deciles. 280 The implications of these findings are twofold. First, in terms of methodology, our results suggests that measuring the true earnings gap between immigrant and native workers is not feasible without accounting for the differences in cognitive competences between the two gr oups, as this leads to a severe overestimation of the unexplained component of the wage gap. Secondly, the results of this study have important policy implications. In particular, the fact that immigrants have lower literacy and numeracy proficiency than native - born workers, combined with the finding that differences in these skills explain a considerable part of the earnings gap, calls for the development and implementation of education and qualification programmes for the immigrant population. Such progr ammes would increase the cognitive competences of the foreign - born workers , which in turn would facilitate the convergence of their earnings towards those of natives. At the same time, however, the existence of a significant earnings gap even after account ing for cognitive abilities implies that presence of labour market discrimination cannot be ruled out, which puts forward the need for stronger legislative provisions against discrimination based on immigration status. 281 APPENDICES 282 APPENDIX E MAIN TABLES AND FIGURES 283 Table E.1 : Descriptive statistics (total sample) Country Austria Belgium (Flanders ) Canada Czech Republic Denmark Estonia Finland France Hou rly wage (in USD, PPP) 21.931 22.145 27.811 8.892 25.267 10.056 20.047 15.820 (0.189) (0.197) (0.188) (0.124) (0.142) (0.130) (0.121) (0.100) Female 0.401 0.417 0.447 0.458 0.469 0.531 0.488 0.440 (0.010) (0.007) (0.005) (0.012) (0.008) (0.008) (0.008) (0.007) Immigrant 0.187 0.061 0.277 0.051 0.101 0.103 0.043 0.109 (0.01 0) (0.006) (0.007) (0.010) (0.003) (0.007) (0.005) (0.005) Age 25 - 34 0.310 0.330 0.316 0.339 0.278 0.349 0.308 0.322 (0.009) (0.008) (0.006) (0.011) (0.007) (0.009) (0.007) (0.007) Age 35 - 54 0.690 0.670 0.684 0.661 0.722 0.651 0.692 0.678 (0.008) (0.008) (0.006) (0.011) (0.007) (0.009) (0.007) (0.007) Education (years) 12.503 13.226 14.009 13.470 13.460 12.682 13.531 12.198 (0.049) (0.062) (0.042) (0.068) (0.044) (0.052) (0.048) (0.059) Experience (years) 20.075 18.304 19.572 17.826 20.639 17.392 16.955 18.016 (0.196) (0.146) (0.131) (0.242) (0.169) (0.168) (0.166) (0.157) Numeracy score (0 - 500) 282.236 294.355 275.880 280.420 291.143 281.001 298.422 267.384 (1.250) (1.176) (0.952) (1.390) (1.148) (0.880) (1.060) (1.040) Literacy score (0 - 500) 275.626 287.860 282.357 278.379 281.788 282.616 303.052 271.045 (1.153) (1.077) (0.910) (1.433) (1.027) (0.856) (0.987) (0.774) Problem solving score (0 - 500) 287.423 288.392 286.636 284.180 290.717 278.516 297.886 - (1.134) (1.081) (0.778) (1.687) (1.062) (0.956) (0.908) No. observations 1,690 1,736 9,280 1,622 2,375 2,221 2,106 2,272 Notes: 1) Means corrected for inverse probability weighted sampling; jack - knife standard errors based on replicate weights (80 replications). Unweighted number of observations reported. 2) France, Italy and Spain did not participate in the survey of p roblem solving skills in technology - rich environment. 284 Table E.1 Country Germany Ireland Italy Japan Korea Nether - lands Norway Poland Hou rly wage (in USD, PPP) 19.278 22.937 15.591 16.895 15.758 23.752 25.979 8.866 (0.178) (0.404) (0.234) (0.211) (0.191) (0.212) (0.190) (0.139) Female 0.372 0.445 0.369 0.359 0.390 0.304 0.441 0.455 (0.011) (0.012) (0.014) (0.009) (0.009) (0.010) (0.008) (0.009) Immigrant 0.146 0.236 0.095 0.001 0.016 0.130 0.143 0.001 (0.01 0) (0.013) (0.015) (0.001) (0.003) (0.008) (0.008) (0.001) Age 25 - 34 0.315 0.438 0.277 0.327 0.348 0.340 0.308 0.415 (0.010) (0.011) (0.013) (0.008) (0.008) (0.011) (0.008) (0.011) Age 35 - 54 0.685 0.562 0.723 0.672 0.652 0.660 0.692 0.585 (0.010) (0.011) (0.013) (0.008) (0.008) (0.011) (0.008) (0.011) Education (years) 13.815 16.054 11.489 13.687 13.816 14.145 14.918 13.673 (0.070) (0.069) (0.128) (0.046) (0.057) (0.066) (0.047) (0.075) Experience (years) 18.513 16.743 17.502 17.161 12.697 18.461 18.216 14.963 (0.208) (0.244) (0.269) (0.189) (0.169) (0.209) (0.143) (0.233) Numeracy score (0 - 500) 282.117 272.592 258.671 300.901 270.574 295.847 294.219 266.092 (1.566) (1.407) (1.759) (1.039) (0.866) (1.358) (1.248) (1.305) Literacy score (0 - 500) 277.274 279.979 257.664 307.503 278.290 297.875 291.693 272.342 (1.430) (1.418) (1.672) (0.777) (0.791) (1.261) (0.988) (1.210) Problem solving score (0 - 500) 286.392 285.190 - 304.649 285.945 297.112 294.446 273.782 (1.685) (1.429) (1.436) (1.253) (0.945) (1.076) (0.908) No. obs ervations 1,669 1,610 1,226 1,877 2,134 1,387 2,073 1,890 285 Table E.1 Country Russia Slovakia Spain UK USA Total OECD 75 Total Sample Hourly wage (in USD, PPP) 4.765 8.960 14.889 20.669 28.735 18.714 18.050 (0.110) (0.161) (0.179) (0.363) (0.524) (0.195) (0.192) Female 0.483 0.484 0.423 0.389 0.471 0.428 0.431 (0.012) (0.010) (0.011) (0.009) (0.010) (0.004) (0.004) Immigrant 0.036 0.014 0.120 0.151 0.174 0.108 0.104 (0.010) (0.003) (0.007) (0.011) (0.011) (0.004) (0.004) Age 25 - 34 0.367 0.338 0.324 0.346 0.350 0.333 0.336 (0.013) (0.009) (0.009) (0.007) (0.008) (0.003) (0.003) Age 35 - 54 0.633 0.662 0.676 0.654 0.650 0.667 0.664 (0.011) (0.008) (0.009) (0.008) (0.008) (0.003) (0.003) Education (years) 14.034 13.799 12.516 13.419 14.044 13.518 13.543 (0.080) (0.074) (0.070) (0.053) (0.062) (0.023) (0.023) Experience (years) 17.046 17.423 16.548 19.555 19.350 17.800 17.765 (0.368) (0.199) (0.208) (0.194) (0.241) (0.096) (0.091) Numeracy score (0 - 500) 274.352 286.337 260.488 277.705 263.070 279.973 279.706 (2.462) (1.077) (1.111) (1.500) (1.387) (0.534) (0.465) Literacy score (0 - 500) 277.505 281.494 264.737 285.191 276.610 281.669 281.471 (2.056) (1.011) (1.183) (1.408) (1.320) (0.505) (0.473) Problem solving score (0 - 500) 280.269 284.146 - 285.191 283.359 288.500 288.109 (4.218) (1.207) (1.408) (1.360) (0.608) (0.639) No. observations 929 1,671 1,695 2,557 1,677 44,768 45,697 75 Excluding the Russian Federation as it is not a member - state. 286 Table E.2 : Descriptive statistics (immigrant/non - immigrant subsamples) Characteristic Immigrant subsample Non - immigrant subsample Demographic Age below 35 25 - 34 0.369 (0.011) 0.332 (0.003) 35 - 54 0.631 (0.011) 0.668 (0.003) Female 0.427 (0.015) 0.431 (0.004) Education in years 13.425 (0.129) 13.562 (0 .025) Education (highest level of schooling) Less than high school 0.190 (0.015) 0.099 (0.002) High school 0.339 (0.014) 0.417 (0.004) Above high school 0.471 (0.016) 0.484 (0.004) Married /partnered 0.818 (0.013) 0.803 (0.005) Has any children 0.707 (0.014) 0.683 (0.006) Immigrant 1.000 0.000 Age at immigration 23.778 (0.361) NA Native language speaker 0.239 (0.019) 0.979 (0.002) Employment history Actual working experience (years of paid work during lifetime) 15.492 (0.330 18.008 (0.084) Sector of employment (current job) Private sector 0.793 (0.015) 0.694 (0.007) Note: Means corrected for inverse probability weighted sampling; jack - knife standard errors based on replicate weights (80 replications). Unweighted number of observations reported. 287 Table E.2 Public sector 0.184 (0.012) 0.283 (0.007) Non - profit organization 0.022 (0.008) 0.023 (0.002) Occupation (current job) Skilled 0.369 (0.017) 0.472 (0.004) Semi - skilled white - collar 0.245 (0.017) 0.248 (0.004) Semi - skilled blue - collar 0.251 (0.015) 0.223 (0.004) Elementary 0.135 (0.013) 0.057 (0.003) Years of formal education required to obtain current job 12.293 (0.147) 13.147 (0.032) On - the - job training 0.376 (0.018) 0.467 (0.005) Hourly wage (in USD, PPD adjusted) 18.622 (0.717) 18.047 (0.207) Skills and competences Skill test score Numeracy skills score (0 - 500) 249.275 (2.774) 283.247 (0.531) Literacy skills score (0 - 500) 252.652 (2.229) 284.876 (0.521) Problem solving in technology - rich environment (0 - 500) 271.178 (2.313) 289.9012 (0.648) Skill use at work (current job) Index of use of information and communications technology (ICT) skills at work 2.180 (0.054) 2.133 (0.014) Index of use of numeracy skills at work 2.096 (0.035) 2.101 (0.011) Index of use of reading skills at work 1.908 (0.047) 2.105 (0.010) Index of use of writing skills at work 2.103 (0.010) 2.190 (0.011) Index of use of task discretion at work 1.728 (0.039) 1.899 (0.008) Index of readiness to learn at work 2.120 (0.042) 2.051 (0.012) 288 Table E.2 Index of learning at work 2.067 (0.039) 2.006 (0.013) Health self evaluation Excellent 0.189 (0.016) 0.167 (0.005) Very good 0.345 (0.017) 0.333 (0.006) Good 0.344 (0.021) 0.358 (0.006) Fair 0.358 (0.006) 0.126 (0.003) Bad 0.126 (0.003) 0.015 (0.003) 289 Figure E.1 : Mean wage conditional on education 290 Figure E.2 : Mean wage conditional on labour market experience 291 Figure E.3 : Mean wage conditional on sector of employment 292 Figure E.4 : Mean wage conditional on skill - based occupation al category 293 Figure E.5 : Mean wage conditional on numeracy test score 294 Figure E.6 : Mean wage conditional on lit eracy test score 295 Figure E.7 : Mean wage conditional on problem solving test score 296 Table E.3 : OLS results (pooled total sample) Controls (A) (B) (C) (D) Immigrant Non - immigrant Immigrant Non - immigrant Immigrant Non - immigrant Immigrant Non - immigrant Education (years) 0.050*** 0.065*** 0.050*** 0.069*** 0.035*** 0.056*** 0.035*** 0.058*** (0.004) (0.001) (0.004 ) (0.001) (0.003) (0.001) (0.002) (0.001) p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 Experience (years) 0.029*** 0.025*** 0.026*** 0.025*** 0.023*** 0.024*** 0.024*** 0.025*** (0.004) (0.012) (0.004) (0.001) (0.003) (0.001) (0.003) (0.001) Experience squared - 0.001*** - 0.000*** - 0.000* ** - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - - - 0.161*** - 0.185*** - 0.141*** - 0.163*** - 0.158*** - 0.180*** (0.044) (0.005) (0.013) (0.005) (0.015) (0.005) Notes: 1) All specif ications include country dummies; in addition, specifications (5) to (7) control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 3) Jack - knife standard err ors based on replicate weights reported in parentheses (80 replications). 4) Japan dropped from the pooled sample since the low number of immigrant - observations in the country leads to insufficient degrees of freedom to estimate the immigrant model with co untry effects. 297 Table E.3 Numeracy score (standardized) - - - - 0.092*** 0.089*** - - (0.011) (0.005) p - value for H 0 : = 0.9918 Literacy score (standardized) - - - - - - 0.092*** 0.080*** (0.008) (0.003) p - value for H 0 : = 0.1246 Problem solving score (standardized) - - - - - - - - Intercept 1.564*** 2.009*** 1.564*** 2.005*** 1.470*** 2.229*** 1.452*** 2.186*** (0.072) (0.025) (0.118) (0.030) (0.137) (0.028) (0.130) (0.028 ) No. observations 5,127 38,226 5,127 38,226 5,126 38,225 5,126 38,225 298 Table E.4 : OLS results (pooled restricted sample) Controls (A) (E) (F) (G) Immigrant Non - immigrant Immigrant Non - immigrant Immigrant Non - immigrant Immigrant Non - immigrant Education (years) 0.050*** 0.065*** 0.044*** 0.059*** 0.034*** 0.056*** 0.038*** 0.055*** (0.003) (0.001) (0.006) (0.001) (0.003) (0.001) (0.006) (0.002) p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 Experience (years) 0.029*** 0.025*** 0.028*** 0.028*** 0.023*** 0.024*** 0.026*** 0.027*** (0.004) (0.012) (0.007) (0.002) (0.005) (0.011) (0.007) (0.002) Experience squared - 0.001*** - 0.000*** - 0.001*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** - 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - - - 0.138*** - 0.167*** - 0.148*** - 0.167*** - 0.134*** - 0.160*** (0.029) (0.011) (0.036) (0.010) (0.024) (0.011) Notes: 1) All specifications include country dummies; in addition, specifications (5) to (7) control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 4) Japan dropped from the pooled sample sin ce the low number of immigrant - observations in the country leads to insufficient degrees of freedom to estimate the immigrant model with country effects. 5) Specifications (A) to (G) estimated on the restricted sample of countries which administered the p roblem solving test. 299 Table E.4 Numeracy score (standardized) - - - - 0.052*** 0.068*** 0.046*** (0.042) 0.042*** (0.008) (0.030) (0.005) p - value for H 0 : = 0.1469 p - value for H 0 : = 0.9725 Literacy score (standardized) - - - - 0.048*** (0.022) 0.027*** (0.005) 0.074*** 0.026** (0.037) (0.011) p - value for H 0 : = 0.0846 p - value for H 0 : = 0.0111 Problem solving score (standardized) - - 0.092*** (0.013) 0.076*** (0.006) - - 0.010 (0.029) 0.033*** (0.008) p - value for H 0 : = 0.0710 p - value for H 0 : = 0.1322 300 Table E.5 : Oaxaca - Blinder decomposition Controls (1) (2) (3) (4) Raw mean gap Explained (Composition effect) Unexplained (Wage structure effect) No. observations (0) Country dummies only 0.065** 0.250*** - 0.185*** 45,697 (0.033) (0.021) (0.022) (A) Education, and Experience (quadratic) 0.065* 0.193*** - 0.127*** 45,271 (0.034) (0.024) (0.022) (B) A + Gender, and Sector of employment 0.065* 0.194*** - 0.129*** 45,226 (0.034) (0.024) (0.022) (C) B + Numeracy test score 0.065* 0.137*** - 0.071*** 45,224 (0.034) (0.026) (0.021) (D) B + Literacy test score 0.065* 0.139*** - 0.073*** 45,224 (0.034) (0.024) 0.022) (E) B + Numeracy and Literacy test scores 0.065* 0.132*** - 0.067*** 45,224 (0.034) (0.025) (0.026) (B) A + Gender, and Sector of employment (restricted sample) 0.118** 0.217*** - 0.099*** 33,889 (0.045) (0.0 34 ) (0.0 26 ) (F) B + Problem solving test score 0.118** 0.178*** - 0.060** 33,889 (0.045) (0.034) (0.023) (G) B + Numeracy, Literacy, Problem solving test scores 0.118** 0.169*** - 0.051** 33,889 (0.045) (0.0 3 5) (0.0 22 ) Note: 1) Reference group: native workers. 2) All specifications (A) to (H) include country dummies. 3) Final sample weight used to correct for inverse probability weighted sampling. Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 301 Figure E.8 : Estimated densities of log - earnings (actual) Notes: 1) Final sample weight used to correct for inverse probability weighted sampling; equal w eight is placed on each country. 2) 9 5% pointwise confidence intervals obtained via bootstrapping. 302 Figure E.9 : Estimated densities of log - earnings (actual immigrants vs. counterfactual natives) Notes: 1) Specification (0) includes country dummies only. Specification (B) controls for education, labour market experience (in quadratics), gender and sector of employment. Specification (E) adds numeracy test and literacy test scores. Both specifications (B) and (E) include country dummies. 2) Final sample weight used to correct for inverse probabilit y weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 303 Figure E.10 : Estimated densities of log - earnings (actual natives vs. counterfactual natives) Notes: 1) Specification (0) includes country dummies only. Specification (B) controls for education, labour market experience (in quadratics), gender and sector of employment. Specification (E) adds numeracy test and literacy test scores. Both specifications (B) and (E) include c ountry dummies. 2) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 304 Figure E.11 : Estimated log - earnings gap (actual vs. unexplained) Notes: 1) Specification (0) includes country dummies only. Specification (B) controls for education, labour market experience (in quadratics), gender and sector of employment. Specification (E) adds numeracy test and literacy test scores. Both specifications (B) and (E) include country dummies. 2) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 305 Table E.6 : DFL decomposition results for selected quantiles Quantile Raw Gap (0) Country dummies only (A) Education, and Experience (quadratic) (B) A + Gender, and Sector of employment (1) Explained (Composition effect) (2) Unexplained (Wage structure effect) (1) Explained (Composition effect) (2) Unexplained (Wage structure effect) (1) Explained (Composition effect) (2) Unexplained (Wage structure effect) 10th 0.287 0.488 - 0.201 0.437 - 0.150 0.437 - 0.150 20th 0.193 0.389 - 0.196 0.328 - 0.135 0.327 - 0.134 30th 0.067 0.305 - 0.238 0.237 - 0.170 0.236 - 0.168 40th 0.025 0.240 - 0.215 0.177 - 0.152 0.177 - 0.152 50th - 0.009 0.201 - 0.210 0.152 - 0.161 0.152 - 0.161 60th - 0.041 0.179 - 0.220 0.112 - 0.153 0.109 - 0.150 70th - 0.046 0.161 - 0.207 0.089 - 0.136 0.089 - 0.135 80th - 0.043 0.140 - 0.183 0.093 - 0.136 0.093 - 0.136 90th 0.040 0.098 - 0.058 0.054 - 0.014 0.059 - 0.019 mean 0.065 0.250 - 0.185 0.193 - 0.127 0.194 - 0.129 Notes: 1) Reference group: native workers. 2) All specifications (C) to (E) include country dummies. 3) Final sample weight used to correct for inverse probability we ighted sampling. 306 Table E.6 Quantile Raw Gap (C) B + Numeracy test score (D) B + Literacy test score (E) B + Numeracy and Literacy test scores (1) Explained (Composition effect) (2) Unexplained (Wage structure effect) (1) Explained (Composition effect) (2) Unexplained (Wage structure effect) (1) Explained (Composition effect) (2) Unexplained (Wage structure effect) 10th 0.287 0.345 - 0.058 0.332 - 0.044 0.331 - 0.044 20th 0.193 0.267 - 0.074 0.258 - 0.065 0.253 - 0.060 30th 0.067 0.187 - 0.120 0.185 - 0.118 0.176 - 0.108 40th 0.025 0.123 - 0.099 0.129 - 0.104 0.117 - 0.092 50th - 0.009 0.083 - 0.092 0.090 - 0.099 0.082 - 0.091 60th - 0.041 0.043 - 0.084 0.047 - 0.088 0.036 - 0.077 70th - 0.046 0.033 - 0.080 0.036 - 0.083 0.025 - 0.071 80th - 0.043 0.040 - 0.083 0.040 - 0.083 0.029 - 0.072 90th 0.040 0.019 0.021 0.019 0.021 0.019 0.021 mean 0.065 0.137 - 0.071 0.139 - 0.073 0.132 - 0.067 307 APPENDIX F SUPPLEMENTARY TABLES AND FIGURES 308 Table F .1A : Country - level OLS results , test scores excluded ( Canada ) Controls Country sample: Canada (share of immigrants 27.7%) (1) (2) I N I N Education (years) 0.055*** 0.066*** 0.048*** 0.065*** (0.005) (0.003) (0.005) (0.003) p - value for H 0 : = 0.0062 p - value for H 0 : = 0.0023 Experience (years) 0.032*** 0.022*** 0.027*** 0.022*** (0.005) (0.003) (0.005) (0.003) Experience squared - 0.001*** - 0.000*** - 0.000*** - 0.000*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.161*** - 0.157*** (0.024) (0.013) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 2.026*** (0.082) 2.090*** 2.239*** 2.063*** (0.082) (0.052) (0.112) (0.073) No. observations 1,852 7,392 1,850 7,381 R - squared 0.19 0.20 0.26 0.26 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 309 Table F .1B : Country - level OLS results , test scores included ( Canada ) Controls Country sample: Canada (share of immigrants 27.7%) (3) (4) (5) I N I N I N Education (years) 0.026*** 0.049*** 0.026*** 0.050*** 0.033*** 0.054*** (0.005) (0.003) (0.006) (0.003) (0.007) (0.003) p - value for H 0 : = 0.0001 p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0030 Experience (years) 0.023*** 0.022*** 0.022*** 0.023*** 0.023*** 0.025*** (0.005) (0.003) (0.005) (0.003) (0.006) (0.004) Experience squared - 0.000** - 0.000*** - 0.000** - 0.000*** - 0.000* - 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0.117*** - 0.133*** - 0.144*** - 0.151*** - 0.124*** - 0.153*** (0.023) (0.013) (0.022) (0.014) (0.027) (0.015) Numeracy score (standardized) 0.122*** 0.075*** - - - - (0.010) (0.008) p - value for H 0 : = 0.0005 Literacy score (standardized) - - 0.120*** 0.065*** - - (0.011) (0.008) p - value for H 0 : = 0.0000 Problem solving score (standardized) - - - - 0.098*** 0.053*** (0.013) (0.007) p - value for H 0 : = 0.0017 Intercept 2.599*** 2.260*** 2.591*** 2.227*** 2.455*** 2.172* ** (0.114) (0.075) (0.115) (0.074) (0.123) (0.078 ) No. observations 1,850 7,381 1,850 7,381 1,520 6,434 R - squared 0.35 0.29 0.34 0.28 0.30 0.27 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 310 Table F .2A : Country - level OLS results , test scores excluded ( Ireland ) Controls Country sample: Ireland (share of immigrants 23.6%) (1) (2) I N I N Education (years) 0.081** 0.083*** 0.077*** 0.080*** (0.010) (0.006) (0.010) (0.006) p - value for H 0 : = 0.7493 p - value for H 0 : = 0.7576 Experience (years) 0.059*** 0.043*** 0.051*** 0.045*** (0.009) (0.008) (0.009 (0.003) Experience squared - 0.001*** - 0.001*** - 0.001*** - 0.001*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.049*** - 0.070*** (0.069) (0.025) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 1.070*** 1.215*** 1.410*** 1.322*** (0.1631) (0.131) (0.167) (0.142) No. observations 347 1,263 347 1,263 R - squared 0.30 0.25 0.34 0.26 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 311 Table F .2B : Country - level OLS results , test scores included ( Ireland ) Controls Country sample: Ireland (share of immigrants 23.6%) (3) (4) (5) I N I N I N Education (years) 0.063** 0.058*** 0.064** 0.061*** 0.063** 0.070*** (0.010) (0.006) (0.009) (0.006) (0.013) (0.008) p - value for H 0 : = 0.6301 p - value for H 0 : = 0.8496 p - value for H 0 : = 0.6108 Experience (years) 0.042*** 0.043*** 0.040 0.043*** 0.057*** 0.049*** (0.009) (0.007) (0.010) (0.002) (0.010) (0.008) Experience squared - 0.001**** - 0.001*** - 0.001 - 0.001*** - 0.001*** - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0. 045*** - 0.029 - 0.063*** - 0.051** - 0.026 - 0.067** (0.036) (0.024) (0.038) (0.024) (0.040) (0.026) Numeracy s core (standardized) 0.098** 0.125*** - - - - (0.031) (0.017) p - value for H 0 : = 0.2851 Literacy score (standardized) - - 0.097*** 0.107*** - - (0.021) (0.018) p - value for H 0 : = 0.7159 Problem solving score (standardized) - - - - 0.105*** 0.081*** (0.027) (0.015) p - value for H 0 : = 0.4635 Intercept 1.706*** 1.665*** 1.768*** 1.580*** 1.580*** 1.410*** (0.172) (0.148) (0.178) (0.138) (0.228) (0.181 No. observations 347 1,263 347 1,263 253 1,074 R - squared 0.39 0.31 0.38 0.30 0.37 0.29 No tes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 312 Table F .3A : Country - level OLS results , test scores excluded ( Austria ) Controls Country sample: Austria (share of immigrants 18.7%) (1) (2) I N I N Education (years) 0.065*** 0.070*** 0.063*** 0.069*** (0.006) (0.003) (0.007) (0.003) p - value for H 0 : = 0.3716 p - value for H 0 : = 0.4473 Experience (years) 0.026*** 0.023*** 0.021** 0.022*** (0.009) (0.004) (0.009) (0.004) Experience squared - 0.001** - 0.000*** 0.000 - 0.000*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.124*** - 0.110*** (0.045) (0.017) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 1.811*** 1.870*** 1.965*** 1.898*** (0.105) (0.050) (0.157) (0.083) No. observations 259 1,431 258 1,427 R - squared 0.31 0.24 0.34 0.26 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 313 Table F .3B : Country - level OLS results , test scores included ( Austria ) Controls Country sample: Austria (share of immigrants 18.7%) (3) (4) (5) I N I N I N Education (years) 0.047*** 0.054*** 0.045*** 0.055*** 0.054*** 0.056*** (0.007) (0.003) (0.007) (0.003) (0.012) (0.004) p - value for H 0 : = 0.4020 p - value for H 0 : = 0.2248 p - value for H 0 : = 0.8051 Experience (years) 0.020** 0.022*** 0.024** 0.023*** 0.028* 0.023*** (0.009) (0.004) (0.009) (0.003) (0.014) (0.004) Experience squared 0.000 - 0.000*** - 0.000* - 0.000*** - 0.001 - 0.000** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0.125*** - 0.088*** - 0.142*** - 0.105*** - 0.078 - 0.092*** (0.039) (0.017) (0.041) (0.017) (0.059) (0.019) Numeracy score (standardized) 0.080*** 0.103*** - - - - (0.012) (0.009) p - value for H 0 : = 0.2516 Literacy score (standardized) - - 0.089*** 0.106*** - - (0.017) (0.009) p - value for H 0 : = 0.3934 Problem solving score (standardized) - - - - 0.108*** 0.085*** (0.032) (0.011) p - value for H 0 : = 0.4374 Intercept 2.236*** 2.033*** 2.228*** 2.019*** 2.077*** 2.030*** (0.152) (0.078) (0.159) (0.081) (0.272) (0.085) No. observations 258 1,427 258 1,427 168 1,229 R - squared 0.39 0.32 0.39 0.32 0.37 0.30 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 314 Table F .4A : Country - level OLS results, test scores excluded ( United Sta tes ) Controls Country sample: United States (share of immigrants 17.4%) (1) (2) I N I N Education (years) 0.083*** 0.103*** 0.089*** 0.109*** (0.006) (0.007) (0.006) (0.007) p - value for H 0 : = 0.0587 p - value for H 0 : = 0.0340 Experience (years) 0.037*** 0.044*** 0.035*** 0.045*** (0.011) (0.007) (0.011) (0.007) Experience squared - 0.001*** - 0.001*** - 0.001*** - 0.001*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.233** - 0.192*** (0.094) (0.028) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 1.591*** 1.273*** 1.595*** 1.267*** (0.111) (0.105) (0.234) (0.126) No. observations 227 1,258 227 1,258 R - squared 0.38 0.28 0.43 0.31 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 315 Table F .4B : Country - level OLS results, test scores included (United States) Controls Country sample: United States (share of immigrants 17.4%) (3) (4) (5) I N I N I N Education (years) 0.081*** 0.082*** 0.076*** 0.085*** 0.078*** 0.083*** (0.010) (0.007) (0.008) (0.007) (0.015) (0.007) p - value for H 0 : = 0.9415 p - value for H 0 : = 0.4651 p - value for H 0 : = 0.7303 Experience (years) 0.034*** 0.044*** 0.034*** 0.045*** 0.024 0.041*** (0.012) (0.007) (0.011) (0.006) (0.017) (0.006) Experience squared - 0.001*** - 0.001*** - 0.001*** - 0.001*** 0.000 - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0.221** - 0.146*** - 0.231** - 0.177*** - 0.310*** - 0.183*** (0.085) (0.029) (0.093) (0.028) (0.062) (0.028) Numeracy score (standardized) 0.033 0.111*** - - - - (0.045) (0.020) p - value for H 0 : = 0.1370 Literacy score (standardized) - - 0.062* 0.113*** - - (0.036) (0.019) p - value for H 0 : = 0.2899 Problem solving score (standardized) - - - - 0.087** 0.118*** (0.044) (0.017) p - value for H 0 : = 0.5177 Intercept 1.751*** 1.640*** 1.868*** 1.579*** 1.970*** 1.615*** (0.368) ( 0.126) (0.321) (0.118) (0.391) (0.125) No. observations 227 1,258 227 1,258 151 1,163 R - squared 0.43 0.34 0.44 0.34 0.36 0.35 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 316 Table F .5A : Country - level OLS results, test scores excluded ( United Kin gdom ) Controls Country sample: United Kingdom (share of immigrants 15.1%) (1) (2) I N I N Education (years) 0.084*** 0.082*** 0.085*** 0.084*** (0.015) (0.005) (0.015) (0.005) p - value for H 0 : = 0.9607 p - value for H 0 : = 0.9813 Experience (years) 0.047*** 0.034*** 0.044*** 0.036*** (0.012) (0.006) (0.013) (0.006) Experience squared - 0.001*** - 0.001*** - 0.001*** - 0.001*** 0.000 0.000 0.000 0.000 Female - - - 0.104 - 0.149*** (0.063) (0.020) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 1.265*** 1.453*** 1.187*** 1.420*** (0.229) (0.081) (0.264) (0.116) No. observations 284 2,153 284 2,150 R - squared 0.20 0.17 0.21 0.20 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 317 Table F .5B : Country - level OLS results, test scores included ( United Kingdom ) Controls Country sample: United Kingdom (share of immigrants 15.1%) (3) (4) (5) I N I N I N Education (years) 0.053*** 0.064*** 0.056*** 0.065*** 0.062*** 0.062*** (0.014) (0.005) (0.014) (0.005) (0.017) (0.006) p - value for H 0 : = 0.4988 p - value for H 0 : = 0.5685 p - value for H 0 : = 0.9974 Experience (years) 0.038*** 0.032*** 0.033*** 0.033*** 0.048*** 0.035*** (0.010) (0.006) (0.011) (0.006) (0.014) (0.006) Experience squared - 0.001*** - 0.001*** - 0.001** - 0.001*** - 0.001*** - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0.012 - 0.106*** - 0.089 - 0.137*** - 0.025 - 0.107*** (0.055) (0.020) (0.063) (0.019) (0.072) (0.019) Numeracy score (standardized) 0.232*** 0.139*** - - - - (0.028) (0.014) p - value for H 0 : = 0.0046 Literacy score (standardized) - - 0.187*** 0.140*** - - (0.029) (0.013) p - value for H 0 : = 0.1648 Problem solving score (standardized) - - - - 0.177*** 0.142*** (0.035) (0.017) p - value for H 0 : = 0.3343 Intercept 1.611*** 1.682*** 1.639*** 1.656*** 1.549*** 1.677*** (0.232) ( 0.121) (0.232) (0.120) (0.287) (0.121) No. observations 284 2,150 284 2,150 249 2,030 R - squared 0.40 0.27 0.35 0.27 0.32 0.27 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 318 Table F .6A : Country - level OLS results, test scores excluded ( Germany ) Controls Country sample: Germany (share of immigrants 14.6%) (1) (2) I N I N Education (years) 0.038*** 0.057*** 0.039*** 0.057*** (0.007) (0.004) (0.007) (0.004) p - value for H 0 : = 0.0111 p - value for H 0 : = 0.0159 Experience (years) 0.020 0.022*** 0.018 0.022*** (0.014) (0.004) (0.015) (0.004) Experience squared 0.000 - 0.000*** 0.000 - 0.000*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.149*** - 0.109*** (0.054) (0.016) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 2.046*** 1.867*** 2.096*** 1.780*** (0.153) (0.071) (0.313) (0.091) No. observations 190 1,459 190 1,459 R - squared 0.23 0.24 0.26 0.27 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 319 Table F .6B : Country - level OLS results, test scores included ( Germany ) Controls Country sample: Germany (share of immigrants 14.6%) (3) (4) (5) I N I N I N Education (years) 0.019** 0.038*** 0.021** 0.041*** 0.039*** 0.046*** (0.009) (0.004) (0.009) (0.004) (0.011) (0.003) p - value for H 0 : = 0.0280 p - value for H 0 : = 0.0173 p - value for H 0 : = 0.4727 Experience (years) 0.017 0.021*** 0.019 0.021*** 0.030* 0.024*** (0.014) (0.004) (0.015) (0.004) (0.016) (0.004) Experience squared 0.000 - 0.000*** 0.000 - 0.000*** 0.000 - 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0.106* - 0.087*** - 0.122** - 0.109*** - 0.147** - 0.110*** (0.057) (0.016) (0.057) (0.016) (0.064) (0.016) Numeracy score (standardized) 0.112*** 0.104*** - - - - (0.025) (0.011) p - value for H 0 : = 0.7567 Literacy score (standardized) - - 0.114*** 0.093*** - - (0.027) (0.011) p - value for H 0 : = 0.4585 Problem solving score (standardized) - - - - 0.053* 0.077*** (0.028) (0.010) p - value for H 0 : = 0.3871 Intercept 2.374*** 2.025*** 2.321*** 2.004*** 2.025*** 1.908*** (0.255) (0.085) (0.290) (0.092) (0.287) (0.076) No. observations 190 1,459 190 1,459 152 1,335 R - squared 0.34 0.32 0.34 0.31 0.36 0.31 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 320 Table F .7A : Country - level OLS results, test scores excluded ( Norway ) Co ntrols Country sample: Norway (share of immigrants 14.3%) (1) (2) I N I N Education (years) 0.046*** 0.052*** 0.048*** 0.065*** (0.007) (0.003) (0.007) (0.003) p - value for H 0 : = 0.0314 p - value for H 0 : = 0.0172 Experience (years) 0.023*** 0.029*** 0.022*** 0.028*** (0.008) (0.003) (0.008) (0.003) Experience squared - 0.000* - 0.001*** - 0.000* - 0.001*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.062* - 0.145*** (0.036) (0.013) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 2.139*** 2.146*** 2.252*** 1.907*** (0.124) (0.057) (0.181) (0.074) No. observations 275 1,796 275 1,796 R - squared 0.25 0.17 0.27 0.29 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 321 Table F .7B : Country - level OLS results, test scores included ( Norway ) Controls Country sample: Norway (share of immigrants 14.3%) (3) (4) (5) I N I N I N Education (years) 0.035*** 0.055*** 0.037*** 0.058*** 0.042*** 0.056*** (0.007) (0.003) (0.007) (0.003) (0.009) (0.003) p - value for H 0 : = 0.0102 p - value for H 0 : = 0.0052 p - value for H 0 : = 0.1107 Experience (years) 0.017** 0.027*** 0.019*** 0.028*** 0.019** 0.030*** (0.007) (0.003) (0.007) (0.003) (0.008) (0.003) Experience squared 0.000 - 0.001*** 0.000 - 0.001*** 0.000 - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0.054 - 0.127*** - 0.073** - 0.140*** - 0.043 - 0.138*** (0.035) (0.013) (0.036) (0.013) (0.039) (0.013) Numeracy score (standardized) 0.073*** 0.057*** - - - - (0.013) (0.008) p - value for H 0 : = 0.2906 Literacy score (standardized) - - 0.072*** 0.049*** - - (0.013) (0.009) p - value for H 0 : = 0.1440 Problem solving score (standardized) - - - - 0.080*** 0.056*** (0.018) (0.009) p - value for H 0 : = 0.2168 Intercept 2.525*** 2.021*** 2.475*** 1.965*** 2.395*** 1.988*** (0.174) (0.074) (0.164) (0.075) (0.209) (0.077) No. observations 275 1,796 275 1,796 210 1,717 R - squared 0.35 0.31 0.34 0.31 0.36 0.31 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 322 Table F .8A : Country - level OLS results, test scores excluded ( Netherlands ) Controls Country sample: Netherlands (share of immigrants 13.0%) (1) (2) I N I N Education (years) 0.047*** 0.085*** 0.044*** 0.086*** (0.008) (0.004) (0.007) (0.004) p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 Experience (years) 0.025** 0.039*** 0.024*** 0.038*** (0.010) (0.004) (0.009) (0.004) Experience squared 0.000 - 0.001*** 0.000 - 0.001*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.128** - 0.062*** (0.055) (0.018) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 1.991*** 1.454*** 2.263*** 1.453*** (0.133) (0.081) (0.162) (0.085) No. observations 124 1,263 123 1,261 R - squared 0.25 0.28 0.30 0.29 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 323 Table F .8B : Country - level OLS results, test scores included ( Netherlands ) Controls Country sample: Netherlands (share of immigrants 13.0%) (3) (4) (5) I N I N I N Education (years) 0.025*** 0.073*** 0.026*** 0.072*** 0.030*** 0.074*** (0.009) (0.005) (0.009) (0.005) (0.009) (0.005) p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0002 Experience (years) 0.030*** 0.037*** 0.027*** 0.036*** 0.036*** 0.037*** (0.009) (0.004) (0.009) (0.004) (0.010) (0.004) Experience squared - 0.000** - 0.001*** - 0.000* - 0.001*** - 0.001** - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0.085 - 0.042** - 0.118** - 0.056*** - 0.032 - 0.059*** (0.056) (0.018) (0.058) (0.018) (0.059) (0.019) Numeracy score (standardized) 0.116*** 0.076*** - - - - (0.027) (0.013) p - value for H 0 : = 0.1820 Literacy score (standardized) - - 0.100*** 0.083*** - - (0.026) (0.012) p - value for H 0 : = 0.5642 Problem solving score (standardized) - - - - 0.098*** 0.071*** (0.026) (0.012) p - value for H 0 : = 0.3227 Intercept 2.506*** 1.608*** 2.498*** 1.621*** 2.267*** 1.596*** (0.178) ( 0.087) (0.183) (0.089) (0.210) (0.094) No. observations 123 1,261 123 1,261 100 1,210 R - squared 0.39 0.31 0.38 0.32 0.36 0.31 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 324 Table F .9A : Country - level OLS results, test scores excluded ( Spain ) Con trols Country sample: Spain (share of immigrants 12.0%) (1) (2) I N I N Education (years) 0.027** 0.076*** 0.030** 0.072*** (0.012) (0.002) (0.012) (0.003) p - value for H 0 : = 0.0002 p - value for H 0 : = 0.0006 Experience (years) 0.023* 0.031*** 0.018 0.029*** (0.011) (0.005) (0.011) (0.004) Experience squared 0.000 - 0.000*** 0.000 - 0.000*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.220*** - 0.169*** (0.065) (0.019) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 1.725*** 1.310*** 1.702*** 1.370*** (0.150) (0.050) (0.457) (0.093) No. observations 200 1,494 200 1,494 R - squared 0.11 0.34 0.18 0.39 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 325 Table F .9B : Country - level OLS results, test scores included ( Spain ) Controls Country sample: Spain (share of immigrants 12.0%) (3) (4) (5) I N I N I N Education (years) 0.024* 0.063*** 0.027** 0.064*** NA NA (0.014) (0.003) (0.013) (0.003) p - value for H 0 : = 0.0043 p - value for H 0 : = 0.0051 Experience (years) 0.018* 0.027*** 0.018 0.028*** (0.010) (0.004) (0.011) (0.004) Experience squared 0.000 - 0.000*** 0.000 - 0.000*** (0.000) (0.000) (0.000) (0.000) Female - 0.207*** - 0.142*** - 0.216*** - 0.154*** (0.072) (0.019) (0.069) (0.020) Numeracy score (standardized) 0.049 0.076*** - - (0.045) (0.014) p - value for H 0 : = 0.5404 Literacy score (standardized) - - 0.029 0.060*** (0.040) (0.012) p - value for H 0 : = 0.4565 Problem solving score (standardized) - - - - Intercept 1.908*** 1.502*** 1.810*** 1.469*** (0.518) (0.092) (0.482) (0.092) No. observations 200 1,494 200 1,494 R - squared 0.19 0.40 0.18 0.40 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 4) Spain did not administer the problem solving test. 326 Table F .10A : Country - level OLS results, test scores excluded ( France ) Controls Country sample: France (share of immigrants 10.9%) (1) (2) I N I N Education (years) 0.034*** 0.058*** 0.036*** 0.060*** (0.004) (0.002) (0.004) (0.002) p - value for H 0 : = 0.0000 p - value for H 0 : = 0.0000 Experience (years) 0.014** 0.021*** 0.013** 0.021*** (0.006) (0.003) (0.006) (0.003) Experience squared 0.000 - 0.000*** 0.000 - 0.000*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.080** - 0.119*** (0.039) (0.013) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 1.962*** 1.699*** 1.842*** 1.652*** (0.065) (0.040) (0.095) (0.053) No. observations 210 2,044 210 2,043 R - squared 0.31 0.26 0.32 0.30 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 327 Table F .10B : Country - level OLS results, test scores included ( France ) Controls Country sample: France (share of immigrants 10.9%) (3) (4) (5) I N I N I N Education (years) 0.026*** 0.045*** 0.029*** 0.050*** NA NA (0.004) (0.002) (0.004) (0.003) p - value for H 0 : = 0.0004 p - value for H 0 : = 0.0001 Experience (years) 0.013** 0.022*** 0.013* 0.022*** (0.006) (0.003) (0.006) (0.003) Experience squared 0.000 - 0.000*** 0.000 - 0.000*** (0.000) (0.000) (0.000) (0.000) Female - 0.068* - 0.097*** - 0.082** - 0.116*** (0.039) (0.013) (0.039) (0.013) Numeracy score (standardized) 0.068*** 0.085*** - - (0.020) (0.007) p - value for H 0 : = 0.3978 Literacy score (standardized) - - 0.057*** 0.072*** (0.019) (0.007) p - value for H 0 : = 0.4594 Problem solving score (standardized) - - - - Intercept 2.063*** 1.824*** 1.995*** 1.762*** (0.080) (0.051) (0.088) (0.053) No. observations 209 2,042 209 2,042 R - squared 0.36 0.34 0.35 0.33 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 4) France did not administer the problem solving test. 328 Table F .11A : Country - level OLS results, test scores excluded ( Estonia ) Controls Country sample: Estonia (share of immigrants 10.3%) (1) (2) I N I N Education (years) 0.026* 0.058*** 0.025* 0.077*** (0.014) (0.004) (0.014) (0.004) p - value for H 0 : = 0.0138 p - value for H 0 : = 0.0004 Experience (years) 0.005 0.035*** 0.01 0.030*** (0.015) (0.005) (0.014) (0.005) Experience squared 0.000 - 0.001*** 0.000 - 0.001*** (0.000) (0.000) (0.000) (0.000) Female - - - 0.527*** - 0.433*** (0.073) (0.021) Numeracy score (standardized) - - - - Literacy score (standardized) - - - - Problem solving score (standardized) - - - - Intercept 1.716*** 1.239*** 1.810** 1.112*** (0.231) (0.071) (0.720) (0.115) No. observations 199 2,019 198 2,014 R - squared 0.04 0.11 0.27 0.28 Notes: 1) Specifications (2) control s for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 329 Table F .11B : Country - level OLS results, test scores included ( Estonia ) Controls Country sample: Estonia (share of immigrants 10.3%) (3) (4) (5) I N I N I N Education (years) 0.006 0.060*** 0.017 0.068*** 0.014 0.058*** (0.015) (0.005) (0.015) (0.004) (0.020) (0.005) p - value for H 0 : = 0.0005 p - value for H 0 : = 0.0010 p - value for H 0 : = 0.0292 Experience (years) 0.016 0.029*** 0.016 0.030*** 0.002 0.040*** (0.014) (0.005) (0.014) (0.005) (0.019) (0.005) Experience squared 0.000 - 0.001*** 0.000 - 0.001*** 0.000 - 0.001*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Female - 0.487*** - 0.402*** - 0.522*** - 0.428*** - 0.555*** - 0.419*** (0.069) (0.022) (0.071) (0.021) (0.078) (0.025) Numeracy score (standardized) 0.175*** 0.118*** - - - - (0.042) (0.014) p - value for H 0 : = 0.2512 Literacy score (standardized) - - 0.094** 0.068*** - - (0.040) (0.013) p - value for H 0 : = 0.5429 Problem solving score (standardized) - - - - 0.076 0.093*** (0.048) (0.014) p - value for H 0 : = 0.7399 Intercept 1.905*** 1.312*** 1.867*** 1.203*** 1.578*** 1.235*** (0.600) (0.114) (0.623) (0.116) (0.385) (0.126) No. observations 198 2,014 198 2,014 145 1,574 R - squared 0.32 0.30 0.29 0.29 0.30 0.29 Notes: 1) All s pecifications control for sector of employment. 2) Final sample weight used to correct for inverse probability weighted sampling. 3) Jack - knife standard errors based on replicate weights reported in parentheses (80 replications). 330 Figure F.1 : Total sample DFL results : specification (0) Notes: 1) Specification (0) includes country dummies only. Specification (A) controls for education and labour market experience (in quadratic); Specification (B) adds controls for gender and sector of employment. Specification (C) adds numeracy test score to the list of covariates in specification (B); specification (D) adds literacy test score to that list, while specification (E) adds both literacy and numeracy test scores. All specifications (A) to (E) include country dummies. 2) Final sample weight used to c orrect for inverse probability weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 331 Figure F.2 : Total sample DFL results : specification (A) Notes: 1) Specification (0) includes country dummies only. Specification (A) controls for education and labour market experience (in quadratic); Specification (B) adds controls for gender and sector of employment. Specification (C) adds numeracy test score to the list of covariates in specification (B); specification (D) adds literacy test score to that list, while specification (E) adds both literacy and numeracy test scores. All specifications (A) to (E) include country dummies. 2) Final sample weight used to c orrect for inverse probability weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 332 Figure F.3 : Total sample DFL results : specification ( B ) Notes: 1) Specification (0) includes country dummies only. Specification (A) controls for education and labour market experience (in quadratic); Specification (B) adds controls for gender and sector of employment. Specification (C) adds numeracy test score to the list of covariates in specification (B); specification (D) adds literacy test score to that list, while specification (E) adds both literacy and numeracy test scores. All specifications (A) to (E) include country dummies. 2) Final sample weight used to c orrect for inverse probability weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 333 Figure F.4 : Total sample DFL results : specification ( C ) Notes: 1) Specification (0) includes country dummies only. Specification (A) controls for education and labour market experience (in quadratic); Specification (B) adds controls for gender and sector of employment. Specification (C) adds numeracy test score to the list of covariates in specification (B); specification (D) adds literacy test score to that list, while specification (E) adds both literacy and numeracy test scores. All specifications (A) to (E) include country dummies. 2) Final sample weight used to c orrect for inverse probability weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 334 Figure F.5 : Total sample DFL results : specification ( D ) Notes: 1) Specification (0) includes country dummies only. Specification (A) controls for education and labour market experience (in quadratic); Specification (B) adds controls for gender and sector of employment. Specification (C) adds numeracy test score to the list of covariates in specification (B); specification (D) adds literacy test score to that list, while specification (E) adds both literacy and numeracy test scores. All specifications (A) to (E) include country dummies. 2) Final sample weight used to c orrect for inverse probability weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 335 Figure F.6 : Total sample DFL results : specification ( E ) Notes: 1) Specification (0) includes country dummies only. Specification (A) controls for education and labour market experience (in quadratic); Specification (B) adds controls for gender and sector of employment. Specification (C) adds numeracy test score to the list of covariates in specification (B); specification (D) adds literacy test score to that list, while specification (E) adds both literacy and numeracy test scores. All specifications (A) to (E) include country dummies. 2) Final sample weight used to c orrect for inverse probability weighted sampling; equal weight is placed on each country. 3) 95% pointwise confidence intervals obtained via bootstrapping. 336 Figure F.7 : Restricted sample DFL results : estimated densities of log - earnings (actual) Notes: 1) Sample restricted to countries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Spec ification (B) adds controls for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sam ple weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 337 Figure F.8 : Restricted sample DFL results : specification (0) Notes: 1) Sample restricted to countries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Specific ation (B) adds controls for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 338 Figure F. 9 : Restricted sample DFL results : specification (A) Notes: 1) Sample restricted to countries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Specification ( B) adds controls for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 339 Figure F. 10 : Restricted sample DFL results : specification (B) Notes: 1) Sample restricted to countries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Specification ( B) adds controls for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 340 Figure F.1 1 : Restricted sample DFL results : specification (C) Notes: 1) Sample restric ted to countries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Specification (B) adds controls for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 341 Figure F.1 2 : Restricted sample DFL results : specification (D) Notes: 1) Sample restricted to countries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Specification (B) adds contro ls for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sample weight used to correc t for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 342 Figure F.1 3 : Restricted sample DFL results : specification (E) Notes: 1) Sample restricted to countries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Specification (B) adds contro ls for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sample weight used to correc t for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 343 Figure F.1 4 : Restricted sample DFL results : specification (F) Notes: 1) Sample restricted to countries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Specification (B) adds contro ls for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 344 Figure F.1 5 : Restricted sample DFL results : specification (G) Notes: 1) Sample restricted to c ountries in PIAAC which administered all test scores. France, Italy and Spain excluded. 2) Specification (0) includes country dummies only. Specification (B) controls for education and labour market experience (in quadratic); Specification (B) adds control s for gender and sector of employment. Specification (G) adds numeracy, literacy and problem solving test score to the list of covariates in specification (B). Both specifications (B) and (G) include country dummies. 3) Final sample weight used to correct for inverse probability weighted sampling; equal weight is placed on each country. 4) 95% pointwise confidence intervals obtained via bootstrapping. 345 BIBLIOGRAPHY 346 BIBLIOGRAPHY Abbott, Michael, and Charles Beach. "Immigrant Earnings Differentials and Birth - Year Effects for Men in Canada: Post War - 1972." Canadian Journal of Economics 26 (August 1993): 505 - 24. Adsera, Alicia, and Barry R. Chiswick. "Are there gender and country of origin differences in immigrant labor market outcomes across European destinations?." Journal of Population Economics 20, no. 3 (2007): 495 - 526. Aydemir, Abdurrahman, and Mikal Skuterud. "Explaining the deteriorating entry earnings of Canada's immigrant cohorts, 1966 2000." Canadian Jou rnal of Economics/Revue canadienne d'économique 38, no. 2 (2005): 641 - 672. Betts, Julian R., and Magnus Lofstrom. "The educational attainment of immigrants: trends and implications." In Issues in the Economics of Immigration , pp. 51 - 116. University of Chicago Press, 2000. Blinder, Alan S. "Wage discrimination: reduced form and structural estimates." Journal of Human resources (1973): 436 - 455. Borjas, George J. "The economics of immigration." Journal of economic literature (1994): 1667 - 1717. Borjas, G eorge J. "The impact of assimilation on the earnings of immigrants: A reexamination of the evidence." (1984). Bratsberg, Bernt, and Dek Terrell . "School quality and returns to education of US immigrants." Economic Inquiry 40, no. 2 (2002): 177 - 198. Chisw ick, Barry R. "The effect of Americanization on the earnings of foreign - born men." The Journal of Political Economy (1978): 897 - 921. Coppola, Lucia, Davide Di Laurea, and Stefano Gerosa. "The Immigrants Wage Gap in Italy." Available at SSRN 2514714 (2013) . DESA, UN. "Trends in international migrant stock: The 2013 revision." (2013). De Silva, Arnold. "Earnings of immigrant classes in the early 1980s in Canada: A reexamination." Canadian Public Policy/Analyse de Politiques (1997): 179 - 202. DiNardo, John. "Propensity score reweighting and changes in wage distributions." University of Michigan, mimeograph (2002). 347 DiNardo, John, Nicole M. Fortin, and Thomas Lemieux. "Labor Market Institutions and the Distribution of Wages, 1973 - 1992: A Semiparametric Approach." Econometrica 64, no. 5 (1996): 1001 - 1044. Kee, P. (1993): The Economic Status of Male Immigrants in the Netherlands . Dissertation, University of Amsterdam. LaLonde, Robert J., and Robert H. Topel. "Labor market adjustments to increased immigration." In Immigration, trade and the labor market , pp. 167 - 199. University of Chicago Press, 1991. Grand, Carl le, and Ryszard Szulkin. "Permanent disadvantage or gradual integration: explaining the immigrant native earnings gap in Sweden." Labour 16, no. 1 (2002): 37 - 64. Lubotsky, Darren. Chutes Or Ladders?: A Longitudinal Analysis of Immigrant Earnings . Industrial Relations Section, Princeton University, 2000. Oaxaca, Ronald. "Male - female wage differential s in urban labor markets." International economic review (1973): 693 - 709. Pischke, Jörn - Steffen. Assimilation and the Earnings of Guestworkers in Germany . No. 92 - 17. ZEW Discussion Papers, 1992. Reitz, Jeffrey G. "Immigrant skill utilization in the Canadian labour market: Implications of human capital research." Journal of International Migration and Integration/Revue de l'integration et de la migration internationale 2, no. 3 (2001): 347 - 378. Tandon, Bankey B. " Earning differentials among native born and foreign born residents of Toronto." International Migration Review (1978): 406 - 410. Wooldridge, Jeffrey M. Econometric analysis of cross section and panel data . The MIT press, 2002.