ESSAYS ON AGRICULTURAL PRODUCTIVITY, YOUTH EMPLOYMENT, AND HUMAN CAPITAL INVESTMENT IN SUB ŒSAHARAN AFRICA By Josephat Koima A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the d egree of Agricultural, Food , and Resource Economics Œ Doctor of Philosophy Economics Œ Dual Major 2021 ABSTRACT ESSAYS ON AGRICULTURAL PRODUCTIVITY, YOUTH EMPLOYMENT, AND HUMAN CAPITAL INVESTMENT IN SUB ŒSAHARAN AFRICA By Josephat Koima This dissertation focuses on the intersection of agricultural productivity, youth employment, and investments in human capital development in Sub ŒSaharan Africa (SSA). Agriculture is a dominant employer and source of income in SSA, and plays an important role in youth empl oyment and educational attainments. In Chapter 1, we study the role of structural transformation in the labor reallocation between the farm and the non Œfarm sector and the consequential impact on worker demographics. Specifically, we investigate whether a gricultural productivity differentially reallocates labor by age and gender. We develop a theoretical model where increased land productivity leads to younger individuals sorting into the non Œfarm sector while older individuals sort into agriculture. We th en use data from Zambia in our empirical analysis. Our main results show some evidence of productivity affecting labor reallocation within recent productivity lags (last 2 years) but not when longer productivity lags (4 or 6) are considered. Specifically, consistent with our model prediction, a 10% increase in a 2 Œyear lagged moving average of productivity decreases the probability of farming by 0.3 percentage points among youth (15 Œ24) and older youth (25 Œ34). We also show that youth (15 Œ24) also exit farm ing following increased productivity. Increased productivity tends to reduce the intensity of farming across all age groups but the reduction is relatively larger among the youth. In addition, young men are more likely to exit business activity as producti vity increases relative to young women Œ across all productivity lags. In the short term (2 Œlags), while youth exit farming, there is no differential outcome between genders. However, among older youth, males are more likely to exit farming compared to wom en. Finally, males mainly drive the reduction in intensity of farming. Overall, while we find some evidence in favor of our hypotheses, the evidence is generally limited to the short term and the marginal effects are quantitatively small. Chapter 2 invest igates the impact of agricultural productivity on human capital investments in Tanzania. Agriculture remains a major source of employment and income in Tanzania. Therefore, any agricultural productivity shocks are likely to affect educational investment decisions. Our results provide evidence that increased agricultural productivity boosts spending on uniform, contributions and total academic expenses. We find positive but statistically non Œsignificant effects of productivity on study times. In addition, we find no evidence of heterogeneous effects by student gender. We show evidence that productivity effects are smaller in female Œheaded households. Finally, we find some evidence that post Œprimary students experience larger impacts compared to primary school students. In Chapter 3, I investigate the impact of primary school electrification on academic outcomes in Kenya. Between 2014 and 2016, the number of primary schools with electricity rose from 56% to 94%. Schools near the grid network were connected to g rid electricity while those further received solar photovoltaics. Using this rapid electrification expansion as a source of identifying variation in a panel fixed effects model, the paper estimates the impact on school test scores, enrollment, and completi on. The paper also attempts to quantify the effects of lighting on education performance by relying on the off Œgrid (solar) electricity coefficients. Using a universe of 8 th grade students in public schools in Kenya, the paper finds no evidence that electr icity affects test scores or enrollment in the short run. However, off Œgrid electrification increases completion by 1%. Using off Œgrid estimates, the paper concludes that lighting has a small positive impact on completion but not on test scores or enrollme nt. iv This dissertation is dedicated to my family, grandfather, and grandmother. Thank you for laying a solid foundation and for all the support and inspiration. v ACKNOWLEDGEMENT S I would like to thank my advisors, Dr. Milu Muyanga, Dr. Thomas Jayne, Dr. Leah Lakdawala, Dr. Jeffrey Wooldridge, Dr. Maria Porter, Dr. Eduardo Nakasone, the students and faculty at the Department of Economics and the Department of Agriculture, Food and Natural Resource Economics, for their feedback. I thank the various stakeho lders that have facilitated my academic journey. These include the Kenya Scholar ŒAthlete Project (KENSAP), Williams College, the University of Maryland, and Michigan State University. I am very grateful to my teachers at Tirip katoi P rimary School, and Keri cho Head ŒQuarters P rim ary S chool for all the work Œ I am here now, thanks in part to you. Many thanks to Kibarasoi Primary School for providing me with a workspace and electricity as I was writing my final dissertation chapter. Thank you for coming through during this COVID Œ19 nightmare. Special thanks to my friends who supported me through this process. Finally, I am grateful to my supportive family. vi TABLE OF CONTENTS LIST OF TABLES ........................................................................................................................ vii Chapter 1 : Effects of Agricultural Productivity on Demographic Composition of Farmers in Zambia ........................................................................................................................................... 1 I. INTRODUCTION ............................................................................................................. 2 II. LITERATURE REVIEW .................................................................................................. 7 III. THEORY AND THE MODEL ....................................................................................... 12 IV. DATA .............................................................................................................................. 29 V. METHODS ...................................................................................................................... 33 VI. RESULTS Œ POOLED PROBIT WITH TIME AND DISTRICT FIXED EFFECTS ... 35 VII. GENERAL DISCUSSION AND DIRECTION FOR FUTURE RESEARCH .............. 49 VIII. CONCLUSION ............................................................................................................... 52 APPENDIX ................................................................................................................................... 53 BIBLIOGRAPHY ......................................................................................................................... 66 Chapter 2 : Does Agricultural Productivity Translate to Increased Human Capital Investment? Impacts of Productivity Shocks on Education Expenses and School Outcomes in Tanzania .................................................................................................................................. 70 I. INTRODUCTION ........................................................................................................... 71 II. LITERATURE REVIEW ................................................................................................ 73 IV. DATA AND METHODS ................................................................................................ 82 V. RESULTS ........................................................................................................................ 86 VI. ROBUSTNESS CHECKS ............................................................................................... 96 VII. CONCLUSION ............................................................................................................. 101 APPENDIX ................................................................................................................................. 104 BIBLIOGRAPHY ....................................................................................................................... 135 Chapter 3 : School Electrification and A cademic Outcomes in Rural Kenya ...................... 138 I. INTRODUCTION ......................................................................................................... 139 II. LITERATURE REVIEW .............................................................................................. 145 III. DATA DESCRIPTION ................................................................................................. 151 IV. IDENTIFICATION STRATEGY ................................................................................. 158 V. MAIN RESULTS .......................................................................................................... 159 VI. HETEROGENEITY BY SUBJECT AND GENDER ................................................... 169 VII. ROBUSTNESS CHECKS ............................................................................................. 174 VIII. CONCLUSION ............................................................................................................. 181 APPENDIX ................................................................................................................................. 183 BIBLIOGRAPHY ....................................................................................................................... 185 vii LIST OF TABLES Table 1.1: Summ ary Statistics (Means) ........................................................................................ 32 Table 1.2: Employment Combinations (Proportion) .................................................................... 33 Table 1.3: Agricultural Productivity and Employment Par ticipation by Age ŒGroup (2 ŒLags Moving Average) .......................................................................................................................... 38 Table 1.4: Agricultural Productivity and Employment Participation by Age ŒGroup (4 ŒLags Moving Average) .......................................................................................................................... 38 Table 1.5: Agricultural Productivity and Employment Participation by Age ŒGroup (6 ŒLags Moving Average) .......................................................................................................................... 39 Table 1.6: Agricultural Productivity and Employment Participati on by Age ŒGroup and Gender (2ŒLags Moving Average) ............................................................................................................ 41 Table 1.7: Agricultural Productivity and Employment Participation by Age ŒGroup and Gender (4ŒLags Moving Average) ............................................................................................................ 41 Table 1.8: Agricultural Productivity and Employment Participation by Age ŒGroup and Gender (6ŒLags Moving Average) ............................................................................................................ 42 Table 1.9: Agricultural Productivity a nd Multiple Employment Participation by Age ŒGroup (2 ŒLags Moving Average) ................................................................................................................. 43 Table 1.10: Agricultural Productivity and Multiple Employment Participation by Age ŒGroup (4 ŒLags Moving Average) ................................................................................................................. 44 Table 1.11: Agricultural Productivity and Multiple Employment Participation by Age ŒGroup (6 ŒLags Moving Average) ................................................................................................................. 44 Table 1.12: Agric ultural Productivity and Multiple Employment Participation by Age ŒGroup and Gender (2 ŒLags Moving Average) ............................................................................................... 46 Table 1.13: Agricultural Productivity and Multiple Employment Participation by Age ŒGroup and Gender (4 ŒLags Moving Average) ............................................................................................... 46 Table 1.14: Agricultural Productivity and Multiple Employment Participation by Age ŒGroup and Gender (6 ŒLags Moving Average) ............................................................................................... 47 Table 1.15: Agricultural Productivity and Employment Participation by Age ŒGroup Œ LPM (2 ŒLags Moving Average) ................................................................................................................. 48 viii Table 1.16: Agricultural Productivity and E mployment Participation by Age ŒGroup Œ LPM (4 ŒLags Moving Average) ................................................................................................................. 49 Table 1.17: Agricultural Productivity and Employment Participation by Age ŒGroup Œ LPM (6 ŒLags Moving Average) ................................................................................................................. 49 Table A 1.1: Agricultural Productivity and Employment Participation Œ First Stage Probit (2 ŒLags Moving Average) ................................................................................................................. 54 Table A 1.2: Agricultural Productivity and Employment Participation Œ First Stage Probit (4 ŒLags Moving Average) ................................................................................................................. 55 Table A 1.3: Agricultural Productivity and Employment Participation Œ First St age Probit (6 ŒLags Moving Average) ................................................................................................................. 56 Table A 1.4: Agricultural Productivity and Employment Participation by Gender Œ First Stage Probit (2 ŒLags Moving Average) ................................................................................................. 57 Table A 1.5: Agricultural Productivity and Employment Participation by Gender Œ First Stage Probit (4 ŒLags Moving Average) ................................................................................................. 58 Table A 1.6: Agricultural Productivi ty and Employment Participation by Gender Œ First Stage Probit (6 ŒLags Moving Average) ................................................................................................. 59 Table A 1.7: Agricultural Productivity and Multiple Employment Participation Œ First Stage Probit (2 ŒLag s Moving Average) ................................................................................................. 60 Table A 1.8: Agricultural Productivity and Multiple Employment Participation Œ First Stage Probit (4 ŒLags Moving Average) ................................................................................................. 61 Table A 1.9: Agricultural Productivity and Multiple Employment Participation Œ First Stage Probit (6 ŒLags Moving Average) ................................................................................................. 62 Table A 1.10: Agricultural Productivity and Multiple Em ployment Participation by Gender Œ First Stage Probit (2 ŒLags Moving Average) ............................................................................... 63 Table A 1.11: Agricultural Productivity and Multiple Employment Participation by Gender Œ First Stage Probit (4 ŒLags Moving Average) ............................................................................... 64 Table A 1.12: Agricultural Productivity and Multiple Employment Participation by Gender Œ First Stage Probit (6 ŒLags Moving Average) ............................................................................... 65 Table 2.1: Summary Statistics ...................................................................................................... 85 Table 2.2: Effects of Land Productivity on School Expenditure and Study Times (FE) .............. 86 Table 2.3: Effects of Labor Productivity Expenditure and Study Times (FE) ............................. 86 Table 2.4: Effects of Land Productivity on School Expenditure and Study Times (IV) .............. 88 ix Table 2.5: Effects of Labor Productivity on School Expenditure and Study Times (IV) ............. 88 Table 2.6: Effects of Land Productivi ty on School Expenditure and Study Times by Gender (FE) ....................................................................................................................................................... 89 Table 2.7: Effects of Labor Productivity on School Expenditure and Study Times by Gender (FE) ............................................................................................................................................... 89 Table 2.8: Effects of Land Productivity on School Expenditure and Study Times by Gender (IV) ....................................................................................................................................................... 90 Table 2.9: Effects of Labor Productivity on School Expenditure and Study T imes by Gender (IV) ....................................................................................................................................................... 90 Table 2.10: Effects of Land Productivity on School Expenditure and Study Times by Gender of Head (FE) ...................................................................................................................................... 91 Tabl e 2.11: Effects of Labor Productivity on School Expenditure and Study Times by Gender of Head (FE) ...................................................................................................................................... 91 Table 2.12: Effects of Land Productivity on School Expenditure and Study Times by Gender of Head (IV) ...................................................................................................................................... 92 Table 2.13: Effects of Labor Productivity on School Expenditure and Study Times by Gender of Head (IV) ...................................................................................................................................... 93 Table 2.14: E ffects of Land Productivity on School Expenditure and Study Times by School ŒLevel (FE) ..................................................................................................................................... 94 Table 2.15: Effects of Labor Productivity on School Expenditure and Study Times by School ŒLevel (FE) ..................................................................................................................................... 94 Table 2.16: Effects of Land Productivity on School Expenditure and Study Times by School ŒLevel (IV) ...................................................................................................................................... 95 Table 2.17: Effects of Labor Productivity on School Expenditure and Study Times by School ŒLevel (IV) ...................................................................................................................................... 95 Table 2.18: Effects of Gross Agricultural Income on School Expenditure and Study Times (FE) ....................................................................................................................................................... 96 Table 2.19: Effects of Gross Agricultural Income on School Expenditure and Study Times (IV) ....................................................................................................................................................... 96 Table 2.20: Effects of Net Agricultural Income per Hectare o n School Expenditure and Study Times (FE) .................................................................................................................................... 97 Table 2.21: Effects of Net Agricultural Income per Hectare on School Expenditure and Study Times (IV) ..................................................................................................................................... 97 x Table 2.22: Effects of Net Agricultural Income on School Expenditure and Study Times (FE) . 98 Table 2.23: Effects of Net Agricultural Income on School Expenditure and Study Time s (IV) .. 98 Table 2.24: Effects of Land Productivity on School Expenditure and Study Times Œ Excluding Other Income (FE) ........................................................................................................................ 99 Table 2.25 : Effects of Labor Productivity on School Expenditure and Study Times Œ Excluding Other Income (FE) ........................................................................................................................ 99 Table 2.26: Effects of Land Productivity on School Expenditure and Study Times Œ Excluding Other Income (IV) ......................................................................................................................... 99 Table 2.27: Effects of Labor Productivity on School Expenditure and Study Times Œ Excluding Other Income (IV) ....................................................................................................................... 100 Table 2.28: Effects of Land Productivity on School Expenditure and Study Times Œ Rainfall Deviations as Instrument (IV) ..................................................................................................... 100 Table 2.29: Effects of Labor Productivity on School Expenditure an d Study Times Œ Rainfall Deviations as Instrument (IV) ..................................................................................................... 101 Table A 2.1: First Stage Regressions Œ Productivity and Rainfall ............................................. 105 Table A 2.2: Effects of Land Productivity on School Expenditure and Study Times (FE) Œ Full Tables .......................................................................................................................................... 106 Table A 2.3: Effects of Labor Productivity on School Expenditure and Study Times (FE) Œ Full Table ........................................................................................................................................... 107 Table A 2.4: Effects of Land Productivity on School Expenditure and Study Times (IV) Œ Full Table ........................................................................................................................................... 108 Table A 2.5: Effects of Labor Productivity on School Expenditure and Study Times (IV) Œ Full Table ........................................................................................................................................... 109 Table A 2.6: Effects of Land Productivity on School Expenditure and Study Times by Gen der (FE) Œ Full Table ......................................................................................................................... 110 Table A 2.7: Effects of Labor Productivity on School Expenditure and Study Times by Gender (FE) Œ Full Table ......................................................................................................................... 111 Table A 2.8: Effects of Land Productivity on School Expenditure and Study Times by Gender (IV) Œ Full Table ......................................................................................................................... 112 Table A 2.9: Effects of Labor Productivity on School Expenditure and Study Times by Gender (IV) Œ Full Table ......................................................................................................................... 113 xi Table A 2.10: Effects of Land Productivity on School Expenditure and Study Times by Gender of Head (FE) Œ Full Table ........................................................................................................... 114 Table A 2.11: Effects of Labor Productivity on School Expenditure and Study Times by Gender of Head (FE) Œ Full Table ........................................................................................................... 115 Table A 2.12: Effects of Land Productivity on School Expenditure and Study Times by Gender of Head (IV) Œ Full Table ........................................................................................................... 116 Table A 2.13: Effects of Labor Productivity on School Expenditure and Study Times by Gender of Head (IV) Œ Full Table ........................................................................................................... 117 Table A 2.14: Effects of Land Productivity on School Expenditure and Study Times by School ŒLevel (FE) Œ Full Table ............................................................................................................... 118 Table A 2.15: Effects of Labor Productivity on School Expenditure and Study Times by School ŒLevel (FE) Œ Full Table ............................................................................................................... 119 Table A 2.16: Effects of Land Productivity on School Expenditure and Study Times by School ŒLevel (IV) Œ Full Table ............................................................................................................... 120 Table A 2.17: Effects of Labor Productivity on School Expenditure and Study Times by School ŒLevel (IV) Œ Full Table ............................................................................................................... 121 Table A 2.18: Effects of Gross Agricultural Income on School Expenditure and Study Times (FE) Œ Full Table ......................................................................................................................... 122 Table A 2.19: Effects of Gross Agricultural Income on School Expenditure and Study Times (IV) Œ Full Table ......................................................................................................................... 123 Table A 2.20: Effects of Net Agricultural Income per Hectare on School Expenditure and Study Times (FE) Œ Full Table .............................................................................................................. 124 Table A 2.21: Effects of Net Agricultural Income per Hectare on School Expenditure and Study Times (IV) Œ Full Table .............................................................................................................. 125 Table A 2.22: Effects of Net Agricultural Income on School Expenditure and Study Times (FE) Œ Full Tables ............................................................................................................................... 126 Table A 2.23: Effects of Net Agricultural Income on School Expenditure and Study Times (IV) Œ Full Table .................................................................................................................................... 127 Table A 2.24: Effects of Land Productivity on School Expenditure and Study Times Œ Excluding Other Income (FE) Œ Full Table .................................................................................................. 128 Table A 2.25: Effects of Labor Productivity on School Expenditure and Study Times Œ Excluding Other Income (FE) Œ Full Table ................................................................................ 129 xii Table A 2.26: Effects of Land Productivity on School Expenditure and Study Times Œ Excluding Other Income (IV) Œ Full Table .................................................................................................. 130 Table A 2.27: Effects of Labor Productivity on School Expenditure and Study Times Œ Excluding Other Income (IV) Œ Full Table ................................................................................. 131 Table A 2.28: Effects of Land Productivity on School Expenditure and Study Times Œ Rainfall Deviations as Instrument (IV) Œ Full Table ................................................................................ 132 Table A 2.29: Effects of Labor Productivity on School Expenditure and Study Times Œ Rainfall Deviations as Instrument (IV) Œ Full Table ................................................................................ 133 Table A 2.30: First Stage Regressions Œ Product ivity and Rainfall Deviations ........................ 134 Table 3.1: Summary Statistics .................................................................................................... 157 Table 3.2: Effects of School Electrific ation on School Test Scores ........................................... 161 Table 3.3: Effects of School Electrification on 8th Grade Enrollment (Dependent variable Œ log of enrollment) .............................................................................................................................. 163 Table 3.4: Effects of School Electrification on 8th Grade Completion (Dependent: Log Completion) ................................................................................................................................ 165 Table 3.5: Test for the Assumption of a Common Trend ........................................................... 168 Table 3.6: Effects of School Electrification on School Mean Test Scores by Subject ............... 171 Table 3.7: Heterogeneous Impacts by Gender Œ Test scores, Enrollmen t, and Completion ...... 173 Table 3.8: School Sample Restriction Based on Ownership and Location (rural/urban) ........... 176 Table 3.9: Inclusion of Regional Time Trends ........................................................................... 178 Table 3.10: Results by Cluster Level Œ School, Zone, Sub Œcounty, and County ...................... 180 Table A 3.1: Complementarity between School Inputs and Electricity ...................................... 184 1 Chapter 1 : Effects of Agricultural Productivity on Demographic Composition of Farmers in Zambia 2 I. INTRODUCTION According to the Food and Agriculture Organiz ation of the United Nations (FAO), one of the major challenges in food security is demographic changes due to population growth, urbanization, and ageing (FAO, 2014). Specifically, by 2050 there will be approximately nine billion people mostly in developin g countries. In addition, due to accelerated urbanization, by 2050 70% of this population will be living in urban areas. FAO estimates that this expanding population will require up to a 60% increase in food production. FAO suggests that African farmers ar e also ageing rapidly. Specifically, the average farmer age is approximately 60 in Africa, as in the highly industrialized United States, despite the fact that 60% of African population is below 24 years of age. Some studies find similar average age of 60 for farmers (Gorman, 2013; Vos, 2015). This ageing threatens the ability of future farmers to meet the increasing demands for food. However, recent work by Yeboah and Jayne (2018), shows that while there is rapid movement from agriculture to non Œfarm secto rs, agriculture remains a major source of employment, and the mean farmer age remains stable generally in sub ŒSaharan Africa. This paper seeks to understand the role of agricultural productivity on the dynamics of age and gender of workers in each respecti ve sector as this process unfolds. It has long been recognized that agriculture can play an important role both in the early and latter stages of economic development (Lewis, 1954; Hirschman, 1958; Johnston and Mellor, 1961). Agriculture tends to be the ma jor source of employment in initial stages of development. As agricultural productivity increases, it can release excess labor from low productivity agricultural activities into the non Œfarm sector. In addition, agriculture also provides strong linkages to the non Œfarm sector. Specifically, high agricultural productivity raises income that subsequently increases demand for non Œfood goods and services. This increased demand 3 eventually stimulates the expansion of the non Œfarm sector. Recent studies have found no alternative development path for Africa without serious agricultural innovation and growth (Diao et al. 2010). The role of agriculture has been recently illustrated by the Green Revolution that contributed to a number of Asian countries moving from slo w to high growth trajectories. Unfortunately, other developing countries, particularly in Africa have not experienced such a revolution illustrating the complex ways in which agricultural driven development can depend on context. Current studies on agricul tural transformation have largely focused on movement of labor between farm and non Œfarm sectors. However, research has paid little attention to the heterogeneous labor reallocation. Foster and Rosenzweig (2007), ar gue that important aspects of exit of lab or from the agriculture to the non Œfarm sector, such as selectivity in human capital, are less studied and less understood. We further argue that other selective aspects such as age and gender, and their interactions with human capital, are equally importa nt. It is thus necessary to understand the impact of structural transformation on demographic composition of farm and non Œfarm workers (age and gender). This will guide economic growth policy formulation. The idea that farmers are ageing in Sub ŒSaharan Afr ica requires a thorough examination before any serious conclusions and policy recommendations can be drawn. Our hypothesis is that, seemingly paradoxically, sustained agricultural productivity will drive young people away from agriculture through three ch annels. First, agricultural productivity relaxes the financial constraints that prohibit individuals from migrating in search of better opportunities. High agricultural productivity thus provides extra resources that allow individuals to migrate. We believ e that young people have higher mobility and are likely to migrate due to the educational attainments that may be required in blue Œcollar and white collar jobs, and also 4 because young people do not have familial constraints ( people with families may have difficulties relocating the entire family). Second, and the focus of our theoretical model, agricultural productivity can generate strong multiplier effects (Snyder et al., 2019) . Agricultural productivity increases disposable income for farmers. These in comes can not only increase regular consumption but also expand the consumption set for the household. Farmers may now demand more food and nonŒfood goods and services. Recent research is showing that increased farm incomes lead to greater food expenditure s of commodities that the farm household does not produce itself (processed foods, tinned fish, coffee, etc. ). Such increased demand can lead to creation of new and expansion of existing businesses. Sustained productivity can therefore lead to structural transformation that can see expansion of non Œfarm job opportunities. These job opportunities may require skills and higher educational levels that are typically abundant among the youth. Consequently, the new job opportunities might draw younger farmers awa y from agriculture. Finally, agricultural productivity also provides households with additional incomes that can make it possible to keep their children consistently in school and attend better quality schools, which increases the probability that these ch ildren will wind up in non Œfarm jobs. Foster and Rosenzweig (2007) find that an increase in agricultural productivity led to increased schooling in India following the introduction of High ŒYielding ŒVariety rice. Overall, we would expect to see increases in agricultural productivity resulting in the average age of a farmer going up, and a wider difference between the mean ages of individuals in non Œfarm versus farm employment. Rather than viewing this as a problem, it would be a positive indicator of structu ral transformation associated with higher living standards. One day, if the mean age of Africans in farming becomes too high, we might need to address it, but as of now, there is no indication that the mean age of 5 people in farming is more than 1 or 2 year s higher than the mean age of people in off Œfarm employment (Yeboah and Jayne, 2018). This paper seeks to understand how structural transformation affects the age and gender composition of farmers (and non Œfarm workers). Instead of simply testing whether t he mean age of farmers is rising, we explore the dynamics of labor movement in and out of farm and non Œfarm sectors by gender and age groups (15 Œ24, 25 Œ34, and 35 Œ65) among working Œage adults . It will provide answers as to whether structural transformation drives young labor away from farming or into farming. This study focuses on Africa as previous research has shown that economic growth experiences can be context specific. For instance, the Green Revolution pushed several Asian countries into new periods of rapid growth while at the same time lifting billions out of poverty. Unfortunately, this technological shock has not been felt in the context of Africa. This paper focuses on age and gender because these demographic characteristics are especially releva nt in Africa. Several studies show evidence of gender effects in diverse settings and it is thus highly probable that structural transformation will have important gender dimensions (Duflo and Udry, 2004; Foster and Rosenzweig, 2002; Goldestein and Udry 20 08). We investigate whether women are left in low Œproductivity agricultural activities and, subsequently, make policy recommendations that ensure equitable benefits of development. Gender dynamics also has potential spillovers. Research on household bargai ning models document evidence that gender of income earners influences household consumption and investment in children as discussed in the next section. In my second dissertation essay, we investigate the impact of agricultural productivity shocks on inve stment in human capital. This work, together, will provide insights on whether structural transformation leaves women in less productive activities while at the same time 6 lowering investments in girls Œ a problem that can lead to a persistent poverty trap. Migration is left out for future research work. To test our hypothesis, this paper focuses on long term agricultural productivity shocks in Zambia. We first develop a model that explains how labor will be reallocated following productivity shocks. We borr ow from previous literature that has developed models that incorporate the roles played by factors such as labor pull, local demand effects , liquidity constraints , and sectoral linkages in determining labor reallocation (Emerick 2018; Harris and Todaro, 19 70; Foster and Rosenzweig, 2007; Bryan et al. 2014). In our model, we focus on the multipliers on the non Œfarm sector generated by increased farm income. These multipliers occur when increased income generate s demand for non Œfarm goods and services, which in turn stimulate s the growth and expansion of firms and enterprises that supply them. The model predicts that high long Œterm productivity will drive youth away from farming resulting in an increase in the mean age of those participating in farming. Conver sely, low productivity in agriculture fails to spur non Œfarm growth and thereby failing to release labor into the non Œfarm sector. In the model, labor is selectively reallocated depending on the stock of human capital acquired (skills and education). This is the primary reason why younger people would be driven away from agriculture. Since women and girls tend to receive less investment in education in Sub ŒSaharan Africa, we hypothesize that women will be disadvantaged relative to men and will experience li ttle movement out of agriculture. In our main empirical results, we find some evidence of productivity affecting labor reallocation within recent productivity lags (last 2 years) but not when longer (4 or 6) productivity lags are considered. Specifically, consistent with our m odel prediction, a 10% increase in a 2 Œyear lagged moving average of productivity decreases the probability of farming by 0.3 percentage 7 points among youth (15 Œ24) and older youth (25 Œ34). We also show that youth (15 Œ24) also exit farm ing following increased productivity. Increased productivity tends to reduce the intensity of farming across all age groups but the reduction is relatively larger among the youth. Considering gender, we find that young men are more likely to exit business activity as productivity increases relative to young women Œ across all productivity lags. In the short term (2 Œlags), while both youth exit farming, there is no differential outcome between genders. However, among older youth, males are more likely to exi t farming compared to women. Finally, males mainly drive the reduction in intensity of farming. Overall, while we find some evidence in favor of our hypotheses, the evidence is generally limited to the short term and the marginal effects are quantitatively small. The rest of the paper is organized as follows. Section 2 reviews the literature. Section 3 provides a brief theoretical motivation for our model and subsequently develops the model tested in this paper. Section 4 discusses the data while section 5 provides our empirical methodology. We present the results in section 6 before a brief discussion on our results and directions for future research in section 7. Section 8 concludes the paper. II. LITERATURE REVIEW The impact of agricultural shocks on employm ent and sectoral labor reallocation have been studied in different contexts with differing and sometime contradictory findings. Below we review related literature and their findings and try to put this study into context and highlight our contributions. The closest work to this paper is Emerick (2018) who investigates the role of agricultural productivity on labor reallocation in rural India. This paper shows that increased productivity caused by abnormally high rainfall leads to an increase in the labor sh are of the non Œagricultural sector. Individuals are more likely to engage in a primary activity in the non Œfarm sector and decrease days devoted to agricultural activities. This is consistent with rainfall shocks increasing 8 agricultural incomes and generat ing positive spillovers into the non Œfarm sector. This paper finds evidence of results being driven by local demand effects that yield multipliers in the non Œfarm sector. These effects are, however, transitory and driven by recent shocks within the previou s two years. Yeboah and Jayne (2018) show that the rate at which labor force is moving out of agriculture is strongly and positively linked to lagged farm productivity growth rate. Our analysis moves further and considers age and gender dynamics in labor r eallocation driven by long Œterm productivity (4 Œlags, and 6 Œlags of productivity). In addition, recent work by Snyder et al. (2019) find that in Zambia, increases in lagged multi Œyear district Œlevel agricultural productivity leads to increased household fa rm income. In addition, increases in district Œlevel agricultural productivity measures among small farms (less than 2 hectares) results in increased off Œfarm incomes. While empirical evidence confirms the presence of agricultural multiplier effects, their strength depends in part on the structure of the economy since nonŒtradable goods will generate more local economic activity (Schneider and Gugerty, 2011). Emerick (2018) shows that the labor reallocated from farming is mostly devoted to the non Œtradable sector, which is not surprising given that the labor reallocation occurs within a short period of the rainfall shock. Structural transformation is a process, however, that takes long and may require sustained productivity levels to affect the tradable sect ors. This implies that we may only detect short Œterm transitory effects, in our context, if labor reallocation is concentrated in the non Œtradeable sector. Agricultural productivity shocks may therefore only cause temporary changes in the structure of the local economy. Similar studies find that high agricultural productivity results in the non Œtradeable sector expanding the most (Foster and Rosenzweig, 2004: McMillan and Harttgen, 2014). Further, Foster and Rosenzweig (2004) show that the positive relation ship between agricultural productivity and nonfarm employment only holds for the local non Œtradeable services sector while the converse 9 holds for the tradeable sector (factory employment). In subsequent work, Foster and Rosenzweig (2007) show that the intr oduction of High ŒYielding ŒVariety (HYV) rice in India resulted in non Œfarm (factory) employment increase in areas with low agricultural productivity. They argue that this is consistent with factories which are labor seeking. This relationship between agric ultural productivity and non Œfarm factory employment highlights the importance of capital mobility. Finally, the Foster and Rosenzweig (2007) results are consistent with the modelling by Matsuyama (1992) that predicts a positive relationship between agricu ltural productivity and growth in closed economies and the opposite in open economies. Our results will therefore be influenced by the prevalence of tradeable vis ŒàŒvis non Œtradeable sectors. The context and the nature of agricultural productivity shock h as important consequences on the labor reallocation within the economy (Bustos et al. , 2016; Irz et al., 2001; Schneider and Gugerty, 2011). Bustos et al. (2016) investigates the impacts of factor Œbiased agricultural technical change in the case of Brazil in an open economy set up. Brazil experienced technical changes stemming from introduction of genetically engineered soybeans (labor saving) and adoption of double planting of maize (land saving). They find that soybean Œgrowing areas experienced rapid grow th in employment and reduction in wages in the industrial sector. On the other hand, maize growing regions experienced the opposite with labor intensity in farming increasing, wages rising, and labor moving out of the industrial sector. These results, howe ver, depend on the strength in complementarities between factors (weak complementarities weaken the results), and labor immobility across regions. Rural towns and urban areas play an important role in structural transformation. Shilpi and Emran (2016), usi ng rainfall as an instrument for agricultural productivity, find a significant positive effect of agricultural productivity growth on growth of informal (small Œscale), 10 manufacturing and skilled services employment, mainly in education and health services. For formal employment, the effect of agricultural productivity growth on employment is found to be largest in the samples that include urban areas and rural towns compared with rural areas alone. Agricultural productivity growth is found to induce structur al transformation within the services sector with employment in formal/skilled services growing at a faster pace than that of low skilled services. These findings suggests that the growth and expansion of rural towns and urban centers can facilitate struct ural transformation. Such findings motivate our assumption that the youth, who are relatively more skilled than the old, will selectively move out of agriculture into the non Œfarm sector as it expands following productivity shocks. Structural transformatio n is particularly important for developing countries where there is a huge labor productivity gap between the farm and the non Œfarm sector. Structural transformation can thus lead to large development gains by reallocating labor away from agriculture espec ially in developing countries where agricultural share of employment is very high (Gollin et al, 2014). The persistence in these gaps can be partly explained by institutional quality, labor mobility, and selection on unobservable skill (Gollin et al., 2014 , Lagakos and Waugh, 2013). An example of institutional failures concerns insecure property rights particularly on land tenure. Gottlieb and Grobovıek (2019) show that, in Ethiopia, communal land ownership weakens individual land rights and the land rental market. This in turn distorts the labor allocation between highly skilled individuals and low Œskilled individuals, and between land rich and land Œpoor individuals. Several studies estimate large increases in agricultural productivity if wedges in land all ocation were removed in order to shift land from unskilled to skilled farmers in Malawi, China and Ethiopia (Restuccia and Santaeulàlia ŒLlopis, 2017; Adamopoulos et al., 2017; Chen et al., 2017). Adamopolous et al (2017) further argue that misallocation of land leads to misallocation of 11 workers between sectors. Taken together, these studies indicate that structural transformation will release labor into the non Œfarm sector from the segment of the population that has skills needed in the non Œfarm sector. The se individuals are likely to be younger with higher education levels ŒŒ Foster and Rosenzweig (2007) show that agricultural productivity resulted in out Œmigration by the highly educated. Such a movement of labor will mitigate some of the distortion inheren t in many countries where land markets are poorly developed or institutionally restricted communally or by the government. Land is a major factor of production in agriculture. However, land market frictions, population growth and intense land subdivision is threatening the future of agriculture as sustainable enterprise and source of employment (Jayne, Mather, and Mghenyi, 2010; Muyanga and Jayne 2014). Kosec et al. (2018) show that in Ethiopia, land market frictions affect migration and employment decisio ns. They find a negative relationship between expected land inheritance and migration and non Œfarm employment. This effect is strongest in areas with low land rental activity and is primarily driven by the youth, and males. These findings indicate that dis tortion in factor markets can result in factor markets dictating labor allocation across sectors in an inefficient manner. The factor markets failure imply that low Œskilled youth with little land access will find themselves in the non Œfarm sector while som e high Œskilled youth may remain in farming. In addition, individuals with comparative advantage in farming may be locked out of farming all together. Some of the studies above document outcomes that differ by gender. We believe that these gendered outcomes are driven by difference in skills, human capital, and incomes. Qian (2008) provides a useful empirical example that motivates our analysis along gender dimension. Qian (2008) finds that in post ŒMao China increases in sex Œspecific agricultural income had sex Œspecific 12 outcomes on survival and education. Specifically, holding total household income fixed, female income improved survival rates for girls, while male income worsened survival rates for girls. In addition female income increased educational attai nment of all children, while male income decreases educational attainment for girls with no impact on boys. Foster and Rosenzweig (2007) find that agricultural productivity resulted in selective out Œmigration by gender and education attainment. Males, and the highly educated, were more likely to migrate. Given the prevalence of gender Œspecific difference in human capital investment in many developing countries, women are more likely to have fewer skills and hence have a lower comparative advantage in the no nŒfarm sector. Male household members in sub ŒSaharan Africa commonly control land and this implies that increases in agricultural income is likely to accrue to male members. The little land access among female household members diminishes agricultural inco me from increased agricultural productivity. This, in addition to the low capital accumulation by female household members, imply little opportunities in the non Œfarm (high skill) sector and low returns to migration. Consequently, female members will gain little from agricultural driven structural transformation. III. THEORY AND THE MODEL In this section, I briefly abstract and motivate a few ways in which agricultural productivity can reallocate labor across economic sectors and influence sorting by age and gen der, before discussing the focus of the model. First, agricultural productivity can relax financial constraints and consequently encourage migration in search of better opportunities. Individuals may desire to migrate from rural areas in search of better Œpaying jobs (typically non Œfarm) in urban areas. However, migration entails costs that can be overcome due to income increases from positive agricultural shocks. To motivate the age and gender dynamics, we assume that the non Œfarm sector requires skills and human capital that are largely endowed to younger individuals and likely more 13 endowed to male than females. This differential endowment in human capital implies the youth (relative to the old), and males (relative to women) have a comparative advantage in nonŒfarm jobs while the older persons (and women) have a comparative advantage in farming. This distribution of human capital endowment should result in sorting of individuals across the farm and the non Œfarm sector. However, the extent of sorting may be limited by other market failures, such as financial constraints and factor misallocation, which may inhibit an efficient sorting. Therefore, young and old individuals may end up misallocating their labor across sectors due to push factors. For instance, financial constraints may lead to excessive presence of youth in farming if the youth cannot invest in the non Œfarm sector or migrate to participate in urban employment. In addition, factor market failures may imply that the older individuals may end up enga ging in suboptimal participation in farming in areas where tenure security is not guaranteed. A positive productivity shock may thus alleviate such land pressures and misallocation by allowing the youth to migrate, or enter the non Œfarm sector while the ol d move into or expand activities in farming. A similar analogy follows along gender dimensions. The model we develop below does not account for migration, financial constraints, and factor market failures. Second, the inherent comparative advantage that is differential between age cohorts can result in differential sorting in the local economy in the short term even without the expansion of the non Œfarm sector. An increase in agricultural productivity will increase demand for agricultural labor and subseque ntly an increase in wages. Assuming that wages are equalized across the local economy, local wages rise and labor supply increases in the agricultural sector. The increase in agricultural labor may lead to sorting. Specifically, agricultural productivity i ncreases will disproportionately attract the labor of individuals with a comparative advantage in agriculture (the old) and less labor from those with comparative advantage in the non Œfarm sector. Overall, while 14 labor supply in agriculture increases, this labor will be largely supplied by the old. On the non Œfarm sector, young individuals reallocate little labor to agriculture while filling the vacancies left by the old in the non Œfarm economy. The result is more older people engaging in the farming sector compared to younger individuals. In other words, a positive agricultural shock results in a higher rate of departure of older individuals from the non Œfarm sector compared to the youth. These types of movements may be transitory in nature if push factors d ictate labor allocation. Once again, we do not model this type of outcome. Instead, we focus on the longer term where agricultural productivity multipliers leads to the reduction in agricultural employment and an increase in non Œfarm sector employment. Our model is motivated by the fact that agriculture may play a very important role in advancing the non Œfarm sector through agricultural income multipliers that result in the growth of the non Œfarm sector. Agricultural productivity sh ocks increase household i ncomes . As incomes grow, the elasticity of demand for food becomes less than one following Engel™s Law . Households start spending on non Œfarm goods and services. This demand can result in the creation and expansion of the non Œfarm sector. With this expansi on come s jobs that, we argue, require higher skills and education levels. Since we assume the youth to be endowed with higher levels of human capital, the non Œfarm sector expansion will draw away young individuals from farming. As the youth leave farming, farming employment opportunities open up for older individuals both on the extensive and intensive margins. If factor markets are inefficient and factors are misallocated across sectors, the agricultural driven non Œfarm expansion may mitigate this problem and lead to outcomes that are more efficient. The overall result is the age of farmers rising with productivity and the age of non Œfarm workers declining. Therefore, observing farmers™ ages rising over time is not necessarily a bad thing as long as this is driven by agricultural productivity. On the other end, 15 declining farmer age may imply some sort of poverty trap if agriculture is persistently unproductive and thus unable to push skilled youth out of agriculture. In the next subsection, we model how agri cultural productivity shocks influences labor reallocation through income multipliers in the non Œfarm sector. We show how this in turn affects sorting of workers by age and gender. This model is a very basic, illustrative model that does not capture all th e intricacies of labor reallocation across various sectors of the economy. Theoretical Model: Agricultural productivity shocks and the non Œfarm sector expansion There are several ways to model agricultural driven transformation (Kongsamut, Rebelo, and Xie m 2001; Gollin, Parentem and Rogerson, 2002; Shilpi and Emran 2016; Emerick, 2018). We follow the Shilpi and Emran (2016) model and adapt it to fit our goals. Specifically, we restrict analysis to the case with labor mobility across sectors but not across regions, and a two Œsector model in a rural setting. Our main innovation to the model is the inclusion of heterogeneous agent types that differ in productivity levels in the non Œfarm sector but are otherwise homogeneous in the agricultural sector. This clos ely mirrors the situation in rural Africa where daily agricultural wage is the same regardless of skill level in the non Œfarm sector. Our model provides hypotheses on how agricultural productivity encourages youth to move into (or stay in) the non Œfarm sec tor while the old move into (or stay in) agriculture using a general equilibrium approach. For simplicity, we describe this process as happening in two stages. First, agricultural productivity stimulates demand for the non Œfarm goods and services that lead to expansion of the non Œfarm sector. In the second stage the youth (old) sort into the non Œfarm (farm) sector based on the comparative advantage. Our assumption is that the youth (old) have comparative advantage in the non Œfarm (farm) sector. We consider the youth as the high Œ16 skill/high Œproductivity type in the non Œfarm sector and the old as the low Œskill/low Œproductivity type in the non Œfarm sector. While both agents are equally productive in the agricultural sector, the high skill type is more productive in the non Œfarm compared to the low Œskill types. We assume the existence of two representative agents for each skill type (high Œskill/young and low Œskill/old), each participating in two sectors of the economy. Each individual is endowed with L units of la bor that can be split between the agricultural and the non Œfarm sectors. The individuals own the production of the farm good. They can also sell labor to the agricultural sector in exchange for wage w or work in the non Œfarm sector and receive wH if highly skilled, or wL otherwise. The agent derives utility from consumption of a farm good Ca and the non Œfarm good Cn. The agricultural good is the numeraire while the non Œfarm good costs p per unit. We assume the utility function is of Stone ŒGeary form. This f unctional from provides a close approximation of structural transformation in the United States (Herrendorf, Rogerson, and Valentinyi, 2013). In our model, household preferences play a significant role in driving our results. Two examples of preferences th at can generate an increase in non Œagricultural sector labor following an increase in agricultural productivity are constant elasticity of substitution (CES) and Stone ŒGeary preferences. For CES, the positive relationship between agricultural productivity and non Œfarm employment will hold if the agricultural good and the non Œagricultural good are complementary. On the other hand, Stone ŒGeary preferences with steep Engel curves will yield similar results. Other preferences such as Cobb ŒDouglas will result in independence between non Œfarm employment and agricultural productivity. In our model, t he objective function for the agent is then: max { ,} U(C,C )=vln(C)+vln(C ) s.t C+pC =I (1) where: 17 I=+w i is the agent type, high Œskill (H) or low Œskill (L), and w is the equilibrium wage for agent type i. can be seen as the subsistence requirements needed. For simplicity we assume that there is no subsistence requirement for the non Œfarm good and thus =0. I represents income and consists of agricultural profits, and labor income derived from wage activity. , are scalar parameters for utilit y weights whose sum equals to unity, +=1. Labor is assumed to be mobile across sectors in the local rural economy so that wages are equalized across the sectors in equilibrium for at least one type of worker (either for the high Œskill or the low Œskill type). Solving the above problem yields the following: C=vI+v (2) C =v(I)p (3) The agricultural sector uses land, A, and la bor to produce the farm good. The pr oduction function is defined by: = where is the productivity parameter. This is the key parameter of interest when investigating the effect of productivity on employment. Sin ce the farmer's objective is to maximize profit, the objective function is given by : max = (4) Solving first order conditions for this equation indicates that the optimal labor input demand , , in the farm sector is: =1 (5) 18 =1 The non Œfarm sector has production function with >1,=1 as the productivity paramete rs and only requires labor as its input: = if =H if = The objective of the firm is to maximize profits as follows: = The firm's optimal labor demand can derived as: =0 <[0,] = > (6) However, with positive consumption of the non Œfarm good, and the market clearing conditions that non Œfarm output must equal non Œfarm good consumption, the equilibrium non Œfarm labor demand is determined by = . Equilibrium Notice that we have two possible labor equilibria since we have two wages in t he non Œfarm sector and one wage in the agricultural sector. At equilibrium, the agricultural wage must equal one of the two non Œfarm wages. Specifically in equilibrium : = = == (7) Equilibrium 1: = = In this equilibrium, the prevailing wage rate in the non Œagricultural sector is equal to the high Œskill wage in the non Œfarm sector. The high skill workers are then indifferent between the two sectors. However, the agricultural wage is higher than the low Œskill non Œfarm wage since =19 = >= and thus the low Œskill workers engage exclusively in the agricultural sector. The equilibrium labor supply is then: [,]=[,0] if =L [,]:[0,] if = (8) All non Œfarm product is paid out to labor as the non Œfarm sector is assumed to be compet itive with zero profits. Hence : = and considering the market clearing requirement that =+==+= , this condition becomes : = where =( ) and hence after substitution : = (+2) (9) Whereas : =0 The market clearing conditions require that consumption is equivalent to output and labor demand is equal to labor supply: =+=+= =+=+= =+ (10) Letting d denote demand, so that are the nonfarm and farm labor demand of type i skill respectively, the equilibrium labor market clearing condition requires labor demand to equal labor supply : +=+ 20 +=+ Using the production first order conditions (5) and (6), the income constraint is given by: =+= 1 + (11) Combining the market clearing conditions for agricult ural good and the optimal consumption conditions: =+=+= =2 1 + +2 (12) We can simplify this further using equation (5) by noting that the aggregate agricultural output is =2, (1)1= (13) Rewriting the above equation (13) as (,)=0 and implicitly differentiating: (,)=(1)1+ =0 =(,)/(,)/ =[1 ](1) +[1 ](1) >0 (14) We can also derive the equilibrium labor allocation, starting from (12): =2 1 + +2=+2 +2 21 and after rearranging: =2 +1 Notice that since the high Œskill (young) and the low Œskill (old) face the same agricultural production function, land access, and agricultural wage, the optim al agricultural labor demand is identical so that : = Since, from production first order conditions, =(+)=(1): =(1)()=211(+) (15) is the aggregate agricultural labor demand. To derive the equilibrium farm labor allocation for each type of agent, we use the fact that in this equilibrium all individuals with low skill in the non Œfarm sector only work in the agricultural sector. This implies that the low Œskill type supplies L while the net amount of labor is supplied by the high Œskill type is: = ==(1)()=2(1)1( + ) (16 ) Similarly, we can use market Œclearing conditions for labor to pin down equilibrium labor allocation in the non Œfarm sector. Note that since all labor not devoted to agriculture goes to the nonŒfarm sector, it is sufficient to derive the equilibrium farm labor supply, which has been accomplished above. = 22 Alternatively, we can simply follow the same steps above, starting from (9) and after some algebra, we can derive: =21(1) (16 ) Note that in this equilibrium we had : =0 By differentiation, we can now inspect the effect of agricultural productivity on agricultural emp loyment : ==0 (17) while : =2(1)[1] =2(1)[1] <0 This is negative because from (14) is positive and labor is fixed at L. On the other hand: =0 and : =2(1)[1] >0 (18) And therefore, a higher agricultural productivity pushes high Œskilled labor out of the agriculture to the non Œagricultural sector . We can investigate the effect of productivity on household income: = 1+=0.5 +=+ 1 23 =1 >0 (19) In this first equilibrium, an increase in productivity results in the high skill individuals leaving agriculture and entering t he non Œfarm sector. Since we assumed, the young individuals have higher Œskills in the non Œfarm sector, a positive productivity shock will therefore result in more young people in the non Œfarm sector. On the other hand, the older individuals devote all thei r labor to the agricultural sector and an increase in agricultural productivity does not affect their labor supply. The overall effect of the departure of the young is the increase in the mean age of farmers and a decline in the mean age of those in the no nŒfarm sector. There is no change in labor supply by older individuals. This type of equilibrium is in line with our hypothesis. Finally, we show the conditions necessary for this equilibrium to hold. Note from (8) that under this first equilibrium the tot al labor supply in the farm sector is =+. Since in equilibrium labor supply must equal labor demand, we can combine this condition with the labor demand equations (5): =+ Which implies : 21 and thus the necessary con dition for equilibrium I is : 1 This equilibrium is consistent with the observed employment behavior of adults. Specifically, in Sub ŒSaharan Africa, there is a high prevalence of youth leaving agricultural employment in search of nonŒfarm sector employment. 24 On the other hand, similar analysis shows that we get the second equilibrium discussed below when: 1 Equilibrium 2: == In this equilibrium, the prevailing wage rate in the non Œagricultural s ector is equal to the low Œskill wage in the non Œfarm sector. The low Œskill workers are then indifferent between the two sectors. However, the agricultural wage is lower than the high Œskill non Œfarm wage since =>== and thus the hi ghŒskill workers engage exclusively in the non Œfarm sector and receive a higher wage . The equilibrium labor supply is then: [,]=[,]:[0,] if =L [0,] if = (20) All non Œfarm product is paid out to labor as the non Œfarm sector is assumed to be competitive with zero profits. Notice, that unlike in the previous equilibrium, there are two wages Œ the high type receive while the low type receive the equilibrium market wage of w. Hence : == = (21) and considering the market clearing requirement that : =+==+= + , this implies : =+ where =() and hence after substitution : =(+2) (22) = 25 The market clearin g conditions require that consumption is equivalent to output and labor demand is equal to labor supply: =+=+= =+=+= =+ (23) +=+ +=+ Using the production first order conditions (5) and (6), the income constraint is g iven by: += 1+ =+= 1+ Combining the market clearing conditions for agricultural good, the budget constraint, and the optimal consumption conditions: = + =+=^_ =+ =_(+)+2 =(2 1+(_+))+2 (24) We can simplify this further using equation (5) by noting that =+=2, (+)2(1 )1=2 Since marke t clearing equilibrium conditions imply = and =, we can substitute = into the equation above to get : 26 (1+)2(1 )1=2 (25) Rewriting equation (25) as (,)=0 and implicitly differentiating yields the effect of productivity on wage: (,)=(1+)2(1)1+2 =(,)/(,)/ =2[1 ](1) (1+) +2[1 ](1)/ >0 (26) We can also derive the equilibrium labor allocation, starting with (24): =2 1+(+)+2=2 +(+)+2 =(1+)+2 1 (27) Since, from production first order conditions, =(+)=(1): =(1)()=(1)1 ((1+)+2 ) (28) is the aggregate labor demand in the farm sector. To derive equilibrium labor deman d in the nonŒfarm sector, we can use market Œclearing conditions for labor. In addition, in this second equilibrium, all labor by those with high non Œfarm skills (the young) is devoted to the nonfarm sector. Note that since all labor not devoted to agricult ure goes to the non Œfarm sector, it is sufficient to derive the equilibrium farm labor supply. Therefore: = 27 ,=,=0 ==(1)1 ((1+) +2 ) ==(1)1 ((1+) +2 ) (29) By differentiation, we can now inspect the effect of agricultural productivity on employment : =(1)1 (1+)2 =2(1)1 <0 (30) This is negative because >0 is positive and labor is fixed at L. Given that = ,=>0 (30) Therefore, agricultural productivity leads to an increase in non Œfarm employment among low Œskill workers, and an overall increase in non Œfarm employment and a decl ine in agricultural employment. We can investigate the effect of productivity on household inco me: =1+=0.5 + =0.5(1+) + 1+ =0.5(1+) 1+ >0 =1+=0.5 +=0.5(1+) +1 + =(0.5(1+)1+ )>0 28 In this second equilibrium, the young always work in the non Œfarm sector. Follo wing an agricultural productivity shock, the old (low skill) reallocate labor away from agriculture towards the non Œfarm sector, with no effect on labor supply among the youth. In this case, it is possible that the mean age in the non Œfarm sector will actu ally rise over time and the differences in the mean age of workers in each sector will grow close together. However, if the level of skill needed in the non Œfarm sector is decreasing in age, then it is likely that the marginal worker that leaves the agricu ltural sector is likely to be among the youngest within the old Œgroup (say a 40 Œyear old instead of a 65 Œyear old). In this case, the mean age in the non Œfarm sector could rise slightly but at a slower pace compared to the mean age within the farm sector. This is likely to be the case because the marginal individual departing from agriculture to enter the non Œfarm sector may require high education and related skills typically found among relatively younger individuals within any age group. In summary, under the assumption that younger individuals have a comparative advantage in the non Œfarm sector, the marginal young worker always has an incentive to reallocate labor away from the farm whenever possible. On the other hand, our model yields an equilibrium in which the marginal old worker either stays in farming or reallocates labor to the non Œfarm sector. We show that under a number of assumptions, the equilibrium average age of farmers goes up with increased productivity. Outcomes along gender dimensions foll ow similar patterns as individuals try to match their skills with a particular sector. Lower skills would discourage labor reallocation. As the literature has documented several instances of women having limited human capital, we would expect their compara tive advantage to be in farming. On the margin, young women may move out of agriculture but at a slower pace compared to their male counterparts due to lack or 29 limited non Œfarm sector skills. This raises an issue of feminization of agriculture and prospect s of women being stuck in poverty and low Œproductivity activities. IV. DATA This paper utilizes data from several sources. First, we use a three Œwave panel survey from the Rural Agricultural Livelihoods Survey (RALS) covering the 2010/11, 2013/14, and the 201 7/18 agricultural years (October ŒSeptember) and the subsequent marketing years (May ŒApril of 2011/12, 2014/15, and 2018/19 respectively). The RALS data were collected in June ŒJuly 2012, 201, and 2019 by the Indaba Agricultural Policy Research Institute (IA PRI) in collaboration with the Zambia Central Statistical Office (CSO) and Ministry of Agriculture (MoA). This is a nationally representative survey of smallholder farm households covering 72 districts initial districts (from 2012) and 8 provinces. The dat aset contains the outcomes of interest and several controls variables used in our analysis. The survey covers approximately 8,000 households. Second, we use the annual Zambian Crop Forecast Surveys (CFS), from the CSO and MoA, covering the years 2006 to 20 18. This survey captures agricultural productivity is representative at the district and national level. We use this dataset to create our measures of agricultural productivity. The data are collected in late March/early April before the main harvest perio d begins in May. They are based on farmers™ expected quantity yields. We can generate productivity values that are representative at the district level. This survey also covers approximately 8,000 to 13,000 households. Third, rainfall data is from version 2.0 of Climate Hazards Center InfraRed Precipitation with Station data (CHIRPS). CHIRPS provide rainfall satellite imagery with a 0.05 resolution 30 that combine both satellite images and rainfall station data. We then process these images into monthly ra infall data values that we subsequently use in our analyses. Table 1.1 below show that summary statistics for the key variables of interest. Productivity, measured as value of all agricultural output per hectare is approximately 4,000 Kwacha (approximatel y $300 in real 2019 values). We do not calculate total factor productivity or labor productivity due to concerns about measurement errors in labor input. There is a low participation in wage activity. Older youth (25 Œ34) and the old (35 Œ65) have a particip ation rate of approximately 12% in wage/salary activity. The young youth (15 Œ24) have the lowest participation rate of 4% in wage activity and 3 % in business activity. It appears that participation in business activity increases with age given that 21% of the older youth engage in business activity compared to 31% of the old (35 Œ65). Participation in farming activities is very high, at 98%, and does not vary among the age groups. However, the intensity of farming shows some variance. Specifically, while a 70% of younger youth (15 Œ24) spend greater than 20 days in farming per agricultural season, over 90% of the older youth and the old spend more than 90 days in farming per agricultural season. A large proportion of rural households in Africa engage in bot h farm and non Œfarm activities. The summary statistics show that farming is the primary major activity, especially for the older individuals. Non Œfarm primary activity decreases with age as 37% of 15 Œ24 year olds, 11% of 24 Œ35 year olds, and only 8% of 35 Œ65 year olds engage in a primary economic activity in the non Œfarm sector. These employment measures are from the demographic module. Table 1.2 reports the different employment combinations as adults in rural areas often engage in multiple employment types . These data are from the employment module. The module contains lists of household members who earned income f rom wage/salary and/or business 31 activities. There is low reported unemployment rate of below 3 percent across all age Œgroups. Exclusive employmen t in farming is the most dominant employment type with a participation rate of 90% among the youth (15 Œ24), 67% among the older youth (25 Œ34) and 29% among the old (35 Œ65). The second most prevalent employment type involves joint farming and business activ ities. However, this type of employment is prevalent among the older youth and the old. It is uncommon for individuals to be employed in all three activities considered (farming, business and wage/salary) and very rare for individuals to participate in bus iness and/or wage/salary activities only. 32 Table 1.1: Summary Statistics (Means) Variable MeanProductivity (Kwacha/Ha): 2-Year Moving Average 4,318 Productivity (Kwacha/Ha): 4-Year Moving Average 4,579 Productivity (Kwacha/Ha): 6-Year Moving Average 4,460 15-24 25-34 35-65 Wage/Salary Activity 0.040.110.12Business Activity 0.030.210.31Farming 0.970.980.98Farming: Less than 20 days 0.260.080.05Farming: Greater than 20 days 0.700.900.93Primary Activity: Farming 0.540.840.88Primary Activity: Non-farm 0.370.110.08Relationship to Head (Base: Head) Spouse 0.040.330.37Child (own/step) 0.700.280.03Relative 0.240.110.03Unrelated 0.010.020.02Marital Status (Base: monogamously married) Polygamously/Divorced/Separated etc 0.030.180.29Never Married 0.850.240.02Highest Education Attained (Base:None) Primary 0.530.520.62Junior High 0.280.210.15Senior High 0.140.150.08College and above 0.010.030.02Head Age 50.5843.3350.61Head Marital Status (Base: monogamously married) Polygamously/Divorced/Separated etc 0.320.280.29Never Married 0.010.010.01Head Education Attained (Base: None) Primary 0.570.550.59Junior High 0.170.190.17Senior High 0.110.120.11College and above 0.040.040.03Household size 8.867.698.10Male childred 2.331.902.14Land owned (Ha) 5.835.165.67Female Head 0.180.150.14Tropical Livestock Units 4.353.273.63Cell Phone 0.720.680.68Solar/Generator 0.450.430.43HH has bank account 0.180.160.17House characteristics Cement floor 0.340.300.31Permanent wall 0.510.470.49Observations 26,98412,19123,684Means by Age Category Standard Deviation 1,590 1,658 1,539 33 Note that the employment shares in Table 1.2 may not necessarily match those in Table 1.1 due to the different modules that a re used to capture the data, and the slight variations in definitions. However, only the data from the employment module is disaggregated enough to allow for the categorizations in Table 1.2 . Table 1.2: Employ ment Combinations (Proportion) V. METHODS To test our hypotheses we implement panel fixed effects methods on pooled probit models. Specifically =+ + + + + + (31) where is the outcome of interest for individual i living in household in district during the survey period . Our primary outcomes of interest include, participation in wage activity, participation in business activity , participation in farming, number of days spent in farming, and the choice of primary activity in either farm or non Œfarm sectors. Note that these outcomes are binary in nature and hence the choice of probit models in our analyses. is a real measure of lagged agricultural productivity at the district level Œ we define productivity as the gross value of production per hectare at the district level. The productivity measure is as simple moving average (MA) of various lags lengths (2 Œyear , 4Œyear or 6 Œyear lagged average): Variable 15-24 25-34 35-65 None 0.030.020.01Farm only 0.900.670.56Farm & Business 0.030.180.29Farm & Wage/Salary 0.040.090.09Farm, Business, & Wage/Salary 0.0050.030.04Business or Wage 0.0020.010.01Observations 16,4406,86515,415Age Category 34 =1 {2,4,6} We generate and separately use each of the three measures of productivity in regressions on equation (31). and are household and individual vector contro ls respectively. and are time and district fixed effects respectively while is the error term. The individual controls include sex, marital status, relationship to household head, and highest education attained. The household c ontrols include household head characteristics (sex, marital status, highest education attainment), household size, number of male children, size of land owned, Tropical Livestock Units (TLUs), assets (phone, solar/generator, bank account), lineage, and ho use characteristics (cement floor, permanent wall, permanent roof). We use lagged productivity to address endogeneity and reverse causality. Using same period productivity measurement can bias results because some unobservable factors that affect producti vity may also affect the employment outcomes. Reverse causality can occur because increased labor intensity in the agricultural sector can increase yields and hence agricultural land productivity. However, our goal is to determine how land productivity aff ects employment. Productivity is calculated by taking the aggregate value of district harvests and dividing by the aggregate land cultivated at the district. We believe that labor reallocation across sectors takes time and requires persistent productivity levels. Therefore, our measure of agricultural productivity relies on district 2 Œyear, 4 Œyear, and 6 Œyear simple moving averages (MA) of productivity lags. As defined above, an N Œyear MA is calculated by taking the average of each district's N lags with equal weighting. However, in future analysis, we plan to experiment with using different weights in generating the productivity MA measures. We chose to focus on productivity at the district level 35 because we believe that structural transformation occurs at a regional level. Using districts for this type of analysis is not new to our paper (see Emerick (2018)). Our approach addresses a number of endogeneity concerns using household and individual controls. In addition, we mitigate identification challenges fro m unobserved factors that may drive the results by using fixed effects. Our time fixed effects control for unobservable factors common to all individuals at a specified time (wave). The district fixed effects absorb unobserved time Œinvariant characteristic s that are common to all individuals within a district. Our identifying assumption is that conditional on the observed controls, the district fixed effects, and the time fixed effects, the impact of productivity on employment outcomes is exogenous. Ideally , an instrumental variables approach would have been preferred. In the literature, rainfall is typically used as an instrument from agricultural productivity. Unfortunately, in our context different measures of rainfall did not yield strong first stage res ults. As a result, our analysis does not provide the benefits of IV techniques. VI. RESULTS Œ POOLED PROBIT WITH TIME AND DISTRICT FIXED EFFECTS Main Results In this section, we report our estimates based on a pooled probit model with time and district fixed effects. Note that all the dependent variable are binary in nature and hence the choice of probit analysis. The analysis is at the individual level but the productivity measure Œ value of yields per hectare Œ is at the district level. We perform our analy sis first using 2 Œyear lagged productivity average, then repeat the analysis using 4 Œyear lagged productivity average, and finally using 6 Œyear lagged productivity average. We use the moving averages because we believe that productivity may have different effects in the short term and in the long run . Our data limits us to a maximum of 6 lags. We estimate the main model (31) among three age categories: 15 Œ24 (youth), 36 25Œ34 (older youth), and 35 Œ65 (the old). Our goal is to test whether agricultural producti vity differentially reallocates labor based on age. The general prediction of our model is that increased agricultural productivity may increase youth participation in the non Œfarm sector while increasing participation in agriculture for the old. In the t ables, below we focus on the marginal effects of productivity shocks for each age Œcategory. Table 1.3 estimates the impact of agricultural productivity on different types of employment based on the average of the previous 2 lags of productivity. In the fir st specification, productivity has a positive but non Œsignificant effect on participation in wage/salary activities. There is no significant different in outcomes across age categories. In the second specification, a 10% increase in productivity reduces th e probability of participation in business activity among the youth by 0.3 percentage points but has no significant effect for older youth and the old. Next, we investigate the impact on farming, and intensity of farming (number of days spent on farming) , and primary economic activity . Due to data limitations resulting from the questionnaire design, the number of days spent on farming are categorical (0, between 0 and 20, greater than 20). In addition, these latter outcomes , in specifications (3) Œ (7), are only captured in the last two survey waves only. We perform our analysis for each of these categories. We define participation in farming as spending more than zero days on farming. Specification (3), consistent with our model, shows a statistically sign ificant decline in participation among the youth and older youth of 0.3 percentage points, following a 10% increase in agricultural productivity, with no effect on the old. In specification (4) , while agricultural productivity increases spending non Œzero b ut less than 20 days in farming, the corresponding marginal effect is 0.3 and 0.4 for the older youth and the old when productivity increases by 10%. However, a 10% increase in productivity decreases the likelihood of spending more than 20 days in farming for each age. While the effect diminishes 37 with rising age, the differences between age groups is not statistically significant. Together with the latter finding, this result suggest farmers reduce the intensity of farming following positive productivity sh ocks. Employment in rural sub ŒSaharan Africa tends be mixed in nature. Specifically, individuals rarely practice one type of activity but may mix both farm and non Œfarm occupations. The final two specifications focus on primary economic activity. Primary a ctivity is the activity on which a person spends the most time. We exclude the cases where the individual devoted equal amounts of time to farm and non Œfarm activities. While we find no statistically significant effects, the results suggest that increased agricultural productivity reallocates labor away from farming as a primary activity to the non Œfarm sector across all age groups. In Table 1.4 and 1.5, we repeat the analysis with a moving average of previous 4 lags and 6 lags of productivity respectively . The results become less precise and estimates are no longer statistically significant. However, a few suggestive patterns continue to hold. The youth are relatively more likely to leave business activity following positive productivity shocks. In additio n, there suggestive evidence that farmers may reduce intensity of farming from more than 20 days to less than 20 days of farming in an agricultural season. Our results are therefore more pronounced in the short term but diminish when additional productivit y lags are considered. Overall, our findings show that productivity changes have significant effects only within the recent short term (2 Œlags). We find younger youth reduce participation in business activities. In addition, as predicted by our model, youn g youth (15 Œ24) and older youth (25 Œ34) decrease participation in farming. While all age Œcategories reduce the number of days devoted to farming following favorable productivity, the effects are relatively higher among the youth and older youth. The exit o f younger youth (15 Œ24) from both business and farming employment may indicate that 38 their participation in employment activities is due to push factors. As agricultural income increase, the push factors and income constraints are relaxed and these younger youth leave employment. This is not surprising since younger youth (15 Œ24) are typically engaged in academic activities and are not yet gainfully employed. Table 1.3: Agricultural Productivity and Employment Participation by Age ŒGroup (2 ŒLags Moving Average) Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications us e all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Table 1.4: Agricultural Productivity and Employment Parti cipation by Age ŒGroup (4 ŒLags Moving Average) Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level ( 72 districts ). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. (1) (2) (3) (4) (5) (6) (7) Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm 15-240.002-0.033*-0.026**0.038-0.057**-0.0530.044(0.10)(-1.83)(-2.19)(1.43)(-2.18)(-0.95)(0.76)25-340.019-0.004-0.026*0.032*-0.048**-0.0470.036(0.65)(-0.18)(-1.67)(1.87)(-2.50)(-1.45)(1.19)35-650.0110.003-0.0130.035**-0.036*-0.0390.046(0.39)(0.17)(-0.61)(2.17)(-1.88)(-1.04)(1.36)Observations63260 63260 38982 38982 38982 38982 38982 Log Value of Yields per Ha - 2 Year MA Non-zero farming days (1) (2) (3) (4) (5) (6) (7) Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm 15-240.003-0.028-0.0080.022-0.028-0.0020.016(0.18)(-1.58)(-0.97)(0.73)(-0.83)(-0.06)(0.40)25-340.017-0.003-0.0070.015-0.021-0.0030.014(0.77)(-0.14)(-0.53)(0.57)(-0.64)(-0.10)(0.56)35-650.0040.0020.0150.015-0.0030.0160.012(0.22)(0.09)(0.97)(0.65)(-0.10)(0.52)(0.48)Observations63260 63260 38982 38982 38982 38982 38982 Log Value of Yields per Ha - 4 Year MA Non-zero farming days 39 Table 1.5: Agricultural Productivity and Employment Participa tion by Age ŒGroup (6 ŒLags Moving Average) Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use all thr ee survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Heterogeneity by Gender Our previous analysis may mask heterogeneous gender outcomes. In the following tables , 1.6 Œ 1.8, we repeat the main analysis and estimate gender effects. We do not report a number of control variables for the ease of readership but full first Œstage probit tables are available in the appendix. Table 1.6 reports results for marginal effects us ing a 2 Œyear measure of productivity. We do not detect any effects on wage/salary activities. We also find no differences in wage/salary employment between genders within any age Œcategory following productivity shocks. Agricultural productivity decreases employment participation among 15 Œ24Œyear Œold male youth, but there is no effect on females in this age category. Specifically, a 10% increase in productivity reduces the likelihood of business activity by 0.5 percent for both male youth. We see no statis tically and quantitatively significant effects among the older youth and old individuals. In farming, the overall impact of productivity increases is a decline in participation and intensity in farming (greater than 20 days). Within each age group, women are more likely to stay in farming and to spend more than 20 days in agriculture compared to their male counterparts but these (1) (2) (3) (4) (5) (6) (7) Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm 15-24-0.002-0.028*-0.0030.021-0.0210.011-0.009(-0.10)(-1.72)(-0.29)(0.58)(-0.52)(0.27)(-0.34)25-340.0130.004-0.0030.017-0.0210.0030.002(0.48)(0.22)(-0.22)(0.56)(-0.55)(0.09)(0.08)35-65-0.0040.0140.0240.0120.0040.026-0.001(-0.14)(0.68)(1.25)(0.45)(0.12)(0.72)(-0.04)Observations63260 63260 38982 38982 38982 38982 38982 Log Value of Yields per Ha - 6 Year MA Non-zero farming days 40 differences tend to be small or statistically non Œsignificant. Finally, specification (6) shows that among the youth (15 Œ24), me n are less likely to engage in farming as a primary economic activity compared to women Œ but the effects are not statistically significant. Table 1.7 repeats the analysis above using 4 Œyear lags of productivity. Though generally statistically insignifican t, the results are qualitatively consistent. We find that among the youth (15 Œ24), women are more likely to engage in business activity compared to men following increased agricultural productivity. Specifically, a 10% increase in agricultural productivity reduces the probability of male youth (15 Œ24) participating in farming by 0.5 percentage points with no effect on women business participation. However, this difference is relatively smaller in magnitude as longer lags of productivity are considered. We also find that productivity gains encourage higher participation in farming, increased intensity of farming, and choice of farming as a primary economic activity among women within each age Œgroup. These differences in employment activity are larger in farm ing than in business activity. Table 1.8 uses six years of productivity. The results here are largely imprecise but the results remain consistent for business activity. Male youth are more likely to leave business employment than female youth following inc reased agricultural productivity. The effects on the other outcomes are statistically non Œsignificant. Judging by point estimates, the overall results suggest that while increased agricultural productivity may encourage exit from business and farming acti vity among the youth, women are more likely to remain in farming compared to men. These results are similar consistent within each age group, and are in Œline with some of our model predictions. However, this evidence is not strong since differences are eit her quantitatively small or not statistically significant. 41 Table 1.6: Agricultural Productivity and Employment Participation by Age ŒGroup and Gender (2ŒLags Moving Average) Table shows probit marginal eff ects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two s pecifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Table 1.7: Agricultural Productivity and E mployment Participation by Age ŒGroup and Gender (4ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed ef fects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0. 05 *** p<0.01. (1) (2) (3) (4) (5) (6) (7) Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm 15-24 # Male 0.001 -0.054** -0.028** 0.044* -0.066** -0.075 0.058 (0.05) (-2.50) (-2.32) (1.65) (-2.45) (-1.25) (0.93) 15-24 # Female 0.003 -0.006 -0.021* 0.026 -0.042 -0.019 0.023 (0.18) (-0.39) (-1.78) (0.96) (-1.58) (-0.38) (0.44) 25-34 # Male 0.028 -0.002 -0.032* 0.028 -0.049** -0.046 0.029 (0.73) (-0.06) (-1.93) (1.54) (-2.29) (-1.36) (0.92) 25-34 # Female 0.010 -0.004 -0.017 0.037* -0.045* -0.047 0.043 (0.46) (-0.18) (-1.03) (1.69) (-1.85) (-1.33) (1.31) 35-65 # Male 0.012 0.006 -0.021 0.038** -0.046** -0.051 0.049 (0.37) (0.28) (-0.86) (2.17) (-2.41) (-1.29) (1.33) 35-65 # Female 0.009 0.001 -0.003 0.032* -0.023 -0.025 0.043 (0.40) (0.04) (-0.15) (1.68) (-1.00) (-0.64) (1.30) Observations 63260 63260 38982 38982 38982 38982 38982 Log Value of Yields per Ha - 2 Year MA Non-Zero Farming Days (1) (2) (3) (4) (5) (6) (7) Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm 15-24 # Male 0.004 -0.050** -0.012 0.028 -0.038 -0.026 0.028 (0.19) (-2.09) (-1.39) (0.92) (-1.07) (-0.63) (0.65) 15-24 # Female 0.002 -0.002 -0.002 0.012 -0.013 0.032 -0.001 (0.14) (-0.11) (-0.21) (0.40) (-0.38) (0.95) (-0.04) 25-34 # Male 0.023 -0.004 -0.011 0.015 -0.026 0.001 0.009 (0.78) (-0.16) (-0.79) (0.56) (-0.78) (0.02) (0.35) 25-34 # Female 0.010 -0.001 0.000 0.014 -0.013 -0.006 0.019 (0.60) (-0.04) (0.00) (0.51) (-0.36) (-0.19) (0.66) 35-65 # Male 0.004 0.010 0.008 0.017 -0.013 0.001 0.016 (0.18) (0.48) (0.45) (0.68) (-0.40) (0.03) (0.61) 35-65 # Female 0.005 -0.007 0.023 0.011 0.010 0.035 0.007 (0.27) (-0.35) (1.35) (0.49) (0.32) (1.01) (0.25) Observations 63260 63260 38982 38982 38982 38982 38982 Log Value of Yields per Ha - 4 Year MA Non-Zero Farming Days 42 Table 1.8: Agricultural Productivity and Employment Participation by Age ŒGroup and Gender (6ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary o utcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use all three survey waves whi le the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Employment Combinations It is typical for rural households to engage in multiple employment pursuits. An alternative means of measur ing employment outcomes is to use the combination of the three main activities: farming, business activity, and wage/salary activity. In this subsection, we divide employment categories into six distinct groups: none (no employment), farming only, farming and business activities, farming and wage/salary activities, farm and business and wage/salary activities, and, farming and/or wage/salary activities. Each individual falls into only one of these six mutually exclusive employment groups. Notice that we lum p together business and wage/salary activities because the share of individuals who only work in business or wage/salary activities is extremely low. Due to data limitations, we can only do this exercise using the last two survey waves only . Table 1.9, using 2 Œyear lags of productivity, indicates that increased productivity increases the likelihood of youth and older youth exiting all employment types. A 10% increase in (1) (2) (3) (4) (5) (6) (7) Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm 15-24 # Male -0.003 -0.050** -0.007 0.028 -0.032 -0.017 0.007 (-0.11) (-2.21) (-0.71) (0.77) (-0.77) (-0.40) (0.23) 15-24 # Female -0.001 -0.004 0.004 0.009 -0.003 0.051 -0.031 (-0.07) (-0.29) (0.34) (0.26) (-0.08) (1.42) (-1.19) 25-34 # Male 0.017 0.009 -0.007 0.019 -0.028 0.009 -0.000 (0.48) (0.33) (-0.46) (0.60) (-0.72) (0.21) (-0.02) 25-34 # Female 0.009 0.003 0.003 0.014 -0.011 -0.001 0.003 (0.42) (0.13) (0.17) (0.44) (-0.27) (-0.02) (0.12) 35-65 # Male -0.005 0.025 0.017 0.014 -0.005 0.008 0.005 (-0.17) (1.07) (0.82) (0.49) (-0.13) (0.22) (0.26) 35-65 # Female -0.002 0.003 0.030 0.009 0.016 0.047 -0.008 (-0.08) (0.13) (1.46) (0.33) (0.44) (1.21) (-0.35) Observations 63260 63260 38982 38982 38982 38982 38982 Log Value of Yields per Ha - 6 Year MA Non-Zero Farming Days 43 agricultural productivity increases the likelihood of reporting no employment by 0.25, and 0.21 percentage points for youth (15 Œ24) and older youth (25 Œ34), respectively, with no effect on the old (35 Œ65). We find no statistically significant differences in effects on farming Œonly employment between age groups. On the other hand, a 10% incr ease in agricultural productivity decreases likelihood of engaging in farming and business activities by 0.3 percentage points among youth. We find no discernible differences between age Œcategories in employment in farming and wage/salary activities. As reported in specification (5), there is a small decline in participation in all three activities among individuals, but the effect is statistically significant for youth only. Finally, we show no evidence of productivity affecting participation in business and/or wage/salary activity. Table 1.9: Agricultural Productivity and Multiple Employment Participation by Age ŒGroup (2 ŒLags Moving Average) Table shows probit marginal effects of productivity on each of th e binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Table 1.10 repeats the analysis using 4 Œyear productivity lags. In this table, we show productivity no longer increases une mployment among the youth and older youth. A 10 % increase in productivity increases the likelihood of engaging in farming only by 0.5 percentage points. Consistent with the previous table, productivity reduces the likelihood of engaging in joint farm (1) (2) (3) (4) (5) (6) None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage 15-24 0.025** 0.006 -0.033** -0.010 -0.017** 0.001 (2.34) (0.19) (-2.09) (-0.64) (-1.96) (0.23) 25-34 0.021* 0.026 -0.012 -0.032 -0.011 0.006 (1.65) (0.69) (-0.52) (-1.53) (-0.85) (1.08) 35-65 0.006 0.009 0.010 -0.028 -0.015 0.002 (0.32) (0.26) (0.55) (-1.52) (-1.51) (0.34) Observations 38982 38982 38982 38982 38777 31998 Log Value of Yields per Ha - 2 Year MA 44 and business activities but only among 15 Œ24 year olds. We find no statistically significant differences in outcomes among the other employment combinations. Under longer productivity lags (6 Œlags), in Table 1.11 , we find the results are consistent as those un der 4 Œlag productivity measures reported in Table 1.10. In addition, increased productivity now has a negative and statistically significant effect on participating in both farm and business activities for both the youth and older adults. The other results remain largely unchanged. Table 1.10: Agricultural Productivity and Multiple Employment Participation by Age ŒGroup (4 ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Table 1.11: Agricultural Productivity and Multiple Employment Participation b y Age ŒGroup (6 ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the d istrict level (72 districts). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. (1) (2) (3) (4) (5) (6) None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage 15-24 0.008 0.019 -0.038** 0.001 -0.006 0.000 (1.04) (0.83) (-2.54) (0.06) (-0.60) (0.08) 25-34 0.002 0.048* -0.025 -0.016 -0.006 0.007 (0.22) (1.74) (-1.47) (-0.61) (-0.53) (1.02) 35-65 -0.021 0.025 0.002 -0.019 -0.010 0.002 (-1.55) (1.05) (0.13) (-0.93) (-1.26) (0.24) Observations 38982 38982 38982 38982 38777 31998 Log Value of Yields per Ha - 4 Year MA (1) (2) (3) (4) (5) (6) None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage 15-24 0.005 0.040 -0.050*** 0.003 -0.006 -0.002 (0.55) (1.40) (-3.00) (0.14) (-0.61) (-0.53) 25-34 0.000 0.072** -0.043** -0.016 -0.006 0.004 (0.03) (2.04) (-2.27) (-0.51) (-0.48) (0.41) 35-65 -0.026 0.036 -0.006 -0.018 -0.011 -0.005 (-1.50) (1.08) (-0.39) (-0.72) (-1.16) (-0.37) Observations 38982 38982 38982 38982 38777 31998 Log Value of Yields per Ha - 6 Year MA 45 Employment Combinati ons by Gender In this sub Œsection, we repeat the preceding analysis but explore heterogeneous effects by gender. Starting with Table 1.12, we find that within each age category, men are more likely than women to be unemployed following an increase in agric ultural productivity. However, the differences tend to be small in magnitude. We find no significant impact of productivity on the likelihood of engaging in farming only both within and across age groups. The table also shows that young males (15 Œ24) tend to reduce likelihood of participation in both farming and business activities following productivity shocks. Productivity tends to reduce the likelihood of participation in three economic activities Œ farming, business, and wage/salary Œ but the coefficien ts are generally statistically insignificant and there are no heterogeneous outcomes among genders. In the remaining employment combinations, we find no statistically significant effects. Table 1.13 uses productivity measures derived from 4 Œyear lags. We show, in the first column, that increases in agricultural productivity generally reduces the likelihood of unemployment among old women. Our results remain consistent for those pursuing farming only. The results on joint farm and business employment remain consistent with young men (15 Œ24) being more likely to abandon this joint activity as productivity increases. In the remaining employment combinations, we find no significant differences in outcomes. These results change a little but generally remain con sistent even when we consider longer lags of productivity Œ as seen in Table 1.14. One notable difference is that productivity encourages participation in farming only for older youth with marginal effect being slightly higher for men. In addition, older y outh males are likely to reduce joint employment in farming and business. 46 Table 1.12: Agricultural Productivity and Multiple Employment Participation by Age ŒGroup and Gender (2 ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data li mitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Table 1.13: Agri cultural Productivity and Multiple Employment Participation by Age ŒGroup and Gender (4 ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, an d, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatis tics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. (1) (2) (3) (4) (5) (6) None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage 15-24 # Male 0.027** 0.021 -0.054*** -0.018 -0.021 0.003 (2.41) (0.65) (-2.81) (-0.96) (-1.60) (0.76) 15-24 # Female 0.023** -0.015 -0.004 -0.001 -0.012* -0.002 (2.03) (-0.48) (-0.25) (-0.05) (-1.72) (-0.84) 25-34 # Male 0.025* 0.031 -0.022 -0.043 -0.015 0.011* (1.69) (0.72) (-0.94) (-1.49) (-0.94) (1.75) 25-34 # Female 0.016 0.020 -0.003 -0.021 -0.007 0.002 (1.20) (0.51) (-0.13) (-1.17) (-0.59) (0.35) 35-65 # Male 0.021 0.012 0.008 -0.037 -0.017 0.001 (0.93) (0.33) (0.36) (-1.54) (-1.50) (0.10) 35-65 # Female -0.005 0.004 0.013 -0.019 -0.013 0.003 (-0.27) (0.12) (0.63) (-1.27) (-1.28) (0.55) Observations 38982 38982 38982 38982 38777 31998 Log Value of Yields per Ha - 2 Year MA (1) (2) (3) (4) (5) (6) None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage 15-24 # Male 0.011 0.033 -0.059*** -0.005 -0.004 0.002 (1.40) (1.14) (-3.06) (-0.22) (-0.29) (0.49) 15-24 # Female 0.003 0.003 -0.009 0.007 -0.007 -0.002 (0.35) (0.13) (-0.59) (0.48) (-1.14) (-0.50) 25-34 # Male 0.004 0.054 -0.031 -0.024 -0.010 0.012 (0.36) (1.56) (-1.45) (-0.71) (-0.82) (1.46) 25-34 # Female -0.002 0.042 -0.020 -0.008 -0.001 0.004 (-0.16) (1.44) (-0.99) (-0.37) (-0.13) (0.54) 35-65 # Male -0.006 0.021 0.006 -0.027 -0.011 0.001 (-0.41) (0.67) (0.34) (-1.01) (-1.17) (0.05) 35-65 # Female -0.033** 0.029 -0.002 -0.010 -0.009 0.004 (-2.16) (1.32) (-0.15) (-0.65) (-1.23) (0.49) Observations 38982 38982 38982 38982 38777 31998 Log Value of Yields per Ha - 4 Year MA 47 Table 1.14: Agricultural Productivity and Multiple Employment Participation by Age ŒGroup and Gender (6 ŒLags Moving Average) Table shows probit marginal eff ects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two s pecifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Sensitivity to Functional Forms Œ Linear Probability Models (LPM) We repeat the analysis in Table 1.3 Œ 1.5 to check whether our results are sensitive to the functional forms employed. Note that to obtain the coefficients for each age category, one must add the main productivity coefficient to the marginal co Œefficient for each age category in the following tables . Table 1.15 repeats the analysis in Table 1.3. Consistent with initial results, we find no effect of productivity on wage/salary employment. We find that younger youth decrease participation in business activity while older youth and the old increase participation in business activities. While both younger and older youth decrease participation in farming following agricultural productivity increases, the old experience no impact of productivity on farming participation (consistent wi th our previous results). Specifications (4) and (5) show that while all age categories reduce intensity in farming, the greatest decline is among the youth relative to the old. Specifications (6) and (7) (1) (2) (3) (4) (5) (6) None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage 15-24 # Male 0.008 0.059 -0.077*** -0.002 -0.005 -0.001 (0.91) (1.62) (-3.61) (-0.09) (-0.33) (-0.17) 15-24 # Female -0.000 0.018 -0.014 0.008 -0.006 -0.004 (-0.01) (0.72) (-0.83) (0.47) (-0.98) (-0.96) 25-34 # Male 0.001 0.083* -0.050** -0.023 -0.011 0.009 (0.10) (1.90) (-2.11) (-0.59) (-0.80) (0.82) 25-34 # Female -0.002 0.063* -0.035 -0.008 -0.000 0.000 (-0.15) (1.75) (-1.59) (-0.32) (-0.03) (0.03) 35-65 # Male -0.011 0.033 -0.003 -0.028 -0.012 -0.008 (-0.55) (0.80) (-0.13) (-0.83) (-1.11) (-0.54) 35-65 # Female -0.037** 0.038 -0.009 -0.008 -0.009 -0.001 (-1.99) (1.31) (-0.48) (-0.44) (-1.11) (-0.08) Observations 38982 38982 38982 38982 38777 31998 Log Value of Yields per Ha - 6 Year MA 48 show that productivity increases the likelihood of older individuals engaging in farming as a primary activity and reducing the likelihood of engaging in non Œfarm primary activity. Results on wage/salary and business activities remain consistent once longer lags are considered (Table 1.16 Œ 1.17). The imp acts of productivity on farming participation weaken among the youth but strengthen among the old. Impact s of productivity on intensity weakens for the youth but the old increase intensity of farming in longer lags. Results remain statistically non Œsignifi cant for the youth on primary activity. However, with longer lags of productivity, older youth report engaging in farming as a primary activity while older individuals report an even higher likelihood of engaging in farming as a primary activity. Overall, while we observe little differences in results over longer lags, we find that our results are quantitatively and qualitatively consistent when we employ LPM techniques. Table 1.15: Agricultural Productivity a nd Employment Participation by Age ŒGroup Œ LPM (2 ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed eff ects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.0 5 *** p<0.01. Wage/Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Productivity: Log Value of Yields per Ha - 2 Year MA Productivity 0.007 -0.050*** -0.029*** 0.065** -0.093*** -0.080 0.072 (0.37) (-2.82) (-2.78) (2.19) (-2.89) (-1.35) (1.18) 25-34 Year Olds # Productivity 0.005 0.061*** 0.018** -0.044 0.062** 0.058 -0.055 (0.46) (3.81) (2.34) (-1.57) (2.08) (1.57) (-1.64) 35-65 Year Olds # Productivity -0.005 0.085*** 0.033*** -0.053** 0.086*** 0.076** -0.057* (-0.46) (3.82) (4.01) (-2.24) (3.11) (2.31) (-1.93) Observations 63260632603898238982389823898238982Non-Zero Farming Days 49 Table 1.16: Agricultural Productivity and Employment Participation by Age ŒGroup Œ LPM (4 ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Table 1.17: Agricultural Productivity and Employment Participation by Age ŒGroup Œ LPM (6 ŒLags Moving Average) Table shows probit marginal effects of productivity on each of the binary outcomes outlined above. Includes household and individual controls, and, survey wave and district fixed effects. Errors are clustered at the district level (72 d istricts). Due to data limitation, the first two specifications use all three survey waves while the remaining specifications use only the latest two waves. TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. VII. GENERAL DISCUSSION AND DIRECTION FOR FU TURE RESEARCH Our main aim was to test our hypothesis that agricultural driven agricultural driven structural transformation would have different effects on the youth compared to the old, and investigate respective gender outcomes. Previous literature usua lly stops at testing for evidence of structural transformation and does not go further at investigating the impact for various demographic groups. Wage/Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Productivity: Log Value of Yields per Ha - 4 Year MA Productivity 0.010 -0.056*** -0.016* 0.055* -0.071* -0.042 0.049 (0.65) (-2.94) (-1.76) (1.76) (-1.93) (-1.12) (1.23) 25-34 Year Olds # Productivity -0.002 0.066*** 0.015* -0.060** 0.075*** 0.070** -0.063** (-0.22) (4.79) (1.83) (-2.34) (2.77) (2.11) (-2.09) 35-65 Year Olds # Productivity -0.016 0.091*** 0.030*** -0.067*** 0.097*** 0.089*** -0.069** (-1.26) (4.16) (3.88) (-2.76) (3.53) (3.07) (-2.51) Observations 63260632603898238982389823898238982Non-Zero Farming Days Wage/Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Productivity: Log Value of Yields per Ha - 6 Year MA Productivity 0.004 -0.063*** -0.010 0.058 -0.068 -0.034 0.031 (0.22) (-3.71) (-1.02) (1.59) (-1.62) (-0.91) (1.12) 25-34 Year Olds # Productivity 0.004 0.082*** 0.012 -0.064** 0.076** 0.075** -0.066** (0.33) (5.96) (1.37) (-2.33) (2.63) (2.13) (-2.14) 35-65 Year Olds # Productivity -0.013 0.122*** 0.030*** -0.076*** 0.106*** 0.098*** -0.072** (-0.99) (5.97) (3.66) (-2.79) (3.48) (3.18) (-2.59) Observations 63260632603898238982389823898238982Non-Zero Farming Days 50 Structural transformation is a gradual process and hence we tested the impact of productivity over varied du rations. Our findings show some evidence of differences between the youth and the old in labor reallocation. However, these differences are not very large and tend to be transitory in nature. We do not find strong evidence that increased agricultural produ ctivity releases young labor into the non Œfarm sector while leaving/drawing old labor in the agricultural sector. We identify a few differences along gender dimensions. Specifically, young males are more likely to exit business activities as agricultural p roductivity increases. We also find suggestive evidence that men are more likely to exit or reduce days devoted to farming in response to favorable agricultural outcomes. However, these differences tend to be quantitatively small. Our results are similar t o those in recent work by Emerick (2018) who finds labor reallocation to be transitory in India following agricultural productivity shocks. While we do not find effects in the longer term, we believe this is partly explained by context. Structural transfor mation requires additional factors such as stronger institutions, markets, and credit access among others. As a developing country, Zambia still faces many economic challenges that may inhibit the process of structural transformation. An alternative explan ation is that productivity have to reach certain thresholds before the positive spillovers in the non Œfarm sector can be achieved. The reliance on rainfed agriculture and low fertilizer usage in Zambia and in sub ŒSaharan Africa may result in lower overall agricultural productivity that keeps the economy from reaching these productivity thresholds and unleashing expansion of the non Œfarm sector and the accompanying economic transformation. Another potential explanation is that, while we try to capture long Œterm productivity effects, our results may be driven by transitory productivity shocks and hence transitory responses are optimal. 51 There are a few potential lines for future research to improve on current work. First, future work can explore the possibili ty that agricultural productivity induces structural transformation only beyond certain thresholds. Such work can use spline and quantile regression techniques to measure effects based on productivity levels. Second, while our probit pooled panel fixed eff ects model addresses some of the common identification challenges, we cannot rule out the presence of other omitted factors that can bias the results. In addition, it is likely that only low productivity individuals will be reallocating labor from one sect or to another and thus mechanically increasing productivity over time. For instance if individuals with low productivity in agriculture exit, then the mean productivity in the agricultural sector increases. Our instrumental variable estimates using rainfal l shocks did not yield satisfactory results and are thus not reported. Part of this failure is likely driven by the fact that rainfall shocks are transitory in nature and thus are not appropriate instruments for long Œterm productivity. Instruments that shi ft total factor productivity in agriculture like the HYV rice in Asia and genetically engineered soybeans in Brazil would be ideal. The introduction of orange Œfleshed sweet potato in Zambia provides a candidate instrument Œ but unfortunately, take up has b een dismal. Finally, future work should consider using alternative measures of productivity. Specifically, conversion of yields into nominal values can induce measurement error that may lead to attenuation bias. One such measure is yields per hectare as a measure or agricultural productivity. If this approach yields differences with our results, then it will indicate that conversion of yields to nominal values requires careful approaches. However, if the results are consistent, then it will imply that when price data is absent or poor, using yields as a measure of productivity may be sufficient to yield unbiased estimates. One candidate is maize yields. Maize is an ideal crop because over the study period between 85% and 90% of the farmers cultivated the cr op (share of 52 cultivated hectares devoted to m aize is between 55% Œ60%). However, a robust measure will require approaches that apportion yields from other crops into the productivity measure. VIII. CONCLUSION This paper investigated heterogeneous movement of labo r between the agricultural and the non Œagricultural sector. We created a model that hypothesizes that increased agricultural productivity may drive youth from the farm into the non Œfarm sector. The consequence of this labor movement is concentration of you nger workers in the non Œfarm sector and an aging farmer population. The overall results suggest that while increased agricultural productivity may encourage exit from business and farming activity among the youth, women are more likely to remain in farming compared to men. These results are consistent within each age group, and are in Œline with some of our model predictions. While we find evidence of differential labor reallocation by age and gender, our empirical tests do not provide sufficient evidence to support our hypothesis. We believe that our results may be driven by the lack of a robust environment to facilitate structural transformation. Specifically, agricultural productivity remains low in Zambia while the institutional environment and market fri ctions may limit the ability of agriculture to be an engine of economic growth. Another potential explanation is that, while we try to capture long Œterm productivity effects, our results may be driven by transitory productivity shocks and hence transitory responses are optimal. Despite our current results, we believe that our paper provides a valuable new extended model that can be used to explore similar topics in different contexts. 53 APPENDIX 54 Table A 1.1: Agricultural Productivity and Employment Participation Œ First Stage Probit (2 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level ( 72 districts ). Due to data limitation, the first two specifications use three survey waves while the remaining specifications use the latest two waves. Wage/Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Log Value of Yields per Ha - 2 Year MA 0.019 -0.208* -0.579** 0.164 -0.245** -0.173 0.157 (0.10) (-1.82) (-2.21) (1.43) (-2.17) (-0.94) (0.76) 25-34 Year Olds # Log Value of Yields per Ha - 2 Year MA 0.108 0.190*** 0.131 0.051 -0.027 -0.033 0.038 (1.42) (2.68) (0.99) (0.45) (-0.25) (-0.30) (0.35) 35-65 Year Olds # Log Value of Yields per Ha - 2 Year MA 0.057 0.222*** 0.433*** 0.050 0.071 0.019 0.067 (0.65) (2.85) (3.95) (0.71) (0.93) (0.23) (0.82) 25-34 Year Olds -0.628 -1.195** -1.244 -0.856 0.562 0.765 -0.872 (-0.98) (-2.02) (-1.13) (-0.89) (0.62) (0.85) (-0.95) 35-65 Year Olds -0.247 -1.388** -4.023*** -0.770 -0.420 0.173 -0.971 (-0.34) (-2.13) (-4.43) (-1.31) (-0.66) (0.25) (-1.41) Female -0.382*** -0.193*** -0.010 0.096*** -0.090*** -0.052** 0.033 (-9.77) (-7.32) (-0.21) (4.68) (-4.51) (-2.15) (1.55) Relation to HH Head (base: Head) Spouse -0.346*** -0.341*** 0.096 -0.254*** 0.230*** 0.231*** -0.280*** (-10.11) (-11.84) (1.55) (-5.81) (5.56) (6.62) (-6.66) Child (own/step) -0.404*** -0.867*** -0.601*** 0.188*** -0.375*** -0.198*** 0.107** (-8.02) (-21.52) (-7.29) (3.60) (-6.91) (-4.54) (2.24) Relative -0.589*** -1.054*** -0.571*** 0.174*** -0.353*** -0.150*** 0.064 (-11.89) (-29.38) (-6.48) (2.82) (-5.71) (-3.11) (1.23) Unrelated -0.217** -0.635*** 0.509*** 0.007 0.066 0.259* -0.232 (-2.28) (-7.86) (3.08) (0.08) (0.87) (1.65) (-1.46) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.093*** -0.062* -0.104 -0.011 -0.041 -0.033 -0.035 (-2.80) (-1.72) (-1.55) (-0.21) (-0.93) (-0.90) (-0.86) Never Married -0.191*** -0.433*** -0.240*** 0.532*** -0.535*** -0.679*** 0.637*** (-5.13) (-17.40) (-4.16) (10.91) (-12.87) (-15.48) (12.51) Education Attained:base-None Primary 0.026 0.151*** 0.333*** 0.130** 0.030 -0.031 0.184*** (0.64) (4.70) (4.48) (2.29) (0.53) (-0.59) (3.59) Junior High 0.074 0.238*** 0.453*** 0.208*** -0.012 -0.113* 0.284*** (1.52) (6.43) (5.70) (3.58) (-0.20) (-1.82) (4.62) Senior High 0.251*** 0.196*** 0.372*** 0.382*** -0.191*** -0.278*** 0.443*** (4.40) (4.99) (4.11) (6.29) (-3.17) (-4.61) (7.47) College and above 1.186*** -0.066 0.014 0.878*** -0.748*** -0.984*** 1.123*** (14.02) (-0.87) (0.12) (8.44) (-8.26) (-10.18) (11.20) Log Head Age -0.176*** -0.264*** -0.060 -0.152*** 0.140** 0.212*** -0.221*** (-4.03) (-6.83) (-0.65) (-2.63) (2.42) (4.24) (-4.52) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.040 0.055 0.025 -0.044 0.049 0.019 -0.003 (1.03) (1.60) (0.32) (-1.16) (1.18) (0.68) (-0.10) Never Married 0.049 0.405*** 0.071 -0.330** 0.307** 0.431*** -0.348*** (0.48) (3.94) (0.27) (-2.28) (2.13) (4.47) (-3.13) Head Education: base-None Primary -0.073** 0.007 0.005 0.017 -0.009 0.001 -0.013 (-2.28) (0.23) (0.07) (0.41) (-0.21) (0.04) (-0.35) Junior High -0.089* 0.005 -0.076 0.087* -0.094* -0.080* 0.071 (-1.74) (0.13) (-0.83) (1.71) (-1.86) (-1.72) (1.44) Senior High -0.112** -0.070* -0.086 0.054 -0.062 -0.050 -0.009 (-1.99) (-1.68) (-0.85) (1.04) (-1.19) (-1.21) (-0.21) College and above -0.074 -0.048 -0.142 0.200** -0.233*** -0.257*** 0.168** (-0.90) (-0.73) (-0.90) (2.33) (-2.66) (-3.50) (2.44) Log Household Size 0.011 -0.023 -0.083 0.037 -0.049 -0.018 0.001 (0.38) (-0.93) (-1.23) (1.08) (-1.46) (-0.50) (0.02) Log Male Children -0.057*** -0.065*** 0.096*** -0.001 0.030 0.018 0.005 (-2.60) (-4.31) (2.71) (-0.03) (1.14) (0.89) (0.20) Log Land Owned (Ha) -0.042*** 0.008 0.033*** -0.035*** 0.041*** 0.034*** -0.031*** (-8.47) (1.41) (2.75) (-3.73) (4.32) (4.59) (-4.59) Female Head 0.142*** 0.226*** 0.005 -0.061 0.062 0.080** -0.071* (4.35) (6.79) (0.06) (-1.36) (1.31) (2.16) (-1.79) Log TLUs -0.159*** -0.010 0.058** -0.029** 0.039*** 0.050*** -0.037** (-11.90) (-1.05) (2.11) (-2.11) (3.19) (2.88) (-2.25) Cell Phone -0.046** 0.095*** 0.001 0.048* -0.043 -0.060*** 0.075*** (-1.96) (4.73) (0.03) (1.68) (-1.51) (-2.63) (3.15) Solar/Generator -0.094*** 0.077*** 0.043 -0.024 0.031 0.035 -0.043* (-4.86) (3.45) (0.83) (-0.81) (0.97) (1.30) (-1.91) HH Bank Account 0.237*** -0.013 -0.118** 0.054 -0.076** -0.096*** 0.098*** (8.26) (-0.51) (-2.19) (1.50) (-2.12) (-3.20) (3.57) House Type Cement Floor -0.068*** -0.003 -0.180*** 0.072*** -0.112*** -0.118*** 0.093*** (-2.72) (-0.13) (-3.62) (2.97) (-4.69) (-4.53) (3.26) Permanent Roof -0.097*** -0.027 0.015 0.037 -0.031 -0.050* 0.053* (-4.39) (-1.27) (0.26) (1.08) (-0.94) (-1.70) (1.87) Observations 63260 63260 38982 38982 38982 38982 38982 55 Table A 1.2: Agricultural Productivity and Employment Pa rticipation Œ First Stage Probit (4 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the fi rst two specifications use three survey waves while the remaining specifications use the latest two waves. Wage/Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Log Value of Yields per Ha - 4 Year MA 0.028 -0.178 -0.182 0.095 -0.121 -0.008 0.056 (0.18) (-1.57) (-0.98) (0.73) (-0.83) (-0.06) (0.40) 25-34 Year Olds # Log Value of Yields per Ha - 4 Year MA 0.082 0.165** 0.067 0.002 0.004 -0.006 0.019 (1.01) (2.48) (0.50) (0.02) (0.03) (-0.06) (0.18) 35-65 Year Olds # Log Value of Yields per Ha - 4 Year MA 0.003 0.186** 0.356*** -0.006 0.107 0.071 0.003 (0.03) (2.34) (3.34) (-0.08) (1.39) (0.79) (0.03) 25-34 Year Olds -0.416 -0.997* -0.713 -0.452 0.304 0.545 -0.715 (-0.60) (-1.78) (-0.63) (-0.50) (0.34) (0.60) (-0.81) 35-65 Year Olds 0.196 -1.099* -3.393*** -0.302 -0.724 -0.261 -0.439 (0.24) (-1.65) (-3.80) (-0.48) (-1.13) (-0.35) (-0.60) Female -0.382*** -0.192*** -0.008 0.095*** -0.090*** -0.052** 0.033 (-9.76) (-7.32) (-0.18) (4.65) (-4.47) (-2.14) (1.54) Relation to HH Head (base: Head) Spouse -0.346*** -0.341*** 0.093 -0.253*** 0.229*** 0.230*** -0.279*** (-10.09) (-11.84) (1.50) (-5.80) (5.53) (6.56) (-6.63) Child (own/step) -0.404*** -0.868*** -0.602*** 0.187*** -0.374*** -0.197*** 0.106** (-8.00) (-21.62) (-7.33) (3.60) (-6.92) (-4.52) (2.21) Relative -0.589*** -1.054*** -0.573*** 0.174*** -0.353*** -0.149*** 0.064 (-11.89) (-29.53) (-6.51) (2.82) (-5.71) (-3.09) (1.22) Unrelated -0.221** -0.632*** 0.523*** -0.007 0.081 0.272* -0.244 (-2.33) (-7.79) (3.05) (-0.08) (1.11) (1.68) (-1.50) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.096*** -0.063* -0.105 -0.013 -0.040 -0.030 -0.039 (-2.85) (-1.72) (-1.57) (-0.26) (-0.88) (-0.82) (-0.95) Never Married -0.192*** -0.431*** -0.241*** 0.530*** -0.534*** -0.678*** 0.636*** (-5.14) (-17.47) (-4.15) (10.88) (-12.81) (-15.61) (12.51) Education Attained:base-None Primary 0.027 0.152*** 0.336*** 0.132** 0.029 -0.033 0.185*** (0.67) (4.71) (4.58) (2.32) (0.52) (-0.61) (3.61) Junior High 0.075 0.239*** 0.457*** 0.210*** -0.013 -0.115* 0.285*** (1.54) (6.41) (5.82) (3.60) (-0.21) (-1.84) (4.62) Senior High 0.252*** 0.197*** 0.376*** 0.383*** -0.190*** -0.279*** 0.444*** (4.41) (4.99) (4.19) (6.31) (-3.17) (-4.60) (7.44) College and above 1.187*** -0.064 0.017 0.878*** -0.746*** -0.984*** 1.123*** (14.06) (-0.85) (0.15) (8.49) (-8.30) (-10.16) (11.18) Log Head Age -0.176*** -0.264*** -0.060 -0.153*** 0.141** 0.212*** -0.221*** (-4.01) (-6.82) (-0.65) (-2.64) (2.43) (4.22) (-4.52) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.041 0.055 0.029 -0.044 0.050 0.019 -0.003 (1.07) (1.60) (0.38) (-1.17) (1.20) (0.69) (-0.11) Never Married 0.049 0.407*** 0.078 -0.330** 0.308** 0.432*** -0.348*** (0.48) (3.96) (0.29) (-2.28) (2.13) (4.51) (-3.16) Head Education: base-None Primary -0.074** 0.007 0.005 0.017 -0.008 0.002 -0.014 (-2.31) (0.22) (0.07) (0.40) (-0.20) (0.06) (-0.36) Junior High -0.090* 0.005 -0.077 0.086* -0.093* -0.079* 0.070 (-1.75) (0.13) (-0.85) (1.70) (-1.84) (-1.69) (1.41) Senior High -0.113** -0.071* -0.087 0.055 -0.063 -0.051 -0.009 (-2.00) (-1.69) (-0.87) (1.05) (-1.20) (-1.21) (-0.20) College and above -0.075 -0.048 -0.147 0.200** -0.234*** -0.257*** 0.168** (-0.92) (-0.74) (-0.93) (2.34) (-2.66) (-3.50) (2.44) Log Household Size 0.011 -0.023 -0.082 0.037 -0.050 -0.018 0.001 (0.38) (-0.93) (-1.21) (1.10) (-1.48) (-0.50) (0.03) Log Male Children -0.057*** -0.064*** 0.096*** -0.001 0.030 0.018 0.004 (-2.60) (-4.31) (2.69) (-0.04) (1.15) (0.89) (0.20) Log Land Owned (Ha) -0.042*** 0.008 0.033*** -0.035*** 0.041*** 0.034*** -0.031*** (-8.42) (1.40) (2.74) (-3.70) (4.28) (4.52) (-4.53) Female Head 0.143*** 0.226*** 0.001 -0.060 0.061 0.079** -0.070* (4.39) (6.74) (0.01) (-1.36) (1.29) (2.16) (-1.80) Log TLUs -0.159*** -0.010 0.057** -0.029** 0.039*** 0.050*** -0.037** (-11.96) (-1.02) (2.10) (-2.10) (3.17) (2.84) (-2.23) Cell Phone -0.047** 0.095*** 0.001 0.048* -0.043 -0.060*** 0.074*** (-1.98) (4.74) (0.01) (1.67) (-1.51) (-2.62) (3.14) Solar/Generator -0.094*** 0.077*** 0.042 -0.024 0.031 0.035 -0.044* (-4.87) (3.45) (0.81) (-0.81) (0.98) (1.30) (-1.91) HH Bank Account 0.237*** -0.013 -0.114** 0.054 -0.075** -0.095*** 0.098*** (8.26) (-0.52) (-2.12) (1.51) (-2.12) (-3.15) (3.57) House Type Cement Floor -0.068*** -0.004 -0.180*** 0.073*** -0.113*** -0.119*** 0.093*** (-2.68) (-0.15) (-3.62) (2.98) (-4.69) (-4.56) (3.30) Permanent Roof -0.097*** -0.027 0.014 0.037 -0.031 -0.050* 0.053* (-4.36) (-1.27) (0.26) (1.09) (-0.95) (-1.69) (1.87) Observations 63260 63260 38982 38982 38982 38982 38982 56 Table A 1.3: Agricultural Productivity and Employment Participation Œ First Stage Probit (6 ŒLags Mov ing Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use three survey wave s while the remaining specifications use the latest two waves. Wage/Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Log Value of Yields per Ha - 6 Year MA -0.020 -0.180* -0.061 0.091 -0.090 0.036 -0.031 (-0.10) (-1.71) (-0.29) (0.58) (-0.52) (0.27) (-0.34) 25-34 Year Olds # Log Value of Yields per Ha - 6 Year MA 0.106 0.199*** 0.008 0.020 -0.028 -0.022 0.040 (1.28) (2.99) (0.05) (0.18) (-0.25) (-0.18) (0.36) 35-65 Year Olds # Log Value of Yields per Ha - 6 Year MA -0.006 0.240*** 0.335*** -0.018 0.111 0.064 0.028 (-0.05) (3.29) (2.88) (-0.21) (1.28) (0.64) (0.30) 25-34 Year Olds -0.617 -1.272** -0.211 -0.601 0.570 0.674 -0.885 (-0.88) (-2.28) (-0.17) (-0.64) (0.61) (0.68) (-0.96) 35-65 Year Olds 0.266 -1.548** -3.222*** -0.206 -0.761 -0.200 -0.649 (0.31) (-2.54) (-3.30) (-0.29) (-1.05) (-0.24) (-0.84) Female -0.382*** -0.192*** -0.008 0.095*** -0.090*** -0.052** 0.033 (-9.77) (-7.32) (-0.18) (4.66) (-4.47) (-2.14) (1.56) Relation to HH Head (base: Head) Spouse -0.346*** -0.343*** 0.093 -0.252*** 0.229*** 0.230*** -0.279*** (-10.10) (-11.88) (1.49) (-5.80) (5.53) (6.56) (-6.64) Child (own/step) -0.405*** -0.868*** -0.603*** 0.187*** -0.374*** -0.197*** 0.106** (-8.01) (-21.63) (-7.34) (3.58) (-6.89) (-4.51) (2.19) Relative -0.590*** -1.055*** -0.573*** 0.174*** -0.353*** -0.149*** 0.063 (-11.88) (-29.60) (-6.52) (2.80) (-5.69) (-3.08) (1.20) Unrelated -0.221** -0.632*** 0.520*** -0.008 0.081 0.270* -0.242 (-2.32) (-7.75) (3.03) (-0.09) (1.11) (1.67) (-1.48) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.096*** -0.061* -0.106 -0.013 -0.039 -0.031 -0.037 (-2.86) (-1.68) (-1.59) (-0.27) (-0.88) (-0.84) (-0.92) Never Married -0.192*** -0.428*** -0.242*** 0.530*** -0.534*** -0.678*** 0.637*** (-5.14) (-17.49) (-4.17) (10.85) (-12.78) (-15.61) (12.55) Education Attained:base-None Primary 0.027 0.150*** 0.336*** 0.132** 0.029 -0.033 0.185*** (0.68) (4.68) (4.57) (2.33) (0.51) (-0.61) (3.61) Junior High 0.075 0.237*** 0.456*** 0.210*** -0.013 -0.115* 0.285*** (1.54) (6.40) (5.82) (3.60) (-0.22) (-1.84) (4.64) Senior High 0.252*** 0.196*** 0.375*** 0.383*** -0.191*** -0.279*** 0.444*** (4.42) (4.98) (4.19) (6.32) (-3.18) (-4.61) (7.46) College and above 1.187*** -0.067 0.016 0.878*** -0.748*** -0.985*** 1.124*** (14.03) (-0.89) (0.14) (8.51) (-8.33) (-10.17) (11.21) Log Head Age -0.176*** -0.262*** -0.060 -0.153*** 0.141** 0.213*** -0.221*** (-4.01) (-6.80) (-0.65) (-2.63) (2.43) (4.22) (-4.50) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.042 0.054 0.029 -0.044 0.050 0.019 -0.004 (1.08) (1.57) (0.38) (-1.17) (1.20) (0.69) (-0.12) Never Married 0.049 0.406*** 0.078 -0.330** 0.309** 0.432*** -0.350*** (0.48) (3.94) (0.29) (-2.29) (2.14) (4.51) (-3.18) Head Education: base-None Primary -0.074** 0.007 0.006 0.017 -0.008 0.002 -0.014 (-2.30) (0.25) (0.08) (0.39) (-0.19) (0.06) (-0.36) Junior High -0.090* 0.006 -0.077 0.086* -0.093* -0.079* 0.070 (-1.76) (0.16) (-0.84) (1.69) (-1.83) (-1.68) (1.40) Senior High -0.113** -0.070* -0.086 0.054 -0.063 -0.051 -0.009 (-2.00) (-1.68) (-0.85) (1.04) (-1.19) (-1.21) (-0.20) College and above -0.075 -0.047 -0.146 0.200** -0.233*** -0.257*** 0.168** (-0.92) (-0.73) (-0.93) (2.34) (-2.66) (-3.49) (2.43) Log Household Size 0.012 -0.023 -0.081 0.037 -0.050 -0.017 0.000 (0.39) (-0.92) (-1.20) (1.10) (-1.47) (-0.49) (0.01) Log Male Children -0.057*** -0.064*** 0.096*** -0.001 0.030 0.018 0.005 (-2.59) (-4.30) (2.68) (-0.04) (1.15) (0.89) (0.21) Log Land Owned (Ha) -0.042*** 0.008 0.033*** -0.035*** 0.041*** 0.034*** -0.031*** (-8.39) (1.40) (2.74) (-3.69) (4.27) (4.52) (-4.52) Female Head 0.142*** 0.226*** 0.001 -0.060 0.061 0.079** -0.070* (4.38) (6.72) (0.01) (-1.37) (1.29) (2.16) (-1.81) Log TLUs -0.159*** -0.010 0.057** -0.029** 0.039*** 0.050*** -0.036** (-12.01) (-1.02) (2.10) (-2.09) (3.15) (2.85) (-2.22) Cell Phone -0.046** 0.095*** -0.000 0.048* -0.043 -0.061*** 0.075*** (-1.96) (4.73) (-0.00) (1.68) (-1.52) (-2.62) (3.14) Solar/Generator -0.094*** 0.077*** 0.041 -0.024 0.031 0.035 -0.043* (-4.86) (3.46) (0.79) (-0.81) (0.97) (1.30) (-1.91) HH Bank Account 0.237*** -0.013 -0.113** 0.054 -0.075** -0.095*** 0.098*** (8.26) (-0.51) (-2.09) (1.50) (-2.10) (-3.16) (3.57) House Type Cement Floor -0.068*** -0.004 -0.180*** 0.073*** -0.113*** -0.119*** 0.093*** (-2.69) (-0.15) (-3.61) (2.97) (-4.67) (-4.55) (3.29) Permanent Roof -0.097*** -0.027 0.015 0.037 -0.030 -0.049* 0.053* (-4.37) (-1.25) (0.26) (1.08) (-0.93) (-1.68) (1.85) Observations 63260 63260 38982 38982 38982 38982 38982 57 Table A 1.4: Agricultural Productivity and Employment Participation by Gender Œ First Stage Probit (2 ŒLags Moving Average) TŒStatistics in par entheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use three survey waves while the remaining specificatio ns use the latest two waves. Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Log Value of Yields per Ha - 2 Year MA 0.019 -0.205* -0.659** 0.189 -0.287** -0.227 0.187 (0.10) (-1.76) (-2.42) (1.54) (-2.37) (-1.16) (0.84) 15-24 # Female # Log Value of Yields per Ha - 2 Year MA 0.001 0.016 0.196** -0.063 0.105** 0.134** -0.075 (0.01) (0.26) (2.16) (-1.19) (2.15) (2.14) (-1.13) 25-34 # Male # Log Value of Yields per Ha - 2 Year MA 0.108 0.175** 0.125 0.053 -0.033 -0.039 0.041 (1.37) (2.50) (0.95) (0.47) (-0.31) (-0.36) (0.37) 25-34 # Female # Log Value of Yields per Ha - 2 Year MA 0.105 0.201** 0.339** -0.003 0.071 0.092 -0.033 (1.32) (2.01) (2.01) (-0.02) (0.50) (0.60) (-0.21) 35-65 # Male # Log Value of Yields per Ha - 2 Year MA 0.056 0.200** 0.417*** 0.056 0.060 0.009 0.071 (0.63) (2.54) (3.80) (0.78) (0.80) (0.11) (0.87) 35-65 # Female # Log Value of Yields per Ha - 2 Year MA 0.059 0.239** 0.635*** -0.018 0.181* 0.153 -0.011 (0.62) (2.48) (4.29) (-0.18) (1.67) (1.25) (-0.09) 25-34 -0.616 -1.111* -1.257 -0.900 0.618 0.831 -0.895 (-0.93) (-1.90) (-1.14) (-0.95) (0.69) (0.94) (-0.98) 35-65 -0.251 -1.306** -3.954*** -0.765 -0.400 0.213 -0.979 (-0.34) (-1.99) (-4.38) (-1.30) (-0.64) (0.31) (-1.41) Female -0.390 -0.414 -1.706** 0.620 -0.979** -1.171** 0.660 (-0.59) (-0.79) (-2.24) (1.42) (-2.41) (-2.25) (1.21) Relation to HH Head (base: Head) Spouse -0.350*** -0.404*** -0.003 -0.205*** 0.153*** 0.181*** -0.245*** (-9.82) (-12.08) (-0.04) (-3.46) (2.79) (3.85) (-4.66) Child (own/step) -0.407*** -0.889*** -0.620*** 0.214*** -0.406*** -0.219*** 0.121** (-8.18) (-22.62) (-7.19) (4.00) (-7.09) (-4.88) (2.54) Relative -0.592*** -1.071*** -0.588*** 0.199*** -0.382*** -0.169*** 0.078 (-11.90) (-30.60) (-6.53) (3.17) (-5.92) (-3.42) (1.47) Unrelated -0.219** -0.670*** 0.479*** 0.026 0.036 0.248 -0.223 (-2.24) (-8.36) (2.73) (0.31) (0.51) (1.60) (-1.40) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.097*** -0.092*** -0.147** 0.006 -0.069 -0.051 -0.022 (-2.82) (-2.59) (-2.01) (0.11) (-1.36) (-1.24) (-0.52) Never Married -0.193*** -0.462*** -0.254*** 0.540*** -0.547*** -0.687*** 0.643*** (-4.98) (-18.14) (-4.37) (10.95) (-13.30) (-16.14) (12.70) Education Attained:base-None Primary 0.027 0.157*** 0.338*** 0.127** 0.035 -0.028 0.181*** (0.67) (4.95) (4.52) (2.23) (0.62) (-0.52) (3.53) Junior High 0.076 0.247*** 0.461*** 0.203*** -0.005 -0.108* 0.280*** (1.54) (6.75) (5.74) (3.49) (-0.08) (-1.74) (4.55) Senior High 0.253*** 0.207*** 0.383*** 0.376*** -0.182*** -0.271*** 0.438*** (4.36) (5.40) (4.18) (6.14) (-3.01) (-4.54) (7.40) College and above 1.188*** -0.047 0.026 0.869*** -0.735*** -0.975*** 1.116*** (14.00) (-0.63) (0.22) (8.33) (-8.08) (-10.05) (11.13) Log Head Age -0.176*** -0.271*** -0.070 -0.149*** 0.133** 0.208*** -0.218*** (-4.01) (-6.84) (-0.76) (-2.59) (2.34) (4.17) (-4.47) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.042 0.075** 0.038 -0.048 0.056 0.024 -0.007 (1.07) (2.14) (0.50) (-1.22) (1.33) (0.85) (-0.20) Never Married 0.050 0.420*** 0.071 -0.324** 0.301** 0.428*** -0.345*** (0.50) (4.05) (0.27) (-2.23) (2.09) (4.45) (-3.10) Head Education: base-None Primary -0.074** 0.004 0.005 0.019 -0.010 0.000 -0.013 (-2.30) (0.13) (0.07) (0.43) (-0.25) (0.00) (-0.33) Junior High -0.090* 0.001 -0.076 0.089* -0.097* -0.083* 0.073 (-1.75) (0.02) (-0.84) (1.75) (-1.91) (-1.78) (1.47) Senior High -0.113** -0.076* -0.087 0.057 -0.066 -0.054 -0.007 (-2.00) (-1.83) (-0.87) (1.08) (-1.25) (-1.29) (-0.16) College and above -0.075 -0.057 -0.144 0.203** -0.237*** -0.261*** 0.171** (-0.91) (-0.87) (-0.91) (2.34) (-2.69) (-3.55) (2.46) Log Household Size 0.013 -0.018 -0.085 0.035 -0.048 -0.016 -0.000 (0.44) (-0.72) (-1.25) (1.03) (-1.41) (-0.46) (-0.00) Log Male Children -0.057*** -0.066*** 0.095*** -0.001 0.028 0.018 0.005 (-2.59) (-4.41) (2.65) (-0.02) (1.09) (0.86) (0.22) Log Land Owned (Ha) -0.042*** 0.008 0.034*** -0.035*** 0.041*** 0.034*** -0.031*** (-8.51) (1.44) (2.77) (-3.74) (4.34) (4.61) (-4.60) Female Head 0.137*** 0.184*** -0.021 -0.050 0.043 0.066* -0.062 (4.19) (5.14) (-0.24) (-1.09) (0.89) (1.74) (-1.53) Log TLUs -0.159*** -0.011 0.057** -0.028** 0.038*** 0.049*** -0.036** (-11.99) (-1.10) (2.10) (-2.04) (3.10) (2.83) (-2.21) Cell Phone -0.047** 0.094*** 0.001 0.048* -0.043 -0.061*** 0.075*** (-1.96) (4.71) (0.02) (1.68) (-1.52) (-2.66) (3.17) Solar/Generator -0.094*** 0.077*** 0.043 -0.024 0.031 0.036 -0.044* (-4.85) (3.45) (0.84) (-0.81) (0.98) (1.32) (-1.91) HH Bank Account 0.237*** -0.013 -0.119** 0.054 -0.076** -0.096*** 0.098*** (8.27) (-0.52) (-2.21) (1.50) (-2.12) (-3.19) (3.57) House Type Cement Floor -0.068*** -0.003 -0.181*** 0.073*** -0.113*** -0.119*** 0.093*** (-2.72) (-0.13) (-3.63) (2.98) (-4.71) (-4.54) (3.26) Permanent Roof -0.097*** -0.027 0.014 0.037 -0.030 -0.049* 0.053* (-4.38) (-1.27) (0.24) (1.07) (-0.93) (-1.68) (1.86) Observations 63260 63260 38982 38982 38982 38982 38982 58 Table A 1.5: Agricultural Productivity and Employment Participation by Gender Œ First Stage Probit (4 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p< 0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use three survey waves while the remaining specifications use the latest two waves. Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Log Value of Yields per Ha - 4 Year MA 0.030 -0.167 -0.266 0.118 -0.163 -0.064 0.082 (0.19) (-1.41) (-1.38) (0.84) (-1.05) (-0.48) (0.55) 15-24 # Female # Log Value of Yields per Ha - 4 Year MA -0.003 -0.006 0.209** -0.058 0.106* 0.139** -0.067 (-0.03) (-0.09) (2.19) (-0.98) (1.94) (2.21) (-1.03) 25-34 # Male # Log Value of Yields per Ha - 4 Year MA 0.082 0.154** 0.062 0.005 -0.002 -0.013 0.021 (0.97) (2.33) (0.46) (0.05) (-0.02) (-0.12) (0.20) 25-34 # Female # Log Value of Yields per Ha - 4 Year MA 0.076 0.157 0.287* -0.045 0.102 0.124 -0.044 (0.91) (1.60) (1.68) (-0.34) (0.76) (0.86) (-0.32) 35-65 # Male # Log Value of Yields per Ha - 4 Year MA 0.002 0.165** 0.340*** -0.001 0.097 0.062 0.006 (0.02) (2.08) (3.19) (-0.02) (1.28) (0.68) (0.07) 35-65 # Female # Log Value of Yields per Ha - 4 Year MA 0.002 0.182* 0.570*** -0.070 0.219** 0.211* -0.067 (0.02) (1.79) (3.79) (-0.67) (1.97) (1.80) (-0.57) 25-34 -0.405 -0.943* -0.720 -0.501 0.363 0.612 -0.734 (-0.57) (-1.70) (-0.64) (-0.56) (0.41) (0.67) (-0.83) 35-65 0.191 -1.030 -3.325*** -0.288 -0.712 -0.228 -0.442 (0.23) (-1.55) (-3.75) (-0.45) (-1.11) (-0.30) (-0.60) Female -0.360 -0.228 -1.819** 0.580 -0.986** -1.222** 0.594 (-0.52) (-0.40) (-2.28) (1.18) (-2.19) (-2.32) (1.11) Relation to HH Head (base: Head) Spouse -0.351*** -0.404*** -0.007 -0.204*** 0.151*** 0.180*** -0.244*** (-9.78) (-12.04) (-0.09) (-3.46) (2.78) (3.83) (-4.69) Child (own/step) -0.407*** -0.890*** -0.622*** 0.214*** -0.406*** -0.218*** 0.121** (-8.17) (-22.71) (-7.22) (4.01) (-7.10) (-4.86) (2.51) Relative -0.592*** -1.072*** -0.589*** 0.199*** -0.382*** -0.168*** 0.077 (-11.88) (-30.69) (-6.55) (3.17) (-5.93) (-3.39) (1.45) Unrelated -0.224** -0.670*** 0.496*** 0.011 0.054 0.264* -0.235 (-2.31) (-8.23) (2.73) (0.13) (0.79) (1.65) (-1.44) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.100*** -0.093*** -0.149** 0.004 -0.068 -0.048 -0.026 (-2.88) (-2.58) (-2.04) (0.06) (-1.33) (-1.17) (-0.60) Never Married -0.195*** -0.461*** -0.255*** 0.539*** -0.546*** -0.686*** 0.641*** (-4.99) (-18.40) (-4.37) (10.93) (-13.25) (-16.27) (12.70) Education Attained:base-None Primary 0.028 0.157*** 0.341*** 0.128** 0.035 -0.029 0.183*** (0.70) (4.96) (4.64) (2.25) (0.61) (-0.54) (3.55) Junior High 0.077 0.248*** 0.465*** 0.205*** -0.005 -0.109* 0.282*** (1.56) (6.73) (5.88) (3.51) (-0.09) (-1.76) (4.56) Senior High 0.253*** 0.208*** 0.387*** 0.377*** -0.182*** -0.272*** 0.440*** (4.38) (5.39) (4.27) (6.16) (-3.02) (-4.54) (7.37) College and above 1.189*** -0.046 0.031 0.868*** -0.733*** -0.974*** 1.117*** (14.05) (-0.62) (0.26) (8.39) (-8.13) (-10.03) (11.12) Log Head Age -0.176*** -0.271*** -0.071 -0.150*** 0.134** 0.208*** -0.218*** (-3.99) (-6.83) (-0.76) (-2.60) (2.34) (4.15) (-4.47) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.044 0.076** 0.042 -0.048 0.057 0.024 -0.007 (1.11) (2.13) (0.55) (-1.24) (1.35) (0.86) (-0.21) Never Married 0.051 0.421*** 0.078 -0.324** 0.301** 0.428*** -0.345*** (0.50) (4.07) (0.29) (-2.23) (2.09) (4.48) (-3.12) Head Education: base-None Primary -0.074** 0.004 0.006 0.018 -0.010 0.001 -0.013 (-2.33) (0.12) (0.07) (0.42) (-0.24) (0.02) (-0.34) Junior High -0.090* 0.001 -0.077 0.088* -0.095* -0.081* 0.071 (-1.77) (0.02) (-0.85) (1.73) (-1.88) (-1.73) (1.44) Senior High -0.113** -0.076* -0.088 0.057 -0.066 -0.054 -0.007 (-2.01) (-1.84) (-0.88) (1.09) (-1.26) (-1.29) (-0.16) College and above -0.076 -0.057 -0.149 0.203** -0.238*** -0.261*** 0.171** (-0.92) (-0.88) (-0.94) (2.35) (-2.70) (-3.55) (2.46) Log Household Size 0.013 -0.018 -0.084 0.036 -0.049 -0.016 0.000 (0.44) (-0.73) (-1.23) (1.05) (-1.42) (-0.45) (0.00) Log Male Children -0.057*** -0.066*** 0.095*** -0.001 0.028 0.017 0.005 (-2.59) (-4.41) (2.63) (-0.03) (1.09) (0.85) (0.23) Log Land Owned (Ha) -0.042*** 0.008 0.033*** -0.035*** 0.041*** 0.034*** -0.031*** (-8.47) (1.43) (2.76) (-3.71) (4.30) (4.53) (-4.54) Female Head 0.137*** 0.183*** -0.025 -0.050 0.043 0.066* -0.062 (4.22) (5.10) (-0.28) (-1.10) (0.89) (1.75) (-1.55) Log TLUs -0.159*** -0.010 0.057** -0.028** 0.038*** 0.049*** -0.036** (-12.05) (-1.07) (2.09) (-2.03) (3.08) (2.80) (-2.20) Cell Phone -0.047** 0.094*** 0.001 0.048* -0.043 -0.061*** 0.075*** (-1.98) (4.72) (0.01) (1.67) (-1.52) (-2.65) (3.17) Solar/Generator -0.094*** 0.078*** 0.042 -0.024 0.031 0.036 -0.044* (-4.86) (3.45) (0.82) (-0.82) (0.99) (1.32) (-1.92) HH Bank Account 0.237*** -0.013 -0.116** 0.054 -0.076** -0.095*** 0.098*** (8.26) (-0.52) (-2.14) (1.51) (-2.12) (-3.15) (3.57) House Type Cement Floor -0.067*** -0.004 -0.181*** 0.073*** -0.113*** -0.119*** 0.094*** (-2.69) (-0.15) (-3.64) (3.00) (-4.71) (-4.56) (3.30) Permanent Roof -0.097*** -0.027 0.014 0.037 -0.030 -0.049* 0.053* (-4.36) (-1.27) (0.24) (1.07) (-0.93) (-1.67) (1.86) Observations 63260 63260 38982 38982 38982 38982 38982 59 Table A 1.6: Agricultural Productivity and Employment Participation by Gender Œ First Stage Probit (6 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and dis trict fixed effects. Errors are clustered at the district level (72 districts). Due to data limitation, the first two specifications use three survey waves while the remaining specifications use the latest two waves. Wage/ Salary Business Farming Farming <20 days Farming >=20 days Primary Activity - Farm Primary Activity - Non-Farm Log Value of Yields per Ha - 6 Year MA -0.021 -0.160 -0.144 0.120 -0.137 -0.031 0.008 (-0.11) (-1.43) (-0.66) (0.71) (-0.74) (-0.22) (0.07) 15-24 # Female # Log Value of Yields per Ha - 6 Year MA 0.006 -0.026 0.202* -0.071 0.116* 0.162** -0.094 (0.07) (-0.38) (1.88) (-1.07) (1.95) (2.54) (-1.42) 25-34 # Male # Log Value of Yields per Ha - 6 Year MA 0.107 0.193*** 0.000 0.023 -0.035 -0.029 0.042 (1.24) (2.94) (0.00) (0.21) (-0.32) (-0.24) (0.38) 25-34 # Female # Log Value of Yields per Ha - 6 Year MA 0.109 0.176* 0.219 -0.040 0.080 0.130 -0.052 (1.27) (1.75) (1.21) (-0.28) (0.56) (0.84) (-0.35) 35-65 # Male # Log Value of Yields per Ha - 6 Year MA -0.007 0.222*** 0.316*** -0.013 0.100 0.053 0.032 (-0.07) (3.02) (2.73) (-0.15) (1.17) (0.53) (0.35) 35-65 # Female # Log Value of Yields per Ha - 6 Year MA 0.002 0.218** 0.539*** -0.094 0.232* 0.225* -0.070 (0.02) (2.19) (3.23) (-0.79) (1.88) (1.76) (-0.56) 25-34 -0.609 -1.265** -0.202 -0.654 0.633 0.747 -0.909 (-0.84) (-2.30) (-0.16) (-0.70) (0.69) (0.75) (-0.98) 35-65 0.264 -1.496** -3.131*** -0.195 -0.740 -0.158 -0.655 (0.31) (-2.45) (-3.22) (-0.27) (-1.02) (-0.19) (-0.85) Female -0.433 -0.058 -1.762* 0.690 -1.074** -1.414*** 0.830 (-0.56) (-0.10) (-1.95) (1.25) (-2.17) (-2.65) (1.51) Relation to HH Head (base: Head) Spouse -0.351*** -0.405*** -0.005 -0.204*** 0.151*** 0.180*** -0.243*** (-9.81) (-11.97) (-0.06) (-3.47) (2.80) (3.84) (-4.69) Child (own/step) -0.408*** -0.890*** -0.622*** 0.213*** -0.405** -0.218*** 0.120** (-8.17) (-22.67) (-7.23) (3.99) (-7.08) (-4.84) (2.49) Relative -0.593*** -1.072*** -0.589*** 0.198*** -0.381** -0.167*** 0.076 (-11.84) (-30.68) (-6.55) (3.16) (-5.91) (-3.38) (1.43) Unrelated -0.223** -0.673*** 0.495*** 0.008 0.055 0.266* -0.237 (-2.28) (-8.19) (2.72) (0.10) (0.82) (1.67) (-1.44) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.100*** -0.091** -0.149** 0.003 -0.068 -0.049 -0.024 (-2.89) (-2.54) (-2.06) (0.06) (-1.33) (-1.20) (-0.57) Never Married -0.195*** -0.458*** -0.256*** 0.539*** -0.546** -0.687*** 0.643*** (-4.99) (-18.41) (-4.39) (10.90) (-13.23) (-16.26) (12.74) Education Attained:base-None Primary 0.028 0.156*** 0.341*** 0.128** 0.034 -0.029 0.182*** (0.71) (4.93) (4.63) (2.26) (0.61) (-0.54) (3.55) Junior High 0.077 0.246*** 0.464*** 0.205*** -0.006 -0.109* 0.281*** (1.56) (6.73) (5.87) (3.52) (-0.09) (-1.77) (4.58) Senior High 0.254*** 0.207*** 0.386*** 0.377*** -0.182** -0.272*** 0.439*** (4.39) (5.38) (4.26) (6.16) (-3.03) (-4.54) (7.39) College and above 1.190*** -0.049 0.029 0.869*** -0.734** -0.974*** 1.116*** (14.04) (-0.66) (0.24) (8.42) (-8.17) (-10.05) (11.14) Log Head Age -0.176*** -0.270*** -0.070 -0.149*** 0.134** 0.208*** -0.218*** (-3.99) (-6.80) (-0.76) (-2.59) (2.34) (4.15) (-4.45) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.044 0.075** 0.042 -0.048 0.057 0.023 -0.007 (1.12) (2.10) (0.55) (-1.24) (1.35) (0.86) (-0.22) Never Married 0.050 0.422*** 0.077 -0.324** 0.302** 0.428*** -0.346*** (0.50) (4.06) (0.29) (-2.23) (2.09) (4.47) (-3.14) Head Education: base-None Primary -0.075** 0.004 0.006 0.018 -0.010 0.001 -0.013 (-2.32) (0.15) (0.08) (0.42) (-0.23) (0.02) (-0.33) Junior High -0.091* 0.002 -0.077 0.088* -0.095* -0.081* 0.071 (-1.77) (0.05) (-0.85) (1.73) (-1.88) (-1.74) (1.44) Senior High -0.113** -0.075* -0.087 0.057 -0.066 -0.054 -0.007 (-2.01) (-1.83) (-0.87) (1.09) (-1.26) (-1.29) (-0.15) College and above -0.076 -0.056 -0.148 0.204** -0.238** -0.261*** 0.171** (-0.92) (-0.86) (-0.94) (2.36) (-2.70) (-3.54) (2.46) Log Household Size 0.013 -0.018 -0.083 0.036 -0.049 -0.016 -0.001 (0.45) (-0.72) (-1.22) (1.05) (-1.42) (-0.45) (-0.02) Log Male Children -0.057*** -0.066*** 0.095*** -0.001 0.028 0.017 0.005 (-2.59) (-4.40) (2.62) (-0.03) (1.09) (0.85) (0.23) Log Land Owned (Ha) -0.042*** 0.008 0.033*** -0.035*** 0.041*** 0.034*** -0.031*** (-8.44) (1.43) (2.77) (-3.70) (4.29) (4.54) (-4.53) Female Head 0.137*** 0.183*** -0.024 -0.050 0.043 0.066* -0.062 (4.23) (5.05) (-0.27) (-1.11) (0.89) (1.76) (-1.55) Log TLUs -0.160*** -0.010 0.057** -0.028** 0.038*** 0.049*** -0.036** (-12.10) (-1.07) (2.09) (-2.02) (3.06) (2.80) (-2.19) Cell Phone -0.047** 0.094*** 0.000 0.048* -0.043 -0.061*** 0.075*** (-1.96) (4.71) (0.00) (1.68) (-1.53) (-2.64) (3.17) Solar/Generator -0.093*** 0.078*** 0.042 -0.024 0.031 0.036 -0.044* (-4.84) (3.46) (0.80) (-0.82) (0.99) (1.32) (-1.92) HH Bank Account 0.237*** -0.013 -0.115** 0.053 -0.075** -0.096*** 0.098*** (8.27) (-0.52) (-2.11) (1.49) (-2.10) (-3.16) (3.57) House Type Cement Floor -0.068*** -0.004 -0.181*** 0.073*** -0.114** -0.119*** 0.093*** (-2.69) (-0.15) (-3.63) (2.99) (-4.70) (-4.56) (3.30) Permanent Roof -0.097*** -0.027 0.014 0.037 -0.030 -0.049* 0.053* (-4.36) (-1.25) (0.25) (1.07) (-0.92) (-1.66) (1.84) Observations 63260 63260 38982 38982 38982 38982 38982 60 Table A 1.7: Agricultural Productivity and Multiple Employment Participation Œ First Stage Probit (2 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clu stered at the district level ( 72 districts ). Due to data limitation, we use only the latest two survey waves . None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage Log Value of Yields per Ha - 2 Year MA 0.648** 0.024 -0.236** -0.103 -0.510** 0.089 (2.34) (0.19) (-2.07) (-0.64) (-2.03) (0.24) 25-34 # Log Value of Yields per Ha - 2 Year MA -0.202 0.059 0.181** -0.127 0.320* 0.288 (-1.30) (0.80) (2.01) (-1.40) (1.71) (0.83) 35-65 # Log Value of Yields per Ha - 2 Year MA -0.565*** 0.003 0.281*** -0.116 0.221 0.012 (-4.24) (0.04) (3.15) (-1.06) (1.31) (0.05) 25-34 1.798 -0.924 -1.079 1.336* -2.343 -2.078 (1.40) (-1.51) (-1.44) (1.77) (-1.52) (-0.72) 35-65 5.108*** -0.530 -1.819** 1.180 -1.574 0.344 (4.63) (-0.74) (-2.47) (1.32) (-1.13) (0.16) Female 0.025 0.272*** -0.212*** -0.365*** -0.275*** -0.169** (0.51) (10.99) (-6.52) (-8.72) (-4.45) (-1.99) Relation to HH Head (base: Head) Spouse 0.145* 0.454*** -0.204*** -0.218*** -0.349*** -0.182* (1.75) (13.71) (-5.03) (-4.87) (-5.60) (-1.66) Child (own/step) 0.953*** 0.706*** -0.766*** -0.174*** -0.491*** -0.193 (12.36) (19.12) (-16.81) (-2.89) (-7.06) (-1.05) Relative 0.941*** 0.863*** -0.946*** -0.343*** -0.609*** -0.413** (11.23) (18.30) (-19.43) (-5.28) (-8.02) (-2.42) Unrelated -0.449* 0.642*** -0.493*** -0.074 -0.251*** -0.130 (-1.72) (10.83) (-6.16) (-0.95) (-3.16) (-0.48) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabi 0.101 0.052 -0.016 -0.089** -0.078 0.097 (1.21) (1.39) (-0.34) (-2.18) (-1.34) (0.72) Never Married 0.249*** 0.205*** -0.389*** -0.134*** -0.331*** 0.118 (3.83) (7.74) (-9.62) (-3.16) (-3.86) (0.58) Education Attained:base-None Primary -0.386*** -0.023 0.138*** -0.054 0.093 -0.007 (-4.69) (-0.70) (3.64) (-1.13) (1.30) (-0.06) Junior High -0.516*** -0.047 0.212*** -0.015 0.060 -0.053 (-5.75) (-1.28) (4.68) (-0.27) (0.76) (-0.38) Senior High -0.422*** -0.073** 0.135*** 0.105* 0.124 -0.011 (-4.18) (-1.98) (2.70) (1.71) (1.32) (-0.06) College and above -0.226 -0.510*** -0.487*** 0.848*** 0.688*** 0.623*** (-1.61) (-8.12) (-3.62) (8.91) (6.45) (2.71) Log Head Age 0.085 0.228*** -0.172*** -0.076 -0.305*** 0.009 (0.90) (4.80) (-3.70) (-1.38) (-3.09) (0.06) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabi -0.035 -0.019 -0.012 0.036 0.086 0.007 (-0.46) (-0.65) (-0.29) (0.76) (1.24) (0.05) Never Married -0.185 -0.112 0.313*** 0.036 0.082 0.228 (-0.72) (-1.01) (3.23) (0.27) (0.38) (0.58) Head Education: base-None Primary -0.005 -0.001 0.015 -0.004 -0.082 -0.092 (-0.06) (-0.03) (0.36) (-0.12) (-1.31) (-0.63) Junior High 0.060 -0.017 -0.020 -0.022 0.013 -0.017 (0.66) (-0.41) (-0.41) (-0.43) (0.17) (-0.12) Senior High 0.079 0.050 -0.082 -0.053 -0.023 -0.011 (0.86) (0.96) (-1.43) (-0.76) (-0.24) (-0.06) College and above 0.163 -0.036 -0.014 -0.086 -0.004 -0.206 (1.03) (-0.46) (-0.19) (-0.81) (-0.04) (-0.93) Log Household Size 0.097 0.002 -0.018 0.009 -0.090* 0.039 (1.42) (0.06) (-0.57) (0.31) (-1.93) (0.37) Log Male Children -0.102*** 0.095*** -0.055** -0.058** -0.054* -0.035 (-2.77) (5.17) (-2.56) (-2.54) (-1.74) (-0.52) Log Land Owned (Ha) -0.028** 0.026*** 0.018** -0.040*** -0.020** -0.033** (-2.20) (4.17) (2.07) (-6.40) (-2.34) (-2.50) Female Head -0.031 -0.135*** 0.248*** 0.046 0.182*** 0.215* (-0.34) (-3.70) (5.49) (1.03) (2.58) (1.69) Log TLUs -0.040 0.078*** 0.000 -0.170*** -0.049** -0.111** (-1.52) (7.10) (0.04) (-10.03) (-2.49) (-2.50) Cell Phone 0.001 -0.034 0.107*** -0.083*** -0.018 -0.026 (0.01) (-1.37) (4.43) (-2.84) (-0.41) (-0.32) Solar/Generator -0.048 0.001 0.103*** -0.120*** -0.086** -0.002 (-0.91) (0.02) (4.96) (-4.22) (-2.36) (-0.03) HH Bank Account 0.064 -0.110*** -0.023 0.167*** 0.130** 0.267*** (1.05) (-4.44) (-0.86) (5.01) (2.47) (3.89) House Type Cement Floor 0.189*** -0.016 0.011 -0.051 -0.117** 0.101 (3.76) (-0.77) (0.40) (-1.64) (-2.33) (1.34) Permanent Roof -0.016 0.042* 0.024 -0.098*** -0.110*** -0.007 (-0.29) (1.85) (0.83) (-3.16) (-2.72) (-0.07) Observations 38982 38982 38982 38982 38777 31998 61 Table A 1.8: Agricultural Productivity and Multiple Employment Participation Œ First Stage Probit (4ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level ( 72 districts ). Due to data limitation, we use only the latest two survey wave s. None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage Log Value of Yields per Ha - 4 Year MA 0.197 0.082 -0.269** 0.010 -0.167 0.041 (1.05) (0.83) (-2.52) (0.06) (-0.60) (0.08) 25-34 # Log Value of Yields per Ha - 4 Year MA -0.150 0.070 0.153 -0.124 0.071 0.417 (-0.97) (0.94) (1.59) (-1.23) (0.35) (1.26) 35-65 # Log Value of Yields per Ha - 4 Year MA -0.479*** -0.004 0.276*** -0.157 -0.027 0.068 (-3.74) (-0.04) (2.80) (-1.38) (-0.14) (0.28) 25-34 1.373 -1.023* -0.848 1.315 -0.282 -3.186 (1.06) (-1.65) (-1.05) (1.56) (-0.17) (-1.15) 35-65 4.415*** -0.476 -1.791** 1.529 0.482 -0.129 (4.13) (-0.67) (-2.19) (1.62) (0.30) (-0.06) Female 0.023 0.272*** -0.211*** -0.365*** -0.275*** -0.169** (0.48) (11.00) (-6.51) (-8.75) (-4.44) (-1.99) Relation to HH Head (base: Head) Spouse 0.148* 0.454*** -0.204*** -0.218*** -0.350*** -0.183* (1.79) (13.75) (-5.04) (-4.87) (-5.60) (-1.66) Child (own/step) 0.956*** 0.706*** -0.768*** -0.174*** -0.492*** -0.193 (12.42) (19.16) (-16.93) (-2.87) (-7.06) (-1.04) Relative 0.943*** 0.864*** -0.947*** -0.342*** -0.607*** -0.414** (11.25) (18.33) (-19.50) (-5.26) (-8.09) (-2.41) Unrelated -0.465* 0.641*** -0.491*** -0.066 -0.239*** -0.129 (-1.73) (10.78) (-6.15) (-0.85) (-2.80) (-0.48) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabi 0.102 0.052 -0.016 -0.091** -0.084 0.098 (1.22) (1.39) (-0.33) (-2.25) (-1.46) (0.73) Never Married 0.251*** 0.204*** -0.385*** -0.135*** -0.336*** 0.118 (3.85) (7.66) (-9.53) (-3.17) (-3.96) (0.58) Education Attained:base-None Primary -0.389*** -0.023 0.139*** -0.053 0.099 -0.008 (-4.79) (-0.71) (3.63) (-1.11) (1.39) (-0.06) Junior High -0.521*** -0.047 0.213*** -0.015 0.065 -0.052 (-5.87) (-1.28) (4.67) (-0.26) (0.82) (-0.37) Senior High -0.426*** -0.073** 0.135*** 0.106* 0.130 -0.011 (-4.27) (-1.98) (2.70) (1.73) (1.40) (-0.07) College and above -0.230 -0.510*** -0.485*** 0.850*** 0.692*** 0.626*** (-1.64) (-8.12) (-3.61) (8.92) (6.53) (2.74) Log Head Age 0.085 0.229*** -0.171*** -0.076 -0.305*** 0.009 (0.90) (4.80) (-3.68) (-1.39) (-3.09) (0.06) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabi -0.041 -0.019 -0.013 0.040 0.094 0.006 (-0.53) (-0.68) (-0.30) (0.84) (1.38) (0.05) Never Married -0.196 -0.110 0.311*** 0.036 0.089 0.228 (-0.76) (-1.00) (3.20) (0.26) (0.41) (0.58) Head Education: base-None Primary -0.006 -0.001 0.015 -0.005 -0.085 -0.091 (-0.08) (-0.03) (0.37) (-0.13) (-1.36) (-0.62) Junior High 0.062 -0.017 -0.020 -0.021 0.011 -0.017 (0.69) (-0.41) (-0.41) (-0.42) (0.16) (-0.11) Senior High 0.080 0.051 -0.083 -0.053 -0.025 -0.010 (0.88) (0.97) (-1.45) (-0.76) (-0.27) (-0.05) College and above 0.168 -0.036 -0.014 -0.087 -0.005 -0.209 (1.07) (-0.46) (-0.20) (-0.82) (-0.06) (-0.94) Log Household Size 0.096 0.002 -0.018 0.009 -0.091* 0.040 (1.40) (0.08) (-0.58) (0.29) (-1.95) (0.38) Log Male Children -0.102*** 0.095*** -0.055** -0.058** -0.054* -0.035 (-2.75) (5.17) (-2.53) (-2.55) (-1.74) (-0.54) Log Land Owned (Ha) -0.028** 0.026*** 0.018** -0.040*** -0.020** -0.033** (-2.19) (4.17) (2.07) (-6.41) (-2.35) (-2.50) Female Head -0.025 -0.135*** 0.248*** 0.045 0.177** 0.216* (-0.27) (-3.70) (5.49) (0.99) (2.54) (1.70) Log TLUs -0.039 0.078*** 0.001 -0.170*** -0.048** -0.111** (-1.49) (7.04) (0.08) (-10.03) (-2.42) (-2.51) Cell Phone 0.002 -0.035 0.108*** -0.082*** -0.018 -0.028 (0.04) (-1.40) (4.44) (-2.85) (-0.42) (-0.34) Solar/Generator -0.047 0.000 0.104*** -0.120*** -0.085** -0.002 (-0.89) (0.01) (4.95) (-4.21) (-2.33) (-0.03) HH Bank Account 0.060 -0.109*** -0.023 0.167*** 0.130** 0.268*** (0.98) (-4.40) (-0.86) (5.02) (2.47) (3.84) House Type Cement Floor 0.189*** -0.015 0.010 -0.052* -0.119** 0.102 (3.78) (-0.74) (0.38) (-1.66) (-2.37) (1.36) Permanent Roof -0.015 0.042* 0.024 -0.098*** -0.108*** -0.007 (-0.28) (1.86) (0.82) (-3.15) (-2.68) (-0.08) Observations 38982 38982 38982 38982 38777 31998 62 Table A 1.9: Agricultural Productivity and Multiple Employment Participation Œ First Stage Probit (6 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level ( 72 districts ). Due to data limitation, we use only the latest two survey waves . None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage Log Value of Yields per Ha - 6 Year MA 0.123 0.169 -0.354*** 0.028 -0.176 -0.318 (0.55) (1.40) (-2.99) (0.14) (-0.61) (-0.52) 25-34 # Log Value of Yields per Ha - 6 Year MA -0.114 0.063 0.155 -0.140 0.077 0.565* (-0.68) (0.77) (1.55) (-1.31) (0.39) (1.65) 35-65 # Log Value of Yields per Ha - 6 Year MA -0.469*** -0.057 0.328*** -0.170 -0.031 0.094 (-3.42) (-0.64) (3.30) (-1.39) (-0.16) (0.37) 25-34 1.071 -0.967 -0.867 1.451 -0.339 -4.453 (0.75) (-1.41) (-1.04) (1.62) (-0.20) (-1.54) 35-65 4.344*** -0.029 -2.229*** 1.633 0.512 -0.356 (3.77) (-0.04) (-2.70) (1.61) (0.32) (-0.16) Female 0.024 0.271*** -0.211*** -0.365*** -0.275*** -0.169** (0.49) (11.01) (-6.51) (-8.75) (-4.44) (-2.00) Relation to HH Head (base: Head) Spouse 0.148* 0.454*** -0.205*** -0.218*** -0.351*** -0.183* (1.80) (13.75) (-5.04) (-4.87) (-5.60) (-1.66) Child (own/step) 0.958*** 0.706*** -0.768*** -0.173*** -0.493*** -0.198 (12.48) (19.17) (-16.92) (-2.87) (-7.08) (-1.06) Relative 0.945*** 0.864*** -0.947*** -0.342*** -0.608*** -0.418** (11.31) (18.34) (-19.50) (-5.25) (-8.10) (-2.42) Unrelated -0.466* 0.638*** -0.490*** -0.066 -0.236*** -0.125 (-1.74) (10.66) (-6.12) (-0.85) (-2.72) (-0.46) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabi 0.102 0.050 -0.013 -0.092** -0.084 0.096 (1.23) (1.33) (-0.28) (-2.27) (-1.45) (0.72) Never Married 0.251*** 0.201*** -0.382*** -0.135*** -0.336*** 0.118 (3.85) (7.60) (-9.46) (-3.17) (-3.97) (0.58) Education Attained:base-None Primary -0.388*** -0.022 0.137*** -0.052 0.099 -0.008 (-4.76) (-0.68) (3.60) (-1.10) (1.40) (-0.06) Junior High -0.520*** -0.046 0.212*** -0.014 0.065 -0.051 (-5.85) (-1.27) (4.66) (-0.26) (0.83) (-0.36) Senior High -0.425*** -0.073** 0.135*** 0.106* 0.131 -0.011 (-4.25) (-1.97) (2.69) (1.73) (1.40) (-0.06) College and above -0.228 -0.509*** -0.487*** 0.850*** 0.692*** 0.632*** (-1.63) (-8.10) (-3.62) (8.92) (6.53) (2.76) Log Head Age 0.085 0.229*** -0.171*** -0.076 -0.305*** 0.009 (0.90) (4.80) (-3.68) (-1.39) (-3.10) (0.06) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabi -0.041 -0.018 -0.014 0.040 0.095 0.009 (-0.53) (-0.62) (-0.35) (0.85) (1.37) (0.07) Never Married -0.195 -0.108 0.309*** 0.036 0.088 0.233 (-0.76) (-0.98) (3.17) (0.27) (0.41) (0.59) Head Education: base-None Primary -0.007 -0.001 0.015 -0.005 -0.085 -0.091 (-0.09) (-0.04) (0.38) (-0.13) (-1.36) (-0.62) Junior High 0.062 -0.018 -0.020 -0.022 0.011 -0.016 (0.68) (-0.42) (-0.40) (-0.43) (0.15) (-0.11) Senior High 0.078 0.051 -0.083 -0.053 -0.025 -0.010 (0.86) (0.97) (-1.45) (-0.77) (-0.27) (-0.05) College and above 0.168 -0.036 -0.015 -0.087 -0.006 -0.210 (1.07) (-0.46) (-0.20) (-0.83) (-0.07) (-0.95) Log Household Size 0.095 0.003 -0.019 0.009 -0.091* 0.040 (1.40) (0.09) (-0.59) (0.29) (-1.95) (0.38) Log Male Children -0.102*** 0.095*** -0.055** -0.058** -0.054* -0.035 (-2.75) (5.17) (-2.52) (-2.55) (-1.75) (-0.53) Log Land Owned (Ha) -0.028** 0.026*** 0.018** -0.040*** -0.020** -0.032** (-2.19) (4.17) (2.07) (-6.41) (-2.35) (-2.48) Female Head -0.025 -0.135*** 0.248*** 0.045 0.177** 0.215* (-0.27) (-3.71) (5.49) (0.99) (2.54) (1.69) Log TLUs -0.039 0.078*** 0.001 -0.170*** -0.048** -0.112** (-1.48) (7.04) (0.08) (-10.04) (-2.43) (-2.56) Cell Phone 0.003 -0.035 0.108*** -0.083*** -0.018 -0.028 (0.05) (-1.41) (4.44) (-2.87) (-0.43) (-0.34) Solar/Generator -0.046 0.000 0.104*** -0.120*** -0.086** -0.001 (-0.88) (0.00) (4.96) (-4.21) (-2.34) (-0.01) HH Bank Account 0.059 -0.110*** -0.023 0.168*** 0.131** 0.267*** (0.96) (-4.41) (-0.84) (5.04) (2.49) (3.91) House Type Cement Floor 0.189*** -0.015 0.010 -0.051* -0.119** 0.102 (3.78) (-0.73) (0.37) (-1.65) (-2.36) (1.35) Permanent Roof -0.016 0.042* 0.024 -0.098*** -0.109*** -0.007 (-0.29) (1.87) (0.81) (-3.15) (-2.68) (-0.07) Observations 38982 38982 38982 38982 38777 31998 63 Table A 1.10: Agricultural Productivity and Multiple Em ployment Participation by Gender Œ First Stage Probit (2 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level ( 72 districts ). Due to da ta limitation, we use only the latest two survey waves . None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage Log Value of Yields per Ha - 2 Year MA 0.732** 0.044 -0.244** -0.129 -0.488* 0.131 (2.56) (0.35) (-2.05) (-0.84) (-1.93) (0.34) 15-24 # Female # Log Value of Yields per Ha - 2 Year MA -0.200** -0.060 0.058 0.072 -0.033 -0.121 (-2.19) (-1.32) (0.90) (0.80) (-0.32) (-0.68) 25-34 # Male # Log Value of Yields per Ha - 2 Year MA -0.198 0.065 0.148* -0.126 0.296 0.301 (-1.29) (0.88) (1.65) (-1.33) (1.58) (0.88) 25-34 # Female # Log Value of Yields per Ha - 2 Year MA -0.418** 0.011 0.224** -0.061 0.301 0.173 (-2.34) (0.13) (2.02) (-0.69) (1.46) (0.47) 35-65 # Male # Log Value of Yields per Ha - 2 Year MA -0.536*** 0.013 0.247*** -0.110 0.193 0.020 (-4.03) (0.16) (2.72) (-1.00) (1.16) (0.08) 35-65 # Female # Log Value of Yields per Ha - 2 Year MA -0.762*** -0.052 0.333*** -0.051 0.209 -0.114 (-4.73) (-0.57) (2.95) (-0.50) (1.02) (-0.38) 25-34 1.836 -0.996 -0.871 1.349* -2.253 -2.175 (1.42) (-1.63) (-1.17) (1.72) (-1.46) (-0.76) 35-65 4.958*** -0.586 -1.648** 1.173 -1.486 0.315 (4.54) (-0.82) (-2.22) (1.30) (-1.09) (0.15) Female 1.755** 0.773** -0.810 -0.922 -0.255 0.897 (2.29) (2.05) (-1.50) (-1.24) (-0.30) (0.60) Relation to HH Head (base: Head) Spouse 0.266** 0.471*** -0.286*** -0.173*** -0.456*** -0.142 (2.47) (11.77) (-5.99) (-3.51) (-6.54) (-1.12) Child (own/step) 0.987*** 0.720*** -0.791*** -0.166*** -0.501*** -0.190 (11.67) (19.24) (-17.64) (-2.75) (-7.08) (-1.04) Relative 0.970*** 0.875*** -0.965*** -0.335*** -0.613*** -0.414** (11.00) (17.66) (-19.83) (-5.10) (-8.15) (-2.46) Unrelated -0.397 0.644*** -0.540*** -0.036 -0.314*** -0.130 (-1.56) (9.74) (-6.63) (-0.45) (-3.75) (-0.46) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.143* 0.058 -0.054 -0.068 -0.142** 0.121 (1.65) (1.54) (-1.14) (-1.52) (-2.31) (0.83) Never Married 0.260*** 0.210*** -0.421*** -0.122*** -0.379*** 0.129 (3.97) (7.57) (-10.15) (-2.76) (-4.39) (0.63) Education Attained:base-None Primary -0.392*** -0.025 0.143*** -0.057 0.103 -0.009 (-4.74) (-0.75) (3.77) (-1.20) (1.44) (-0.07) Junior High -0.525*** -0.049 0.221*** -0.020 0.078 -0.058 (-5.79) (-1.35) (4.93) (-0.36) (0.96) (-0.41) Senior High -0.435*** -0.075** 0.147*** 0.100 0.144 -0.018 (-4.25) (-2.02) (2.99) (1.62) (1.53) (-0.11) College and above -0.240* -0.515*** -0.468*** 0.843*** 0.713*** 0.615*** (-1.69) (-8.23) (-3.48) (8.92) (6.59) (2.66) Log Head Age 0.095 0.229*** -0.183*** -0.067 -0.328*** 0.019 (1.00) (4.84) (-3.93) (-1.28) (-3.32) (0.12) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.046 -0.022 0.013 0.024 0.131* -0.006 (-0.61) (-0.72) (0.31) (0.49) (1.83) (-0.04) Never Married -0.189 -0.112 0.328*** 0.032 0.110 0.226 (-0.74) (-1.01) (3.35) (0.24) (0.50) (0.58) Head Education: base-None Primary -0.006 -0.000 0.012 -0.003 -0.088 -0.091 (-0.07) (-0.01) (0.30) (-0.08) (-1.41) (-0.63) Junior High 0.059 -0.016 -0.024 -0.020 0.003 -0.014 (0.66) (-0.39) (-0.49) (-0.40) (0.04) (-0.10) Senior High 0.079 0.052 -0.089 -0.050 -0.036 -0.006 (0.86) (1.00) (-1.55) (-0.73) (-0.39) (-0.03) College and above 0.163 -0.034 -0.022 -0.085 -0.019 -0.201 (1.03) (-0.43) (-0.30) (-0.80) (-0.21) (-0.91) Log Household Size 0.101 -0.000 -0.014 0.008 -0.085* 0.036 (1.47) (-0.02) (-0.46) (0.26) (-1.82) (0.35) Log Male Children -0.102*** 0.094*** -0.058*** -0.056** -0.059* -0.033 (-2.77) (5.23) (-2.65) (-2.52) (-1.88) (-0.49) Log Land Owned (Ha) -0.029** 0.025*** 0.018** -0.040*** -0.020** -0.033** (-2.24) (4.15) (2.10) (-6.40) (-2.28) (-2.48) Female Head -0.012 -0.126*** 0.196*** 0.069 0.100 0.238* (-0.13) (-3.42) (3.96) (1.52) (1.21) (1.81) Log TLUs -0.040 0.079*** -0.000 -0.170*** -0.050** -0.110** (-1.51) (7.10) (-0.03) (-9.99) (-2.55) (-2.48) Cell Phone 0.000 -0.034 0.107*** -0.082*** -0.019 -0.025 (0.01) (-1.37) (4.42) (-2.84) (-0.43) (-0.31) Solar/Generator -0.048 0.000 0.104*** -0.120*** -0.085** -0.003 (-0.92) (0.02) (4.98) (-4.20) (-2.33) (-0.05) HH Bank Account 0.065 -0.110*** -0.023 0.167*** 0.131** 0.269*** (1.06) (-4.47) (-0.86) (5.00) (2.51) (3.90) House Type Cement Floor 0.192*** -0.016 0.011 -0.051 -0.118** 0.100 (3.81) (-0.78) (0.38) (-1.63) (-2.33) (1.33) Permanent Roof -0.015 0.042* 0.024 -0.098*** -0.110*** -0.007 (-0.27) (1.83) (0.82) (-3.16) (-2.72) (-0.07) Observations 38982 38982 38982 38982 38777 31998 64 Table A 1.11: Agricultural Productivity and Multiple Employment Participation by Gender Œ First Stage Probit (4 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Errors are clustered at the district level ( 72 districts ). Due to data limitation, we use only the latest two survey waves . None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage Log Value of Yields per Ha - 4 Year MA 0.293 0.089 -0.260** -0.022 -0.145 0.051 (1.51) (0.84) (-2.24) (-0.14) (-0.52) (0.10) 15-24 # Female # Log Value of Yields per Ha - 4 Year MA -0.236** -0.020 0.009 0.088 -0.044 -0.031 (-2.50) (-0.34) (0.12) (0.91) (-0.40) (-0.17) 25-34 # Male # Log Value of Yields per Ha - 4 Year MA -0.147 0.070 0.130 -0.125 0.049 0.423 (-0.95) (0.95) (1.37) (-1.18) (0.25) (1.28) 25-34 # Female # Log Value of Yields per Ha - 4 Year MA -0.402** 0.057 0.157 -0.042 0.044 0.385 (-2.23) (0.59) (1.26) (-0.44) (0.19) (1.06) 35-65 # Male # Log Value of Yields per Ha - 4 Year MA -0.450*** 0.002 0.249** -0.152 -0.053 0.075 (-3.53) (0.02) (2.52) (-1.32) (-0.29) (0.31) 35-65 # Female # Log Value of Yields per Ha - 4 Year MA -0.712*** -0.023 0.285** -0.076 -0.048 0.033 (-4.41) (-0.22) (2.19) (-0.68) (-0.20) (0.11) 25-34 1.409 -1.049* -0.723 1.342 -0.217 -3.218 (1.08) (-1.71) (-0.91) (1.53) (-0.13) (-1.16) 35-65 4.261*** -0.493 -1.675** 1.526 0.553 -0.157 (4.04) (-0.69) (-2.07) (1.61) (0.36) (-0.08) Female 2.064*** 0.440 -0.405 -1.062 -0.160 0.140 (2.61) (0.87) (-0.64) (-1.31) (-0.17) (0.09) Relation to HH Head (base: Head) Spouse 0.273** 0.470*** -0.285*** -0.174*** -0.459*** -0.147 (2.54) (11.75) (-5.96) (-3.53) (-6.55) (-1.16) Child (own/step) 0.989*** 0.719*** -0.792*** -0.165*** -0.502*** -0.193 (11.71) (19.30) (-17.79) (-2.74) (-7.06) (-1.05) Relative 0.972*** 0.875*** -0.965*** -0.335*** -0.611*** -0.416** (11.00) (17.67) (-19.86) (-5.08) (-8.23) (-2.46) Unrelated -0.417 0.647*** -0.543*** -0.026 -0.308*** -0.120 (-1.59) (9.81) (-6.59) (-0.32) (-3.43) (-0.42) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.146* 0.058 -0.053 -0.070 -0.151** 0.120 (1.68) (1.53) (-1.11) (-1.57) (-2.46) (0.82) Never Married 0.262*** 0.209*** -0.416*** -0.123*** -0.387*** 0.129 (4.00) (7.46) (-10.07) (-2.78) (-4.50) (0.63) Education Attained:base-None Primary -0.395*** -0.025 0.144*** -0.056 0.109 -0.010 (-4.87) (-0.76) (3.77) (-1.18) (1.53) (-0.08) Junior High -0.530*** -0.049 0.222*** -0.019 0.082 -0.057 (-5.94) (-1.35) (4.91) (-0.35) (1.02) (-0.40) Senior High -0.440*** -0.075** 0.148*** 0.101* 0.150 -0.018 (-4.35) (-2.02) (2.98) (1.66) (1.59) (-0.11) College and above -0.246* -0.514*** -0.468*** 0.846*** 0.717*** 0.618*** (-1.74) (-8.22) (-3.49) (8.99) (6.65) (2.69) Log Head Age 0.096 0.230*** -0.183*** -0.068 -0.328*** 0.019 (1.00) (4.84) (-3.90) (-1.28) (-3.33) (0.12) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.051 -0.023 0.013 0.028 0.142** -0.007 (-0.67) (-0.74) (0.30) (0.57) (2.01) (-0.05) Never Married -0.200 -0.111 0.327*** 0.031 0.118 0.224 (-0.78) (-1.00) (3.34) (0.23) (0.54) (0.57) Head Education: base-None Primary -0.006 -0.000 0.012 -0.004 -0.091 -0.090 (-0.08) (-0.01) (0.30) (-0.11) (-1.45) (-0.62) Junior High 0.061 -0.016 -0.024 -0.020 0.002 -0.015 (0.68) (-0.39) (-0.49) (-0.40) (0.03) (-0.10) Senior High 0.079 0.052 -0.089 -0.051 -0.038 -0.007 (0.88) (1.00) (-1.55) (-0.74) (-0.41) (-0.03) College and above 0.169 -0.034 -0.022 -0.087 -0.019 -0.205 (1.08) (-0.43) (-0.30) (-0.82) (-0.21) (-0.93) Log Household Size 0.100 0.000 -0.015 0.007 -0.086* 0.038 (1.46) (0.01) (-0.48) (0.24) (-1.84) (0.37) Log Male Children -0.102*** 0.094*** -0.057*** -0.056** -0.059* -0.034 (-2.75) (5.23) (-2.63) (-2.53) (-1.88) (-0.51) Log Land Owned (Ha) -0.029** 0.025*** 0.018** -0.040*** -0.020** -0.033** (-2.22) (4.15) (2.10) (-6.41) (-2.29) (-2.49) Female Head -0.006 -0.126*** 0.197*** 0.067 0.094 0.238* (-0.06) (-3.42) (3.96) (1.47) (1.14) (1.81) Log TLUs -0.039 0.078*** 0.000 -0.170*** -0.049** -0.111** (-1.48) (7.02) (0.02) (-10.02) (-2.46) (-2.50) Cell Phone 0.002 -0.035 0.108*** -0.082*** -0.019 -0.027 (0.03) (-1.40) (4.42) (-2.85) (-0.44) (-0.33) Solar/Generator -0.047 0.000 0.104*** -0.120*** -0.085** -0.003 (-0.90) (0.00) (4.98) (-4.20) (-2.30) (-0.04) HH Bank Account 0.062 -0.110*** -0.023 0.167*** 0.131** 0.269*** (1.00) (-4.42) (-0.87) (5.01) (2.52) (3.84) House Type Cement Floor 0.192*** -0.015 0.010 -0.052* -0.120** 0.102 (3.83) (-0.75) (0.36) (-1.66) (-2.37) (1.35) Permanent Roof -0.015 0.042* 0.023 -0.098*** -0.109*** -0.007 (-0.27) (1.85) (0.81) (-3.15) (-2.69) (-0.07) Observations 38982 38982 38982 38982 38777 31998 65 Table A 1.12: Agricultural Productivity and Multiple Employment Participation by Gender Œ First Stage Probit (6 ŒLags Moving Average) TŒStatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes survey wave and district fixed effects. Erro rs are clustered at the district level ( 72 districts ). Due to data limitation, we use only the latest two survey waves . None Farm only Farm & Business Farm & Wage/Salary Farm, Business, & Wage/Salary Business or Wage Log Value of Yields per Ha - 6 Year MA 0.219 0.181 -0.348*** -0.007 -0.157 -0.319 (0.96) (1.39) (-2.75) (-0.04) (-0.53) (-0.52) 15-24 # Female # Log Value of Yields per Ha - 6 Year MA -0.232** -0.033 0.025 0.095 -0.021 -0.003 (-2.15) (-0.49) (0.30) (0.87) (-0.18) (-0.01) 25-34 # Male # Log Value of Yields per Ha - 6 Year MA -0.109 0.065 0.129 -0.141 0.051 0.570* (-0.65) (0.79) (1.30) (-1.26) (0.26) (1.65) 25-34 # Female # Log Value of Yields per Ha - 6 Year MA -0.359* 0.039 0.171 -0.052 0.068 0.560 (-1.90) (0.36) (1.33) (-0.51) (0.29) (1.46) 35-65 # Male # Log Value of Yields per Ha - 6 Year MA -0.438*** -0.050 0.298*** -0.164 -0.063 0.103 (-3.20) (-0.56) (2.94) (-1.32) (-0.34) (0.40) 35-65 # Female # Log Value of Yields per Ha - 6 Year MA -0.695*** -0.087 0.349*** -0.081 -0.035 0.088 (-3.92) (-0.81) (2.68) (-0.64) (-0.14) (0.28) 25-34 1.084 -1.003 -0.716 1.479 -0.232 -4.472 (0.76) (-1.48) (-0.86) (1.58) (-0.14) (-1.55) 35-65 4.172*** -0.060 -2.085** 1.626 0.632 -0.392 (3.66) (-0.08) (-2.50) (1.59) (0.41) (-0.18) Female 2.026** 0.546 -0.534 -1.120 -0.352 -0.097 (2.25) (0.98) (-0.77) (-1.21) (-0.36) (-0.06) Relation to HH Head (base: Head) Spouse 0.271** 0.470*** -0.285*** -0.174*** -0.460*** -0.148 (2.52) (11.74) (-5.95) (-3.53) (-6.55) (-1.17) Child (own/step) 0.991*** 0.719*** -0.791*** -0.164*** -0.503*** -0.198 (11.76) (19.32) (-17.78) (-2.73) (-7.07) (-1.07) Relative 0.973*** 0.875*** -0.965*** -0.334*** -0.612*** -0.420** (11.04) (17.68) (-19.83) (-5.07) (-8.23) (-2.47) Unrelated -0.421 0.642*** -0.539*** -0.024 -0.300*** -0.112 (-1.62) (9.72) (-6.53) (-0.30) (-3.26) (-0.39) Marital Status:base -Monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit 0.146* 0.055 -0.050 -0.071 -0.151** 0.119 (1.68) (1.46) (-1.05) (-1.59) (-2.46) (0.81) Never Married 0.263*** 0.206*** -0.413*** -0.124*** -0.387*** 0.130 (4.00) (7.39) (-9.99) (-2.78) (-4.51) (0.63) Education Attained:base-None Primary -0.394*** -0.024 0.143*** -0.056 0.109 -0.010 (-4.83) (-0.73) (3.74) (-1.17) (1.54) (-0.08) Junior High -0.528*** -0.049 0.222*** -0.019 0.082 -0.056 (-5.91) (-1.33) (4.90) (-0.34) (1.03) (-0.39) Senior High -0.438*** -0.075** 0.147*** 0.102* 0.150 -0.017 (-4.32) (-2.01) (2.98) (1.67) (1.60) (-0.10) College and above -0.243* -0.513*** -0.469*** 0.846*** 0.717*** 0.625*** (-1.72) (-8.20) (-3.49) (9.00) (6.67) (2.72) Log Head Age 0.095 0.230*** -0.182*** -0.068 -0.328*** 0.019 (1.00) (4.83) (-3.90) (-1.29) (-3.34) (0.12) Head Marital Status: base monogamously married Polygamous/Widowed/Divorced/Separated/Cohabit -0.051 -0.020 0.011 0.028 0.141** -0.004 (-0.67) (-0.68) (0.25) (0.57) (2.00) (-0.03) Never Married -0.198 -0.109 0.324*** 0.032 0.117 0.228 (-0.77) (-0.98) (3.30) (0.23) (0.54) (0.58) Head Education: base-None Primary -0.007 -0.001 0.013 -0.004 -0.091 -0.090 (-0.09) (-0.02) (0.31) (-0.11) (-1.45) (-0.61) Junior High 0.061 -0.017 -0.024 -0.020 0.001 -0.015 (0.68) (-0.39) (-0.49) (-0.41) (0.02) (-0.10) Senior High 0.078 0.052 -0.089 -0.051 -0.039 -0.007 (0.86) (1.00) (-1.55) (-0.75) (-0.41) (-0.03) College and above 0.168 -0.034 -0.022 -0.087 -0.021 -0.206 (1.07) (-0.43) (-0.30) (-0.82) (-0.23) (-0.93) Log Household Size 0.100 0.001 -0.016 0.007 -0.086* 0.038 (1.46) (0.02) (-0.49) (0.24) (-1.84) (0.36) Log Male Children -0.102*** 0.094*** -0.057*** -0.056** -0.059* -0.034 (-2.74) (5.23) (-2.62) (-2.53) (-1.88) (-0.50) Log Land Owned (Ha) -0.029** 0.025*** 0.018** -0.040*** -0.020** -0.032** (-2.24) (4.16) (2.11) (-6.41) (-2.29) (-2.47) Female Head -0.007 -0.127*** 0.198*** 0.067 0.094 0.237* (-0.07) (-3.44) (3.98) (1.47) (1.15) (1.79) Log TLUs -0.039 0.078*** 0.000 -0.170*** -0.049** -0.112** (-1.47) (7.02) (0.02) (-10.03) (-2.48) (-2.55) Cell Phone 0.002 -0.035 0.108*** -0.082*** -0.019 -0.028 (0.04) (-1.41) (4.43) (-2.86) (-0.45) (-0.34) Solar/Generator -0.046 -0.000 0.104*** -0.120*** -0.085** -0.002 (-0.89) (-0.00) (4.98) (-4.20) (-2.31) (-0.03) HH Bank Account 0.061 -0.110*** -0.023 0.168*** 0.132** 0.268*** (0.98) (-4.42) (-0.85) (5.02) (2.53) (3.91) House Type Cement Floor 0.192*** -0.015 0.010 -0.051* -0.120** 0.102 (3.82) (-0.74) (0.34) (-1.65) (-2.36) (1.35) Permanent Roof -0.016 0.042* 0.023 -0.098*** -0.109*** -0.007 (-0.28) (1.85) (0.81) (-3.15) (-2.69) (-0.07) Observations 38982 38982 38982 38982 38777 31998 66 BIBLIOGRAPHY 67 BIBLIOGRAPHY Adamopoulos, T., Brandt, L., Leight, J., & Restuccia, D. (2017). Misallocation, selection and productivity: A quantitative analysis with panel data from China (No. w23039). National Bureau of Economic Research. Bryan, G., Chowdhury, S., & Mobarak, A. M. (2014). Underinvestment in a profitable technology: The case of seasonal migration in Bangladesh. Econometri ca, 82(5), 1671 Œ1748. Bustos, P., Caprettini, B., & Ponticelli, J. (2016). Agricultural productivity and structural transformation: Evidence from Brazil. American Economic Review , 106(6), 1320 Œ65. Chen, C., Restuccia, D., & Santaeulàlia ŒLlopis, R. (2017) . The effects of land markets on resource allocation and agricultural productivity (No. w24034). National Bureau of Economic Research. Diao, X., Hazell, P., & Thurlow, J. (2010). The role of agriculture in African development. World development, 38(10), 1 375 Œ1383. Duflo, E., & Udry, C. (2004). Intrahousehold resource allocation in Cote d'Ivoire: Social norms, separate accounts and consumption choices (No. w10498). National Bureau of Economic Research. Emerick, K. (2018). Agricultural productivity and the sectoral reallocation of labor in rural India. Journal of Development Economics , 135, 488 Œ503. Food and Agriculture Organization of the United Nations (2014). https://www.un.org/en/ecosoc/integration/pdf/foodandagricultureorganization.pdf Foster, A. D., & Rosenzweig, M. R. (2002). Household division and rural economic growth. The Review of Economic Studies , 69(4), 839 Œ869. Foster, A. D., & Rosenzweig, M. R. (2004). Agricultural productivity growth, rural economic diversity, and economic reforms: India, 1970Œ2000. Economic Development and Cultural Change , 52(3), 509 Œ542. Foster, A. D., & Rosenzweig, M. R. (2007). Economic development and the decline of agricultural employment. Handbook of development economics , 4, 3051Œ3083. Goldstein, M., & Udry, C. (2 008). The profits of power: Land rights and agricultural investment in Ghana. Journal of political Economy , 116(6), 981 Œ1022. Gollin, D., Lagakos, D., & Waugh, M. E. (2014). The agricultural productivity gap. The Quarterly Journal of Economics, 129(2), 93 9Œ993. 68 Gollin, D., Parente, S., & Rogerson, R. (2002). The role of agriculture in development. American economic review , 92(2), 160 Œ164. Gorman, M. (2013). Older women, older farmers Šthe hidden face of agriculture. Briefing, London: HelpAge. Retrieved fr om http://www.helpage.org/silo/files/older Œwomen Œolder Œfarmers Œthe Œhidden Œface ŒofŒagriculture.pdf Gottlieb, C., & Grobovıek, J. (2019). Communal land and agricultural productivity. Journal of Development Economics , 138, 135Œ152. Harris, J. R., & Todaro, M. P. (1970). Migration, unemployment and development: a two Œsector analysis. The American economic review , 60(1), 126 Œ142. Herrendorf, B., Rogerson, R., & Valentinyi, A. (2013). Two perspectives on preferences and structural transformation. American Econo mic Review , 103(7), 2752 Œ89. Hirschman, A. O. (1958). The Strategy of Economic. Development. Yale University Press, Connecticut . Irz, X., Lin, L., Thirtle, C., & Wiggins, S. (2001). Agricultural productivity growth and poverty alleviation. Development po licy review , 19(4), 449 Œ466. Jayne, T., D. Mather, & E. Mghenyi. 2010. Principal Challenges Confronting Smallholder Agriculture in Sub ŒSaharan Africa. World Development 38 (10): 1384 Œ98. Johnston, B. F., & Mellor, J. W. (1961). The role of agriculture in economic development. The American Economic Review, 51(4), 566 Œ593. Kongsamut, P., Rebelo, S., & Xie, D. (2001). Beyond balanced growth. The Review of Economic Studies , 68(4), 869 Œ882. Kosec, K., Ghebru, H., Holtemeyer, B., Mueller, V., & Schmidt, E. (2 018). The effect of land access on youth employment and migration decisions: Evidence from rural Ethiopia. American Journal of Agricultural Economics , 100(3), 931 Œ954. Lagakos, D., & Waugh, M. E. (2013). Selection, agriculture, and cross Œcountry productiv ity differences. American Economic Review , 103(2), 948 Œ80. Lewis, W. A. (1954). Economic development with unlimited supplies of labour. Matsuyama, K. (1992). Agricultural productivity, comparative advantage, and economic growth. Journal of economic theor y, 58(2), 317 Œ334. McMillan, M. S., & Harttgen, K. (2014). What is Driving the'African Growth Miracle'? (No. w20077). National Bureau of Economic Research. 69 Muyanga, M., & T. Jayne. 2014. Effects of Rising Rural Population Density on Smallholder Agricultu re in Kenya. Food Policy 48 (Supplement C): 98 Œ113. Qian, N. (2008). Missing women and the price of tea in China: The effect of sex Œspecific earnings on sex imbalance. The Quarterly Journal of Economics, 123(3), 1251 Œ1285. Restuccia, D., & Santaeulalia ŒLlopis, R. (2017). Land misallocation and productivity (No. w23128). National Bureau of Economic Research. Schneider, K., & Gugerty, M. K. (2011). Agricultural productivity and poverty reduction: Linkages and pathways. Libraries Test Journal, 1(1), 56 Œ74. Shilpi, F., & Emran, S. (2016). Agricultural productivity and non Œfarm employment: evidence from Bangladesh . The World Bank. Snyder, J., Jayne, T., Mason, N., & Samboko, P. (2019). Agricultural Productivity and rural household incomes: Micro Œlevel evide nce from Zambia (No. 1879 Œ2020Œ456). Vos, R. (2015). Thought for Food: Strengthening global governance of food security. committee for development policy background paper no. 29. UN Department of Economic and Social Affairs, New York, NY. Retrieved from http://www.un.org/en/development/desa/policy/cdp/cdp \_background \_papers/bp2015 \_29.pdf Yeboah, F. K., & Jayne, T. S. (2018). Africa™s evolving emplo yment trends. The Journal of Development Studies , 54(5), 803 Œ832. 70 Chapter 2 : Does Agricultural Productivity Translate to Increased Human Capital Investment? Impacts of Productivity Shocks on Education Expenses and School Outcomes in Tanzania 71 I. INTRODUCTION The World Bank data indicate that agriculture remains a major source of livelihood in Tanzania and a number of countries in Sub ŒSaharan Africa (World Bank, 2020). In Tanzania, the World Bank estimates that in 2013 agriculture contributed 27% of value add ed to the GDP and accounted for 69% of employment. While agriculture continues to play an important role, high population growth rate and a bulging youth population imply that agriculture alone will not be sufficient to meet the future needs of the contine nt. Countries that have successfully developed have done so, in part, through increased productivity and investment in human capital. Consequently, this paper investigates the impact of agricultural productivity shocks on investments in human capital and school outcomes. Rural households in Sub ŒSaharan Africa are highly dependent on agriculture and thus any expenditure decisions are highly tied to agricultural income. In this paper, I investigate whether positive productivity shocks , at the household leve l, lead to increased expenditure on education and study time. Unlike in the previous chapter where we focused on productivity at the district level, in this chapter we focus on productivity at the household level. Agriculture can affect outcomes through tw o major channels. First, increased productivity relaxes financial constraints facing the household. Assuming that education is a normal good, higher productivity can induce increased schooling expenditures on items such as fees, uniforms, books, among othe rs. Second, child labor is very common among developing countries. Increased productivity can have a number of countervailing effects on school attendance, progression, and study times. If increased farm productivity creates a demand within the household for additional farm labor, the demand for child labor could increase, leading to absenteeism, withdrawal, and even drop outs (Beegle et al., 2006; Dureya and Arends ŒKuenning, 2003; Kruger, 2007). On the other hand, increased farm 72 productivity might entail the adoption of new labor Œsaving technologies such as mechanization, hence freeing up child labor and encouraging school attendance. In addition, higher incomes resulting from gains in farm productivity could enable households to hire labor to substitute for child family labor and encourage school attendance. The net impact of changes in agricultural productivity on child labor, and school expenditure, is thus an empirical question. Few related studies on these issues have access to high quality nationall y representative data. Finally, this paper looks at the role of gender of students in outcomes and on whether productivity shocks have differential impacts depending on the gender of the household head. Specifically, are boys and girls affected differently ? Are the results different when female Œheaded households experience productivity shocks than when male Œheaded households experience similar shocks? This is in the spirit of the literature that has found gender to be important in affecting household decisi ons (Duflo, 2003; Thomas, 1990, 1994). We make a few contributions to the existing literature . First, we provide empirical estimates on the role of agricultural productivity growth in human capital investment particularly in Sub ŒSaharan Africa . Second, we provide insights on the dynamics of agricultural productivity and child labor and its consequences on study time. Third, we develop a theoretical model and empirically explore the impacts of productivity along gender lines of school Œage children. Fourth, we contribute to the literature on household bargaining models by investigating whether the productivity shocks have any differential effects on investment in education and education outcomes depending on the gender of the household head. Finally, we contr ibute to the debate on whether agricultural productivity shocks are procyclical in early childhood and countercyclical later by comparing effects of productivity on primary school versus post Œprimary school students. 73 Our results provide evidence that incre ased agricultural productivity boosts spending on uniform, contributions and total academic expenses. A 10% increase in land productivity results in 9%, 16%, and 6% increase in uniform, contribution, and total school expenditures respectively. In addition , a 10% increase in labor productivity leads to a 19%, 31%, and 12% increase in uniform, contribution, and total school expenditure . We find positive but statistically non Œsignificant effects of productivity on study times. In addition, we find no evidence of heterogeneous effects by s tudent gender. We show evidence that productivity effects are smaller in female Œheaded households. Finally, we find some evidence that post Œprimary students experience larger impacts compared to primary school students. The re st of the paper is organized as follows. In section 2, we review related literature to motivate our analysis and put our contributions in context. We then provide a brief theoretic model to guide our empirical analysis in section 3. In section 4, we discus s the data and the empirical methodology. Section 5 provides results while section 6 discusses our robustness checks. Section 7 concludes the analysis. II. LITERATURE REVIEW The goal of this paper is to estimate the direct impact of productivity shocks on ho usehold investment in child education expenses, and time devoted to studying. In addition, we will investigate heterogeneous effects by gender of the child as well as the gender of the household head. We briefly review some of the existing literature relat ed to our study. The literature focuses on the various mechanisms, such as income, nutrition, health, and child labor, through which agricultural income shocks affect schooling outcomes. They also provide evidence that the gender of the household head and the child may lead to heterogeneous impacts of shocks. 74 In Tanzania, as in many developing countries, agriculture is a major source of income. Consequently, any productivity shocks will affect income and hence spending on education. The role of income in de termining access to education has been documented in the literature (e.g. Deininger, 2003; Grimm 2011). Deininger (2003) shows that the introduction of free primary education in Uganda increased enrollment and decreased academic expenditure. While the qual ity of education arguably declined due to congestion, the study shows that costs of schooling can be an impediment to school access. Grimm (2011), similarly, shows that a decline in household income by 10% decreases enrollment rates of children (6 Œ13 year s) by 2.5 percentage points for boys and 3 percentage points for girls. In other words, shocks to agricultural income can not only reduce school investment but also lead to complete withdrawal from school. While income shocks may have immediate impact on s chool outcomes, some studies show long Œterm effect s. In Indonesia, Maccini and Ya ng (2009) show that early life negative rainfall shocks have adverse impacts on women™s long Œterm health, assets, and education attainment. They attribute these to the positiv e impacts of rainfall on agricultural output that result in higher incomes, food access, and better health for infant girls. Some studies show that investment in human capital is procyclical in early life but become countercyclical afterwards (Shah and Ste inberg , 2017; DeSalle, 2020). Shah and Steinberg (2017) show that Indian children report lower likelihood of work in drought years and higher school attendance. In the long run, adults who experienced higher rainfall during school years have lower total ye ars of schooling and lower wages. Rosales ŒRueda (2018), finds that children exposed to El Nino flooding (in Œutero) experience poor health, are shorter in stature five to seven years later, and score lower on cognitive tests in Ecuador. Child labor is anot her important channel through which agricult ural productivity shocks affect education outcomes. The extent of child labor largely depends on the income and substitution 75 effects. Specifically, increased productivity can increase household incomes leading to increased school expenditure and school attendance. On the other hand, increased productivity can be accompanied by agricultural wage increases that increase the opportunity cost of school attendance for children and hence can decrease enrollment. The net effect depends on the strength of the income and substitution effects. Dureya and Arends ŒKuenning (2003) provide evidence that improved local labor market conditions increases opportunity costs of schooling and results in higher incidence of child labor a nd discontinuation of schooling in Brazil. Kruger (2007) finds short Œterm variation in local economic conditions Œ proxied by coffee production Œ led to more child labor among middle Œincome boys and girls and school withdrawals of poor children in Brazil. In Tanzania, the focus of our study, DeSalle (2020) finds that favorable early life productivity shocks have a positive effect on development of future cognitive skills of the child but if positive shocks occur during school Œage time, they increase child l abor and reduce academic performance. The most common evidence in the literature tend to find that child labor is driven by push factors due to negative productivity shocks in agricultural households. In essence, these shocks can constrain household income s and force households to deploy children to pursue paid work or use child labor as a substitute for adult labor in household chores. Child labor tends to act as insurance against negative shocks either due to lack of access to credit or incomplete credit markets (Jacoby and Skoufias , 1997 ; Beegle et al., 2006). Jacoby and Skoufias (1997) show that child labor is a form of household self Œinsurance for poor households in rural India. Beegl e et al. (2006) find that, in Ka gera region of Tanzania, child labor a cts as a buffer against transitory shocks. They show that access to credit can reduce the need for child labor during periods of crop failure. One positive finding is that school enrolment decreased less than expected because many children were 76 able to co mbine school and work. However, the study does not provide any evidence that increased child labor does not affect child school performance. More evidence from Edmonds (2005) documents the impact of the introduction of the elderly pension program to black families in Post ŒApartheid South Africa. When an elderly male becomes eligible for the pension program in a black family, there are large increases in school attendance and decline in child labor. This evidence points to liquidity constraints among black families. The paper further shows that the impacts were more sensitive to income for boys than for girls. The authors argue that their findings indicate differences in credit access by gender. Given our interest in heterogeneity by gender, we provide addit ional empirical evidence that show that productivity shocks can have different impacts depending on the gender of the household head as well as the gender of the child. Dammert (2010) studies gender and sibling differences in time allocation within househo lds in Nicaragua and Guatemala. She shows that older boys spend more time engaged in market and domestic work, while older girls spend more time in domestic work compared to their younger siblings. While this paper does not investigate impacts of agricult ural productivity shocks, it indicates that the impacts can vary by gender given the documented evidence of gender differences in time allocation. Cameron and Worswick (2001) find that households with girls have a higher propensity to cut back on schooling expenditure following crop loss shocks in Indonesi a. Marchetta et al. (2019) show that negative rainfall deviations and cyclones reduce test scores, and school enrollment in Madagascar. While both boys and girls are likely to engage in work following the se shocks, the show that girls experience a larger adverse effect. Some evidence, following from household bargaining theory, suggest that the gender of the household head has important consequences on investment in education. The seminal paper by 77 Qian (2 008) provides strong empirical evidence of the significance of gender. She shows that an increase in sex Œspecific incomes has different impacts on boys and girls in China. While an increase in female income increases survival rates for girls, the impact is adverse when male income increases. Female income increases educational attainment for girls. However, male income worsens attainment among girls with no impact on boys. Related literature, show that women empowerment has positive impacts on child welfare (Saenz and Thompson, 2016; Wiig, 2013; Reggio, 2011). Saenz and Thompson (2016) show that the Zambian crop input subsidy program resulted in a greater reduction in crop diversification in male Œheaded households than in female Œheaded households. This refle cts different cropping decision s by gender and possibly differential impact of weather shocks. It is likely that female Œheaded households may be less susceptible to crop losses due to weather and pests, and consequently these shocks have different impacts on educational outcomes for their children. Wiig (2013) finds evidence that joint titling of land improved women empowerment in rural Peru. This evidence suggests that empowered women may play a large role in income allocation compared to less Œempowered women. This will likely result in different decision making during periods of agricultural productivity shocks. For instance, Reggio (2011) shows that, in Mexico, an increase in a mother™s bargaining power results in lower working hours for daughters than f or sons. Our study has a few strengths. First, we take advantage of a large nationally representative individual panel data spanning three survey waves. We add to the few literature on impacts of agricultural shocks on education outcomes in Africa. Unlike several past studies, this study provides evidence on study time and disaggregated individual Œlevel school expenditure data. Our paper is one of the few studies that focus on causal effects of productivity shocks on schooling outcomes in Sub ŒSaharan Afric a (Beegle et al., 2006; Boozer and Suri, 2001). Others related 78 studies directly focus on impacts of weather shocks on school outcomes (Jensen, 2000; World Bank, 2007; Marchetta et al., 2019). Finally, more broadly, we contribute to the literature on househ old bargaining models. III. THEORETICAL MODEL We start with a simple two Œperiod household model where a household head decides the level of household consumptions, education investment, and labor input. We assume that preferences for education differ according to the gender of the household head and the gender of the child. Specifically, we assume male heads prefer investing in male children while female heads put equal value in educating a child regardless of gender. These preferences are assumed to be driven b y two factors. First, educating a child yields immediate bragging rights in period 1. Second, investment in education yields returns in the second period and these investments will vary based on the expected return to education for a particular child. In o ther words, both gender discrimination/bias and differential expectation of returns influence levels of investments in education levels. We assume a representative house hold with a boy and a girl. The general household problem is as follows: max ,,,UC,C,L,L =UC,L+L +EU(C) s.t. + (+)=+(2) =+{.}() + {,} >0 ,>01 (1) 1 Lb and L g are assumed to be strictly positive to reflect the mandated minimum education requ irements per child. G overnments in developing countries have attempted to put in place mandatory school attendanc e for children espe cially for pre Œsecondary levels. 79 where a household head p in period 1 chooses consumption for period 1 and 2 ( ,), and investment in education by choosing labor allocated to school activities for boys ( ) and girls () subject to the intertemporal household budget constraint. First, the household must meet subsistence consumption needs ,. Second, the household faces a budget const raint where in each period i it receives an exogenous income . is a hicks Œneutral productivity shifter. In the empirical analysis, we use rainfall, which is arguably a hicks Œneutral technical factor, as an instrument for agricultural productivity . In addition, the household derives wage income from time not spent doing education related chores while in the second period the household derives additional income from returns to schooling ,(.), where G is a concave function with a normally distributed error term with a mean of zero. We further assume that education returns are increasing in the level of exogenous income levels in period 1 to reflect the idea that higher household income provides quality education etc. The expected returns to e ducation depend on gender of the head and the gender of the student b (boy)/g (girl) . The household head™s investment choices in education depends on beliefs on expected returns to education for a child of each gender. The direct costs of schooling is assu ming to be fixed at f with an indirect opportunity cost of w per unit of labor allocated to education activities Œ this can be the agricultural wage or the shadow agricultural wage. For simplicity, we assume the functional form of the returns to education as ,()=,() We specify a simple utility function similar to those of Stone ŒGeary preferences. For simplicity, we assume away the error term and let G to be deterministic. In particular, a household head solves the objective funct ion in equation (2). denotes the weight on the utility derived from educating a child of gender i. Notice that the optimal consumption and investment levels are pinned downed by choosing ,: 80 max ,(,,,) =() + () + () + () (2) s.t. =+2+ =+,()+ , >0 ,>0 The first order conditions are given by: L:+ =+1, (3) L:+ =+1, (4) Equations (3) a nd (4) indicate that the optimal education labor investment equalizes the marginal utility cost of academic labor and the marginal utility benefit of education. The marginal benefit is the sum of period 1 marginal utility from bragging rights and the perio d 2 discounted marginal utility from consumption from returns to education. If either the bragging rights or returns to education are higher, then the investments in education are high. Combining the first order conditions: =1[,,](5) Equation (5a) shows the tradeoffs between investing in education for a boy and a girl. This equation pins down the relationship between L and L. At the optimum, t he marginal benefit of a boy's education net of foregone marginal benefit s of investing in a girl's education must equal the discounted difference in the net benefit of education returns between boys and girls in period 2. Alternatively, the marginal utili ty from investing in a boy must equal the marginal utility in investing in a girl. 81 +1, =+1, (5) If a parent (or household) receives larger bragging benefits from educating a boy, and/or have expectations of higher returns from educating a boy, then the parent (or household) will invest more in boys than in girls. The converse is true. If a parent (or household head) places equal weights on both the bragging benefits and expected returns from educating a child, regardless of gender, then the househol d will invest equal amounts in both genders. Finally, we can perform comparative statics to investigate the impact of productivity shocks on the investment in boys and girls. From (3) and (4) we can show , respectively, that: L=+()+()++(),>0 L=+()+()++(),>0 For simplicity, we can assume that and are relatively smaller compared to , and ,. Then we can show that under our assumptions while boys enjoy higher levels of education investments, girls ™ investments are more responsive to productivity shocks : LL=+()++(),+()++(),>1 (5) Empirically, differences in levels of investments in education between boys and girls will indicate a likelihood of existence of discriminatory bragging rights and/or different expected returns to edu cation by gender. In addition, we expect girls ™ education expenses to be relatively more responsive to productivity shocks. This theoretical exercise is meant to motivate our 82 empirical analysis but is not meant to perfectly model all factors at play in hou sehold decisions on education. Our empirical analysis cannot distinguish the roles of discrimination and differences in expected returns to education in investment choices. IV. DATA AND METHODS Data for this analysis is from the World Bank™s Living Standards M easurements Survey (LSMS) in Tanzania. We use data from the first three waves 2008/09, 2010/11, and 2012/13. This data has a rich education module that captures individual level school expenses over the previous 12 months and study time in the previous 7 d ays from the date of the survey. Consequently, we restrict our sample to individuals who are currently in school or were in school the previous academic year. The study time variable is only available for students who are enrolled in school. We use these waves to create individual panels. Productivity is measured as the gross value of all crop output produced on the farm per hectare planted or per labor Œday (family and hired). Note that all labor including child labor, both family and hired, is treated equa lly. We also use alternative measures of productivity for robustness checks Œ gross crop income, net crop income per hectare planted, and net crop income. The net values subtracts explicit cash costs from gross values. We measure productivity at the househ old level. Table 2.1 shows the summary statistics for the main outcomes and dependent variables as well as the control variables. We estimate a panel fixed effects model at the individual level =+ + +++ (6) Where y are the set of outcomes Œ education expenses and study time Œ , iht represents individual i in household h at time t. prod is the household harvest value per hectare or per labor Œday. X is a vector of hou sehold and individual controls, is an individual fixed effect while is a time fixed effect. Two measures of productivity are land productivity (Tsh/Ha) and labor productivity 83 (Tsh/Days of Labor). The time fixed effects, , control for factors that are invariant across all individuals within time t. On the other hand, the individual fixed effects, , controls for individual factors that are constant over time. The fixed effects identification relies on the assumption that condition al on the observed characteristics and fixed effects, the unobserved components are orthogonal to productivity. Unfortunately, the identification strategy above does not control for relevant unobserved time Œvarying factors that are correlated with producti vity and affect outcomes of interest. This is one shortcoming of our models (FE and IV). To address some endogeneity concerns, we complement the FE estimation strategy above by using rainfall to instrument for productivity. The IV also addresses simultanei ty concerns since labor use, and hence time use, and productivity are simultaneously determined. Increased productivity can lead to increased demand for labor resulting in decreased study Œtime, school participation, and school expenses. On the other hand, increasing farm labor input, and hence reduced study time, school participation, and expenses, can lead to increased productivity. Similarly, cutting down on school expenses to hire more inputs can increase land productivity. In addition, there can be a m echanical relationship between labor and productivity measures. Specifically, if school child labor leaves then land productivity drops. In a similar vein, and considering that child labor is relatively less productive, reduction in child labor mechanicall y increases productivity. T hese introduce identification problems as they also contributes to reverse causality. The mechanical relationship between productivity and schoolchild labor may not be severe if children are able to combine school and work in pre sence of shocks. Beegle et al. (2006) argue that when households experienced crop loss shocks in Tanzania, school enrolment decreased less than expected because many children were able to combine school and work. To solve the concerns raised above we instr ument for productivity using district rainfall during the wettest 84 quarter during the long rain seasons (March to May). We also use district quarterly deviations of rainfall as a robustness check. These rainfall shocks create exogenous variations in the pro ductivity measures. We expect that the variations in productivity due to weather shocks to be relatively larger than those driven mechanically by labor changes. The corresponding first stage equation is given by: =++ +++ (7) Where i is individual, is the productivity at household h in district d at time t, Rain is total rainfall during the months of March to May (wettest quart er during long rains), X are a vector of household and individual characteristics, are individual fixed effects and are the time fixed effects. Note that productivity only varies at the household level while rainfall varies at the district level. Table A 2.1 shows that the instrument is a strong predictor of agricultural productivity. The second requirement for instrument validity is that rainfall is exogenous. That is, the only way through which rainfall affects educational outcomes is only through its impact on agricultural productivity. Tanzania is a largely agricultural middle income country that relies heavily on rainfall for agricultural activities. Since our study relies on rainfall deviations and not extreme weather events, we argue t hat the main impact of rainfall on educational activities and spending is mostly through its impact on agricultural productivity and incomes. Several studies have also used rainfall as an instrument for agricultural productivity (e.g. Emerick, 2019). Our second stage equation is given by: =+ + +++ (8) 85 Table 2.1: Summary Statistics Variable Obs Mean Std. Dev. Land Productivity (Tsh/Ha) 11,061529,663 791,106 Labor Productivity (Tsh/Man Days) 11,0614,356 6,643 Student Characteristics Fee Expenses (Tsh) 11,06125,297 120,175 Book Expenses (Tsh) 11,0617,677 13,289 Uniform Expenses (Tsh) 11,06111,513 11,054 Transport Expenses (Tsh) 11,0611,894 12,412 Tuition Expenses (Tsh) 11,0613,283 12,904 Contribution Expenses (Tsh) 11,0616,534 15,386 Food Expenses (Tsh) 11,0615,802 25,978 Total School Expenses (Tsh) 11,06166,504 183,636 Study time (Minutes) 7,88289 209 In School 11,0610.91 0.29 Female 11,0610.49 0.50 Relationship to Head Child 11,0610.76 0.43 Grandchild 11,0610.16 0.37 Relative 11,0610.07 0.26 Non-Relative 11,0610.005 0.068 Marital Status Married 11,0610.003 0.05 Other Status 11,0610.003 0.05 Female Head 11,0610.21 0.40 Adult Equivalent 11,0616.35 3.23 Head Characteristics Primary 11,0610.66 0.47 Secondary 11,0610.11 0.31 Post Secondary 11,0610.005 0.07 Age 11,06150 13.00 Formal Land Rights 11,0610.18 0.38 Tropical Livestock Units 11,0613.32 23.88 Farm Size 11,0612.73 3.38 Value of Crop Yields 11,061914,089 1,503,416 Wage Income 11,06152,973 199,753 Income Transfers 11,06148,832 201,552 Non-Farm Income 11,0613,786,434 13,500,000 86 V. RESULTS a. Overall Results The following tables are shortened but full tabl es are available in the appendix section. Table 2.2 shows the effects of land productivity on education expenditure and study time. The results indicate that while productivity has positive effects on expenditure, it is only statistically significant for s chool fees. In particular, a 10% increase in land productivity yields a 0.6% increase in education spending on fees. The effect on study times is quantitative small and statistically insignificant. Table 2.3 shows consistent findings for labor productivity . The fixed effects estimates are qualitatively similar to those for land productivity. Table 2.2: Effects of Land Productivity on School Expenditure and Study Times (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.3: Effects of Labor Productivity Expenditure and Study Times (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the hous ehold level. Errors clustered at the individual level. Table 2.4 shows the IV estimates. These estimates are more encouraging. Increased land productivity results in increased spending on uniform, school contribution, and total school (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.059*** 0.001 0.026 0.018 -0.026 0.020 0.020 -0.007 (3.26) (0.04) (1.26) (1.42) (-1.13) (1.08) (1.51) (-0.34) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.055* 0.012 0.007 0.025 -0.022 0.040 0.026 -0.013 (1.87) (0.50) (0.23) (1.17) (-0.62) (1.36) (1.33) (-0.37) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 87 expenditures. A 10% increase in land productivity results in 9%, 16%, and 6% increases in uniform, contribution, and total school expenditures respectively. The results also indicate that land productivity has a positive impact on book spending and total study time but the r esults are not statistically significant. Table 2.5, shows that the IV estimates using labor productivity are qualitatively similar to those for land productivity but the estimates are much larger. Specifically, a 10% increase in labor productivity leads to a 19%, 31%, and 12% increase in uniform, contribution and total school expenditure. In addition, a 10% increase in labor productivity increases study time by 10% but latter finding is not statistically significant. Taken together and focusing on the IV estimates, we find that both land and labor productivity results in increased expenditure on school with a potential to increase study times. In addition, labor productivity has a larger impact on outcomes. The labor productivity coefficients tend to be larger compared to the land productivity coefficients. One potential explanation is that the benefits of high labor productivity tend to be immediate and allows households to respond quickly and easily to positive shocks. Labor productivity frees up resour ces that can be used to increase school expenses and study times. On the hand, the benefits of increased yields (land productivity) may not materialize until harvests are completed and hence the impacts on educational investments may be relatively lower. 88 Table 2.4: Effects of Land Productivity on School Expenditure and Study Times (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth f ixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.5: Effects of Labor Productivity on School Expenditure and Study Times (IV) TŒsta tistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. b. Results by Student G ender The overall results may hide heterogeneity i n effects by gender. Table 2.6 Œ2.9 repeats the previous analysis while interacting the productivity variable with the gender dummy vari able. In Table 2.6 Œ2.7, the fixed effects estimate show that land and l abor productivity tend to have statistically non Œsignificant effects on outcomes. We therefore focus on the pre ferred IV estimates in Table 2.8 Œ2.9. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.217 0.320 0.938** 0.205 1.552*** 0.267 0.617* 0.354 (-0.65) (1.00) (2.21) (1.08) (2.64) (0.80) (1.88) (1.28) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.433 0.638 1.873** 0.410 3.097** 0.532 1.232* 1.029 (-0.64) (0.99) (2.02) (1.06) (2.31) (0.79) (1.77) (1.17) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 89 Table 2.6: Effects of Land Productivity on School Expendit ure and Study Times by Gender (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at th e individual level. Table 2.7: Effects of Labor Productivity on School Expenditure and Study Times by Gender (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.8 shows that a 10% increase in agricultural land productivity academic spending on books, uniform, contributions and total expenses by 6% 10 % 13%, and 7.8% respectively. On the other hand, there is suggestive evidence that land productivity may have differential effects by gender. Specifically, for girls, the impacts are higher for transport, contribu tions, and food, while lower for books, uniform, and total expenses. In addition, girls appear to experience larger positive impacts on time spent studying. However, these gender effects are not statist ically significant. Table 2.9 repeats the analysis u sing labor productivity as the dependent variable. The results are qualitatively similar to those in Table 2.5 with few differences. First, the coefficients are generally larger in magnitude. Second, only coefficients on uniform, contributions, and total e xpenses (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.043* 0.006 0.000 -0.008 -0.033 0.036 0.015 -0.040 (1.89) (0.25) (0.01) (-0.47) (-1.02) (1.48) (0.72) (-1.51) Female X Log Labor Productivity 0.033 -0.010 0.052 0.052** 0.013 -0.032 0.011 0.068* (0.89) (-0.33) (1.28) (2.15) (0.29) (-0.86) (0.41) (1.71) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.063 0.051 -0.037 -0.016 -0.043 0.073* 0.022 -0.026 (1.52) (1.40) (-0.82) (-0.53) (-0.85) (1.73) (0.73) (-0.56) Female X Log Labor Productivity -0.015 -0.074 0.085 0.077** 0.040 -0.063 0.006 0.024 (-0.26) (-1.62) (1.39) (1.97) (0.59) (-1.14) (0.17) (0.36) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 90 remain statistically significant. Finally, increased labor productivity appears to have a large adverse impact on study times for girls Œ however, this coefficient is imprecisely estimated and thus not statistically significant. Overall, where the re are some differences in outcomes between boys and girls, these differences are statistically non Œsignificant. Table 2.8: Effects of Land Productivity on School Expenditure and Study Times by Gender (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.9: Effects of Labor Productivity on School Expenditure and Study Times by Gender (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed e ffects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. c. Results by Gender of Household Head Existing literature show evidence indicating that household spending may vary depending on the gender of the household head. In this section, we investigate the significance of the gender of the household head in academic outcomes following productivity shocks in Table 2.10Œ2.13. Tables (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.178 0.643* 1.028** 0.095 1.300** -0.272 0.783** -0.409 (-0.56) (1.66) (2.30) (0.47) (2.23) (-0.77) (2.06) (-0.27) Log Land Productivity X Female -0.031 -1.074 -0.298 0.367 0.837 1.788 -0.550 5.161 (-0.03) (-1.31) (-0.26) (0.59) (0.42) (1.22) (-0.75) (0.32) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.334 1.205 1.945** 0.185 2.477* -0.491 1.477* 1.660 (-0.56) (1.56) (2.14) (0.47) (1.93) (-0.71) (1.95) (0.23) Log Labor Productivity X Female -0.102 -1.832 -0.235 0.727 2.005 3.307 -0.793 -13.717 (-0.06) (-1.19) (-0.10) (0.58) (0.43) (1.05) (-0.55) (-0.22) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 91 2.10 and 2.11 provide results from panel fixed effects estimation. We focus the discussion o n the IV estimates in Table 2.12 Œ2.13. Table 2.10: Effects of Land Productivity on School Expenditure and Study Times by Gender of Head (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.11: Effect s of Labor Productivity on School Expenditure and Study Times by Gender of Head (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.12 shows that gender has important consequences on the impacts of land productivity on the outcomes of interest. Generally, the impact of land productivity is lower among female Œheaded households. Specifically, a 10% increase in land productivity increases spending on uniform, contributions and total expenses by 15%, 24%, and 9% respectively among male Œheaded households. On the other hand, for female Œheaded households, the aggreg ate impact of a 10% increase in land productivity on uniform, contributions and total expenses is 0.05%, 1.6%, and 1% respectively. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.027 0.008 0.016 0.016 -0.016 0.021 0.025 -0.003 (1.27) (0.39) (0.66) (1.04) (-0.60) (0.98) (1.59) (-0.10) Female Head X Log Land Productivity 0.111*** -0.025 0.037 0.008 -0.035 -0.003 -0.015 -0.015 (2.77) (-0.81) (0.83) (0.31) (-0.70) (-0.07) (-0.56) (-0.35) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.016 0.012 -0.024 0.017 -0.004 0.048 0.021 -0.036 (0.48) (0.43) (-0.67) (0.70) (-0.09) (1.43) (0.96) (-0.85) Female Head X Log Labor Productivity 0.153** 0.000 0.122* 0.030 -0.072 -0.034 0.018 0.078 (2.27) (0.01) (1.77) (0.72) (-0.94) (-0.54) (0.42) (1.06) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 92 Table 2.13, using labor productivity, shows qualitatively similar results with larger differences between male Œheaded and female Œheaded households. However, these latter set of results have larger standard errors. A 10% increase in labor productivity increases spending on uniforms, contributions, and total expenses by 30%, 48%, and 19%, respectively, for male Œheaded household s. The corresponding estimates for female Œheaded households are a decline of 7%,8%, and 2% in uniform, contribution and total expenses respectively Œ the impact on total expenses is not statistically significant for either gender. Overall, we find some evi dence that while agricultural productivity tends to have positive impacts on academic spending, the impact tends to be lower among female Œheaded households. One potential explanation is that female Œheaded households in Tanzania may be disadvantaged in nume rous ways that may limit their ability to capitalize on positive productivity shocks in increasing education investments. For instance, women may have lower access to credit markets and therefore cannot access credit even in the anticipation of good harves ts. This may explain why our results are contrary to the literature that find that women tend to make better investments in children compared to men (e.g. Qian, 2008). Table 2.12: Effects of Land Productivity on School Expenditure and Study Times by Gender of Head (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household lev el. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.396 0.542 1.487** 0.158 2.371** 0.462 0.923* 0.419 (-0.96) (1.13) (2.22) (0.59) (2.46) (0.96) (1.86) (0.83) Female Head X Log Land Productivity 0.556 -0.600 -1.482** 0.128 -2.212** -0.526 -0.828* -0.126 (1.37) (-1.08) (-2.32) (0.43) (-2.46) (-1.15) (-1.81) (-0.26) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 93 Table 2.13: Effects of Labor Productivity on School Expenditure and Study Times by Gender of Head (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0. 01. Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. d. Results by Level of Schooling: Primary versus Post ŒPrimary Some studi es indicate that the relationship between agricultural productivity and education outcomes are pro Œcyclical for younger children and counter Œcyclical for older children. We explore this hypothesis by comparing outcomes between primary school students (youn ger) versus post Œsecondary students (relatively older). There may also be other heterogeneous factors at play at different education levels. We define a primary school dummy, that equals one if student in primary or lower level and equals zero if in post Œprimary level, and then interact the dummy with productivity. We re port our findings in Tables 2.14 Œ2.17. Starting with the fixed effects estimates in Tables 2.14 Œ2.15, we find suggestive evidence that both land and labor productivity have larger positive e ffects for post Œprimary students. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.816 1.107 3.036* 0.321 4.840* 0.943 1.885 1.170 (-0.95) (1.05) (1.81) (0.56) (1.96) (0.91) (1.58) (0.98) Female Head X Log Labor Productivity 1.396 -1.520 -3.771* 0.288 -5.648* -1.331 -2.117 -0.568 (1.26) (-1.08) (-1.88) (0.34) (-1.94) (-1.06) (-1.54) (-0.39) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 94 Table 2.14: Effects of Land Productivity on School Expenditure and Study Times by School ŒLevel (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes hou sehold and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.15: Effects of Labor Productiv ity on School Expenditure and Study Times by School ŒLevel (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Our pre ferred estimations in Table 2.16 Œ2.17, using IV, show similar results, though the differences are smaller in magnitude. While we observe large differences in expenditure on fees and food, there are o nly small differences in aggregate expenditure. While land productivity has positive effects on expenses, a 10% increase in land productivity has an effect that is lower by 6%, 0.5%, 0.5%, 11%, and 0.8% on fees, uniform, transport, food, and total expenses respectively. The effects are qualitatively similar when using labor productivity as a measure of agricultural productivity. However, the coefficients are larger while the differences in effects by school level are slightly larger. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.463*** 0.006 0.064*** 0.059*** -0.004 0.095*** 0.075*** 0.009 (19.42) (0.33) (2.70) (3.56) (-0.14) (4.04) (4.63) (0.35) Primary X Log Land Productivity (TSH/Ha) -0.543*** -0.008 -0.050*** -0.055*** -0.030* -0.101*** -0.074*** -0.019 (-35.87) (-0.76) (-3.77) (-5.74) (-1.65) (-6.41) (-9.07) (-1.04) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.711*** 0.015 0.057 0.087*** 0.016 0.173*** 0.110*** 0.013 (19.35) (0.53) (1.59) (3.07) (0.38) (4.66) (4.74) (0.30) Primary X Log Labor Productivity (TSH/Day) -0.884*** -0.004 -0.067*** -0.084*** -0.051* -0.180*** -0.113*** -0.032 (-38.25) (-0.26) (-3.10) (-5.36) (-1.76) (-6.92) (-8.71) (-1.05) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 95 Table 2.16: Effects of Land Productivity on School Expenditure and Study Times by School ŒLevel (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed ef fects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.17: Effects of Labor Productivity on School Expenditure and Study Times by School ŒLevel (I V) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Focusing o n the IV estimates, important differences emerge on fee and food expenses. Generally, primary school fees tend to be lower but post Œprimary school fees tend to be significantly higher. Therefore, post Œprimary fees are likely to be more responsive to shocks to household incomes. In addition, food expenses are likely to decrease for primary school because these students typically attend day schools and therefore consume most of their food at home. Positive productivity shocks can therefore lower food expenses at home, which translate to lower food expense s for primary school students. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.892*** 0.338 1.031** 0.307 1.610*** 0.473 0.763** 0.373 (3.21) (1.08) (2.42) (1.57) (2.73) (1.41) (2.34) (1.35) Primary X Log Land Productivity (TSH/Ha) -0.585*** -0.010 -0.049*** -0.053*** -0.031 -0.109*** -0.077*** -0.018 (-33.19) (-0.89) (-2.87) (-5.26) (-1.21) (-6.45) (-6.94) (-0.87) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 1.502*** 0.662 2.008** 0.580 3.156** 0.883 1.468** 1.055 (2.64) (1.05) (2.18) (1.47) (2.38) (1.31) (2.12) (1.20) Primary X Log Labor Productivity (TSH/Day) -0.945*** -0.011 -0.066* -0.083*** -0.029 -0.172*** -0.116*** -0.035 (-32.15) (-0.58) (-1.93) (-4.86) (-0.55) (-5.97) (-5.02) (-0.94) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 96 VI. ROBUSTNESS CHECKS a. Alternative Measures of Productivity The remaining tables perform sensitivity analyses using different measures of productivity. Given that we have confidence in our instrument, we focus on the IV results. Table 2.18 Œ2.19 use the value of crop income as the measure of productivity Œ instead of gross crop income per hectare. The coefficients are larger but qualitatively similar to those in Table 2.2Œ2.5. The IV estimates fall between the estimates from labor productivity and land productivity. Specifically, a 10% increase in value of crop income leads to a 13%, 22% and 9% increase in uniform, contributions, and total expenses. Table 2.18: Effects of Gross Agricultural Income on School Expenditure and Study Times (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/ agricultural income is at the household level. Errors clustered at the individual level. Table 2.19: Effects of Gross Agricultural Income on School Expenditure and Study Times (IV) TŒstatistics in parenth eses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.20 Œ2.21 using net crop revenue per hectare instead of gross crop revenue per hectare. Net revenue is the gross revenue net of expli cit production costs. Table 2.20 reports the (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Value of Crop Income (TSH) 0.058** -0.006 0.011 0.024 0.025 0.019 0.012 0.002 (2.41) (-0.29) (0.43) (1.41) (0.85) (0.78) (0.70) (0.09) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Value of Crop Income (TSH) -0.307 0.454 1.330** 0.291 2.201** 0.378 0.875* 0.712 (-0.63) (1.00) (2.05) (1.05) (2.32) (0.79) (1.84) (1.20) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 97 fixed effect results that are qualitatively similar those in Table 2.2, which uses gross revenue per hectare. The IV estimates show that a 10% increase in net revenue per hectare leads to a 10%, 18%, and 7% in uniform, contributions, and total expenses respectively. These estimates are quantitatively and qualitatively similar to those in Table 2.4, which uses gros s income per hectare as a measure of productivity. Table 2.22 Œ2.23 are closely related to Table 2.20Œ2.21. The measure of productivity is net crop income instead of net crop income per hectare. The results are similar to those in Table 2.20Œ2.21, and qual itatively consistent with our previous set of results. Table 2.20: Effects of Net Agricultural Income per Hectare on School Expenditure and Study Times (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.21: Effect s of Net Agricultural Income per Hectare on School Expenditure and Study Times (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Ln Net Revenue per Ha (TSH/Ha) 0.050* 0.008 -0.002 0.027 -0.006 0.057* 0.004 0.039 (1.69) (0.34) (-0.09) (1.11) (-0.16) (1.65) (0.28) (1.31) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Ln Net Revenue per Ha (TSH/Ha) -0.179 0.416 0.988** 0.227 1.771*** 0.264 0.729** 0.315 (-0.52) (1.19) (2.20) (1.09) (3.13) (0.71) (2.02) (0.96) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 98 Table 2.22: Effects of Net Agricultural Income on School Expenditure and Study Times (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.0 5 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.23: Effects of Net Agricultural Income on School Expenditure and Study Times (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. b. Exclusion of Household Income Non Œagricultural household incomes, such as agricultural wage income, play an important role in determining the school expenses and study times. However, control ling for these variables in our estimation can yield unbiased estimates because non Œagricultural incomes can be potentially endogenous. For instance, high agricultural wage incomes may indicate higher participation of children in paid wage activities or su bstitution of adult labor for child labor in household chores. At the same time, these factors may affect household agricultural productivity (land/labor). In addition, higher agricultural wage income may indicate higher participation of household members in wage activities and hence lower labor input and agricultural productivity in the family farm. Generally, high agricultural productivity may induce a household to decrease non Œagricultural (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Ln Net Crop Income (Tsh) 0.042 0.020 0.007 0.026 0.026 0.069** 0.007 0.010 (1.49) (0.96) (0.29) (1.14) (0.79) (2.19) (0.54) (0.35) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Ln Net Crop Income (Tsh) -0.247 0.365 1.071** 0.234 1.772*** 0.305 0.705* 0.426 (-0.65) (1.02) (2.29) (1.10) (3.02) (0.81) (1.91) (1.32) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 99 labor supply and hence lower non Œagricultural incomes. On the oth er hand, lower agricultural productivity may force a household to supplement income by engaging in non Œfarm employment activities. We repeat our main analysis in Table 2.2Œ2.5 while excluding these incomes in the regressions. Our results are robust to the exclusion of household non Œagr icultural income. Table 2.24: Effects of Land Productivity on School Expenditure and Study Times Œ Excluding Other Income (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.25: Effect s of Labor Productivity on School Expenditure and Study Times Œ Excluding Other Income (FE) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. Table 2.26: Effects of Land Productivity on School Expenditure and Study Times Œ Excluding Other Income (IV) TŒstatistics in par entheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.060*** 0.001 0.026 0.018 -0.027 0.020 0.021 -0.007 (3.28) (0.08) (1.25) (1.40) (-1.16) (1.10) (1.51) (-0.36) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.055* 0.012 0.007 0.024 -0.024 0.040 0.026 -0.013 (1.88) (0.51) (0.22) (1.14) (-0.66) (1.37) (1.34) (-0.36) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.190 0.364 0.905** 0.205 1.539*** 0.281 0.597* 0.307 (-0.58) (1.16) (2.19) (1.09) (2.66) (0.85) (1.86) (1.14) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 100 Table 2.27: Effects of Labor Productivity on School Expenditure and Study Times Œ Excluding Other Income (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effe cts. Productivity/agricultural income is at the household level. Errors clustered at the individual level. c. Alternative Rainfall Measure Œ Quarterly Deviations from Historical Mean Our primary instrumental variable is the rainfall levels during the wettes t quarter during growing season (March Œ May). We repeat our analysis to test for sensitivity of our results to alternative measures of rainfall. We instrument for productivity using standardized quarterly deviations of rainfall from a long Œterm decadal tr end (2007 Œ 2017). The first stage is show in T able A 2.30 . While the results in Tables 2.28 Œ2.29 are imprecise, the coefficients are generally in l ine with our initial analysis. Table 2.28: Effects of Land P roductivity on School Expenditure and Study Times Œ Rainfall Deviations as Instrument (IV) TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.384 0.737 1.829** 0.415 3.112** 0.567 1.207* 0.946 (-0.57) (1.13) (2.01) (1.07) (2.30) (0.84) (1.75) (1.05) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.563 0.453* 0.994** -0.011 0.854* 0.219 0.227 0.490** (-1.58) (1.74) (2.49) (-0.06) (1.93) (0.68) (1.10) (2.37) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 101 Table 2.29: Effects of Labor Productivity on School Expenditure and Study Times Œ Rainfall Deviations as Instrument (IV) TŒsta tistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01 . Includes household and individual controls, and interview Œmonth fixed effects. Productivity/agricultural income is at the household level. Errors clustered at the individual level. d. Other Concerns Œ Fun ctional Forms The use of log transformation of variables has many benefits. For example, it mitigates the effects of outliers, and provides coefficients that are easy to translate. However, log transformations are not possible for observations that are zer os. One way around this is to add a small positive value to all observations before log Œtransformation Œ we follow this approach in our analysis . When the sample includes few observations, this may not be problematic. However, if a sizeable portion of the observations is zeros, then the results will be biased. We consider this a potential weakness of our analysis. Future versions of this paper will test for robustness of our results to the use of alternative functional forms such as the poisson or exponenti al functions with fixed effects. VII. CONCLUSION Agriculture is a significant employment and income source in Sub ŒSaharan Africa. In addition, credit access and insurance markets are either missing or incomplete. These, in conjunction with the reliance on rain Œfed agriculture, imply that agricultural household are likely to suffer from adverse shocks to agricultural productivity. Due to rapid population growth and the continued subdivision of land, the role of agriculture as a major employer and source of income is likely to subside in the near future. Consequently, other sources of income and economic growth outside of (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.497 0.130 2.098** 0.250 1.278 1.146 0.596 1.080 (-0.70) (0.25) (2.23) (0.64) (1.38) (1.50) (1.28) (1.60) Individual Fixed Effects YYYYYYYYTime Fixed Effects YYYYYYYYOutcomes in Logs 102 the agricultural sector will become increasingly important. Economic literature has shown that human capital is an important component of economi c growth. This paper has provided empirical evidence on the relationship between agricultural productivity and investment in human capital development. Our study provides four main findings. First, increases in agricultural income has large positive effec ts on academic spending. Second, while we find evidence that expenses for female students tend to be relatively lower, the differences are not statistically significant. Third, we find that productivity impacts tend to be very low in female Œheaded househol ds. Finally, we show evidence that some academic expenditure may be more sensitive for post Œprimary school students. The quality and quantity of human capital developed is arguably positively correlated with the levels of education expenditure and study time. Our findings imply that adverse agricultural shocks are likely to have long Œterm economic effects by lowering the quantity and quality of education attained. This paper provides several policy suggestions given our findings. First, the government and the private sector should invest in measures that shield students from adverse agricultural income shocks. These may include elimination or reduction of school fees for primary and pre Œprimary students, and provision of food in presence of adverse weather shocks. Second, the government and the private sector should develop and encourage the take Œup of weather Œbased insurance (e.g. crop insurance). Third, policies, that pay special attention to female Œheaded household, should be designed to ensure that thes e households are not significantly affected by shocks. Fourth, to encourage post Œprimary education attainment, policies should be designed to shield post Œprimary students in agricultural households from agricultural income shocks. Such policies may includ e fee deferral during periods of adverse weather shocks. Generally, development of credit markets and crop Œinsurance markets can help households to smooth 103 consumption and minimize effects of disruptive shocks. Access to credit and education on crop Œinsura nce can be targeted at the households that are most sensitive to agricultural shocks Œ e.g. female Œheaded households, and households with post Œprimary students. 104 APPENDIX 105 Table A 2.1: First Stage Regre ssions Œ Productivity and Rainfall TŒstatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave and interview month FE . (1) (2) (3) (4) (5) Ln Land Productivity (Tsh/Ha) Ln Labor Productivity (Tsh/Day) Ln Gross Crop Income (Tsh) Ln Net Crop Income per Ha (Tsh) Ln Net Crop Income (Tsh) Ln Total Rain (March - May) 0.582*** 0.292*** 0.413** 0.528*** 0.513*** (3.34) (2.69) (2.57) (4.41) (3.69) Observations 11054 11054 11054 11054 11054 F Statistic 9.1813.8510.2713.59.88Cragg-Donald Wald F statistic 13.817.799.8220.8017.88Kleibergen-Paap Wald rk F statistic 11.237.296.6219.4413.62106 Table A 2.2: Effects of Land Productivity on School Expendi ture and Study Times (FE) Œ Full Tables TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level . Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.059*** 0.001 0.026 0.018 -0.026 0.020 0.020 -0.007 (3.26) (0.04) (1.26) (1.42) (-1.13) (1.08) (1.51) (-0.34) Relationship to Head (Excludes Spouse) Child 0.559 -0.715 3.030* -0.532 5.900*** -0.984** 1.280 1.202 (0.29) (-0.43) (1.83) (-1.31) (3.84) (-2.07) (0.67) (1.53) Grandchild 0.621 -1.286 2.845 -0.594 5.419*** -0.465 1.113 1.047 (0.31) (-0.76) (1.59) (-1.26) (3.26) (-0.69) (0.58) (1.36) Relative 0.409 -1.374 2.554 -0.620 5.490*** -1.174** 0.987 0.313 (0.21) (-0.81) (1.48) (-1.49) (3.44) (-2.10) (0.51) (0.49) Non-relative -0.056 -1.881 1.564 -0.985** 4.114** -1.758** 0.128 0.000 (-0.03) (-1.08) (0.84) (-2.00) (2.26) (-2.50) (0.06) (.) Marital Status (Excludes Never Married) Married 1.316 2.392 1.854 -1.529* 6.765*** -1.565*** 0.075 -5.102*** (0.94) (1.13) (0.98) (-1.75) (3.19) (-2.86) (0.04) (-15.31) Other status -1.838* -1.441** -1.171 -1.214** -0.354 -2.519*** -1.958** 0.005 (-1.82) (-2.08) (-1.13) (-2.11) (-0.30) (-3.68) (-2.38) (0.01) Female Head 0.240 -0.078 -0.521 -0.216 -0.653 0.129 -0.321 -0.121 (0.78) (-0.29) (-1.54) (-1.40) (-1.61) (0.47) (-1.36) (-0.34) Head Education (Excludes - No Education) Primary -0.073 -0.116 0.469** 0.025 0.443* -0.153 0.098 -0.485** (-0.48) (-0.72) (2.11) (0.29) (1.72) (-0.94) (0.68) (-2.27) Secondary -0.373 -0.161 -0.081 -0.148 0.700 -0.608 -0.064 -0.157 (-0.89) (-0.52) (-0.22) (-0.73) (1.39) (-1.58) (-0.24) (-0.35) Post-Secondary 0.966 1.689 -3.320 1.445 3.552*** 3.375 1.610 5.804*** (0.58) (1.27) (-1.14) (1.00) (2.86) (1.20) (0.92) (11.82) Head Marital Status (Excludes- Never Married Married -0.583 -1.275** -1.165 -0.957** -1.180 -0.392 -1.212* 0.653 (-0.84) (-2.10) (-1.52) (-2.17) (-1.31) (-0.51) (-1.95) (1.01) Other status -0.682 -1.152** -0.920 -0.799* -0.844 -0.793 -1.106* 0.439 (-1.01) (-1.97) (-1.23) (-1.90) (-0.96) (-1.07) (-1.86) (0.68) Ln Head Age -0.508 0.004 0.510 0.204 1.051 0.161 0.150 -0.233 (-0.75) (0.01) (0.68) (0.62) (1.32) (0.31) (0.30) (-0.37) Land Rights Document 0.105 -0.224** 0.048 0.055 -0.180 0.161 0.011 0.014 (0.85) (-2.35) (0.37) (0.76) (-1.23) (1.24) (0.14) (0.10) Ln TLUs 0.048 0.080 0.051 0.074* -0.021 0.318*** 0.036 -0.010 (0.58) (1.12) (0.49) (1.67) (-0.18) (3.78) (0.56) (-0.09) Ln Adult Equivalent -0.434 0.639** 0.602* -0.164 0.233 -0.303 0.401 0.050 (-1.34) (2.43) (1.67) (-0.87) (0.58) (-0.96) (1.51) (0.13) Ln Land Size (ha) -0.024 0.110 0.120 -0.015 0.280* 0.038 0.039 -0.100 (-0.20) (1.21) (0.89) (-0.22) (1.72) (0.30) (0.50) (-0.60) Ln Ag Wage Income -0.006 -0.001 -0.012 0.002 0.002 0.008 -0.005 -0.021* (-0.62) (-0.15) (-1.20) (0.37) (0.16) (0.83) (-0.75) (-1.84) Ln Transfer Income -0.000 -0.004 -0.009 -0.007 -0.023* 0.003 0.002 0.010 (-0.04) (-0.58) (-0.80) (-1.35) (-1.90) (0.30) (0.35) (0.94) Ln Nonfarm Income 0.011 0.020*** 0.003 0.002 0.010 0.001 -0.001 0.003 (1.43) (3.00) (0.34) (0.42) (0.95) (0.15) (-0.24) (0.28) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 107 Table A 2.3: Effects of Labor Productiv ity on School Expenditure and Study Times (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.055* 0.012 0.007 0.025 -0.022 0.040 0.026 -0.013 (1.87) (0.50) (0.23) (1.17) (-0.62) (1.36) (1.33) (-0.37) Relationship to Head (Excludes Spouse) Child 0.544 -0.732 3.048* -0.549 5.904*** -1.020** 1.265 1.211 (0.28) (-0.45) (1.84) (-1.34) (3.86) (-2.11) (0.66) (1.54) Grandchild 0.622 -1.306 2.875 -0.608 5.414*** -0.501 1.102 1.052 (0.31) (-0.78) (1.60) (-1.29) (3.27) (-0.74) (0.57) (1.36) Relative 0.407 -1.394 2.583 -0.636 5.487*** -1.212** 0.975 0.320 (0.21) (-0.83) (1.50) (-1.52) (3.45) (-2.14) (0.50) (0.50) Non-relative -0.073 -1.894 1.576 -0.999** 4.119** -1.788** 0.114 0.000 (-0.04) (-1.09) (0.84) (-2.02) (2.27) (-2.52) (0.06) (.) Marital Status (Excludes Never Married) Married 1.217 2.394 1.804 -1.556* 6.810*** -1.592*** 0.042 -5.091*** (0.85) (1.12) (0.95) (-1.76) (3.20) (-2.83) (0.02) (-16.87) Other status -1.878* -1.443** -1.186 -1.227** -0.336 -2.537*** -1.973** 0.011 (-1.86) (-2.08) (-1.14) (-2.11) (-0.29) (-3.69) (-2.40) (0.02) Female Head 0.228 -0.070 -0.538 -0.214 -0.646 0.138 -0.321 -0.123 (0.74) (-0.26) (-1.59) (-1.39) (-1.59) (0.50) (-1.36) (-0.34) Head Education (Excludes - No Education) Primary -0.070 -0.115 0.468** 0.026 0.442* -0.150 0.099 -0.488** (-0.46) (-0.71) (2.10) (0.31) (1.72) (-0.92) (0.69) (-2.29) Secondary -0.363 -0.160 -0.078 -0.144 0.695 -0.603 -0.060 -0.158 (-0.86) (-0.52) (-0.21) (-0.71) (1.38) (-1.56) (-0.23) (-0.35) Post-Secondary 1.004 1.692 -3.307 1.459 3.536*** 3.393 1.625 5.788*** (0.60) (1.27) (-1.14) (1.01) (2.84) (1.21) (0.93) (11.76) Head Marital Status (Excludes- Never Married Married -0.545 -1.274** -1.149 -0.945** -1.197 -0.378 -1.198* 0.646 (-0.79) (-2.10) (-1.50) (-2.16) (-1.34) (-0.50) (-1.93) (0.99) Other status -0.639 -1.152** -0.900 -0.786* -0.863 -0.779 -1.091* 0.432 (-0.95) (-1.97) (-1.20) (-1.88) (-0.99) (-1.05) (-1.83) (0.67) Ln Head Age -0.504 0.018 0.492 0.215 1.051 0.187 0.160 -0.228 (-0.75) (0.04) (0.65) (0.65) (1.32) (0.36) (0.31) (-0.36) Land Rights Document 0.099 -0.221** 0.041 0.056 -0.177 0.164 0.011 0.013 (0.81) (-2.31) (0.31) (0.76) (-1.21) (1.27) (0.14) (0.10) Ln TLUs 0.078 0.078 0.067 0.082* -0.035 0.325*** 0.046 -0.013 (0.95) (1.10) (0.64) (1.86) (-0.30) (3.88) (0.71) (-0.11) Ln Adult Equivalent -0.422 0.638** 0.608* -0.161 0.227 -0.300 0.405 0.052 (-1.30) (2.42) (1.68) (-0.85) (0.56) (-0.95) (1.53) (0.14) Ln Land Size (ha) -0.073 0.107 0.103 -0.032 0.301* 0.016 0.021 -0.096 (-0.60) (1.17) (0.77) (-0.46) (1.85) (0.13) (0.26) (-0.57) Ln Ag Wage Income -0.006 -0.001 -0.012 0.002 0.002 0.008 -0.005 -0.021* (-0.60) (-0.14) (-1.19) (0.38) (0.16) (0.83) (-0.74) (-1.85) Ln Transfer Income -0.000 -0.004 -0.008 -0.007 -0.023* 0.003 0.002 0.010 (-0.04) (-0.59) (-0.79) (-1.37) (-1.90) (0.28) (0.34) (0.95) Ln Nonfarm Income 0.012 0.020*** 0.003 0.002 0.010 0.001 -0.001 0.003 (1.46) (3.00) (0.35) (0.44) (0.94) (0.16) (-0.22) (0.27) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 108 Table A 2.4: Effect s of Land Productivity on School Expenditure and Study Times (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricu ltural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.217 0.320 0.938** 0.205 1.552*** 0.267 0.617* 0.354 (-0.65) (1.00) (2.21) (1.08) (2.64) (0.80) (1.88) (1.28) Relationship to Head (Excludes Spouse) Child 0.861 -1.064 2.033 -0.737* 4.175* -1.253* 0.627 1.246 (0.41) (-0.68) (0.98) (-1.72) (1.65) (-1.74) (0.35) (1.51) Grandchild 1.075 -1.812 1.342 -0.903 2.819 -0.871 0.130 0.847 (0.49) (-1.07) (0.58) (-1.62) (1.02) (-0.91) (0.07) (1.01) Relative 0.847 -1.880 1.108 -0.918* 2.987 -1.565* 0.041 0.152 (0.39) (-1.13) (0.50) (-1.82) (1.13) (-1.78) (0.02) (0.22) Non-relative 0.157 -2.126 0.862 -1.129** 2.899 -1.948** -0.332 0.000 (0.07) (-1.30) (0.38) (-2.23) (1.05) (-2.24) (-0.17) (.) Marital Status (Excludes Never Married) Married 0.766 3.029 3.675 -1.154* 9.916** -1.073 1.266 -1.551 (0.42) (1.17) (1.63) (-1.77) (2.40) (-0.97) (0.62) (-0.57) Other status -1.976* -1.280* -0.713 -1.119** 0.440 -2.395*** -1.658* -0.050 (-1.93) (-1.73) (-0.63) (-2.01) (0.31) (-3.81) (-1.85) (-0.07) Female Head 0.018 0.179 0.211 -0.065 0.616 0.327 0.159 0.140 (0.04) (0.46) (0.39) (-0.30) (0.78) (0.85) (0.41) (0.31) Head Education (Excludes - No Education) Primary -0.086 -0.101 0.512* 0.034 0.519 -0.141 0.126 -0.712** (-0.54) (-0.59) (1.94) (0.37) (1.39) (-0.84) (0.73) (-2.35) Secondary -0.346 -0.192 -0.169 -0.166 0.547 -0.632 -0.122 -0.285 (-0.79) (-0.59) (-0.38) (-0.75) (0.78) (-1.64) (-0.40) (-0.59) Post-Secondary 1.079 1.559 -3.692 1.369 2.908* 3.275 1.367 5.664*** (0.66) (1.12) (-1.22) (0.93) (1.80) (1.13) (0.73) (10.81) Head Marital Status (Excludes- Never Married Married -0.422 -1.462** -1.700* -1.067** -2.106* -0.537 -1.562** 0.901 (-0.56) (-2.26) (-1.79) (-2.22) (-1.78) (-0.67) (-2.25) (1.30) Other status -0.476 -1.391** -1.602* -0.939** -2.024* -0.977 -1.552** 0.646 (-0.63) (-2.15) (-1.68) (-1.98) (-1.69) (-1.24) (-2.24) (0.96) Ln Head Age -0.795 0.336 1.459 0.399 2.691** 0.417 0.771 0.036 (-1.01) (0.49) (1.37) (0.94) (2.00) (0.67) (1.03) (0.05) Land Rights Document 0.014 -0.119 0.349* 0.117 0.341 0.242 0.208 0.048 (0.08) (-0.82) (1.73) (1.17) (1.24) (1.43) (1.50) (0.34) Ln TLUs 0.232 -0.133 -0.556* -0.051 -1.073** 0.154 -0.361 -0.304 (0.98) (-0.57) (-1.73) (-0.37) (-2.38) (0.63) (-1.48) (-1.20) Ln Adult Equivalent -0.356 0.548* 0.342 -0.218 -0.216 -0.373 0.231 -0.075 (-1.02) (1.89) (0.78) (-1.02) (-0.37) (-1.13) (0.71) (-0.18) Ln Land Size (ha) -0.185 0.296 0.651** 0.094 1.200*** 0.182 0.387* 0.225 (-0.79) (1.42) (2.25) (0.75) (2.87) (0.75) (1.83) (0.72) Ln Ag Wage Income -0.005 -0.001 -0.013 0.002 0.001 0.008 -0.006 -0.022* (-0.58) (-0.17) (-1.10) (0.33) (0.04) (0.79) (-0.71) (-1.81) Ln Transfer Income 0.002 -0.007 -0.015 -0.008 -0.034** 0.001 -0.002 0.013 (0.15) (-0.80) (-1.16) (-1.51) (-2.04) (0.13) (-0.24) (1.11) Ln Nonfarm Income 0.013 0.019*** -0.001 0.001 0.004 0.000 -0.004 -0.000 (1.51) (2.70) (-0.06) (0.25) (0.29) (0.03) (-0.58) (-0.04) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 109 Table A 2.5: Effects of Labor Productivity on School Expenditure and Study Times (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.433 0.638 1.873** 0.410 3.097** 0.532 1.232* 1.029 (-0.64) (0.99) (2.02) (1.06) (2.31) (0.79) (1.77) (1.17) Relationship to Head (Excludes Spouse) Child 1.258 -1.649 0.315 -1.112 1.334 -1.741 -0.502 0.567 (0.54) (-0.99) (0.10) (-1.39) (0.28) (-1.33) (-0.22) (0.58) Grandchild 1.474 -2.400 -0.384 -1.281 -0.036 -1.362 -1.006 0.337 (0.60) (-1.29) (-0.11) (-1.38) (-0.01) (-0.88) (-0.41) (0.31) Relative 1.265 -2.497 -0.702 -1.314 -0.006 -2.080 -1.149 -0.482 (0.51) (-1.36) (-0.21) (-1.45) (-0.00) (-1.38) (-0.48) (-0.50) Non-relative 0.485 -2.611 -0.561 -1.440* 0.546 -2.352* -1.267 0.000 (0.21) (-1.59) (-0.18) (-1.85) (0.11) (-1.81) (-0.56) (.) Marital Status (Excludes Never Married) Married 1.052 2.607 2.437 -1.425** 7.869* -1.425 0.452 -0.459 (0.62) (1.04) (1.14) (-2.18) (1.65) (-1.25) (0.22) (-0.12) Other status -1.785* -1.563** -1.543 -1.301** -0.933 -2.631*** -2.204*** -0.519 (-1.73) (-2.27) (-1.40) (-2.10) (-0.62) (-4.01) (-2.68) (-0.53) Female Head -0.083 0.329 0.652 0.031 1.344 0.452 0.448 0.433 (-0.15) (0.64) (0.88) (0.11) (1.23) (0.89) (0.83) (0.68) Head Education (Excludes - No Education) Primary -0.118 -0.054 0.650** 0.064 0.746* -0.102 0.217 -0.602** (-0.69) (-0.29) (2.18) (0.65) (1.70) (-0.56) (1.07) (-1.98) Secondary -0.405 -0.106 0.084 -0.110 0.966 -0.560 0.045 -0.318 (-0.92) (-0.31) (0.16) (-0.50) (1.20) (-1.36) (0.12) (-0.56) Post-Secondary 0.883 1.848 -2.842 1.555 4.313*** 3.516 1.926 6.833** (0.53) (1.34) (-0.93) (1.06) (2.82) (1.26) (1.05) (6.68) Head Marital Status (Excludes- Never Married Married -0.574 -1.236* -1.038 -0.922** -1.012 -0.349 -1.127 1.571 (-0.73) (-1.89) (-0.99) (-2.05) (-0.61) (-0.46) (-1.41) (1.22) Other status -0.629 -1.164* -0.937 -0.794* -0.925 -0.788 -1.115 1.307 (-0.82) (-1.83) (-0.91) (-1.83) (-0.56) (-1.08) (-1.43) (1.03) Ln Head Age -1.087 0.767 2.722* 0.675 4.781** 0.776 1.601 -0.209 (-1.00) (0.75) (1.72) (1.08) (2.21) (0.80) (1.39) (-0.22) Land Rights Document -0.028 -0.057 0.529* 0.157 0.639 0.293 0.327 0.092 (-0.13) (-0.29) (1.84) (1.20) (1.50) (1.35) (1.58) (0.54) Ln TLUs 0.161 -0.027 -0.248 0.017 -0.562* 0.241 -0.158 -0.249 (1.14) (-0.19) (-1.11) (0.20) (-1.73) (1.64) (-0.98) (-1.07) Ln Adult Equivalent -0.383 0.587** 0.458 -0.192 -0.024 -0.340 0.308 -0.303 (-1.12) (2.03) (0.96) (-0.92) (-0.04) (-1.04) (0.90) (-0.54) Ln Land Size (ha) 0.050 -0.051 -0.367 -0.129 -0.484 -0.108 -0.283 0.027 (0.24) (-0.26) (-1.19) (-1.03) (-1.06) (-0.52) (-1.29) (0.12) Ln Ag Wage Income -0.006 0.000 -0.009 0.003 0.008 0.009 -0.003 -0.017 (-0.67) (0.00) (-0.68) (0.49) (0.43) (0.91) (-0.33) (-1.21) Ln Transfer Income 0.004 -0.010 -0.024 -0.010 -0.050** -0.001 -0.008 0.006 (0.32) (-0.98) (-1.48) (-1.60) (-2.24) (-0.11) (-0.75) (0.48) Ln Nonfarm Income 0.012 0.020*** 0.004 0.002 0.011 0.002 -0.001 0.012 (1.40) (2.87) (0.33) (0.45) (0.70) (0.17) (-0.10) (0.88) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 110 Table A 2.6: Effects of Land Productivity on School Expenditure and Study Times by Gender (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The un it of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.043* 0.006 0.000 -0.008 -0.033 0.036 0.015 -0.040 (1.89) (0.25) (0.01) (-0.47) (-1.02) (1.48) (0.72) (-1.51) Female X Log Labor Productivity 0.033 -0.010 0.052 0.052** 0.013 -0.032 0.011 0.068* (0.89) (-0.33) (1.28) (2.15) (0.29) (-0.86) (0.41) (1.71) Relationship to Head (Excludes Spouse) Child 0.551 -0.712 3.016* -0.546 5.896*** -0.975** 1.277 1.212 (0.29) (-0.43) (1.82) (-1.39) (3.84) (-2.08) (0.67) (1.54) Grandchild 0.612 -1.283 2.832 -0.608 5.415*** -0.457 1.110 1.046 (0.31) (-0.76) (1.58) (-1.32) (3.26) (-0.69) (0.57) (1.36) Relative 0.403 -1.372 2.544 -0.631 5.487*** -1.168** 0.985 0.327 (0.21) (-0.81) (1.47) (-1.56) (3.44) (-2.11) (0.51) (0.51) Non-relative -0.072 -1.876 1.539 -1.010** 4.107** -1.743** 0.122 0.000 (-0.04) (-1.07) (0.82) (-2.09) (2.25) (-2.49) (0.06) (.) Marital Status (Excludes Never Married) Married 1.343 2.383 1.896 -1.486* 6.776*** -1.591*** 0.083 -4.762*** (0.96) (1.12) (1.01) (-1.80) (3.19) (-2.96) (0.05) (-11.86) Other status -1.826* -1.444** -1.154 -1.196** -0.349 -2.530*** -1.954** 0.006 (-1.81) (-2.08) (-1.11) (-2.10) (-0.30) (-3.68) (-2.37) (0.01) Female Head 0.245 -0.079 -0.514 -0.209 -0.651 0.124 -0.319 -0.104 (0.80) (-0.30) (-1.52) (-1.37) (-1.60) (0.46) (-1.35) (-0.30) Head Education (Excludes - No Education) Primary -0.079 -0.114 0.460** 0.015 0.441* -0.147 0.096 -0.495** (-0.51) (-0.71) (2.06) (0.18) (1.71) (-0.91) (0.67) (-2.31) Secondary -0.388 -0.156 -0.105 -0.172 0.694 -0.594 -0.069 -0.180 (-0.93) (-0.50) (-0.29) (-0.84) (1.38) (-1.55) (-0.26) (-0.39) Post-Secondary 0.948 1.695 -3.349 1.416 3.544*** 3.393 1.604 5.758*** (0.57) (1.27) (-1.15) (0.98) (2.85) (1.21) (0.91) (11.70) Head Marital Status (Excludes- Never Married Married -0.573 -1.278** -1.149 -0.941** -1.176 -0.402 -1.209* 0.671 (-0.83) (-2.11) (-1.50) (-2.15) (-1.31) (-0.52) (-1.94) (1.03) Other status -0.675 -1.154** -0.909 -0.788* -0.842 -0.800 -1.104* 0.449 (-0.99) (-1.97) (-1.22) (-1.89) (-0.96) (-1.08) (-1.85) (0.70) Ln Head Age -0.489 -0.002 0.541 0.235 1.058 0.143 0.157 -0.231 (-0.72) (-0.00) (0.72) (0.71) (1.32) (0.28) (0.31) (-0.36) Land Rights Document 0.104 -0.224** 0.047 0.055 -0.180 0.161 0.011 0.009 (0.85) (-2.35) (0.36) (0.75) (-1.23) (1.25) (0.14) (0.07) Ln TLUs 0.047 0.081 0.049 0.071 -0.022 0.320*** 0.036 -0.020 (0.56) (1.12) (0.46) (1.61) (-0.19) (3.80) (0.55) (-0.17) Ln Adult Equivalent -0.436 0.639** 0.599* -0.166 0.233 -0.302 0.401 0.061 (-1.35) (2.43) (1.66) (-0.88) (0.58) (-0.95) (1.51) (0.16) Ln Land Size (ha) -0.025 0.111 0.119 -0.016 0.280* 0.039 0.039 -0.097 (-0.21) (1.21) (0.88) (-0.23) (1.72) (0.30) (0.50) (-0.58) Ln Ag Wage Income -0.006 -0.001 -0.012 0.002 0.002 0.008 -0.005 -0.021* (-0.60) (-0.15) (-1.18) (0.41) (0.17) (0.81) (-0.74) (-1.82) Ln Transfer Income -0.000 -0.004 -0.009 -0.007 -0.023* 0.003 0.002 0.010 (-0.04) (-0.58) (-0.81) (-1.37) (-1.90) (0.31) (0.35) (0.96) Ln Nonfarm Income 0.012 0.020*** 0.003 0.002 0.010 0.001 -0.001 0.003 (1.44) (3.00) (0.35) (0.45) (0.95) (0.14) (-0.23) (0.27) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 111 Table A 2.7: Effects of Labor Productivity on School Expenditure and Study Times by Gender (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and intervi ewŒmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustere d at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.063 0.051 -0.037 -0.016 -0.043 0.073* 0.022 -0.026 (1.52) (1.40) (-0.82) (-0.53) (-0.85) (1.73) (0.73) (-0.56) Female X Log Labor Productivity -0.015 -0.074 0.085 0.077** 0.040 -0.063 0.006 0.024 (-0.26) (-1.62) (1.39) (1.97) (0.59) (-1.14) (0.17) (0.36) Relationship to Head (Excludes Spouse) Child 0.551 -0.694 3.005* -0.588 5.883*** -0.988** 1.262 1.216 (0.29) (-0.42) (1.81) (-1.47) (3.83) (-2.08) (0.66) (1.54) Grandchild 0.631 -1.265 2.829 -0.650 5.392*** -0.466 1.098 1.053 (0.32) (-0.74) (1.57) (-1.40) (3.25) (-0.70) (0.57) (1.36) Relative 0.414 -1.356 2.539 -0.676 5.466*** -1.179** 0.972 0.326 (0.21) (-0.79) (1.46) (-1.64) (3.43) (-2.11) (0.50) (0.51) Non-relative -0.064 -1.849 1.525 -1.046** 4.095** -1.750** 0.110 0.000 (-0.03) (-1.05) (0.81) (-2.14) (2.25) (-2.48) (0.05) (.) Marital Status (Excludes Never Married) Married 1.218 2.401 1.796 -1.562* 6.806*** -1.587*** 0.042 -5.041*** (0.85) (1.13) (0.96) (-1.85) (3.21) (-2.94) (0.02) (-14.57) Other status -1.878* -1.443** -1.186 -1.228** -0.336 -2.537*** -1.973** 0.009 (-1.86) (-2.09) (-1.14) (-2.11) (-0.29) (-3.69) (-2.40) (0.01) Female Head 0.224 -0.087 -0.518 -0.196 -0.636 0.123 -0.319 -0.117 (0.73) (-0.33) (-1.54) (-1.29) (-1.57) (0.45) (-1.36) (-0.33) Head Education (Excludes - No Education) Primary -0.069 -0.108 0.459** 0.018 0.438* -0.143 0.098 -0.491** (-0.45) (-0.66) (2.06) (0.21) (1.70) (-0.88) (0.69) (-2.30) Secondary -0.359 -0.140 -0.101 -0.164 0.685 -0.586 -0.061 -0.165 (-0.85) (-0.45) (-0.28) (-0.80) (1.36) (-1.52) (-0.23) (-0.36) Post-Secondary 1.008 1.711 -3.329 1.439 3.525*** 3.409 1.624 5.793*** (0.60) (1.29) (-1.15) (1.00) (2.83) (1.22) (0.92) (11.77) Head Marital Status (Excludes- Never Married Married -0.545 -1.270** -1.153 -0.949** -1.199 -0.375 -1.199* 0.650 (-0.79) (-2.10) (-1.51) (-2.16) (-1.34) (-0.49) (-1.93) (1.00) Other status -0.637 -1.145* -0.909 -0.793* -0.867 -0.773 -1.092* 0.435 (-0.94) (-1.96) (-1.22) (-1.89) (-0.99) (-1.04) (-1.83) (0.68) Ln Head Age -0.511 -0.017 0.532 0.251 1.070 0.158 0.162 -0.217 (-0.76) (-0.03) (0.70) (0.75) (1.34) (0.30) (0.32) (-0.34) Land Rights Document 0.100 -0.219** 0.039 0.054 -0.178 0.166 0.011 0.013 (0.81) (-2.28) (0.29) (0.72) (-1.22) (1.29) (0.14) (0.09) Ln TLUs 0.079 0.080 0.066 0.080* -0.036 0.326*** 0.046 -0.015 (0.95) (1.13) (0.63) (1.82) (-0.31) (3.91) (0.71) (-0.13) Ln Adult Equivalent -0.422 0.637** 0.610* -0.160 0.228 -0.301 0.405 0.052 (-1.30) (2.42) (1.69) (-0.85) (0.57) (-0.95) (1.53) (0.14) Ln Land Size (ha) -0.072 0.112 0.097 -0.037 0.298* 0.021 0.020 -0.096 (-0.59) (1.23) (0.72) (-0.54) (1.83) (0.16) (0.26) (-0.57) Ln Ag Wage Income -0.005 -0.001 -0.012 0.002 0.002 0.008 -0.005 -0.021* (-0.60) (-0.13) (-1.20) (0.37) (0.15) (0.84) (-0.74) (-1.85) Ln Transfer Income -0.000 -0.004 -0.008 -0.007 -0.023* 0.003 0.002 0.010 (-0.04) (-0.58) (-0.80) (-1.38) (-1.90) (0.29) (0.34) (0.96) Ln Nonfarm Income 0.012 0.020*** 0.003 0.002 0.010 0.001 -0.001 0.003 (1.46) (2.98) (0.37) (0.47) (0.95) (0.14) (-0.22) (0.26) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 112 Table A 2.8: Effects of Land Productivity on School Expenditure and Study Times by Gender (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed a cademic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.178 0.643* 1.028** 0.095 1.300** -0.272 0.783** -0.409 (-0.56) (1.66) (2.30) (0.47) (2.23) (-0.77) (2.06) (-0.27) Log Land Productivity X Female -0.031 -1.074 -0.298 0.367 0.837 1.788 -0.550 5.161 (-0.03) (-1.31) (-0.26) (0.59) (0.42) (1.22) (-0.75) (0.32) Relationship to Head (Excludes Spouse) Child 0.755 -0.535 2.179 -0.917 3.763 -2.133 0.898 2.318 (0.36) (-0.25) (1.06) (-1.28) (1.10) (-0.86) (0.47) (0.33) Grandchild 0.906 -1.179 1.518 -1.119 2.326 -1.924 0.454 0.000 (0.41) (-0.52) (0.65) (-1.30) (0.62) (-0.71) (0.23) (.) Relative 0.695 -1.325 1.262 -1.107 2.554 -2.491 0.326 0.591 (0.32) (-0.60) (0.57) (-1.39) (0.71) (-0.96) (0.16) (0.11) Non-relative 0.034 -1.436 1.053 -1.364* 2.361 -3.097 0.022 0.142 (0.02) (-0.65) (0.47) (-1.70) (0.64) (-1.19) (0.01) (0.03) Marital Status (Excludes Never Married) Married 0.792 1.718 3.311 -0.707 10.938* 1.110 0.594 41.754 (0.37) (0.77) (1.25) (-0.66) (1.84) (0.27) (0.29) (0.29) Other status -1.980* -1.761** -0.846 -0.955 0.815 -1.594 -1.905** -0.289 (-1.75) (-2.16) (-0.68) (-1.55) (0.43) (-1.52) (-2.07) (-0.20) Female Head 0.016 -0.142 0.122 0.045 0.866 0.861 -0.006 2.667 (0.03) (-0.28) (0.18) (0.13) (0.73) (1.05) (-0.01) (0.30) Head Education (Excludes - No Education) Primary -0.095 0.079 0.562* -0.028 0.379 -0.441 0.219 -2.603 (-0.42) (0.35) (1.75) (-0.20) (0.73) (-1.24) (1.05) (-0.38) Secondary -0.355 0.327 -0.026 -0.343 0.142 -1.496 0.144 -2.668 (-0.55) (0.64) (-0.04) (-0.84) (0.12) (-1.61) (0.33) (-0.33) Post-Secondary 1.087 2.255 -3.499 1.131 2.366 2.115 1.724 1.473 (0.60) (1.61) (-1.12) (0.72) (1.05) (0.61) (0.92) (0.10) Head Marital Status (Excludes- Never Married Married -0.456 -1.662** -1.755* -0.999** -1.950 -0.204 -1.664** 3.491 (-0.61) (-2.29) (-1.80) (-2.13) (-1.64) (-0.25) (-2.29) (0.39) Other status -0.517 -1.456** -1.620* -0.917** -1.974* -0.869 -1.586** 2.446 (-0.70) (-2.05) (-1.67) (-2.00) (-1.67) (-1.07) (-2.22) (0.38) Ln Head Age -0.786 -0.524 1.220 0.693 3.362 1.850 0.330 1.514 (-0.72) (-0.52) (0.81) (0.93) (1.39) (1.07) (0.35) (0.24) Land Rights Document 0.008 -0.175 0.334 0.137 0.385 0.336 0.180 -0.103 (0.05) (-1.05) (1.52) (1.15) (1.11) (1.28) (1.24) (-0.19) Ln TLUs 0.202 0.068 -0.501 -0.119 -1.229 -0.180 -0.258 -2.463 (0.66) (0.22) (-1.15) (-0.53) (-1.64) (-0.33) (-0.91) (-0.33) Ln Adult Equivalent -0.331 0.654* 0.371 -0.254 -0.298 -0.549 0.286 0.170 (-0.92) (1.96) (0.81) (-1.06) (-0.44) (-1.20) (0.85) (0.12) Ln Land Size (ha) -0.165 0.191 0.622* 0.130 1.282** 0.357 0.333 2.049 (-0.64) (0.77) (1.86) (0.78) (2.29) (0.89) (1.49) (0.31) Ln Ag Wage Income -0.007 -0.005 -0.014 0.003 0.004 0.014 -0.008 -0.008 (-0.68) (-0.58) (-1.14) (0.50) (0.20) (1.07) (-0.93) (-0.12) Ln Transfer Income 0.001 -0.003 -0.014 -0.009 -0.037* -0.004 -0.000 0.038 (0.10) (-0.38) (-1.05) (-1.52) (-1.88) (-0.28) (-0.04) (0.45) Ln Nonfarm Income 0.013 0.017** -0.001 0.002 0.005 0.003 -0.004 -0.027 (1.54) (2.36) (-0.11) (0.34) (0.36) (0.27) (-0.71) (-0.28) Observations 11061 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 113 Table A 2.9: Effects of Labor Productivity on School Expenditure and Study Times by Gender (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in sch ool the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.334 1.205 1.945** 0.185 2.477* -0.491 1.477* 1.660 (-0.56) (1.56) (2.14) (0.47) (1.93) (-0.71) (1.95) (0.23) Log Labor Productivity X Female -0.102 -1.832 -0.235 0.727 2.005 3.307 -0.793 -13.717 (-0.06) (-1.19) (-0.10) (0.58) (0.43) (1.05) (-0.55) (-0.22) Relationship to Head (Excludes Spouse) Child 1.118 -0.153 0.507 -1.706 -0.304 -4.443 0.146 2.146 (0.40) (-0.05) (0.13) (-0.93) (-0.04) (-0.84) (0.06) (0.37) Grandchild 1.272 -0.720 -0.168 -1.947 -1.875 -4.393 -0.279 4.238 (0.42) (-0.23) (-0.04) (-0.96) (-0.21) (-0.77) (-0.10) (0.27) Relative 1.069 -0.856 -0.492 -1.964 -1.803 -5.042 -0.439 1.236 (0.36) (-0.28) (-0.12) (-0.99) (-0.21) (-0.90) (-0.17) (0.21) Non-relative 0.352 -1.065 -0.362 -2.053 -1.147 -5.143 -0.598 0.000 (0.12) (-0.38) (-0.09) (-1.14) (-0.14) (-0.99) (-0.25) (.) Marital Status (Excludes Never Married) Married 1.099 2.633 2.440 -1.435 7.840 -1.472 0.463 -58.381 (0.67) (1.30) (1.23) (-1.63) (1.20) (-0.35) (0.27) (-0.23) Other status -1.798* -1.479** -1.532 -1.334** -1.025 -2.782*** -2.168*** 4.054 (-1.74) (-2.01) (-1.39) (-2.00) (-0.59) (-2.91) (-2.68) (0.21) Female Head -0.107 -0.340 0.566 0.297 2.076 1.659 0.159 -6.161 (-0.11) (-0.40) (0.43) (0.44) (0.83) (0.98) (0.20) (-0.21) Head Education (Excludes - No Education) Primary -0.126 0.098 0.669** 0.003 0.580 -0.376 0.282 1.810 (-0.61) (0.42) (2.01) (0.02) (1.02) (-1.08) (1.27) (0.17) Secondary -0.399 0.351 0.143 -0.291 0.466 -1.385 0.243 5.017 (-0.66) (0.69) (0.19) (-0.72) (0.34) (-1.47) (0.51) (0.21) Post-Secondary 0.896 2.217* -2.795 1.408 3.909** 2.850 2.085 -2.534 (0.51) (1.66) (-0.91) (0.95) (2.15) (0.98) (1.12) (-0.06) Head Marital Status (Excludes- Never Married Married -0.592 -1.167 -1.029 -0.950* -1.088 -0.475 -1.097 -6.812 (-0.77) (-1.55) (-0.99) (-1.89) (-0.58) (-0.45) (-1.42) (-0.19) Other status -0.654 -0.981 -0.914 -0.866* -1.125 -1.118 -1.036 -6.325 (-0.85) (-1.31) (-0.87) (-1.72) (-0.59) (-1.03) (-1.35) (-0.19) Ln Head Age -1.077 -0.561 2.551 1.202 6.233 3.172 1.027 -6.513 (-0.60) (-0.33) (0.95) (0.89) (1.23) (0.93) (0.63) (-0.22) Land Rights Document -0.028 -0.099 0.524* 0.173 0.684 0.368 0.309 0.048 (-0.14) (-0.44) (1.70) (1.10) (1.26) (1.02) (1.47) (0.07) Ln TLUs 0.140 0.077 -0.234 -0.025 -0.677 0.053 -0.113 2.493 (0.74) (0.41) (-0.79) (-0.17) (-1.21) (0.14) (-0.61) (0.21) Ln Adult Equivalent -0.378 0.594* 0.459 -0.195 -0.032 -0.352 0.311 1.852 (-1.13) (1.86) (0.97) (-0.87) (-0.04) (-0.77) (0.92) (0.21) Ln Land Size (ha) 0.056 0.177 -0.338 -0.219 -0.733 -0.519 -0.184 -0.518 (0.17) (0.56) (-0.68) (-0.88) (-0.77) (-0.82) (-0.61) (-0.20) Ln Ag Wage Income -0.008 0.001 -0.009 0.002 0.006 0.007 -0.002 -0.033 (-0.81) (0.14) (-0.67) (0.37) (0.32) (0.50) (-0.27) (-0.43) Ln Transfer Income 0.003 -0.005 -0.024 -0.012 -0.055* -0.010 -0.006 -0.021 (0.21) (-0.42) (-1.27) (-1.43) (-1.75) (-0.49) (-0.52) (-0.13) Ln Nonfarm Income 0.012 0.017** 0.003 0.004 0.015 0.009 -0.002 -0.041 (1.26) (1.97) (0.27) (0.63) (0.75) (0.64) (-0.32) (-0.19) Observations 11061 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 114 Table A 2.10: Effects of Land Productivity on School Expenditure and Study Times by Gender of Head (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either a re currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.027 0.008 0.016 0.016 -0.016 0.021 0.025 -0.003 (1.27) (0.39) (0.66) (1.04) (-0.60) (0.98) (1.59) (-0.10) Female Head X Log Land Productivity 0.111*** -0.025 0.037 0.008 -0.035 -0.003 -0.015 -0.015 (2.77) (-0.81) (0.83) (0.31) (-0.70) (-0.07) (-0.56) (-0.35) Relationship to Head (Excludes Spouse) Child 0.458 -0.692 2.996* -0.539 5.932*** -0.981** 1.294 1.199 (0.24) (-0.42) (1.83) (-1.32) (3.86) (-2.05) (0.67) (1.52) Grandchild 0.510 -1.261 2.808 -0.602 5.453*** -0.462 1.128 1.051 (0.26) (-0.74) (1.58) (-1.27) (3.28) (-0.69) (0.58) (1.36) Relative 0.289 -1.347 2.514 -0.629 5.528*** -1.171** 1.004 0.313 (0.15) (-0.80) (1.47) (-1.50) (3.46) (-2.08) (0.51) (0.49) Non-relative -0.175 -1.854 1.524 -0.993** 4.151** -1.755** 0.144 0.000 (-0.09) (-1.06) (0.82) (-2.00) (2.28) (-2.48) (0.07) (.) Marital Status (Excludes Never Married) Married 1.200 2.418 1.815 -1.537* 6.801*** -1.562*** 0.091 -5.057*** (0.86) (1.13) (0.96) (-1.75) (3.21) (-2.84) (0.05) (-13.78) Other status -1.850* -1.438** -1.176 -1.215** -0.350 -2.519*** -1.956** 0.004 (-1.82) (-2.08) (-1.13) (-2.11) (-0.30) (-3.68) (-2.38) (0.01) Female Head -1.027* 0.208 -0.943 -0.305 -0.257 0.162 -0.145 0.048 (-1.85) (0.47) (-1.59) (-0.87) (-0.35) (0.32) (-0.38) (0.08) Head Education (Excludes - No Education) Primary -0.082 -0.114 0.466** 0.024 0.446* -0.152 0.099 -0.484** (-0.54) (-0.71) (2.09) (0.28) (1.73) (-0.94) (0.69) (-2.26) Secondary -0.404 -0.154 -0.091 -0.150 0.709 -0.608 -0.059 -0.158 (-0.97) (-0.50) (-0.25) (-0.74) (1.42) (-1.58) (-0.23) (-0.35) Post-Secondary 0.926 1.698 -3.333 1.442 3.565*** 3.376 1.616 5.803*** (0.56) (1.27) (-1.15) (1.00) (2.86) (1.20) (0.92) (11.82) Head Marital Status (Excludes- Never Married Married -0.606 -1.270** -1.173 -0.959** -1.173 -0.392 -1.209* 0.653 (-0.88) (-2.09) (-1.53) (-2.17) (-1.30) (-0.51) (-1.94) (1.01) Other status -0.700 -1.148** -0.926 -0.800* -0.839 -0.793 -1.103* 0.439 (-1.04) (-1.96) (-1.24) (-1.90) (-0.96) (-1.07) (-1.85) (0.68) Ln Head Age -0.429 -0.014 0.537 0.210 1.026 0.159 0.139 -0.240 (-0.64) (-0.03) (0.71) (0.63) (1.28) (0.31) (0.27) (-0.38) Land Rights Document 0.108 -0.225** 0.049 0.056 -0.181 0.161 0.011 0.015 (0.87) (-2.36) (0.37) (0.76) (-1.23) (1.24) (0.14) (0.11) Ln TLUs 0.065 0.076 0.057 0.075* -0.026 0.318*** 0.034 -0.012 (0.78) (1.07) (0.54) (1.68) (-0.22) (3.77) (0.53) (-0.11) Ln Adult Equivalent -0.456 0.644** 0.594* -0.166 0.240 -0.302 0.404 0.051 (-1.41) (2.45) (1.65) (-0.88) (0.59) (-0.96) (1.53) (0.13) Ln Land Size (ha) -0.022 0.110 0.121 -0.015 0.280* 0.038 0.039 -0.099 (-0.18) (1.20) (0.90) (-0.22) (1.72) (0.30) (0.49) (-0.59) Ln Ag Wage Income -0.007 -0.001 -0.013 0.002 0.002 0.008 -0.005 -0.021* (-0.74) (-0.11) (-1.23) (0.35) (0.19) (0.83) (-0.72) (-1.83) Ln Transfer Income -0.001 -0.004 -0.009 -0.007 -0.023* 0.003 0.002 0.010 (-0.06) (-0.57) (-0.81) (-1.36) (-1.89) (0.30) (0.36) (0.94) Ln Nonfarm Income 0.012 0.020*** 0.003 0.002 0.010 0.001 -0.001 0.003 (1.54) (2.97) (0.37) (0.44) (0.93) (0.14) (-0.26) (0.28) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 115 Table A 2.11: Effects of Labor Productivity on School Expenditure and Study Times by Gender of Head (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Samp le restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.016 0.012 -0.024 0.017 -0.004 0.048 0.021 -0.036 (0.48) (0.43) (-0.67) (0.70) (-0.09) (1.43) (0.96) (-0.85) Female Head X Log Labor Productivity 0.153** 0.000 0.122* 0.030 -0.072 -0.034 0.018 0.078 (2.27) (0.01) (1.77) (0.72) (-0.94) (-0.54) (0.42) (1.06) Relationship to Head (Excludes Spouse) Child 0.434 -0.732 2.961* -0.570 5.955*** -0.995** 1.252 1.216 (0.23) (-0.45) (1.85) (-1.39) (3.88) (-2.05) (0.66) (1.54) Grandchild 0.502 -1.306 2.780 -0.632 5.471*** -0.474 1.088 1.048 (0.26) (-0.78) (1.59) (-1.33) (3.30) (-0.70) (0.56) (1.35) Relative 0.291 -1.395 2.491 -0.658 5.542*** -1.186** 0.961 0.320 (0.15) (-0.83) (1.48) (-1.57) (3.48) (-2.08) (0.50) (0.50) Non-relative -0.179 -1.894 1.491 -1.020** 4.169** -1.765** 0.102 0.000 (-0.09) (-1.09) (0.82) (-2.06) (2.29) (-2.48) (0.05) (.) Marital Status (Excludes Never Married) Married 1.159 2.394 1.757 -1.567* 6.837*** -1.579*** 0.036 -5.202*** (0.82) (1.12) (0.92) (-1.77) (3.23) (-2.77) (0.02) (-16.17) Other status -1.861* -1.443** -1.173 -1.224** -0.344 -2.541*** -1.971** 0.021 (-1.84) (-2.08) (-1.13) (-2.11) (-0.29) (-3.70) (-2.40) (0.03) Female Head -0.852 -0.073 -1.396** -0.423 -0.138 0.376 -0.446 -0.677 (-1.49) (-0.16) (-2.43) (-1.20) (-0.20) (0.76) (-1.25) (-1.05) Head Education (Excludes - No Education) Primary -0.068 -0.115 0.470** 0.027 0.442* -0.150 0.099 -0.490** (-0.45) (-0.71) (2.11) (0.31) (1.71) (-0.93) (0.69) (-2.31) Secondary -0.392 -0.160 -0.101 -0.149 0.709 -0.597 -0.063 -0.153 (-0.93) (-0.52) (-0.28) (-0.73) (1.42) (-1.54) (-0.24) (-0.34) Post-Secondary 0.953 1.692 -3.348 1.449 3.560*** 3.404 1.619 5.751*** (0.57) (1.27) (-1.16) (1.01) (2.86) (1.21) (0.92) (11.72) Head Marital Status (Excludes- Never Married Married -0.571 -1.274** -1.169 -0.950** -1.185 -0.372 -1.201* 0.647 (-0.84) (-2.10) (-1.54) (-2.17) (-1.32) (-0.49) (-1.93) (0.99) Other status -0.662 -1.152** -0.919 -0.790* -0.853 -0.774 -1.094* 0.430 (-0.99) (-1.97) (-1.24) (-1.89) (-0.97) (-1.04) (-1.83) (0.67) Ln Head Age -0.419 0.018 0.560 0.232 1.011 0.169 0.169 -0.207 (-0.63) (0.04) (0.74) (0.70) (1.27) (0.32) (0.33) (-0.32) Land Rights Document 0.104 -0.221** 0.045 0.057 -0.179 0.163 0.012 0.015 (0.84) (-2.30) (0.34) (0.77) (-1.22) (1.26) (0.15) (0.11) Ln TLUs 0.079 0.078 0.068 0.082* -0.035 0.325*** 0.046 -0.015 (0.95) (1.10) (0.64) (1.86) (-0.30) (3.88) (0.71) (-0.13) Ln Adult Equivalent -0.442 0.638** 0.593 -0.165 0.237 -0.296 0.403 0.030 (-1.36) (2.42) (1.64) (-0.87) (0.59) (-0.94) (1.52) (0.08) Ln Land Size (ha) -0.056 0.107 0.116 -0.028 0.293* 0.013 0.023 -0.089 (-0.46) (1.16) (0.87) (-0.41) (1.79) (0.10) (0.29) (-0.53) Ln Ag Wage Income -0.006 -0.001 -0.012 0.002 0.002 0.008 -0.005 -0.021* (-0.63) (-0.14) (-1.22) (0.37) (0.17) (0.84) (-0.74) (-1.85) Ln Transfer Income -0.001 -0.004 -0.009 -0.007 -0.023* 0.003 0.002 0.010 (-0.08) (-0.59) (-0.82) (-1.38) (-1.88) (0.29) (0.34) (0.93) Ln Nonfarm Income 0.012 0.020*** 0.003 0.002 0.010 0.001 -0.001 0.002 (1.52) (3.01) (0.40) (0.46) (0.92) (0.15) (-0.21) (0.25) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 116 Table A 2.12: Effects of Land Productivity on School Expenditure and Study Times by Gender of Head (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.396 0.542 1.487** 0.158 2.371** 0.462 0.923* 0.419 (-0.96) (1.13) (2.22) (0.59) (2.46) (0.96) (1.86) (0.83) Female Head X Log Land Productivity 0.556 -0.600 -1.482** 0.128 -2.212** -0.526 -0.828* -0.126 (1.37) (-1.08) (-2.32) (0.43) (-2.46) (-1.15) (-1.81) (-0.26) Relationship to Head (Excludes Spouse) Child 0.287 -0.568 3.256 -0.842** 6.001* -0.819 1.310 0.356 (0.14) (-0.33) (1.09) (-1.98) (1.69) (-0.93) (0.60) (0.62) Grandchild 0.413 -1.294 2.619 -1.013* 4.725 -0.418 0.843 0.000 (0.20) (-0.72) (0.84) (-1.91) (1.28) (-0.39) (0.37) (.) Relative 0.161 -1.307 2.525 -1.040** 5.102 -1.062 0.832 -0.729 (0.08) (-0.73) (0.83) (-2.18) (1.42) (-1.08) (0.37) (-1.10) Non-relative -0.537 -1.518 2.363 -1.258** 5.139 -1.415 0.506 -0.866 (-0.26) (-0.82) (0.75) (-2.36) (1.32) (-1.33) (0.22) (-1.05) Marital Status (Excludes Never Married) Married 0.162 3.757 5.472* -1.309 12.598** -0.435 2.269 -0.882 (0.08) (1.24) (1.74) (-1.51) (2.28) (-0.26) (0.88) (-0.18) Other status -2.053* -1.187 -0.483 -1.139** 0.784 -2.314*** -1.530 -0.066 (-1.93) (-1.55) (-0.38) (-2.00) (0.46) (-3.76) (-1.62) (-0.09) Female Head -6.373 7.087 17.268** -1.537 26.070** 6.384 9.681* 1.593 (-1.33) (1.07) (2.26) (-0.44) (2.42) (1.17) (1.75) (0.28) Head Education (Excludes - No Education) Primary -0.152 -0.048 0.644** 0.022 0.716 -0.094 0.200 -0.721** (-0.90) (-0.25) (2.10) (0.23) (1.60) (-0.51) (0.99) (-2.21) Secondary -0.524 -0.030 0.231 -0.200 1.144 -0.490 0.102 -0.298 (-1.17) (-0.08) (0.44) (-0.86) (1.39) (-1.26) (0.29) (-0.59) Post-Secondary 0.936 1.761 -3.194 1.326 3.651* 3.451 1.645 5.644*** (0.56) (1.21) (-1.00) (0.90) (1.89) (1.17) (0.84) (10.31) Head Marital Status (Excludes- Never Married Married -0.541 -1.366** -1.463 -1.088** -1.752 -0.452 -1.429** 0.920 (-0.73) (-2.10) (-1.37) (-2.31) (-1.34) (-0.56) (-2.07) (1.30) Other status -0.567 -1.329** -1.450 -0.952** -1.797 -0.923 -1.467** 0.669 (-0.77) (-2.04) (-1.35) (-2.05) (-1.36) (-1.14) (-2.12) (0.96) Ln Head Age -0.418 -0.046 0.515 0.480 1.284 0.082 0.244 -0.001 (-0.57) (-0.07) (0.51) (1.13) (0.94) (0.14) (0.38) (-0.00) Land Rights Document 0.011 -0.118 0.351 0.117 0.343 0.243 0.209 0.062 (0.07) (-0.79) (1.57) (1.17) (1.09) (1.41) (1.40) (0.39) Ln TLUs 0.314 -0.255 -0.858* -0.025 -1.522** 0.047 -0.529 -0.346 (1.13) (-0.80) (-1.83) (-0.14) (-2.22) (0.14) (-1.55) (-0.89) Ln Adult Equivalent -0.433 0.654** 0.604 -0.240 0.175 -0.280 0.378 -0.077 (-1.26) (2.24) (1.34) (-1.12) (0.30) (-0.83) (1.19) (-0.18) Ln Land Size (ha) -0.179 0.311 0.688** 0.091 1.255*** 0.195 0.407* 0.260 (-0.77) (1.40) (2.12) (0.70) (2.61) (0.76) (1.77) (0.63) Ln Ag Wage Income -0.012 0.005 0.003 0.000 0.024 0.013 0.003 -0.021 (-1.21) (0.48) (0.18) (0.05) (1.17) (1.23) (0.34) (-1.52) Ln Transfer Income (0.02) (-0.68) (-0.92) (-1.55) (-1.71) (0.21) (-0.07) (1.11) Ln Nonfarm Income 0.017* 0.014 -0.012 0.002 -0.014 -0.004 -0.010 -0.001 (1.84) (1.62) (-0.96) (0.40) (-0.77) (-0.40) (-1.25) (-0.07) Observations 11061 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 117 Table A 2.13: Effects of Labor Productivity on School Expenditure and Study Times by Gender of Head (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.816 1.107 3.036* 0.321 4.840* 0.943 1.885 1.170 (-0.95) (1.05) (1.81) (0.56) (1.96) (0.91) (1.58) (0.98) Female Head X Log Labor Productivity 1.396 -1.520 -3.771* 0.288 -5.648* -1.331 -2.117 -0.568 (1.26) (-1.08) (-1.88) (0.34) (-1.94) (-1.06) (-1.54) (-0.39) Relationship to Head (Excludes Spouse) Child 0.142 -0.679 2.724 -1.296* 4.942 -0.891 0.851 0.166 (0.07) (-0.31) (0.54) (-1.75) (0.71) (-0.56) (0.26) (0.26) Grandchild 0.194 -1.345 2.235 -1.481* 3.885 -0.438 0.464 0.000 (0.09) (-0.58) (0.42) (-1.73) (0.54) (-0.25) (0.13) (.) Relative 0.031 -1.489 1.800 -1.505* 3.741 -1.197 0.256 -0.844 (0.01) (-0.65) (0.35) (-1.84) (0.53) (-0.71) (0.08) (-1.09) Non-relative -0.649 -1.649 1.827 -1.623** 4.121 -1.510 0.073 -0.380 (-0.30) (-0.73) (0.35) (-2.13) (0.58) (-0.91) (0.02) (-0.37) Marital Status (Excludes Never Married) Married 0.487 3.212 3.939 -1.540** 10.118 -0.895 1.295 0.248 (0.27) (1.05) (1.12) (-2.01) (1.55) (-0.52) (0.44) (0.05) Other status -1.642 -1.741** -1.985 -1.267** -1.594 -2.787*** -2.452*** -0.581 (-1.50) (-2.33) (-1.51) (-2.02) (-0.81) (-3.98) (-2.68) (-0.54) Female Head -9.993 11.104 27.392* -2.013 41.391* 9.887 15.463 4.470 (-1.23) (1.08) (1.85) (-0.32) (1.92) (1.07) (1.51) (0.42) Head Education (Excludes - No Education) Primary -0.135 -0.067 0.618* 0.066 0.699 -0.113 0.199 -0.581** (-0.74) (-0.33) (1.79) (0.66) (1.40) (-0.58) (0.89) (-2.05) Secondary -0.722 0.191 0.820 -0.167 2.068* -0.301 0.458 -0.350 (-1.40) (0.38) (1.04) (-0.57) (1.77) (-0.58) (0.84) (-0.58) Post-Secondary 0.388 2.378 -1.528 1.454 6.281*** 3.980 2.664 7.078*** (0.21) (1.55) (-0.46) (0.95) (2.91) (1.41) (1.34) (4.63) Head Marital Status (Excludes- Never Married Married -0.871 -0.977 -0.394 -0.972* -0.047 -0.122 -0.765 1.544 (-1.16) (-1.38) (-0.33) (-1.93) (-0.03) (-0.14) (-1.03) (1.23) Other status -0.904 -0.939 -0.379 -0.836* -0.088 -0.591 -0.801 1.302 (-1.25) (-1.41) (-0.34) (-1.75) (-0.06) (-0.71) (-1.17) (1.04) Ln Head Age -0.361 0.017 0.862 0.817 1.995 0.120 0.557 -0.365 (-0.39) (0.02) (0.61) (1.36) (1.09) (0.14) (0.61) (-0.35) Land Rights Document -0.016 -0.082 0.467 0.161 0.545 0.271 0.292 0.075 (-0.08) (-0.40) (1.37) (1.29) (1.09) (1.26) (1.27) (0.47) Ln TLUs 0.149 -0.043 -0.287 0.020 -0.621 0.228 -0.180 -0.226 (1.04) (-0.25) (-0.94) (0.22) (-1.36) (1.35) (-0.86) (-1.13) Ln Adult Equivalent -0.549 0.778** 0.932 -0.228 0.685 -0.173 0.574 -0.134 (-1.48) (2.35) (1.61) (-0.99) (0.85) (-0.47) (1.50) (-0.26) Ln Land Size (ha) 0.227 -0.240 -0.837 -0.093 -1.188 -0.274 -0.547 -0.029 (0.76) (-0.69) (-1.46) (-0.47) (-1.40) (-0.81) (-1.37) (-0.14) Ln Ag Wage Income -0.011 0.003 -0.002 0.002 0.018 0.011 0.001 -0.017 (-1.06) (0.30) (-0.11) (0.35) (0.80) (1.07) (0.12) (-1.24) Ln Transfer Income -0.000 -0.007 -0.018 -0.011* -0.039 0.001 -0.004 0.008 (-0.04) (-0.71) (-0.96) (-1.71) (-1.62) (0.09) (-0.39) (0.62) Ln Nonfarm Income 0.016* 0.016* -0.008 0.003 -0.006 -0.003 -0.007 0.013 (1.76) (1.84) (-0.52) (0.55) (-0.30) (-0.26) (-0.79) (0.87) Observations 11061 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 118 Table A 2.14: Effects of Land Productivity on School Expenditure and Study Times by School ŒLevel (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth f ixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the i ndividual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.463*** 0.006 0.064*** 0.059*** -0.004 0.095*** 0.075*** 0.009 (19.42) (0.33) (2.70) (3.56) (-0.14) (4.04) (4.63) (0.35) Primary X Log Land Productivity (TSH/Ha) -0.543*** -0.008 -0.050*** -0.055*** -0.030* -0.101*** -0.074*** -0.019 (-35.87) (-0.76) (-3.77) (-5.74) (-1.65) (-6.41) (-9.07) (-1.04) Relationship to Head (Excludes Spouse) Child 0.525 -0.715 3.026* -0.535 5.898*** -0.990* 1.275 1.243 (0.25) (-0.43) (1.83) (-1.43) (3.90) (-1.84) (0.67) (1.58) Grandchild 0.461 -1.288 2.830 -0.611 5.410*** -0.495 1.091 1.065 (0.22) (-0.76) (1.58) (-1.37) (3.30) (-0.70) (0.57) (1.38) Relative 0.242 -1.377 2.538 -0.637 5.481*** -1.205** 0.965 0.336 (0.11) (-0.81) (1.47) (-1.63) (3.49) (-1.96) (0.50) (0.52) Non-relative -0.571 -1.888 1.516 -1.037** 4.086** -1.854** 0.058 0.000 (-0.26) (-1.08) (0.81) (-2.15) (2.27) (-2.47) (0.03) (.) Marital Status (Excludes Never Married) Married 1.380 2.393 1.859 -1.522* 6.768*** -1.553** 0.083 -4.931*** (1.35) (1.12) (0.99) (-1.84) (3.22) (-2.48) (0.05) (-13.36) Other status -0.996 -1.429** -1.093 -1.129** -0.307 -2.364*** -1.844** 0.049 (-1.40) (-2.06) (-1.05) (-2.07) (-0.26) (-3.32) (-2.25) (0.07) Female Head 0.020 -0.081 -0.542 -0.238 -0.665 0.088 -0.350 -0.131 (0.09) (-0.30) (-1.61) (-1.58) (-1.63) (0.33) (-1.50) (-0.37) Head Education (Excludes - No Education) Primary -0.098 -0.117 0.466** 0.022 0.442* -0.157 0.094 -0.480** (-0.77) (-0.72) (2.10) (0.26) (1.72) (-0.98) (0.66) (-2.25) Secondary -0.519 -0.163 -0.095 -0.162 0.692 -0.636 -0.084 -0.151 (-1.55) (-0.53) (-0.26) (-0.79) (1.38) (-1.63) (-0.32) (-0.33) Post-Secondary 1.281 1.694 -3.291 1.477 3.569*** 3.433 1.653 5.820*** (0.63) (1.27) (-1.13) (1.01) (2.83) (1.23) (0.92) (11.86) Head Marital Status (Excludes- Never Married Married -0.243 -1.270** -1.133 -0.923** -1.162 -0.329 -1.166* 0.675 (-0.46) (-2.10) (-1.48) (-2.07) (-1.29) (-0.44) (-1.90) (1.05) Other status -0.202 -1.145** -0.875 -0.750* -0.818 -0.704 -1.041* 0.462 (-0.39) (-1.96) (-1.17) (-1.76) (-0.93) (-0.96) (-1.77) (0.72) Ln Head Age -0.597 0.003 0.502 0.195 1.046 0.145 0.138 -0.258 (-1.27) (0.01) (0.67) (0.61) (1.31) (0.28) (0.28) (-0.41) Land Rights Document 0.234** -0.222** 0.060 0.069 -0.173 0.185 0.029 0.020 (2.41) (-2.33) (0.46) (0.94) (-1.18) (1.44) (0.36) (0.15) Ln TLUs -0.010 0.079 0.046 0.068 -0.024 0.307*** 0.029 -0.014 (-0.15) (1.10) (0.44) (1.54) (-0.21) (3.69) (0.44) (-0.12) Ln Adult Equivalent 0.083 0.646** 0.650* -0.112 0.261 -0.207 0.471* 0.069 (0.32) (2.46) (1.80) (-0.59) (0.65) (-0.66) (1.79) (0.18) Ln Land Size (ha) 0.027 0.111 0.125 -0.010 0.283* 0.048 0.046 -0.093 (0.28) (1.22) (0.93) (-0.14) (1.74) (0.38) (0.59) (-0.55) Ln Ag Wage Income 0.001 -0.001 -0.012 0.002 0.002 0.009 -0.004 -0.021* (0.14) (-0.13) (-1.14) (0.52) (0.19) (0.96) (-0.62) (-1.85) Ln Transfer Income 0.001 -0.004 -0.008 -0.007 -0.023* 0.003 0.002 0.010 (0.12) (-0.58) (-0.79) (-1.33) (-1.89) (0.33) (0.38) (0.87) Ln Nonfarm Income -0.001 0.020*** 0.002 0.001 0.009 -0.001 -0.003 0.002 (-0.16) (2.98) (0.21) (0.16) (0.89) (-0.13) (-0.56) (0.24) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 119 Table A 2.15: Effects of Labor Productivity on School Expenditure and Study Times by School ŒLevel (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes i ndividual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed ac ademic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.711*** 0.015 0.057 0.087*** 0.016 0.173*** 0.110*** 0.013 (19.35) (0.53) (1.59) (3.07) (0.38) (4.66) (4.74) (0.30) Primary X Log Labor Productivity (TSH/Day) -0.884*** -0.004 -0.067*** -0.084*** -0.051* -0.180*** -0.113*** -0.032 (-38.25) (-0.26) (-3.10) (-5.36) (-1.76) (-6.92) (-8.71) (-1.05) Relationship to Head (Excludes Spouse) Child 0.815 -0.731 3.068* -0.523 5.919*** -0.965* 1.299 1.250 (0.45) (-0.45) (1.87) (-1.39) (3.90) (-1.81) (0.70) (1.58) Grandchild 0.655 -1.305 2.878 -0.605 5.416*** -0.494 1.106 1.063 (0.35) (-0.77) (1.62) (-1.36) (3.30) (-0.70) (0.59) (1.38) Relative 0.469 -1.394 2.588 -0.630 5.491*** -1.199** 0.983 0.341 (0.26) (-0.83) (1.51) (-1.60) (3.48) (-1.97) (0.52) (0.53) Non-relative -0.313 -1.895 1.558 -1.022** 4.105** -1.837** 0.083 0.000 (-0.16) (-1.09) (0.84) (-2.13) (2.27) (-2.46) (0.04) (.) Marital Status (Excludes Never Married) Married 0.850 2.393 1.776 -1.590* 6.788*** -1.666*** -0.005 -4.960** (0.76) (1.12) (0.94) (-1.90) (3.21) (-2.60) (-0.00) (-15.21) Other status -1.064 -1.439** -1.124 -1.150** -0.289 -2.372*** -1.869** 0.058 (-1.61) (-2.08) (-1.08) (-2.03) (-0.25) (-3.21) (-2.30) (0.08) Female Head 0.105 -0.071 -0.547 -0.226 -0.653 0.113 -0.336 -0.128 (0.47) (-0.27) (-1.62) (-1.50) (-1.60) (0.43) (-1.45) (-0.36) Head Education (Excludes - No Education) Primary -0.107 -0.115 0.465** 0.023 0.440* -0.157 0.094 -0.484** (-0.83) (-0.71) (2.10) (0.27) (1.71) (-0.98) (0.66) (-2.28) Secondary -0.493 -0.161 -0.088 -0.156 0.688 -0.629 -0.076 -0.153 (-1.50) (-0.52) (-0.24) (-0.76) (1.37) (-1.61) (-0.29) (-0.34) Post-Secondary 1.501 1.695 -3.269 1.506 3.564*** 3.494 1.689 5.844** (0.77) (1.27) (-1.12) (1.03) (2.83) (1.26) (0.94) (11.81) Head Marital Status (Excludes- Never Married Married -0.266 -1.272** -1.128 -0.919** -1.181 -0.321 -1.163* 0.679 (-0.50) (-2.10) (-1.48) (-2.10) (-1.32) (-0.43) (-1.89) (1.04) Other status -0.236 -1.150** -0.870 -0.748* -0.840 -0.697 -1.040* 0.465 (-0.46) (-1.97) (-1.17) (-1.79) (-0.96) (-0.96) (-1.77) (0.72) Ln Head Age -0.509 0.018 0.491 0.215 1.051 0.187 0.159 -0.244 (-1.13) (0.04) (0.66) (0.66) (1.32) (0.36) (0.32) (-0.39) Land Rights Document 0.162* -0.221** 0.046 0.062 -0.173 0.177 0.019 0.017 (1.67) (-2.30) (0.35) (0.84) (-1.18) (1.38) (0.24) (0.13) Ln TLUs -0.007 0.078 0.061 0.073* -0.040 0.307*** 0.035 -0.017 (-0.11) (1.10) (0.58) (1.67) (-0.34) (3.72) (0.54) (-0.15) Ln Adult Equivalent 0.137 0.640** 0.651* -0.108 0.260 -0.187 0.476* 0.074 (0.54) (2.43) (1.80) (-0.57) (0.64) (-0.59) (1.81) (0.19) Ln Land Size (ha) 0.000 0.107 0.108 -0.025 0.305* 0.031 0.030 -0.089 (0.00) (1.17) (0.81) (-0.36) (1.87) (0.25) (0.39) (-0.53) Ln Ag Wage Income -0.002 -0.001 -0.012 0.002 0.002 0.009 -0.005 -0.021* (-0.27) (-0.14) (-1.17) (0.46) (0.17) (0.92) (-0.67) (-1.86) Ln Transfer Income 0.005 -0.004 -0.008 -0.007 -0.023* 0.004 0.003 0.010 (0.61) (-0.59) (-0.75) (-1.27) (-1.87) (0.39) (0.45) (0.90) Ln Nonfarm Income -0.002 0.020*** 0.002 0.001 0.009 -0.001 -0.003 0.002 (-0.30) (2.99) (0.23) (0.17) (0.87) (-0.17) (-0.55) (0.21) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 120 Table A 2.16: Effects of Land Productivity on School Expenditure and Study Times by School ŒLevel (IV) Œ Full Table TŒstatistics in parentheses. *p <0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or wer e in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.892*** 0.338 1.031** 0.307 1.610*** 0.473 0.763** 0.373 (3.21) (1.08) (2.42) (1.57) (2.73) (1.41) (2.34) (1.35) Primary X Log Land Productivity (TSH/Ha) -0.585*** -0.010 -0.049*** -0.053*** -0.031 -0.109*** -0.077*** -0.018 (-33.19) (-0.89) (-2.87) (-5.26) (-1.21) (-6.45) (-6.94) (-0.87) Relationship to Head (Excludes Spouse) Child 0.087 -1.077 1.968 -0.807* 4.134 -1.397 0.526 1.285 (0.04) (-0.69) (0.93) (-1.93) (1.62) (-1.62) (0.30) (1.55) Grandchild -0.207 -1.833 1.235 -1.020* 2.751 -1.109 -0.039 0.862 (-0.10) (-1.08) (0.53) (-1.85) (0.99) (-1.04) (-0.02) (1.02) Relative -0.402 -1.901 1.003 -1.032** 2.922 -1.797* -0.123 0.170 (-0.19) (-1.14) (0.45) (-2.05) (1.10) (-1.79) (-0.07) (0.24) Non-relative -0.917 -2.144 0.772 -1.227** 2.843 -2.147** -0.473 0.000 (-0.42) (-1.30) (0.34) (-2.39) (1.02) (-2.15) (-0.25) (.) Marital Status (Excludes Never Married) Married 2.180 3.053 3.793 -1.025* 9.991** -0.810 1.452 -1.344 (1.48) (1.17) (1.62) (-1.79) (2.38) (-0.58) (0.68) (-0.49) Other status -0.732 -1.259* -0.609 -1.006* 0.506 -2.164*** -1.494 -0.010 (-0.92) (-1.70) (-0.53) (-1.93) (0.35) (-3.39) (-1.63) (-0.01) Female Head 0.323 0.184 0.237 -0.037 0.632 0.383 0.199 0.134 (1.00) (0.48) (0.43) (-0.17) (0.79) (0.99) (0.51) (0.30) Head Education (Excludes - No Education) Primary -0.080 -0.101 0.513* 0.034 0.519 -0.140 0.127 -0.711** (-0.60) (-0.59) (1.91) (0.36) (1.38) (-0.82) (0.71) (-2.34) Secondary -0.569* -0.196 -0.188 -0.186 0.535 -0.674* -0.151 -0.281 (-1.74) (-0.60) (-0.42) (-0.82) (0.75) (-1.72) (-0.48) (-0.58) Post-Secondary 1.142 1.560 -3.687 1.375 2.911* 3.286 1.375 5.677** (0.54) (1.12) (-1.20) (0.91) (1.78) (1.12) (0.71) (10.82) Head Marital Status (Excludes- Never Married Married -0.450 -1.462** -1.702* -1.070** -2.108* -0.542 -1.566** 0.926 (-0.78) (-2.26) (-1.76) (-2.16) (-1.76) (-0.68) (-2.22) (1.34) Other status -0.464 -1.390** -1.601* -0.938* -2.024* -0.975 -1.551** 0.671 (-0.80) (-2.15) (-1.65) (-1.91) (-1.67) (-1.24) (-2.20) (1.00) Ln Head Age -0.190 0.346 1.509 0.454 2.723** 0.530 0.850 0.017 (-0.35) (0.51) (1.40) (1.05) (2.00) (0.83) (1.13) (0.02) Land Rights Document 0.375*** -0.113 0.379* 0.150 0.360 0.309* 0.256* 0.054 (2.74) (-0.78) (1.86) (1.48) (1.29) (1.79) (1.80) (0.38) Ln TLUs -0.280 -0.141 -0.599* -0.098 -1.100** 0.059 -0.428* -0.312 (-1.33) (-0.61) (-1.85) (-0.69) (-2.42) (0.24) (-1.74) (-1.23) Ln Adult Equivalent 0.008 0.554* 0.372 -0.184 -0.197 -0.305 0.279 -0.059 (0.03) (1.90) (0.83) (-0.85) (-0.33) (-0.92) (0.83) (-0.14) Ln Land Size (ha) 0.262 0.304 0.688** 0.135 1.223*** 0.265 0.446** 0.236 (1.37) (1.46) (2.35) (1.05) (2.90) (1.09) (2.07) (0.76) Ln Ag Wage Income 0.001 -0.001 -0.012 0.002 0.001 0.009 -0.005 -0.022* (0.16) (-0.16) (-1.03) (0.44) (0.06) (0.91) (-0.58) (-1.81) Ln Transfer Income -0.002 -0.007 -0.015 -0.009 -0.034** 0.001 -0.002 0.012 (-0.20) (-0.81) (-1.16) (-1.54) (-2.04) (0.07) (-0.29) (1.05) Ln Nonfarm Income -0.003 0.019*** -0.002 -0.000 0.003 -0.003 -0.006 -0.001 (-0.53) (2.66) (-0.20) (-0.04) (0.22) (-0.32) (-0.90) (-0.08) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 121 Table A 2.17: Effects of Labor Productivity on School Expenditure and Study Times by School ŒLevel (IV) Œ Full T able TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 1.502*** 0.662 2.008** 0.580 3.156** 0.883 1.468** 1.055 (2.64) (1.05) (2.18) (1.47) (2.38) (1.31) (2.12) (1.20) Primary X Log Labor Productivity (TSH/Day) -0.945*** -0.011 -0.066* -0.083*** -0.029 -0.172*** -0.116*** -0.035 (-32.15) (-0.58) (-1.93) (-4.86) (-0.55) (-5.97) (-5.02) (-0.94) Relationship to Head (Excludes Spouse) Child -0.259 -1.668 0.209 -1.246 1.288 -2.017 -0.687 0.612 (-0.14) (-1.01) (0.06) (-1.48) (0.27) (-1.38) (-0.31) (0.63) Grandchild -0.646 -2.426 -0.532 -1.468 -0.100 -1.746 -1.265 0.351 (-0.32) (-1.31) (-0.15) (-1.51) (-0.02) (-1.05) (-0.52) (0.32) Relative -0.841 -2.523 -0.849 -1.499 -0.070 -2.462 -1.407 -0.457 (-0.42) (-1.38) (-0.25) (-1.57) (-0.01) (-1.50) (-0.59) (-0.48) Non-relative -1.184 -2.632 -0.677 -1.587* 0.495 -2.655* -1.472 0.000 (-0.62) (-1.61) (-0.21) (-1.92) (0.10) (-1.82) (-0.65) (.) Marital Status (Excludes Never Married) Married 1.078 2.607 2.439 -1.423** 7.870 -1.420 0.455 -0.326 (0.64) (1.03) (1.09) (-2.23) (1.63) (-0.94) (0.21) (-0.08) Other status -1.150* -1.555** -1.499 -1.245** -0.914 -2.516*** -2.127*** -0.465 (-1.74) (-2.26) (-1.35) (-2.02) (-0.60) (-3.63) (-2.58) (-0.48) Female Head 0.572 0.337 0.698 0.089 1.364 0.571 0.529 0.426 (1.36) (0.66) (0.93) (0.30) (1.24) (1.11) (0.96) (0.67) Head Education (Excludes - No Education) Primary -0.037 -0.053 0.655** 0.071 0.749* -0.087 0.226 -0.597** (-0.25) (-0.28) (2.16) (0.70) (1.69) (-0.47) (1.08) (-1.98) Secondary -0.437 -0.106 0.082 -0.113 0.965 -0.566 0.041 -0.312 (-1.25) (-0.31) (0.16) (-0.49) (1.19) (-1.31) (0.11) (-0.55) Post-Secondary 1.722 1.859 -2.784 1.629 4.338*** 3.668 2.028 6.892** (0.85) (1.35) (-0.91) (1.09) (2.82) (1.32) (1.08) (6.69) Head Marital Status (Excludes- Never Married Married -0.203 -1.232* -1.012 -0.890* -1.000 -0.281 -1.081 1.606 (-0.31) (-1.88) (-0.93) (-1.88) (-0.59) (-0.36) (-1.26) (1.23) Other status -0.223 -1.159* -0.909 -0.758* -0.913 -0.715 -1.065 1.342 (-0.35) (-1.81) (-0.85) (-1.65) (-0.54) (-0.94) (-1.27) (1.05) Ln Head Age 0.382 0.785 2.824* 0.805 4.825** 1.043 1.781 -0.227 (0.45) (0.78) (1.77) (1.26) (2.23) (1.04) (1.53) (-0.23) Land Rights Document 0.362** -0.053 0.556* 0.191 0.651 0.364 0.374* 0.096 (1.97) (-0.27) (1.92) (1.43) (1.54) (1.64) (1.78) (0.56) Ln TLUs -0.139 -0.031 -0.269 -0.010 -0.571* 0.187 -0.195 -0.252 (-1.02) (-0.22) (-1.19) (-0.11) (-1.76) (1.24) (-1.17) (-1.09) Ln Adult Equivalent 0.116 0.593** 0.493 -0.148 -0.009 -0.250 0.369 -0.278 (0.40) (2.04) (1.01) (-0.69) (-0.01) (-0.75) (1.03) (-0.50) Ln Land Size (ha) -0.183 -0.054 -0.383 -0.149 -0.491 -0.150 -0.311 0.034 (-1.00) (-0.28) (-1.23) (-1.17) (-1.07) (-0.70) (-1.37) (0.16) Ln Ag Wage Income -0.000 0.000 -0.008 0.003 0.008 0.010 -0.002 -0.017 (-0.03) (0.01) (-0.63) (0.58) (0.44) (1.00) (-0.23) (-1.22) Ln Transfer Income -0.001 -0.010 -0.025 -0.011 -0.050** -0.002 -0.009 0.006 (-0.10) (-0.98) (-1.48) (-1.60) (-2.23) (-0.18) (-0.77) (0.44) Ln Nonfarm Income -0.003 0.020** 0.003 0.001 0.011 -0.001 -0.002 0.011 (-0.37) (2.83) (0.24) (0.19) (0.66) (-0.12) (-0.32) (0.83) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 122 Table A 2.18: Effects of Gross Agricultural Income on School Expenditure and Study Times (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Samp le restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Value of Crop Income (TSH) 0.058** -0.006 0.011 0.024 0.025 0.019 0.012 0.002 (2.41) (-0.29) (0.43) (1.41) (0.85) (0.78) (0.70) (0.09) Relationship to Head (Excludes Spouse) Child 0.534 -0.705 3.042* -0.550 5.833*** -0.991** 1.284 1.203 (0.28) (-0.43) (1.84) (-1.36) (3.80) (-2.07) (0.67) (1.53) Grandchild 0.610 -1.274 2.868 -0.610 5.330*** -0.467 1.125 1.041 (0.31) (-0.75) (1.60) (-1.29) (3.21) (-0.69) (0.58) (1.35) Relative 0.383 -1.362 2.573 -0.642 5.397*** -1.181** 0.996 0.309 (0.20) (-0.80) (1.49) (-1.54) (3.39) (-2.10) (0.51) (0.48) Non-relative -0.092 -1.872 1.569 -1.005** 4.059** -1.769** 0.127 0.000 (-0.05) (-1.07) (0.84) (-2.04) (2.23) (-2.50) (0.06) (.) Marital Status (Excludes Never Married) Married 1.304 2.380 1.821 -1.520* 6.862*** -1.571*** 0.055 -5.011*** (0.93) (1.12) (0.96) (-1.76) (3.22) (-2.87) (0.03) (-13.63) Other status -1.867* -1.441** -1.185 -1.223** -0.340 -2.529*** -1.968** 0.004 (-1.85) (-2.08) (-1.14) (-2.10) (-0.29) (-3.67) (-2.39) (0.01) Female Head 0.252 -0.084 -0.532 -0.205 -0.606 0.132 -0.325 -0.114 (0.82) (-0.31) (-1.57) (-1.34) (-1.49) (0.48) (-1.37) (-0.32) Head Education (Excludes - No Education) Primary -0.075 -0.116 0.468** 0.024 0.445* -0.153 0.097 -0.490** (-0.49) (-0.72) (2.10) (0.28) (1.73) (-0.95) (0.67) (-2.30) Secondary -0.363 -0.161 -0.078 -0.144 0.699 -0.605 -0.061 -0.160 (-0.87) (-0.52) (-0.21) (-0.71) (1.39) (-1.57) (-0.23) (-0.35) Post-Secondary 0.967 1.692 -3.313 1.443 3.531*** 3.376 1.614 5.805*** (0.58) (1.27) (-1.14) (1.00) (2.83) (1.20) (0.92) (11.80) Head Marital Status (Excludes- Never Married Married -0.585 -1.271** -1.156 -0.962** -1.211 -0.392 -1.207* 0.659 (-0.85) (-2.09) (-1.51) (-2.19) (-1.35) (-0.51) (-1.94) (1.02) Other status -0.678 -1.148* -0.908 -0.802* -0.881 -0.791 -1.099* 0.444 (-1.01) (-1.96) (-1.21) (-1.91) (-1.01) (-1.07) (-1.84) (0.69) Ln Head Age -0.494 -0.004 0.497 0.217 1.110 0.165 0.144 -0.229 (-0.73) (-0.01) (0.66) (0.65) (1.40) (0.32) (0.28) (-0.36) Land Rights Document 0.105 -0.226** 0.043 0.058 -0.163 0.160 0.009 0.014 (0.85) (-2.36) (0.33) (0.79) (-1.11) (1.24) (0.11) (0.11) Ln TLUs 0.065 0.083 0.064 0.076* -0.048 0.324*** 0.046 -0.017 (0.78) (1.16) (0.61) (1.73) (-0.42) (3.88) (0.71) (-0.15) Ln Adult Equivalent -0.447 0.642** 0.604* -0.171 0.213 -0.307 0.401 0.046 (-1.38) (2.44) (1.67) (-0.91) (0.53) (-0.97) (1.52) (0.12) Ln Land Size (ha) -0.101 0.114 0.097 -0.043 0.277* 0.013 0.019 -0.094 (-0.83) (1.23) (0.72) (-0.62) (1.69) (0.10) (0.23) (-0.56) Ln Ag Wage Income -0.006 -0.001 -0.012 0.002 0.002 0.008 -0.005 -0.021* (-0.63) (-0.14) (-1.20) (0.36) (0.16) (0.82) (-0.75) (-1.84) Ln Transfer Income -0.001 -0.004 -0.009 -0.007 -0.023* 0.003 0.002 0.010 (-0.08) (-0.57) (-0.80) (-1.39) (-1.94) (0.29) (0.35) (0.95) Ln Nonfarm Income 0.012 0.020*** 0.003 0.002 0.010 0.001 -0.001 0.003 (1.45) (3.00) (0.35) (0.43) (0.94) (0.15) (-0.22) (0.28) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 123 Table A 2.19: Effects of Gross Agricultural Income on School Expenditure and Study Times (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Value of Crop Income (TSH) -0.307 0.454 1.330** 0.291 2.201** 0.378 0.875* 0.712 (-0.63) (1.00) (2.05) (1.05) (2.32) (0.79) (1.84) (1.20) Relationship to Head (Excludes Spouse) Child 1.103 -1.421 0.986 -0.966 2.443 -1.551 -0.061 0.971 (0.48) (-0.93) (0.42) (-1.51) (0.67) (-1.44) (-0.03) (1.06) Grandchild 1.290 -2.129 0.412 -1.107 1.281 -1.135 -0.482 0.509 (0.54) (-1.27) (0.16) (-1.47) (0.33) (-0.89) (-0.25) (0.49) Relative 1.140 -2.313 -0.162 -1.195 0.888 -1.926 -0.794 -0.071 (0.47) (-1.35) (-0.06) (-1.56) (0.23) (-1.47) (-0.40) (-0.09) Non-relative 0.420 -2.515 -0.279 -1.378** 1.011 -2.272* -1.082 0.000 (0.18) (-1.58) (-0.11) (-2.00) (0.27) (-1.93) (-0.58) (.) Marital Status (Excludes Never Married) Married 0.638 3.217 4.226 -1.034 10.828* -0.916 1.629 1.425 (0.32) (1.15) (1.46) (-1.57) (1.94) (-0.66) (0.69) (0.26) Other status -1.869* -1.438** -1.175 -1.221** -0.325 -2.527*** -1.962** -0.146 (-1.85) (-2.03) (-1.08) (-2.05) (-0.25) (-3.89) (-2.30) (-0.18) Female Head -0.123 0.387 0.822 0.069 1.626 0.500 0.561 0.498 (-0.21) (0.70) (1.03) (0.21) (1.39) (0.89) (0.97) (0.76) Head Education (Excludes - No Education) Primary -0.080 -0.111 0.484* 0.027 0.473 -0.149 0.108 -0.693** (-0.50) (-0.66) (1.84) (0.30) (1.26) (-0.89) (0.62) (-2.20) Secondary -0.389 -0.128 0.017 -0.125 0.856 -0.579 0.001 -0.175 (-0.89) (-0.39) (0.04) (-0.56) (1.11) (-1.42) (0.00) (-0.34) Post-Secondary 1.113 1.508 -3.840 1.336 2.662* 3.232 1.269 6.887*** (0.68) (1.10) (-1.26) (0.91) (1.72) (1.13) (0.69) (6.71) Head Marital Status (Excludes- Never Married Married -0.359 -1.555** -1.972* -1.127** -2.556 -0.614 -1.741** 1.134 (-0.44) (-2.21) (-1.84) (-2.31) (-1.61) (-0.74) (-2.12) (1.42) Other status -0.426 -1.463** -1.815* -0.986** -2.377 -1.038 -1.692** 0.905 (-0.52) (-2.09) (-1.69) (-2.05) (-1.48) (-1.27) (-2.07) (1.14) Ln Head Age -0.972 0.598 2.226 0.567 3.960** 0.635 1.275 -0.704 (-1.01) (0.68) (1.60) (1.02) (2.11) (0.77) (1.29) (-0.76) Land Rights Document -0.019 -0.071 0.489* 0.148 0.573 0.282 0.301 0.057 (-0.09) (-0.39) (1.84) (1.20) (1.48) (1.37) (1.61) (0.39) Ln TLUs 0.210 -0.100 -0.460 -0.030 -0.913** 0.181 -0.298 -0.354 (1.00) (-0.49) (-1.47) (-0.24) (-1.98) (0.85) (-1.32) (-1.17) Ln Adult Equivalent -0.264 0.412 -0.057 -0.305 -0.876 -0.487 -0.031 -0.516 (-0.62) (1.14) (-0.10) (-1.17) (-1.12) (-1.22) (-0.08) (-0.79) Ln Land Size (ha) 0.167 -0.223 -0.873* -0.239 -1.321* -0.251 -0.615 -0.182 (0.44) (-0.63) (-1.67) (-1.08) (-1.71) (-0.69) (-1.63) (-0.89) Ln Ag Wage Income -0.005 -0.002 -0.016 0.001 -0.004 0.007 -0.008 -0.019 (-0.50) (-0.29) (-1.26) (0.20) (-0.25) (0.69) (-0.90) (-1.43) Ln Transfer Income 0.004 -0.011 -0.027* -0.011 -0.054** -0.002 -0.010 0.010 (0.37) (-1.03) (-1.65) (-1.64) (-2.49) (-0.17) (-0.91) (0.83) Ln Nonfarm Income 0.012 0.020*** 0.001 0.002 0.007 0.001 -0.002 0.005 (1.48) (2.79) (0.11) (0.34) (0.49) (0.09) (-0.36) (0.45) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 124 Table A 2.20: Effects of Net Agricultural Income per Hectare on School Expenditure and Study Times (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individu al. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Ln Net Revenue per Ha (TSH/Ha) 0.050* 0.008 -0.002 0.027 -0.006 0.057* 0.004 0.039 (1.69) (0.34) (-0.09) (1.11) (-0.16) (1.65) (0.28) (1.31) Relationship to Head (Excludes Spouse) Child 0.686 -0.745 3.056* -0.398 5.763*** -0.779* 1.304 1.233 (0.35) (-0.46) (1.84) (-0.97) (3.76) (-1.68) (0.68) (1.55) Grandchild 0.761 -1.326 2.867 -0.450 5.373*** -0.316 1.118 1.109 (0.38) (-0.79) (1.59) (-0.95) (3.24) (-0.47) (0.58) (1.43) Relative 0.508 -1.500 2.583 -0.472 5.321*** -0.976* 0.969 0.344 (0.26) (-0.89) (1.49) (-1.12) (3.35) (-1.74) (0.49) (0.54) Non-relative 0.065 -1.923 1.550 -0.886* 3.997** -1.602** 0.113 0.000 (0.03) (-1.11) (0.82) (-1.76) (2.19) (-2.24) (0.06) (.) Marital Status (Excludes Never Married) Married 1.216 2.411 1.835 -1.467 6.808*** -1.481*** 0.075 -4.783*** (0.84) (1.15) (0.95) (-1.60) (3.18) (-2.79) (0.04) (-18.39) Other status -1.910* -1.521** -1.211 -0.896* -0.377 -2.167*** -2.039** 0.021 (-1.82) (-2.13) (-1.13) (-1.77) (-0.31) (-3.45) (-2.40) (0.03) Female Head 0.292 -0.195 -0.732** -0.235 -0.667 0.172 -0.453** 0.010 (0.93) (-0.73) (-2.18) (-1.45) (-1.60) (0.61) (-1.97) (0.03) Head Education (Excludes - No Education) Primary -0.065 -0.187 0.437* 0.023 0.319 -0.099 0.065 -0.482** (-0.41) (-1.11) (1.91) (0.25) (1.20) (-0.59) (0.43) (-2.19) Secondary -0.303 -0.209 -0.085 -0.165 0.552 -0.549 -0.042 -0.147 (-0.70) (-0.65) (-0.22) (-0.78) (1.05) (-1.47) (-0.15) (-0.31) Post-Secondary 1.039 1.594 -3.336 1.444 3.379*** 3.465 1.591 5.933*** (0.63) (1.21) (-1.14) (1.00) (2.70) (1.24) (0.91) (12.00) Head Marital Status (Excludes- Never Married Married -0.665 -1.336** -0.883 -0.769* -1.302 -0.250 -1.270* 0.690 (-0.95) (-2.09) (-1.09) (-1.89) (-1.38) (-0.30) (-1.94) (1.05) Other status -0.768 -1.208* -0.612 -0.609 -0.959 -0.698 -1.169* 0.497 (-1.12) (-1.95) (-0.78) (-1.59) (-1.04) (-0.86) (-1.86) (0.77) Ln Head Age -0.761 0.155 0.570 0.241 1.297 -0.043 0.183 -0.457 (-1.11) (0.29) (0.73) (0.70) (1.59) (-0.08) (0.34) (-0.69) Land Rights Document 0.082 -0.174* 0.060 0.056 -0.119 0.096 0.023 0.044 (0.65) (-1.80) (0.44) (0.74) (-0.80) (0.73) (0.28) (0.32) Ln TLUs 0.066 0.071 -0.016 0.077* -0.047 0.346*** 0.026 -0.003 (0.76) (0.95) (-0.15) (1.65) (-0.39) (3.93) (0.39) (-0.02) Ln Adult Equivalent -0.404 0.601** 0.601 -0.210 0.236 -0.305 0.407 -0.167 (-1.22) (2.23) (1.62) (-1.08) (0.57) (-0.94) (1.51) (-0.43) Ln Land Size (ha) 0.018 0.133 0.094 0.006 0.337** 0.065 0.030 -0.058 (0.14) (1.41) (0.66) (0.08) (1.99) (0.48) (0.37) (-0.33) Ln Ag Wage Income -0.008 0.000 -0.012 0.001 0.002 0.004 -0.003 -0.018 (-0.87) (0.04) (-1.11) (0.30) (0.12) (0.39) (-0.46) (-1.57) Ln Transfer Income -0.001 -0.003 -0.007 -0.007 -0.023* 0.001 0.002 0.012 (-0.13) (-0.41) (-0.60) (-1.27) (-1.85) (0.05) (0.29) (1.10) Ln Nonfarm Income 0.012 0.018*** 0.003 0.002 0.011 0.003 -0.002 0.007 (1.51) (2.71) (0.36) (0.51) (1.07) (0.41) (-0.34) (0.73) Observations 10840 10840 10840 10840 10840 10840 10840 6887 Outcomes in Logs 125 Table A 2.21: Effects of Net Agricultural Income per Hectare on School Expenditure and Study Times (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed ef fects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individu al level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Ln Net Revenue per Ha (TSH/Ha) -0.179 0.416 0.988** 0.227 1.771*** 0.264 0.729** 0.315 (-0.52) (1.19) (2.20) (1.09) (3.13) (0.71) (2.02) (0.96) Relationship to Head (Excludes Spouse) Child 0.706 -0.963 2.528 -0.504 4.816** -0.890* 0.917 -0.095 (0.36) (-0.62) (1.36) (-1.19) (2.36) (-1.65) (0.50) (-0.16) Grandchild 0.693 -1.467 2.524 -0.519 4.759** -0.388 0.867 0.000 (0.35) (-0.91) (1.27) (-1.10) (2.22) (-0.54) (0.46) (.) Relative 0.518 -1.783 1.895 -0.611 4.086* -1.120* 0.465 -0.773 (0.26) (-1.10) (0.97) (-1.36) (1.92) (-1.70) (0.25) (-1.16) Non-relative 0.030 -2.150 0.998 -0.997* 3.008 -1.718** -0.291 -1.106 (0.01) (-1.29) (0.48) (-1.91) (1.22) (-2.19) (-0.15) (-1.37) Marital Status (Excludes Never Married) Married 1.157 2.453 1.939 -1.447 6.993*** -1.459** 0.150 -4.555*** (0.81) (1.18) (1.30) (-1.63) (3.32) (-2.36) (0.10) (-12.26) Other status -1.937* -1.476** -1.103 -0.874* -0.183 -2.144*** -1.960** 0.055 (-1.87) (-2.01) (-0.96) (-1.67) (-0.14) (-3.54) (-2.16) (0.07) Female Head 0.196 -0.046 -0.371 -0.163 -0.020 0.247 -0.189 -0.048 (0.56) (-0.15) (-0.91) (-0.91) (-0.04) (0.79) (-0.65) (-0.13) Head Education (Excludes - No Education) Primary -0.108 -0.138 0.557** 0.047 0.534 -0.074 0.152 -0.415* (-0.67) (-0.75) (2.02) (0.49) (1.41) (-0.42) (0.83) (-1.78) Secondary -0.416 -0.079 0.230 -0.102 1.117* -0.483 0.188 0.104 (-0.92) (-0.22) (0.50) (-0.47) (1.69) (-1.27) (0.55) (0.19) Post-Secondary 0.945 1.755 -2.945 1.522 4.081*** 3.547 1.878 6.509*** (0.55) (1.34) (-1.03) (1.05) (3.30) (1.27) (1.08) (7.88) Head Marital Status (Excludes- Never Married Married -0.581 -1.527** -1.348 -0.863** -2.134* -0.347 -1.610** 0.946 (-0.78) (-2.26) (-1.51) (-2.02) (-1.66) (-0.41) (-2.13) (1.17) Other status -0.659 -1.451** -1.203 -0.728* -2.019 -0.822 -1.602** 0.664 (-0.89) (-2.16) (-1.35) (-1.75) (-1.57) (-0.97) (-2.14) (0.86) Ln Head Age -0.929 0.445 1.274 0.382 2.559** 0.104 0.698 -0.946 (-1.22) (0.68) (1.27) (0.94) (2.17) (0.17) (0.95) (-1.03) Land Rights Document 0.011 -0.081 0.286 0.101 0.286 0.144 0.189 0.079 (0.07) (-0.64) (1.59) (1.11) (1.25) (0.93) (1.55) (0.55) Ln TLUs 0.089 -0.000 -0.190 0.042 -0.358** 0.310*** -0.101 -0.071 (0.84) (-0.00) (-1.34) (0.69) (-2.01) (2.77) (-0.98) (-0.49) Ln Adult Equivalent -0.228 0.343 -0.025 -0.336 -0.886 -0.436 -0.051 -0.398 (-0.57) (0.97) (-0.05) (-1.30) (-1.31) (-1.08) (-0.13) (-0.82) Ln Land Size (ha) -0.239 0.605 1.239** 0.237 2.390*** 0.305 0.868** 0.303 (-0.57) (1.46) (2.32) (0.97) (3.48) (0.66) (2.06) (0.66) Ln Ag Wage Income -0.006 -0.006 -0.027** -0.002 -0.027 0.000 -0.015 -0.021* (-0.54) (-0.62) (-2.04) (-0.30) (-1.59) (0.04) (-1.53) (-1.73) Ln Transfer Income 0.001 -0.009 -0.020 -0.009 -0.046*** -0.002 -0.008 0.011 (0.11) (-0.91) (-1.43) (-1.55) (-2.75) (-0.19) (-0.84) (0.99) Ln Nonfarm Income 0.013 0.018** 0.001 0.002 0.008 0.003 -0.003 0.010 (1.62) (2.54) (0.12) (0.42) (0.61) (0.36) (-0.52) (0.97) Observations 10847 10840 10840 10840 10840 10840 10840 6887 Outcomes in Logs 126 Table A 2.22: Effects of Net Agricultural Income on School Expenditure and Study Times (FE) Œ Full Tables TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Err ors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Ln Net Crop Income (Tsh) 0.042 0.020 0.007 0.026 0.026 0.069** 0.007 0.010 (1.49) (0.96) (0.29) (1.14) (0.79) (2.19) (0.54) (0.35) Relationship to Head (Excludes Spouse) Child 0.594 -0.728 3.053* -0.531 5.853*** -1.011** 1.297 1.195 (0.31) (-0.45) (1.85) (-1.28) (3.81) (-2.11) (0.67) (1.52) Grandchild 0.695 -1.295 2.884 -0.579 5.362*** -0.470 1.142 1.041 (0.35) (-0.77) (1.61) (-1.22) (3.23) (-0.70) (0.59) (1.35) Relative 0.463 -1.392 2.588 -0.617 5.424*** -1.209** 1.013 0.307 (0.24) (-0.83) (1.50) (-1.45) (3.40) (-2.14) (0.52) (0.48) Non-relative -0.039 -1.894 1.579 -0.989** 4.076** -1.791** 0.139 0.000 (-0.02) (-1.09) (0.85) (-1.98) (2.23) (-2.53) (0.07) (.) Marital Status (Excludes Never Married) Married 1.200 2.391 1.801 -1.563* 6.818*** -1.603*** 0.034 -5.035*** (0.83) (1.13) (0.94) (-1.73) (3.22) (-2.89) (0.02) (-19.70) Other status -1.866* -1.440** -1.184 -1.222** -0.340 -2.527*** -1.968** 0.004 (-1.85) (-2.08) (-1.14) (-2.10) (-0.29) (-3.68) (-2.39) (0.01) Female Head 0.213 -0.068 -0.539 -0.217 -0.619 0.147 -0.334 -0.118 (0.69) (-0.26) (-1.60) (-1.38) (-1.52) (0.54) (-1.42) (-0.33) Head Education (Excludes - No Education) Primary -0.072 -0.114 0.468** 0.026 0.447* -0.147 0.097 -0.488** (-0.47) (-0.71) (2.10) (0.31) (1.73) (-0.91) (0.68) (-2.28) Secondary -0.351 -0.153 -0.076 -0.136 0.707 -0.580 -0.059 -0.150 (-0.84) (-0.50) (-0.21) (-0.67) (1.41) (-1.51) (-0.22) (-0.33) Post-Secondary 1.001 1.694 -3.307 1.459 3.547*** 3.400 1.620 5.824*** (0.60) (1.27) (-1.14) (1.02) (2.85) (1.22) (0.92) (11.83) Head Marital Status (Excludes- Never Married Married -0.575 -1.287** -1.154 -0.963** -1.212 -0.424 -1.204* 0.665 (-0.83) (-2.13) (-1.50) (-2.19) (-1.35) (-0.56) (-1.93) (1.03) Other status -0.669 -1.166** -0.906 -0.805* -0.883 -0.830 -1.096* 0.447 (-0.99) (-2.00) (-1.21) (-1.92) (-1.01) (-1.12) (-1.83) (0.70) Ln Head Age -0.534 0.021 0.489 0.208 1.100 0.200 0.135 -0.244 (-0.79) (0.04) (0.65) (0.63) (1.38) (0.39) (0.26) (-0.38) Land Rights Document 0.095 -0.220** 0.041 0.056 -0.165 0.170 0.006 0.016 (0.77) (-2.30) (0.31) (0.75) (-1.13) (1.32) (0.08) (0.12) Ln TLUs 0.082 0.078 0.068 0.082* -0.042 0.322*** 0.049 -0.018 (0.99) (1.09) (0.65) (1.88) (-0.36) (3.87) (0.76) (-0.16) Ln Adult Equivalent -0.448 0.625** 0.604* -0.178 0.207 -0.347 0.402 0.039 (-1.38) (2.37) (1.66) (-0.95) (0.51) (-1.09) (1.52) (0.10) Ln Land Size (ha) -0.071 0.104 0.102 -0.033 0.288* 0.006 0.025 -0.095 (-0.58) (1.14) (0.77) (-0.48) (1.76) (0.05) (0.32) (-0.57) Ln Ag Wage Income -0.006 -0.001 -0.012 0.001 0.002 0.007 -0.005 -0.021* (-0.68) (-0.19) (-1.21) (0.29) (0.13) (0.70) (-0.76) (-1.85) Ln Transfer Income -0.001 -0.005 -0.008 -0.007 -0.024* 0.002 0.002 0.010 (-0.06) (-0.62) (-0.79) (-1.40) (-1.94) (0.22) (0.36) (0.94) Ln Nonfarm Income 0.012 0.020*** 0.003 0.002 0.010 0.001 -0.001 0.003 (1.44) (3.00) (0.35) (0.42) (0.94) (0.14) (-0.22) (0.29) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 127 Table A 2.23: Effects of Net Agricultural Income on School Expenditure and Study Times (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous co mpleted academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Ln Net Crop Income (Tsh) -0.247 0.365 1.071** 0.234 1.772*** 0.305 0.705* 0.426 (-0.65) (1.02) (2.29) (1.10) (3.02) (0.81) (1.91) (1.32) Relationship to Head (Excludes Spouse) Child 0.800 -0.973 2.299 -0.678 4.615** -1.178* 0.803 -0.156 (0.39) (-0.65) (1.30) (-1.48) (2.07) (-1.95) (0.46) (-0.25) Grandchild 0.854 -1.485 2.300 -0.694 4.404* -0.599 0.760 0.000 (0.41) (-0.95) (1.21) (-1.37) (1.90) (-0.78) (0.42) (.) Relative 0.742 -1.726 1.560 -0.819 3.736 -1.437* 0.339 -0.795 (0.36) (-1.09) (0.83) (-1.63) (1.60) (-1.95) (0.19) (-1.13) Non-relative 0.161 -2.133 0.843 -1.133** 2.868 -1.953** -0.344 -0.971 (0.08) (-1.32) (0.42) (-2.07) (1.09) (-2.38) (-0.18) (-1.16) Marital Status (Excludes Never Married) Married 1.190 2.403 1.838 -1.556* 6.878*** -1.595** 0.058 -5.100*** (0.81) (1.16) (1.26) (-1.75) (3.07) (-2.47) (0.04) (-18.39) Other status -1.875* -1.429** -1.150 -1.215** -0.283 -2.520*** -1.946** 0.002 (-1.88) (-2.03) (-1.04) (-2.05) (-0.23) (-3.77) (-2.25) (0.00) Female Head 0.070 0.103 -0.011 -0.114 0.248 0.263 0.013 -0.207 (0.19) (0.31) (-0.02) (-0.60) (0.42) (0.80) (0.04) (-0.53) Head Education (Excludes - No Education) Primary -0.100 -0.081 0.571** 0.046 0.616 -0.124 0.165 -0.418* (-0.62) (-0.47) (2.07) (0.51) (1.63) (-0.73) (0.93) (-1.88) Secondary -0.464 -0.019 0.339 -0.055 1.387** -0.488 0.212 0.255 (-1.02) (-0.05) (0.72) (-0.26) (2.06) (-1.21) (0.61) (0.43) Post-Secondary 0.932 1.776 -3.054 1.508 3.964*** 3.456 1.787 6.808*** (0.55) (1.35) (-1.07) (1.05) (3.30) (1.25) (1.03) (7.55) Head Marital Status (Excludes- Never Married Married -0.392 -1.506** -1.828* -1.095** -2.318* -0.573 -1.646** 0.983 (-0.49) (-2.23) (-1.84) (-2.32) (-1.66) (-0.70) (-2.11) (1.09) Other status -0.451 -1.427** -1.709* -0.962** -2.201 -1.008 -1.622** 0.635 (-0.57) (-2.11) (-1.71) (-2.06) (-1.57) (-1.24) (-2.09) (0.73) Ln Head Age -0.784 0.320 1.410 0.388 2.611** 0.404 0.739 -0.955 (-0.99) (0.47) (1.30) (0.92) (2.04) (0.64) (0.95) (-1.04) Land Rights Document 0.027 -0.138 0.291 0.105 0.245 0.225 0.170 0.071 (0.18) (-1.08) (1.57) (1.14) (1.05) (1.46) (1.37) (0.49) Ln TLUs 0.122 0.029 -0.081 0.053 -0.287* 0.289*** -0.049 -0.113 (1.23) (0.32) (-0.59) (0.95) (-1.68) (2.85) (-0.51) (-0.84) Ln Adult Equivalent -0.238 0.374 -0.167 -0.329 -1.058 -0.518 -0.103 -0.354 (-0.54) (1.00) (-0.31) (-1.22) (-1.46) (-1.23) (-0.25) (-0.69) Ln Land Size (ha) 0.015 0.002 -0.213 -0.095 -0.230 -0.064 -0.182 -0.126 (0.09) (0.01) (-0.99) (-0.95) (-0.81) (-0.40) (-1.22) (-0.72) Ln Ag Wage Income -0.002 -0.007 -0.030** -0.002 -0.027 0.003 -0.017* -0.023** (-0.14) (-0.71) (-2.20) (-0.37) (-1.62) (0.24) (-1.72) (-1.99) Ln Transfer Income 0.004 -0.010 -0.024* -0.010* -0.050*** -0.001 -0.008 0.008 (0.33) (-1.02) (-1.69) (-1.68) (-2.83) (-0.11) (-0.85) (0.73) Ln Nonfarm Income 0.012 0.019*** 0.000 0.001 0.006 0.001 -0.003 0.008 (1.51) (2.81) (0.04) (0.31) (0.43) (0.07) (-0.47) (0.73) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 128 Table A 2.24: Effects of Land Productivity on School Expenditure and Study Times Œ Excluding Other Income (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are current ly in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) 0.060*** 0.001 0.026 0.018 -0.027 0.020 0.021 -0.007 (3.28) (0.08) (1.25) (1.40) (-1.16) (1.10) (1.51) (-0.36) Relationship to Head (Excludes Spouse) Child 0.493 -0.806 3.020* -0.517 5.923*** -0.987** 1.271 1.214 (0.26) (-0.50) (1.83) (-1.29) (3.89) (-2.11) (0.66) (1.57) Grandchild 0.560 -1.377 2.856 -0.580 5.446*** -0.480 1.110 1.154 (0.28) (-0.82) (1.59) (-1.24) (3.31) (-0.72) (0.57) (1.50) Relative 0.359 -1.441 2.549 -0.607 5.517*** -1.177** 0.979 0.330 (0.18) (-0.86) (1.48) (-1.47) (3.49) (-2.13) (0.50) (0.52) Non-relative -0.100 -1.936 1.555 -0.971** 4.140** -1.757** 0.118 0.000 (-0.05) (-1.13) (0.83) (-1.98) (2.29) (-2.51) (0.06) (.) Marital Status (Excludes Never Married) Married 1.284 2.329 1.877 -1.525* 6.768*** -1.587*** 0.082 -5.086*** (0.91) (1.10) (1.00) (-1.75) (3.16) (-2.94) (0.05) (-15.49) Other status -1.874* -1.488** -1.179 -1.205** -0.337 -2.519*** -1.965** -0.046 (-1.86) (-2.13) (-1.13) (-2.10) (-0.29) (-3.66) (-2.38) (-0.06) Female Head 0.237 -0.082 -0.520 -0.214 -0.649 0.127 -0.321 -0.129 (0.77) (-0.31) (-1.54) (-1.39) (-1.60) (0.47) (-1.36) (-0.36) Head Education (Excludes - No Education) Primary -0.070 -0.115 0.476** 0.025 0.446* -0.157 0.100 -0.466** (-0.46) (-0.71) (2.14) (0.29) (1.73) (-0.97) (0.69) (-2.18) Secondary -0.358 -0.140 -0.086 -0.154 0.685 -0.604 -0.062 -0.170 (-0.86) (-0.45) (-0.24) (-0.76) (1.36) (-1.57) (-0.23) (-0.37) Post-Secondary 1.012 1.767 -3.313 1.449 3.579*** 3.381 1.607 5.716*** (0.61) (1.33) (-1.14) (1.00) (2.86) (1.21) (0.92) (11.93) Head Marital Status (Excludes- Never Married Married -0.595 -1.302** -1.170 -0.966** -1.211 -0.394 -1.207* 0.697 (-0.85) (-2.13) (-1.51) (-2.17) (-1.34) (-0.52) (-1.94) (1.09) Other status -0.694 -1.178** -0.947 -0.816* -0.905 -0.784 -1.102* 0.486 (-1.02) (-2.00) (-1.26) (-1.92) (-1.03) (-1.06) (-1.84) (0.77) Ln Head Age -0.512 0.001 0.489 0.197 1.024 0.173 0.148 -0.312 (-0.76) (0.00) (0.65) (0.60) (1.29) (0.33) (0.29) (-0.49) Land Rights Document 0.108 -0.222** 0.045 0.051 -0.193 0.162 0.013 0.023 (0.88) (-2.33) (0.34) (0.69) (-1.32) (1.26) (0.17) (0.17) Ln TLUs 0.048 0.079 0.048 0.072 -0.027 0.319*** 0.037 -0.011 (0.58) (1.11) (0.46) (1.63) (-0.23) (3.81) (0.57) (-0.10) Ln Adult Equivalent -0.414 0.676*** 0.614* -0.156 0.266 -0.304 0.398 0.019 (-1.28) (2.59) (1.70) (-0.83) (0.66) (-0.97) (1.51) (0.05) Ln Land Size (ha) -0.022 0.119 0.113 -0.014 0.283* 0.044 0.036 -0.114 (-0.18) (1.31) (0.84) (-0.21) (1.74) (0.34) (0.46) (-0.67) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 129 Table A 2.25: Effects of Labor Productivity on School Expenditure and Study Times Œ Excludin g Other Income (FE) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sampl e restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) 0.055* 0.012 0.007 0.024 -0.024 0.040 0.026 -0.013 (1.88) (0.51) (0.22) (1.14) (-0.66) (1.37) (1.34) (-0.36) Relationship to Head (Excludes Spouse) Child 0.476 -0.822 3.038* -0.534 5.928*** -1.022** 1.255 1.223 (0.25) (-0.51) (1.84) (-1.32) (3.91) (-2.15) (0.65) (1.58) Grandchild 0.561 -1.396 2.886 -0.594 5.443*** -0.516 1.098 1.159 (0.28) (-0.84) (1.61) (-1.26) (3.32) (-0.77) (0.57) (1.51) Relative 0.355 -1.461 2.578 -0.621 5.516*** -1.214** 0.966 0.337 (0.18) (-0.88) (1.49) (-1.49) (3.50) (-2.17) (0.49) (0.53) Non-relative -0.118 -1.949 1.567 -0.986** 4.146** -1.787** 0.104 0.000 (-0.06) (-1.14) (0.84) (-2.00) (2.30) (-2.53) (0.05) (.) Marital Status (Excludes Never Married) Married 1.183 2.331 1.827 -1.553* 6.814*** -1.614*** 0.050 -5.069*** (0.82) (1.10) (0.97) (-1.76) (3.18) (-2.91) (0.03) (-17.12) Other status -1.915* -1.491** -1.194 -1.218** -0.319 -2.537*** -1.980** -0.040 (-1.90) (-2.14) (-1.15) (-2.10) (-0.27) (-3.68) (-2.41) (-0.06) Female Head 0.225 -0.075 -0.536 -0.213 -0.642 0.136 -0.321 -0.130 (0.73) (-0.28) (-1.59) (-1.38) (-1.58) (0.50) (-1.36) (-0.36) Head Education (Excludes - No Education) Primary -0.067 -0.113 0.475** 0.026 0.445* -0.154 0.101 -0.470** (-0.44) (-0.70) (2.13) (0.31) (1.72) (-0.95) (0.70) (-2.20) Secondary -0.347 -0.139 -0.082 -0.150 0.680 -0.599 -0.058 -0.171 (-0.83) (-0.45) (-0.23) (-0.74) (1.35) (-1.55) (-0.22) (-0.37) Post-Secondary 1.051 1.770 -3.300 1.463 3.562*** 3.400 1.622 5.699*** (0.63) (1.33) (-1.14) (1.01) (2.83) (1.21) (0.93) (11.86) Head Marital Status (Excludes- Never Married Married -0.557 -1.300** -1.155 -0.954** -1.228 -0.380 -1.193* 0.691 (-0.80) (-2.13) (-1.49) (-2.16) (-1.36) (-0.50) (-1.92) (1.08) Other status -0.651 -1.178** -0.927 -0.803* -0.924 -0.770 -1.087* 0.480 (-0.96) (-2.00) (-1.23) (-1.90) (-1.05) (-1.04) (-1.82) (0.76) Ln Head Age -0.508 0.014 0.470 0.208 1.024 0.199 0.158 -0.307 (-0.75) (0.03) (0.63) (0.63) (1.29) (0.38) (0.31) (-0.48) Land Rights Document 0.103 -0.219** 0.038 0.051 -0.191 0.166 0.013 0.022 (0.83) (-2.29) (0.29) (0.69) (-1.30) (1.28) (0.17) (0.17) Ln TLUs 0.079 0.078 0.065 0.080* -0.041 0.326*** 0.046 -0.014 (0.95) (1.10) (0.62) (1.81) (-0.35) (3.91) (0.72) (-0.12) Ln Adult Equivalent -0.401 0.676*** 0.621* -0.153 0.260 -0.301 0.402 0.020 (-1.24) (2.59) (1.72) (-0.81) (0.65) (-0.96) (1.52) (0.05) Ln Land Size (ha) -0.070 0.115 0.096 -0.031 0.304* 0.022 0.018 -0.109 (-0.58) (1.27) (0.72) (-0.44) (1.87) (0.17) (0.23) (-0.64) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 130 Table A 2.26: Effects of Land Productivity on School Expenditure and Study Times Œ Excluding Other Income (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricu ltural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.190 0.364 0.905** 0.205 1.539*** 0.281 0.597* 0.307 (-0.58) (1.16) (2.19) (1.09) (2.66) (0.85) (1.86) (1.14) Relationship to Head (Excludes Spouse) Child 0.756 -1.188 2.095 -0.715* 4.274* -1.260* 0.663 1.260 (0.37) (-0.78) (1.02) (-1.69) (1.70) (-1.77) (0.37) (1.57) Grandchild 0.960 -1.959 1.447 -0.881 2.936 -0.897 0.186 0.994 (0.45) (-1.19) (0.63) (-1.61) (1.07) (-0.95) (0.10) (1.21) Relative 0.746 -2.005 1.184 -0.898* 3.085 -1.581* 0.084 0.198 (0.35) (-1.24) (0.54) (-1.81) (1.17) (-1.81) (0.04) (0.29) Non-relative 0.085 -2.205 0.903 -1.111** 2.978 -1.950** -0.310 0.000 (0.04) (-1.38) (0.40) (-2.21) (1.08) (-2.25) (-0.16) (.) Marital Status (Excludes Never Married) Married 0.778 3.065 3.658* -1.145* 9.941** -1.060 1.251 -1.982 (0.43) (1.17) (1.66) (-1.76) (2.41) (-0.95) (0.62) (-0.75) Other status -2.005** -1.297* -0.717 -1.106** 0.485 -2.382*** -1.662* -0.090 (-1.96) (-1.73) (-0.63) (-1.99) (0.34) (-3.80) (-1.85) (-0.12) Female Head 0.036 0.211 0.190 -0.063 0.615 0.337 0.144 0.096 (0.09) (0.54) (0.36) (-0.30) (0.79) (0.88) (0.38) (0.22) Head Education (Excludes - No Education) Primary -0.082 -0.097 0.519** 0.034 0.523 -0.144 0.128 -0.664** (-0.52) (-0.56) (1.98) (0.38) (1.41) (-0.86) (0.74) (-2.25) Secondary -0.330 -0.180 -0.183 -0.174 0.512 -0.633 -0.126 -0.287 (-0.76) (-0.54) (-0.42) (-0.79) (0.73) (-1.64) (-0.42) (-0.59) Post-Secondary 1.118 1.611 -3.688 1.369 2.910* 3.270 1.360 5.566*** (0.69) (1.16) (-1.22) (0.92) (1.78) (1.13) (0.73) (10.90) Head Marital Status (Excludes- Never Married Married -0.448 -1.515** -1.687* -1.076** -2.131* -0.547 -1.546** 0.918 (-0.59) (-2.31) (-1.78) (-2.23) (-1.80) (-0.69) (-2.23) (1.36) Other status -0.503 -1.455** -1.616* -0.959** -2.098* -0.982 -1.541** 0.672 (-0.67) (-2.22) (-1.69) (-2.00) (-1.74) (-1.24) (-2.23) (1.02) Ln Head Age -0.769 0.375 1.393 0.390 2.637** 0.440 0.742 -0.079 (-0.99) (0.55) (1.33) (0.92) (1.98) (0.71) (1.00) (-0.10) Land Rights Document 0.027 -0.104 0.330* 0.112 0.315 0.246 0.201 0.054 (0.17) (-0.73) (1.68) (1.13) (1.16) (1.47) (1.48) (0.38) Ln TLUs 0.215 -0.164 -0.539* -0.054 -1.075** 0.145 -0.349 -0.267 (0.92) (-0.71) (-1.72) (-0.39) (-2.41) (0.61) (-1.46) (-1.08) Ln Adult Equivalent -0.343 0.572* 0.362 -0.210 -0.182 -0.378 0.233 -0.101 (-0.99) (1.95) (0.83) (-0.99) (-0.31) (-1.15) (0.72) (-0.24) Ln Land Size (ha) -0.166 0.329 0.622** 0.094 1.189*** 0.194 0.370* 0.169 (-0.72) (1.60) (2.21) (0.77) (2.88) (0.81) (1.79) (0.56) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 131 Table A 2.27: Effects of Labor Productivity on School Expenditure and Study Times Œ Excluding Other Income (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.384 0.737 1.829** 0.415 3.112** 0.567 1.207* 0.946 (-0.57) (1.13) (2.01) (1.07) (2.30) (0.84) (1.75) (1.05) Relationship to Head (Excludes Spouse) Child 1.108 -1.864 0.417 -1.096 1.420 -1.781 -0.444 0.612 (0.48) (-1.15) (0.13) (-1.37) (0.30) (-1.35) (-0.20) (0.62) Grandchild 1.317 -2.642 -0.249 -1.266 0.051 -1.423 -0.933 0.469 (0.54) (-1.45) (-0.07) (-1.37) (0.01) (-0.92) (-0.39) (0.43) Relative 1.119 -2.719 -0.589 -1.301 0.069 -2.131 -1.086 -0.422 (0.46) (-1.51) (-0.18) (-1.44) (0.01) (-1.41) (-0.46) (-0.43) Non-relative 0.376 -2.762* -0.481 -1.425* 0.623 -2.379* -1.223 0.000 (0.16) (-1.72) (-0.15) (-1.82) (0.13) (-1.82) (-0.55) (.) Marital Status (Excludes Never Married) Married 1.028 2.586 2.468 -1.415** 7.917* -1.429 0.466 -0.836 (0.62) (1.00) (1.18) (-2.21) (1.65) (-1.21) (0.23) (-0.21) Other status -1.838* -1.618** -1.512 -1.286** -0.868 -2.629*** -2.187*** -0.548 (-1.79) (-2.36) (-1.38) (-2.09) (-0.57) (-4.03) (-2.66) (-0.56) Female Head -0.057 0.389 0.631 0.037 1.365 0.474 0.436 0.388 (-0.11) (0.75) (0.86) (0.13) (1.23) (0.93) (0.81) (0.61) Head Education (Excludes - No Education) Primary -0.110 -0.043 0.653** 0.065 0.752* -0.102 0.217 -0.577* (-0.65) (-0.22) (2.21) (0.66) (1.70) (-0.56) (1.07) (-1.94) Secondary -0.381 -0.083 0.058 -0.120 0.921 -0.558 0.033 -0.303 (-0.87) (-0.24) (0.11) (-0.55) (1.14) (-1.34) (0.09) (-0.55) Post-Secondary 0.943 1.948 -2.852 1.559 4.333*** 3.530 1.912 6.712*** (0.57) (1.41) (-0.93) (1.05) (2.79) (1.26) (1.04) (6.33) Head Marital Status (Excludes- Never Married Married -0.580 -1.263* -1.061 -0.934** -1.068 -0.353 -1.133 1.542 (-0.75) (-1.86) (-1.02) (-2.06) (-0.64) (-0.47) (-1.43) (1.22) Other status -0.633 -1.206* -0.999 -0.819* -1.048 -0.791 -1.134 1.289 (-0.83) (-1.82) (-0.98) (-1.86) (-0.63) (-1.08) (-1.47) (1.04) Ln Head Age -1.030 0.874 2.633* 0.671 4.745** 0.825 1.560 -0.294 (-0.96) (0.85) (1.69) (1.08) (2.19) (0.85) (1.37) (-0.32) Land Rights Document -0.010 -0.034 0.504* 0.151 0.611 0.300 0.315 0.094 (-0.05) (-0.17) (1.81) (1.17) (1.46) (1.40) (1.57) (0.56) Ln TLUs 0.154 -0.046 -0.248 0.013 -0.580* 0.236 -0.157 -0.233 (1.09) (-0.32) (-1.12) (0.15) (-1.76) (1.59) (-0.98) (-0.98) Ln Adult Equivalent -0.368 0.622** 0.485 -0.182 0.028 -0.340 0.314 -0.276 (-1.09) (2.11) (1.03) (-0.88) (0.04) (-1.04) (0.93) (-0.52) Ln Land Size (ha) 0.041 -0.067 -0.363 -0.129 -0.486 -0.111 -0.280 0.004 (0.19) (-0.34) (-1.20) (-1.03) (-1.06) (-0.53) (-1.29) (0.02) Observations 11054 11054 11054 11054 11054 11054 11054 7050 Outcomes in Logs 132 Table A 2.28: Effects of Land Productivity on School Expenditure and Study Times Œ Rainfall Deviations as Instrument (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes indiv idual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academ ic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Land Productivity (TSH/Ha) -0.563 0.453* 0.994** -0.011 0.854* 0.219 0.227 0.490** (-1.58) (1.74) (2.49) (-0.06) (1.93) (0.68) (1.10) (2.37) Relationship to Head (Excludes Spouse) Child 1.166 -1.046 1.914 -0.545 4.664** -1.333* 1.113 0.502 (0.51) (-0.68) (0.94) (-1.12) (2.49) (-1.93) (0.61) (0.81) Grandchild 1.555 -1.836 1.055 -0.619 3.550* -0.911 0.856 0.000 (0.65) (-1.12) (0.46) (-1.03) (1.71) (-0.96) (0.46) (.) Relative 1.334 -1.779 0.868 -0.636 3.660* -1.648* 0.759 -0.562 (0.56) (-1.10) (0.40) (-1.15) (1.84) (-1.89) (0.41) (-0.77) Non-relative 0.351 -2.027 0.611 -1.016* 2.921 -2.100** 0.020 -0.687 (0.15) (-1.23) (0.27) (-1.83) (1.35) (-2.42) (0.01) (-0.81) Marital Status (Excludes Never Married) Married 0.140 3.259 3.912 -1.540 8.463*** -1.222 0.455 0.616 (0.06) (1.12) (1.63) (-1.54) (2.80) (-1.21) (0.25) (0.29) Other status -2.151** -1.179 -0.607 -1.222** 0.107 -2.396*** -1.831** 0.102 (-2.04) (-1.54) (-0.52) (-2.07) (0.08) (-3.79) (-2.18) (0.13) Female Head -0.279 0.248 0.278 -0.254 0.096 0.342 -0.189 0.285 (-0.57) (0.69) (0.51) (-1.11) (0.15) (0.88) (-0.65) (0.58) Head Education (Excludes - No Education) Primary -0.137 -0.104 0.581** 0.016 0.482 -0.158 0.082 -0.806*** (-0.73) (-0.58) (2.16) (0.18) (1.58) (-0.94) (0.57) (-2.74) Secondary -0.340 -0.247 -0.178 -0.148 0.651 -0.646* -0.145 -0.310 (-0.72) (-0.72) (-0.40) (-0.73) (1.12) (-1.70) (-0.56) (-0.63) Post-Secondary 1.241 1.776 -3.594 1.417 3.091** 3.240 1.607 5.531*** (0.74) (1.12) (-1.18) (1.00) (2.03) (1.10) (0.84) (11.29) Head Marital Status (Excludes- Never Married Married -0.202 -1.484** -1.794* -0.937** -1.719* -0.515 -1.314** 0.912 (-0.24) (-2.29) (-1.83) (-2.03) (-1.75) (-0.65) (-2.06) (1.34) Other status -0.207 -1.434** -1.721* -0.776* -1.538 -0.972 -1.253** 0.725 (-0.25) (-2.23) (-1.74) (-1.70) (-1.55) (-1.23) (-2.01) (1.08) Ln Head Age -1.278 0.416 1.862* 0.205 2.341** 0.366 0.248 0.159 (-1.40) (0.64) (1.78) (0.47) (2.18) (0.55) (0.46) (0.18) Land Rights Document -0.116 -0.056 0.363* 0.046 0.067 0.216 0.075 0.081 (-0.68) (-0.43) (1.85) (0.50) (0.31) (1.33) (0.77) (0.55) Ln TLUs 0.379 -0.213 -0.649** 0.100 -0.552* 0.181 -0.116 -0.350* (1.57) (-1.12) (-2.22) (0.79) (-1.74) (0.83) (-0.78) (-1.92) Ln Adult Equivalent -0.161 0.560* 0.288 -0.170 -0.061 -0.312 0.364 -0.122 (-0.42) (1.84) (0.64) (-0.83) (-0.13) (-0.93) (1.30) (-0.27) Ln Land Size (ha) -0.363 0.368** 0.710** -0.035 0.781** 0.165 0.166 0.274 (-1.43) (2.02) (2.53) (-0.28) (2.43) (0.70) (1.20) (1.04) Ln Ag Wage Income -0.004 -0.007 -0.019 0.002 -0.005 0.004 -0.009 -0.021* (-0.35) (-0.77) (-1.56) (0.36) (-0.38) (0.45) (-1.29) (-1.66) Ln Transfer Income 0.004 -0.005 -0.014 -0.008 -0.026* 0.005 0.002 0.015 (0.32) (-0.65) (-1.06) (-1.43) (-1.85) (0.43) (0.25) (1.22) Ln Nonfarm Income 0.015* 0.017** 0.001 0.002 0.006 0.000 -0.003 -0.001 (1.73) (2.38) (0.07) (0.48) (0.51) (0.05) (-0.48) (-0.14) Observations 10957 10957 10957 10957 10957 10957 10957 6979 Outcomes in Logs 133 Table A 2.29: Effects of Labor Productivity on School Expenditure and Study Times Œ Rainfall Deviations as Instrument (IV) Œ Full Table TŒstatistics in parentheses. *p<0.1 ** p<0.05 *** p<0.01. Includes individual, wave, and interview Œmonth fixed effects . The unit of analysis is individual. Productivity/agricultural income is at the household level. Sample restricted to children who either are currently in school or were in school the previous completed academic year. Errors clustered at the individual level. (1) (2) (3) (4) (5) (6) (7) (8) Fees Books Uniform Transport Contributions Food Total Expenses Minutes Studying Log Labor Productivity (TSH/Day) -0.497 0.130 2.098** 0.250 1.278 1.146 0.596 1.080 (-0.70) (0.25) (2.23) (0.64) (1.38) (1.50) (1.28) (1.60) Relationship to Head (Excludes Spouse) Child 1.242 -0.729 0.067 -0.908 3.813 -2.699 0.527 0.163 (0.52) (-0.41) (0.02) (-1.26) (1.42) (-1.43) (0.28) (0.24) Grandchild 1.447 -1.286 -0.849 -1.066 2.813 -2.504 0.221 0.000 (0.57) (-0.68) (-0.23) (-1.24) (0.96) (-1.18) (0.11) (.) Relative 1.243 -1.239 -1.085 -1.088 2.889 -3.262 0.110 -0.710 (0.49) (-0.65) (-0.30) (-1.31) (1.01) (-1.57) (0.05) (-0.81) Non-relative 0.428 -1.764 -0.994 -1.327* 2.173 -3.276* -0.487 -0.231 (0.18) (-0.97) (-0.29) (-1.83) (0.77) (-1.70) (-0.24) (-0.23) Marital Status (Excludes Never Married) Married 1.083 2.432 2.557 -1.456** 7.167*** -1.346 0.175 0.813 (0.64) (1.02) (1.04) (-2.04) (2.62) (-0.68) (0.10) (0.25) Other status -1.765* -1.443** -1.501 -1.259** -0.570 -2.711** -2.056** -0.388 (-1.71) (-2.07) (-1.30) (-2.10) (-0.47) (-4.15) (-2.55) (-0.41) Female Head -0.139 -0.046 0.854 -0.077 0.237 0.930 0.021 0.480 (-0.23) (-0.11) (1.06) (-0.24) (0.29) (1.52) (0.05) (0.84) Head Education (Excludes - No Education) Primary -0.160 -0.123 0.793** 0.052 0.590* -0.016 0.147 -0.592** (-0.83) (-0.70) (2.42) (0.49) (1.80) (-0.07) (0.89) (-2.00) Secondary -0.445 -0.193 0.143 -0.121 0.868 -0.499 -0.058 -0.242 (-1.01) (-0.62) (0.27) (-0.58) (1.48) (-1.08) (-0.20) (-0.43) Post-Secondary 0.841 2.026 -2.564 1.478 3.837*** 3.648 1.873 6.766*** (0.48) (1.31) (-0.83) (1.03) (2.59) (1.26) (0.98) (8.25) Head Marital Status (Excludes- Never Married Married -0.598 -1.179** -1.036 -0.932** -1.094 -0.316 -1.136* 1.506 (-0.75) (-1.97) (-0.92) (-2.14) (-1.04) (-0.38) (-1.75) (1.28) Other status -0.658 -1.064* -0.958 -0.791* -0.869 -0.823 -1.082* 1.358 (-0.84) (-1.84) (-0.87) (-1.89) (-0.84) (-1.03) (-1.72) (1.16) Ln Head Age -1.260 0.040 3.464** 0.553 3.016* 1.635 0.771 -0.070 (-1.03) (0.05) (1.99) (0.82) (1.95) (1.30) (0.94) (-0.07) Land Rights Document -0.068 -0.163 0.586** 0.113 0.126 0.438* 0.156 0.113 (-0.31) (-1.02) (2.00) (0.91) (0.44) (1.82) (1.12) (0.69) Ln TLUs 0.132 0.037 -0.448* 0.045 -0.278 0.093 -0.092 -0.313 (0.81) (0.29) (-1.78) (0.48) (-1.18) (0.49) (-0.76) (-1.45) Ln Adult Equivalent -0.317 0.700*** 0.499 -0.187 0.148 -0.302 0.406 -0.330 (-0.92) (2.63) (0.99) (-0.94) (0.32) (-0.83) (1.43) (-0.60) Ln Land Size (ha) 0.086 0.074 -0.389 -0.091 -0.032 -0.247 -0.114 -0.028 (0.39) (0.47) (-1.23) (-0.74) (-0.11) (-1.04) (-0.76) (-0.14) Ln Ag Wage Income -0.008 -0.004 -0.008 0.002 0.002 0.009 -0.006 -0.011 (-0.80) (-0.51) (-0.61) (0.45) (0.18) (0.79) (-0.87) (-0.76) Ln Transfer Income 0.003 -0.003 -0.022 -0.010 -0.030* -0.002 -0.001 0.008 (0.28) (-0.37) (-1.33) (-1.59) (-1.91) (-0.18) (-0.13) (0.61) Ln Nonfarm Income 0.013 0.019*** 0.006 0.002 0.010 0.002 -0.001 0.012 (1.55) (2.75) (0.47) (0.51) (0.83) (0.22) (-0.23) (0.92) Observations 10957 10957 10957 10957 10957 10957 10957 6979 Outcomes in Logs 134 Table A 2.30: First Stage Regressions Œ Productivity and Rainfall Deviations TŒstatistics in parentheses. * p<0.1 ** p<0.05 *** p<0.01. Includes individual, and wave FE. Rainfall deviations are defined as normalized quarterly deviations from a decade mean (2007 Œ2017) (1) (2) Ln Land Productivity (Tsh/Ha) Ln Labor Productivity (Tsh/Day) Q1 Rainfall deviation (Jan-Mar) 0.133*** 0.064** (3.12) (2.31) Q2 Rainfall deviation (Apr-Jun) 0.178*** 0.073** (3.31) (2.31) Q3 Rainfall deviation (Jul-Sep) -0.028 -0.019 (0.45) (0.40) Q4 Rainfall deviation (Oct-Dec) -0.054 -0.012 (0.97) (0.29) Observations 11054 11054 F-Stat 10.616.79135 BIBLIOGRAPHY 136 BIBLIOGRAPHY Beegle, K., Dehejia, R. H., & Gatti, R. ( 2006). Child labor and agricultural shocks. Journal of Development economics , 81(1), 80 Œ96. Boozer, M., & Suri, T. (2001). Child labor and schooling decisions in Ghana. Manuscript, Yale University . Cameron, L. A., & Worswick, C. (2001). Education expenditu re responses to crop loss in Indonesia: A gender bias. Economic development and cultural change , 49(2), 351 Œ363. Dammert, A. C. (2010). Siblings, child labor, and schooling in Nicaragua and Guatemala. Journal of Population Economics , 23(1), 199 Œ224. Deinin ger, K. (2003). Does cost of schooling affect enrollment by the poor? Universal primary education in Uganda. Economics of Education review , 22(3), 291 Œ305. Delesalle, E. (2019). Good Or Bad Timing?: The Effects of Productivity Shocks on Education and on Schooling Performance (No. UCL ŒUniversité Catholique de Louvain). Institut de recherche économiques et sociales, UC Louvain. Rosales ŒRueda (2018), allocation in South Africa. The World Bank Economic Review , 17(1), 1 Œ25. Duryea, S., & Arends ŒKuenning, M. (2003). School attendance, child labor and local labor market fluctuations in urban Brazil. World Development , 31(7), 1165 Œ1178. Edmonds, E. V. (2006). Child labor and schooling responses to anticipated income in South Africa. Journal of development Economics , 81(2), 386 Œ414. Grimm, M. (2011). Does household income m atter for children™s schooling? Evidence for rural Sub ŒSaharan Africa. Economics of Education Review , 30(4), 740 Œ754. Jacoby, H. G., & Skoufias, E. (1997). Risk, financial markets, and human capital in a developing country. The Review of Economic Studies , 64(3), 311 Œ335. Jensen, R. (2000). Agricultural volatility and investments in children. American Economic Review , 90(2), 399 Œ404. Kruger, D. I. (2007). Coffee production effects on child labor and schooling in rural Brazil. Journal of development Economics , 82(2), 448 Œ463. Maccini, S., & Yang, D. (2009). Under the weather: Health, schooling, and economic consequences of early Œlife rainfall. American Economic Review , 99(3), 1006 Œ26. Marchetta, F., Sahn, D. E., & Tiberti, L. (2019). The role of weather on sch ooling and work of young adults in Madagascar. American Journal of Agricultural Economics , 101(4), 1203 Œ1227. Qian, N. (2008). Missing women and the price of tea in China: The effect of sex Œspecific earnings on sex imbalance. The Quarterly Journal of Econo mics , 123(3), 1251 Œ1285. 137 Reggio, I. (2011). The influence of the mother's power on her child's labor in Mexico. Journal of development economics , 96(1), 95 Œ105. Saenz, M., & Thompson, E. (2017). Gender and policy roles in farm household diversification in Zambia. World Development , 89, 152Œ169. Shah, M., & Steinberg, B. M. (2017). Drought of opportunities: Contemporaneous and long Œterm impacts of rainfall shocks on human capital. Journal of Political Economy , 125(2), 527 Œ561. Thomas, D. (1990). Intra Œhouse hold resource allocation: An inferential approach. Journal of human resources , 635Œ664. Thomas, D. (1994). Like father, like son; like mother, like daughter: Parental resources and child height. Journal of human resources , 950Œ988. World Bank (2007) Malawi Poverty and Vulnerability Assessment: Investing in Our Future . Washington, D.C. https://openknowledge.worldbank.org/handle/10986/7909 World Bank (2020). https://databank.worldbank.org/source/world Œdevelopment Œindicators Wiig, H. (2013). Joint titling in rural Peru: Impact on women™s participation in household decision Œmaking. World Development , 52, 104Œ119 . 138 Chapter 3 : School Electrification and Academic Outcomes in Rural Kenya 139 I. INTRODUCTION 2 A number of developing countries are embarking on extensive rural electrification projects in an effort to improve household incomes and welfare . Electrification can affect in comes directly and indirectly. Electricity expands the set of possibl e income generating activities and provides light for extended working hour s, and consequently increase s income (Khandker et al ., 2009b) Œ this is an example of a direct mechanism . Indire ct mechanisms may include positive electrification effect s on education and health, which in turn can lead to higher incomes . Specifically, i mproved and cheaper electric lighting increases time available for studying and hence can improve educational outco mes. There is a general consensus that education is a crucial investment in human capital and thus has long run impacts on labor market outcomes. There is also evidence of excessive indoor household pol lution resulting from excessive use of kerosene or fi rewood lighting which emit harmful gases and soot (Baron and Torero , 2017; Lam et al ., 2012(b); Bates et al ., 2013). Pollution is likely to affect both short Œterm educational attainment due to sickness and affect long Œterm health outcomes (Lim et al ., 2012 ). Baron and Torero (2017) finds evidence that grid electrification significantly reduces indoor po llution in El Salvador . Electrification can therefore improve health outcomes by reducing respiratory and pollution Œrelated illnesses. Bernard (2012) argues there has been little impact evaluation to study the effects of rural electrification on the sectors, such as education, that are generally used to justify the funding of electrification projects in Sub ŒSaharan Africa. This paper zeroes in on education an d attempts to quantify the effect s of school electrification on educational outcomes Œtest scores, enrollment , and completion Œamong 8th grade primary school students. In Kenya, schoo ls close to the grid network 2 Abbreviations: Rural Electrification Authority (REA), Kenya Electricity Transmission Company Limited (KETRACO), the Kenya Certificate of Primary Education (KCPE). Kenya Certificate of Secondary School ( KCSE). Kenya National Examination Council (KNEC), Randomized Encouragement Designs (RED). 140 were connected to the grid electricity while those farther away were provided with off Œgrid ( solar ) power by the government. Consequently, this study also aims to test whether grid and off Œgrid (solar) electricity have different effects on the outcomes listed above. There are several channels throug h which electrification can affect academic outcome s. This paper offers three examples. First, electrification outside of school can increase parents™ participation in income generating activities (Khandker et al ., 2009a: Dinkleman, 2011 , World Bank 2008 ). This can translate to improved outcomes to the extent that more income results in increased purchase of school inputs. In addition, increased income can allow parents to reduce demand for child labor which frees more time for studying and school attendanc e. Second, electricity provides light that allows more studying hours after sunset and before sunrise. In addition, it may increase study time by reducing time spent cooking or fetching firewood (Khandker et al . 2014, World Bank 2008 ). Third, substituting wood Œbased or kerosene Œbased lighting with electric lighting can improve health outcome by reducing respiratory and eye illnesses caused by toxic soot and gases emitted by non Œelectric light sources. This in turn increases school attendance and also perfor mance. With m ore study time, and good health , students are likely to attend school mo re and perform better at school. These changes can translate to more academic progression and consequently higher rates of school completion. While there are many mechanis ms at play, the mechanism of interest in this study is lighting. To achieve this, this paper argues that solar electricity only affects school outcomes through its light , and has no impact outside of the school (unlike electricity). Thus solar coefficients provide estimates of the impact of electrification through lighting. Estimating the causal impact of rural electrification on a number of economic outcomes is challenging due to the presence of confounding factors arising from policy decisions and 141 socioec onomic factors. For instance, governments are more likely to develop infrastructure in areas with great economic potential. Additionally, political connections and influences are likely to influence these decisions. Besides, those who are politically conne cted are likely to be wealthy individuals. These factors make it difficult to quantify the causal impacts of electrification. The literature on rural electrification focuses mainly on economic outcomes such as income and employment with the effects on educ ation usually not being the main focus of the studies. Other related literature also face a number of shortcomings in their identification strategies mainly due to absence of natural experiments. A number of studies offer suggestive evidence that rural ele ctrification improves welfare growth of rural households but based on descriptive and correlational studies between rural electrification and development ( Asian Development Bank, 2010; Barnes, Peskin and Fitzgerald 2003; Cockburn 2005; Khandker 1996; Marti ns 2005). The Kenyan situation provides an ideal setting for this study. In Kenya, electricity supply expansion has been an important government goal. Beginning in mid Œ2013, the government through the Rural Electrification Authority (REA) engaged in an amb itious project to connect all public primary schools with electricity to support it™s the government™s Digital Learning Programme. This was implemented by extending grid electricity to schools close to the grid network and installation of solar photovoltai cs in off Œgrid areas. This project saw the rise of primary schools with electricity from 48% in 2014 to 80% in 2016. Schools with solar power rose from 7% in 2014 to 13% in 2016. In aggregate, schools with power rose from approximately 56% in 2013 to app roximately 94% in 2016. While households did not receive similar coverage, households with electricity increased from 27% in 2013 to 55% in 2016 following electrification of an addition of 1.3 million households. This policy shock provides a convenient e nvironment to study the effect of electrification. The rapid nature of the project reduces the likelihood of 142 confounding policy factors that may affect academic outcomes. In determining routes for new electric transmission lines, the government "first loo ks at major corridors, such as existing utility lines, roads, and railroads before considering other areas" Œ according to the government Œowned Kenya Electricity Transmission Company Limited (KETRACO). Most of these utility lines, roads, and railroads have been in existence for a long while before 2013. While the government has invested large sums on infrastructure, most of these funds are channeled towards upgrading or repair ing of existing infrastructure. As a result, concurrent infrastructure are unlikel y to have been completed in time to alter the existing network of utilities, roads, and railroads in a mann er that affected the trajectory of the grid network . There might be concerns of schools endogenously selecting where to locate. However, public schoo ls do not have flexibility in choosing where to locate. Typically, the government and the community agree on a location of a school based on t he population density. Generally , in rural areas, new schools are built equidistantly from the nearest two or more schools to balance the distribution of schools across a geographic location. Given the speed of electricity rollout and these rigidities in infrastructure development and the location of public schools, the connection of schools to the grid is likely to h ave been exogenous conditional on school fixed effects . The data used in this study is an unbalanced panel of the universe of all schools in Kenya, with 8 th grade students, from 2014 to 2016. The main source of variation in school electrification is driven by the government™s push to electrify public schools. Specifically, 73% of private schools are connected to the grid electricity compared to 44% of public schools in 2014. However, over time, the number of public schools rises to 78% and catches up with p rivate schools by 2016. Initially, the same share (7%) of both private and public schools have off Œgrid electricity but the share of public schools rises to about 13% by 2016. Given that private schools decisions on 143 location and electrification status are likely to be endogenous, this paper restricts analysis to public schools only. The analysis is only for 8 th grade students since examination data is only available and is nationally representative for 8 th grade primary school students. Panel fixed effects provides flexibility in handling endogeneity concerns and is thus used as the primary identification strategy. The panel fixed effects model, finds no statistically significant effects of either grid or off Œgrid electrification on test scores and enrollmen t. However, there is some evidence that off Œgrid electrification may increase school completion by 1%. This is a small effect. This result may suggest that solar effects only provide positive benefits to a school while grid electrification potentially has negative spillovers outside of school. For instance, while light may encourage more completion, this effect is offset by students dropping out to exploit employment opportunities created by arrival of grid electrification. On net, these effects cancel out. Regarding the mechanisms at play, and relying on the solar coefficients, lighting has small but limited effects on academic outcomes. Therefore lighting alone may not be sufficient to boost educational outcomes without complementary academic inputs. The f indings are robust to inclusion of private schools, exclusion of urban schools, and variation in clustering of standard errors. Overall, this paper finds weak evidence that electrification improves academic outcomes. To the best of my knowledge, this paper is the first to quantify the impact of electrification through lighting on education outcomes . In addition, I have not come across any study on education that jointly estimate s the effects of solar and grid electrification. Finally, unlike many papers tha t focus on electrification at the household Œlevel, this paper focus es on electrification at school . These contributions are important for several reasons. First, many papers do not attempt to isolate the channels through which electrification affect outcom es such as educational outcomes. Thus, this paper attempts to tackle this missing part of analysis by a ttempting to isolate the effect 144 of lighting. Second, the effects of grid and solar are likely to be different. Solar power at school level can only affec t outcomes mostly through light ing and has no additional benefits outside of school. On the other hand, grid electricity at school can be an indicator of electrification outside of school. As noted in the literature below, electrification can have income effects, which can ultimately affect education outcomes. Consequently, the solar coefficient will provide an ideal estimate for the effect of lighting . Jointly estimating the effects of solar and grid electricity allows for estimation of non Œlighting effect s of electrification. The underlying assumption is that , any differences between solar and grid electricity coefficients is a measure of additional effects of electricity outside of the lighting channel. One must, however, not push this idea too far becaus e grid electrification may be more reliable and provide higher quality light than solar power. Quality concerns are mediated by technological advances that have improved the quality of solar power illumination. Still , some of the results could be driven by these differences in quality and reliability of these power sources . Taking these and other caveats into consideration, the estimates of solar will provide lower bounds for the coefficients of interest. Household level studies, as reviewed below, do not a ttempt or are unable to easily isolate the mechanisms through which electricity affect s education outcomes . The remaining sections are organized as follows. Section 2 is a review of the existing literature . Section 3 , is a brief description of the context and the data. Section 4 , is a methodology section. This is followed by the results in section 5. Section 6 explore heterogeneity by subject and gender. Section 7 reports robustness checks, and finally section 8 is the conclusion . 145 II. LITERATURE REVIEW In revi ewing the existing literature, this paper splits the review into two major parts Œ grid and off Œgrid studies Œ and within each part provides literature on non Œexperimental and experimental studies. The first part begins with non Œexperimental followed by ex perimental studies on grid electrification. The bulk of the literature on grid use non Œexperimental methods given the difficulty in randomizing grid network. The second part focuses on off Œgrid electrification, which is largely dominated by experimental me thods. A common feature of these studies is that they are done at the household level. This paper diverges by focusing on electrification at school. Several studies have investigated the impact of grid electrification on incomes and education outcomes. Comparing Vietnam communes with and without electricity, Khandker et al . (2009a) find that electricity has positive effect on both economic and educational outcomes. Electrification increases household™s farm cash income by 30 percent, with no effect on non Œfarm income. Furthermore, it increases enrollment by about 10% for both boys and girls. The increase in years of schooling is limited to boys only with electricity increasing years of schooling by 0.52 (about 12% increase relative to year 2002 baseline). I n related literature, Khand ker et al . (2009 b), show electricity increases total household income by betw een 9 percent and 30 percent in rural Bangladesh . Educational outcomes also improve but the results are sensitive to the estimation approach . On the oth er hand, Dasso et al . (2015), find that grid electrification do es not lead to substantial improvements in educational outcome. Taking advantage of a rapid expansion of electricity in rural Peru (Pro grama de Electrificación Rural) and relying household surv ey panel data , they find that rural electrification in Peru increases female enrollment but the effect does not translate to improved attendance. Surprisingly, using school Œlevel panel data, electrification reduces learning in Math and Reading. However, lo nger exposure among treated schools increases 146 scores in Reading among boys and girls but Math improvement is only observed among boys . This finding is consistent with the literature that show that technological innovations may take time before impacting st udent school outcomes (Kho, Lakdawala, and Nakasone, 2018) Some studies rely on geographical influence on grid electrification process to overcome identification issues. Libscomb et al . (2013) exploit the heavy reliance of hydropower and the geographic con siderations that influence the location of hydro Œelectricity dams to study the effec t of electricity on development in Brazil. Using water flow and river gradients to instrument for electrification, they find large positive effects on income, and , educatio nal literacy and school enrollment. Results show that going from no electricity to full electri fication in a county leads to reduction s in the illiteracy rate of 8 percentage points (25 percent drop at the mean) and reduction in the proportion of the popul ation with less than four years of education of 21 percentage points (32 percent decrease at the mean). However, the largest gains were expe rienced in years of schooling, which increased by two years (about a 72 percent increase at the mean ). This suggests that more children obtained post Œprimary (or grade four) education, which may have ultimately led to labor productivity increases. In a similar spirit, Dinkelman (2011) studies the impacts of electrification on employment in rural South Africa. Using land gradient as an instrument together with a fixed effects model, she finds that electricity increases female employment in treated areas . Experimental evidence of impacts of electrification on educational outcomes are rare. This is largely driven by the fac t that it is difficult to randomize grid electrification. Fortunately, certain policies and technological advancements have created opportunities for experimental interventions. For instance, Randomi zed Encouragement Designs (RED) can be employed to create exogenous variation in electricity access. Bernard and Torero (2013) are the first to implement this design on electrification in a developing country. Subsequently, it is employed by 147 Barron and Torero (2014) in El Salvador. They find evidence of grid le ading to increased time allocated to educational activities, increased participation in non Œfarm income generation activities but also children engaging more in household chores. Technological innovations and desire for sustainable energy sources has also led to rise in use of portable sources of power such as solar panels and solar Œpowered devices including solar lamps. Consequently, t here is a nascent literature that provide experimental and non Œexperi mental evidence on the effect s of solar power or solar Œpowered lanterns on education performance. Generall y, except for a few studies, the findings tend to support the hypothesis that solar power leads to improvement in a number of measures of school outcomes. However, there is mixed evidence on the effects o n academic performance. These papers include Furukawa (2014), Barron and Torero (2014) , Arráiz and Calero (2015), Kudo et al. (2017), Hassan and Lucchino (2017), Aevarsdottir, Barton, and Bold (2017). Non Œexperimental studies on solar power included works by Arráiz and Calero (2015). Using household Œlevel and individual Œlevel data and employing propensity score matching techniques, Arrái z and Calero (2015) estimate that solar Œpowered home systems (SHSs) in rural Peru increases children study time, years of schooling (among elementary school students) and higher rates of enrollment (in secondary school). Specifically, enrollment increases by 12 percentage points for those enrolled in high/middle school. In addition, it leads to an increase in years of schooli ng by 0.4 from a base of 3.2 years, and increase in time spent studying by 9 minutes from a baseline of 84 minutes per day. The most common experimental study on the effects of solar power involve the use of solar lamps. Furukawa (2014) conduct a randomize d experiment involving 204 participants in Uganda where some participants received solar lamps among 5 th and 7 th grade students . After 5 months, 148 the paper reports some evidence that the solar lamps increased daily study times by 30 minutes but surprisingly lowered academic performance. In particular, test scores for mathematics and English declined by 0.25 standard deviations, with high performing students (top quintile) experiences largest declines of 0.8 standard deviations. The author explains that these results could be driven by measurement error of study times as students lacked watches/clocks at home, inadequate charging of lamps leading to flickering lights, and possible intra Œhousehold factors that limited the use of the lamps for studying. These re sults are also limited by the small sample size, short observation time of 5 months, and also due to the adverse weather occasioned by the rainy season that minimize ability to charge solar lamps. While this study conducts the experiment at the school leve l, the use of solar lamps is not restricted to school. These solar lamps are available at home and are subject to be used for other purposes besides studying . Unlike Furukawa (2014), Hassan and Lucchino (2017) find positive effects and spillovers among 7 th grade students in a similar experiment in Kenya. The authors report improved math scores of 0.88 standard deviations among treated students in a class with average treatment intensity (43%). In addition, there is evidence of spillovers as an increasing t he share of treated students by 10% leads to a 0.22 standard deviation increase in scores of control students. The study provides some evidence suggesting that this spillover is largely driven by within Œschool interactions through co Œstudying after sunset. The co Œstudying spillovers are likely to be larger in a school setting than in households becau se schools provide larger avenues and central location for studying. Small sample sizes are common in experiments due to logistical or funding constraints. Aev arsdottir, Barton and Bold (2017) conduct a solar lamp experiment with a large sample involving treating 1800 households with students in one of 60 schools in Tanzania. The experiment randomly provided full, partial, or no subsidies towards purchase of a s olar lamp with the 149 capability of charging mobile phones. They find that purchase of a solar lamp leads in a 25% increase in income. Adult labor participation on both the extensive and intensive margin rise by between 10% and 20%. Unlike in Barron and Torer o (2014), improvement in labor force participation by adults does not lead to increase in child labor participation. Unfortunately, the study finds no evidence of improvement in education outcomes such as enrollment, attendance, and time spent studying. While the It is thus unlikely that the treatment would have had any positive effects on academic performance. These results are similar to those reported in an experimental setting in Bangladesh where outcomes were muted (Kudo, Shonchoy and Takahashi, 2017). Among 4 thŒ8th grade students, s olar lamps initially led to increases in attendance but this effect diminished over time. In addition, treated students experienced an increase in night study time of 20 Œ25 minutes a day but the treatment had no statisticall y significant impact on school progression. As evidenced by the literature above, most of the studies look at the impact of electrification at the household level. These studies also document significant impact s of electrification on incomes and labor dema nd, which are likely to have also influence d the findings on educational outcomes. As such, these studies are unable to quantify the direct (non Œmonetary) impact of electrification on education and also cannot distinguish the channels through which electri fication affects educational outcomes. In the latter experimental studies, the use of solar lamps limits the ability to isolate the impact of solar power, as these lamps are portable and available for use outside of school. The use of solar lamps at home i s subject to competing uses at home and may underestimate the true impacts of solar light on academic outcomes. Besides, if household chores are prioritized when the solar lamp is being used, and given that solar lamps typically provide power for a few hou rs, by the time students get the chance to study, the solar lamp light will likely 150 be dimmer. In addition, solar lamps at home may lead to improvement in incomes either through charging of phones or through extended time engaging in income Œgenerating activ ities (Aevarsdottir, Barton and Bold, 2017). This increased income can lead to more purchase of inputs for students. Finally, the solar lamps are likely to provide weaker illu mination compared to both grid and photovoltaic solar panels (used in Kenyan sch ools). This paper on the other hand overcomes these challenges by studying presence of solar power that is used only at the school and by relying on solar photovoltaic power that provides higher quality lighting than solar lamps. The presence of these alte rnative sources of lighting will benefit students in the early evenings and early mornings at school. While power may provide additional benefits during daytime at school, I believe th e key benefits of light will be in the early evenings and mornings. Idea lly, if time use data is available, it would be easy to quantify how much time i s spent at school and whether students are studying using school lighting. However, I do not have this kind of data. There are, however, reasons to believe that electrification at school allows students to study in the early mornings and evenings more so at school than at home. Specifically, in areas with close to the grid network , connection to electricity requires substantial fees of approximately $300 and thus few families wo uld afford to get connection. Second, safety, particularly in rural areas where the bulk of the sample schools are located, is not a big issue. Communities are homogenous and it is common practice for 8th grade students to stay at school till 6 PM and to r eport to sch ool before 7 AM. However, safety may be a larger concern for female students and this might lead to heterogeneous responses by gender. Thus, effects for girls may represent a lower bound of the impact of electrification if electrification is on ly at school and girls spend less additional time studying at dawn and dusk. 151 III. DATA DESCRIPTION Context The Kenyan education follows an 8 Œ4Œ4 system. The 8 Œ4Œ4 is designed so that ideally a student spends 8 years in primary school, 4 years in secondary scho ol, and 4 years in university. Students start school in pre Œschool, which lasts three years before the 8 Œ4Œ4 system kicks in . Following the completion of pre Œschool, students enroll in primary schools for a period of eight years. Each school year is split into three semesters with school sessions starting in January and lasting three months with a oneŒmonth intervening break. Primary school education culminates in the final national exam Œ the Kenya Certificate of Primary Education (KCPE). T his is a very co mpetitive standard national exam whose results are used to admit students to secondary school. Secondary school lasts four years. After four years, secondary school students must sit the national exam ŒKenya Certificate of Secondary School (KCSE). This exa m determines entry into university and the type of majors that student are eligible to pursue. This study focuses on the KCPE examination results for the 8 th grade students. This is because it is the only nationally representative examination results for p rimary school students. In addition, most secondary schools already have electricity, and hence has little variation in school electrification. Completion is defined as taking the KCPE exams while enrollment represents the number of 8 th grade students at t he beginning of the year. The national examination scores and school completion data was provided by the Kenya National Examination Council. School completion is defined as having taken the 8 th grade national examination data. Examination covers five subj ects, English, Kiswahili, Mathematics, Science, and Social Studies. The maximum score for each subject is 100 while the minimum score is 0. In the regression analysis, the test scores are standardized to have a mean 0 and a standard deviation of 1. 152 The a dministrative structure of the schools is organized as follows. At the national level, all schools and academic institutions are under the Ministry of Education. Secondary, primary school educations and pre Œschool fall under the Department of Basic E ducati on while post Œsecondary education is under the D epartment of Higher L earning. The ministry of education delegates some of its duties to the County Education Offices, which supervise various Sub ŒCounty Education Offices. Under the Sub Œcounty education offic es, schools are grouped into school Zones. Finally, within each school zone are several schools headed by a head teacher (principal). Thus the geographical hierarchy of primary school is National ŒCounty ŒSub ŒCounty ŒZone ŒSchool. This study uses data fro m all primary schools that had 8 th grade students during the period of study (2014 Œ2016). The unit of analysis is the school. Data on school electrification and school characteristics were obtained from the Ministry of Education Kenya, which liaises with t he school principals in collecting these data. The data is typically collected between October and November each year through a national primary school census. Data on school characteristics was gathered from the Ministry of Education. The school character istics avai lable include infrastructure Œtemporary and permanent classrooms, toilet facilities, primary sources of water, number of private ly and public ly hired teacher s, number of students (enrollment ), school location (rural/urban), school ownership (pri vate/public), school accommodation type (day, boarding or day and boarding), school gender (girls only, boys only, or mixed). Test score and school completion data is available from the Kenya National Examination Council (KNEC ), which administers and grade s the primary and secondary national exams. KNEC is an independent entity within the Ministry of Education. 153 School electrification variation is largely driven by the nationwide campaign to provide electricity to all public schools. This project started in an attempt to implement the government™s Digital Learning Programme. The government intended to supply laptops to every first grade student in primary school and provided digital access of educational content. Th is was a major campaign promise that the pr esident had pledged in the run up to the 2013 elections. Upon winning the elections, the new administration embarked on an ambitious program to electrify schools to enable its digital learning program and also to improve access of households to electricity . Unlike in previous cases, the government intended to supply electricity specifically to schools and other public facilities. As of June 2013, out of 21,222 primary schools in the country, 48% had access to electricity. However, by 2016, 80% of the 34,12 4 schools had electricity. Public schools largely drive the changes in elect rification during this period . Specifically, by 2014, 8,522 public schools had grid electricity while 1 ,582 had solar. By 2015, the number of public schools with grid increased to 12,970 while solar schools doubled to 3 ,604. Finally, by 2016, 16 ,403 public schools had grid electricity while the number with solar remains steady at 3 ,543. Meanwhile, the total number of public schools only rose by less than 2000 from 21,625 in 2014 to 23,439 in 2016. The rapid nature of the project reduces the likelihood of confounding policy factors that may affect academic outcomes. In determining routes for new electric transmission lines, the government "first looks at major corridors, such as exi sting utility lines, roads, and railroads before considering other areas" Œ according to the government Œowned Kenya Electricity Transmission Company Limited (KETRACO). Most of these utility lines, roads, and railroads have been in existence for a long whil e before 2013. While the government has invested large sums on infrastructure, most of these funds are channeled towards upgrading or repairing of existing infrastructure. As a result, concurrent infrastructure are unlikely to have been completed in time t o alter the existing network 154 of utilities, roads, and railroads in a manner that affected the trajectory of the grid network. In addition, public schools do not have flexibility in choosing where to locate. Typically, the government and the community agre e on a location of a school based on the population density. Generally speaking, in rural areas, new schools are built equidistantly from the nearest two or more schools to balance the distribution of schools across a geographic location. Given the speed o f electricity rollout and these rigidities in infrastructure development and the location of public schools, the connection of schools to the grid is likely to have been exogenous. However, t his paper takes additional steps to address potential endogeneity issues using panel fixed effects at the school level and by controlling for a number of school level observables. In addition, it includes variables to absorb school and regional time varying unobservable. Electrification projects tend to be implemented r egionally and as argued the main factors influencing electricity rollout were likely fixed within the short period of 2014 Œ2016. This paper argues that the factors that could have influenced electrification remain largely unchanged at the school level and thus the identifying assumption is that conditional on school fixed effects, electrification was largely exogenous. Summary statistics The data contains an unbalanced panel of three years from 2014 to 2016 for the main analysis. These were the only years in which the government had digitized records of school data. Table 3 .1 below shows the summary statistics of the main variable s of interest. The statistics are derived from the observations used in the panel analysis, which restricts the sample to only p ublic schools. Any observations not used in the regression analysis are excluded. This paper uses the universe of 155 all 8 th grade schools that have all the data available. This summary is for the 2014 and 2016, which correspond to the beginning and the end of the study period. The test score s summary shows that schools with grid electricity outperform those with off Œgrid and those without electricity. Similarly, schools with off Œgrid electricity generally outperform those without electricity though by a sma ll margin and sometimes the difference is not statistically significant . Schools with grid electricity tend to have higher enrollment and completion while those with off Œgrid have slightly lower enrollment and completion compared to schools without electri city . This suggests that grid electrification is installed first in areas with high population densities and possibly in proximity to other amenities and infrastructure. School inputs are reported in student Œinput ratio for easy comparability across school s. Schools with off Œgrid and without electricity have similar student Œbook ratio while grid schools sometimes has slightly has a better rati o. Generally, by subject, 3 Œ4 students share a singl e book. Compared to the control schools off Œgrid schools have a higher student Œteacher ratio while grid has a similar student Œteacher ratio to the control schools . As shown by the student Œclassroom ratio, schools with either form of electricity initially are more crowded by about 2 extra students per class resulting in an average class size of 38 students. Water is useful for both consumption, cleaning and related sanitary conditions of the s chool. The statistics show that electrification is generally associated with access to better water sources (tap and borehole) wit h tap water being the largest predictor of electrification. To highlight the importance of s chool ownership on electrification status, this paper includes statistics of private schools in the summary but not in the regression. Public schools account for 92 % of schools without electricity in 2014 but this share declines to 77% by 2016, largely driven by increase in number of public schools receiving grid electrification. During this 156 period, public schools accounts for 82% of schools with off Œgrid electricit y in 2014 but this hare rises modestly to 85% by 2016. Further insights can be gleaned from looking at the distribution of schools with electricity within each school type. For instance, 72 % of private schools are connected to the grid electricity compared to 44% of public schools in 2014. However, over time, the number of public schools rises and catches up with private schools by 2016. Initially, the same share (7%) of both private and public schools have off Œgrid electricity but the share of public schoo ls rises to about 11 % by 2016 with little change to the share of private schools . Finally, school location in rural areas is negatively associated with grid electrification but positively associated with off Œgrid electrification. Another way of looking at the data is to focus on a specific location and examine electrification. Only 42 % of public schools in rural areas had grid el ectricity in 2014 compared to 80 % of public schools in urban areas. This gap, however, decreases as the government electrificatio n project continues through 2016 (to 78% and 96% respectively) . On the other hand, rural schools are more likely to have off Œgrid electricity compared to those in urban areas. The share of rural schools with off Œgrid electrification rises from 7% in 2014 t o 12% in 2016 while the share of urban schools with off Œgrid power remains at 1 % throughout the period. Overall, while there are some differences between schools based on electrification status there is no consistent pattern of differences between schools with electricity and those without. In addition, the differences in attributes tend to be minor particularly for school student Œinput ratios. In addition, schools without electricity and those with off Œgrid electricity are qualitatively similar in characte ristics. 157 Table 3.1: Summary Statistics Table includes T Œtests performed on the differences in variables between treated and non Œtreated schools . * p<0.10, ** p<0.05, *** p<0.01. Statistic No light Off-grid Grid Off-grid - No Light Grid - No light No light Off-grid Grid Off-grid - No Light Grid - No light School Mean Score (out of 500) 2382412483***10***2362352451.5 9***Enrolment (Total) 3673764999132***349335431-14* 81***Enrolment (8th Grade) 333151-2*** 18***312845-3*** 13.35*** Completion 323050-1.88*** 18***312844-2.75*** 12.87*** Pupil-Books Ratio (4-8 grade) Math 2.99 2.97 2.97 -0.02 -0.02 2.97 2.8 2.77 -0.17 -0.2 English 2.8 2.84 2.69 0.04 -0.11* 2.62 2.74 2.55 0.12 -0.07 Kiswahili 3.06 3.06 30-0.06 2.66 2.83 2.65 0.17 -0.01 Science 4.28 4.36 4.39 0.08 0.11 4.35 4.41 3.75 0.06 -0.6*** Social studies 4.08 4.34 3.87 0.26 -0.21* 4.02 4.16 3.53 0.14 -0.49** Main Source of Water No water 10%6%4%-4% -6%*** 12%8%5%-4%*** 7%*** Rain 32%28%24%-4% -8%*** 31%28%25%-3%** -6%*** River 24%25%16%2%* -8%*** 22%27%16%5%*** -6%*** Tap 19%17%38%-2% 19%*** 17%14%32%-3%** 16%*** Borehole 16%24%18%8%*** 2%*** 18%24%22%5%*** 3%*** Government Teachers -----17.15 14.16 21.78 -2.99*** 4.63*** Private Teachers -----4.64 5.73 4.89 1.09*** 0.25* Total Teachers -----20.03 18.32 24.73 -1.71*** 4.7*** Students-Teacher Ratio -----17.62 18.97 17.42 1.35*** -0.2 Permanent Classrooms 8.3 7.76 11.4 -0.54 3.1*** 7.95 7.63 10.32 -0.32*** 2.37*** Temporary Classrooms 2.26 2.35 2.07 0.09 -0.19*** 2.26 1.76 1.81 -0.5*** -0.45*** Total Classrooms 10.32 9.88 13.23 -0.44 2.91*** 9.32 8.84 11.23 -0.48*** 1.91*** Students-Classrooms Ratio 3638.28 37.63 2.28*** 1.63*** 37.5 37.23 37.6 -0.27*** 0.1 Teacher-Toilet Ratio 9.39 9.17 10.12 -0.22* 0.73*** Student-Toilet Ratio 38.9 46.13 34.82 7.23*** -4*** 39.25 45.03 34.67 5.78*** -4.58*** Ownership Private 8%18%25%9.9%*** 17%*** 23%15%21%-9%*** -3%*** Public 92%82%75%-9.9%*** -17%*** 77%85%79%9%*** 3%*** Rural 98%99%91%1.2%*** -7%*** 99%99%94%1%*** -4%*** Urban 2%1%9%-1% 7%*** 1%1%6%-1%*** 4%*** Obs 8,655 1,241 7,794 1,883 1,978 14,024 20142016158 IV. IDENTIFICATION STRATEGY If the electrification process was random, the impacts could be estimated using t he naïve OLS specified as follows: =+ + (1) where is the outcome of interest at school s, in county c, in z one z, and at time t. ELEC is a vector of electricity dummy variable s for grid electricity and off Œgrid electricity . The counterfactual is having no electricity. The is the error term. In this model would be the coefficient s of interes t estimating the average treatment effects of electrification. However, electrification was not randomized and thus e stimating equation (1) is likely to yield contaminate d coefficient s of interest due to omitted variables that are likely to be correlated w ith the electricity connection and also affect outcomes of interests. To address these issues, I add school Œlevel controls and time and region fixed effects as follows: =+++ ++ (2) where are observable school characteristics such as infrastructure, teacher and student demographics and characteristics. are zone fixed effect s which capture factors that are common across schools within a zone that are fixed over time. are year fixed effects which control for factors that are fixed for all schools within time t. Specification (2), however, does not a ddress unobserved school Œlevel fixed factors . Consequently, I use a panel fixed effects model as follows: =+++++ (3) where are observable scho ol characteristics such as infrastructure, teacher and student demographics and characteristics. I include which are school fixed effects which capture time 159 invariant characteristics of the school while are year fixed effects which control for factors that are fixed for all schools within time t. The underlying identificatio n assumption in specification (3 ) is that the omitted variables are time i nvariant at the school Œlevel . While specification (3 ) addresses most of the endogeneity concern s raised previously, it does not address the issue of time Œvarying omitted factors that are likely to be correlated with electricity connection and the outcome of interest. Following, previous literature , I argue that the time variant characteristics are likely to be correlated with baseline school characteristics (Almond et al. , 2011; Acemoglu et al. , 2004; Hoynes and Schazenbach, 2009). The preferred specification s (4) therefore includes a linear time t rend that allows baseline characteristics to differe ntially affect outcomes with time. Thus, the main identifying assumption of this paper is that, conditional on these set of controls, school electrification was exogenous. =++++++ (4) V. MAIN RESULTS The findings based on panel fix ed effects. The tables below report outcomes on test scores, attendance, and c ompletion for 8 th grade students in Kenya. The unit of observation is school. These regressions are based on variations of Specification (3) . Part 1: School test scores The general format of the tables starts with a simple panel fixed effects regression o f the outcome variable on electrification variable and then proceeds with additi on of controls and clustering of errors by school . In Table 3. 2, specification (1) regresses test scores on electrification status only. To address potential omitted variable b ias, specification (2) adds school level controls to specification (1). However, specifications (2 ) does not account for important omitted time varying 160 school Œlevel factors . There is no obvious or best method to address this issue. However, time varying c onfounding factors are likely to be correlated with the characteristics of the school. In line with some existing literature, specification (3) includes an interaction of initial school characteristics and year. This will absorb some of the time varying co nfounders. Finally, all specifications Specification (3) cluster standard errors at the school level since outcomes are likely to be correlated within the school over time. Failing to cluster will result in inflated/deflated standard errors leading to misl eading p Œvalues and inference interference. The remaining tables follow the same format. The fiYfl in the tables indicates a fiYesfl. All test Œscores are standardized so that the national mean is 0 with a standard deviation of 1 every year. Thus, all coefficie nts should be interpreted as changes in standard deviations. Specification (1) indicates that electrification reduces school mean scores by 0.02 standard deviations. These estimates are statistically significant only for grid but they are also likely to suffer from omitted variable bias. Specification (2) confirms this suspicion as estimates increase by half to Œ0.01 and become statistically insignificant. Time varying confounders appear to also play a role in the estimates since estimates increase further when an interaction between 2014 school characteristics and time are included in the regression (specification 3). Taken together, the preferred specification (3) shows that off Œgrid electrification has a small positive but statistically insignificant effe ct on test scores while grid electrification has a small negative but statistically insignificant effect. The negative effects of grid electrification may suggest that grid electrification may have negative impacts on test Œscores outside of school. Howeve r, we cannot push this point too far as the estimates are quite small. These findings are surprising, as one would expect electrification to improve school outcomes. It is, however, possible that electrification affects the composition of 161 Table 3.2: Effects of School Electrification on School Test Scores Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 . (1) (2) (3) Off-Grid Electricity -0.021 -0.0127 0.0020(0.0142) (0.0138) (0.0140) Grid Electricity -0.0230*** -0.0107 -0.0018 (0.0071) (0.0069) (0.0079) Enrolment Boys (8th Grade) -0.0075*** -0.0121*** (0.0024) (0.0012) Enrolment Girls (8th Grade) -0.0119*** -0.0092*** (0.0007) (0.0008) Enrolment Boys (1st -7th Grade) 0.00020.0006***(0.0001) (0.0002) Enrolment Girls (1st -7th Grade) 0.0006***0.0004* (0.0001) (0.0002) Books 4-8th Grade (,00s) -0.00002 -0.00008 (0.0001) (0.0001) Total Classrooms -0.0004 -0.0007 (0.0009) (0.0011) Rain Water 0.0290*0.0249(0.0150) (0.0202) River Water 0.01430.0353* (0.0153) (0.0208) Tap Water 0.0283*0.0319(0.0167) (0.0216) Borehole Water -0.0031 0.0246(0.0163) (0.0218) Toilets - Boys 0.00005***0.0008(0.0000) (0.0009) Toilet - Girls 0.0004-0.0002 (0.0007) (0.0009) Toilet - Male Teachers -0.0071 -0.0045 (0.0046) (0.0066) Toilet - Female Teachers 0.0002-0.0002 (0.0008) (0.0005) Constant -0.229*** -0.0046 0.0098(0.0049) (0.0416) (0.0390) N524925249252492R-Squared 0.000.05300.06School Fixed Effects YYY Year Fixed Effects YYY Controls YY Initial controls x year Y School Cluster YYY 162 students if it leads to lower dropouts. Specifically, if electrification prevents dropou ts among lowest achieving students then it can lead to lower average school test scores. On the other hand, if electrification also benefits the high achieving students we can expect zero to positive effects on test scores. However, as shown below, I find no strong evidence of electricity increasing enrollment or completion Œ only solar power has a positive and statistically significant effect of increasing enrollment by one percent. Another potential explanation could be that the effects take time and give n the short nature of the panel data, observations within school are not sufficient to result in improved test scores. Kho, Lakdawala, and Nakasone (2018) provide evidence indicating that the effects of technological improvements may take time before affec ting student performance. In summary, b ased on these results, grid and off Œgrid electrification have no differential impacts on tes t scores, at least in the short Œterm. Part 2 : Enrollment This part repeats the analysis of part 2 but focusing on enrollment as the outcome of interest. If electrification creates more study time and more study time results in improved performance, schools with electricity are likely to experience increases in enrollment . In addition, improved performance could lead to lower le vels of dropping out. While Table 3. 2 finds no effects on test scores, it is possible that anticipated improved test score by students following electrification can encourage students to enroll and stay at school. The estimates below test whether enrollmen t increases following electrification. Table 3. 3 reports panel fixed effects estimates. Unlike test scores, enrollment is in log forms. The format of the results is as in Table 3. 2. Specification (1), which omits controls, indicates that both grid and non Œgrid electrification increases enrollment by 2.5% . Addition of school 163 controls to the model increases estimates slightly to 2.6% and 2.8% for off Œgrid and grid electricity respectively. However, it appears that time varying confounders also affect enroll ment in a Table 3.3: Effects of School Electrification on 8th Grade Enrollment (Dependent variable Œ log of enrollment) Errors clustered at school level (specification 4). Standard errors in parenthesis. * p<0.10, ** p<0.05, *** p<0.01. (1) (2) (3) Off-Grid Electricity 0.0251***0.0264***0.0071(0.0068) (0.0068) (0.0069) Grid Electricity 0.0246***0.0281***0.0026(0.0036) (0.0036) (0.0041) Enrolment Boys (1-7th Grade) 0.0003***0.0005***(0.0001) (0.0001) Enrolment Girls (1-7th Grade) 0.0005***0.0003***(0.0001) (0.0001) Books 4-8th Grade (,00s) -0.00001 -0.00001 (0.0000) (0.0001) Total Classrooms 0.0020***0.0033***(0.0007) (0.0011) Rain Water -0.0108 -0.0151 (0.0072) (0.0098) River Water -0.00321 -0.0189* (0.0074) (0.0101) Tap Water -0.00659 -0.0139 (0.0081) (0.0106) Borehole Water 0.00792-0.0172 (0.0078) (0.0107) Toilets - Boys 0.00000.0006(0.00003) (0.0008) Toilet - Girls 0.0013***0.0020***(0.0005) (0.0008) Toilet - Male Teachers 0.0094***0.0093** (0.0024) (0.0038) Toilet - Female Teachers -0.0001 -0.0001 (0.0003) (0.0002) Constant 3.515***3.313***3.304***(0.0025) (0.0229) (0.0251) N523665236652366R-Squared 0.000.010.02School Fixed Effects YYY Year Fixed Effects YYY Controls YY Initial controls x year Y School Cluster Y Y Y 164 significant manner . Once an interaction of initial school characteristics and time are included in specification (3), the coefficients shrink and become statistically insignificant. Specification (3) shows that electrificati on has a positive effect on enrollment of less than 1%. The results are robust to clustering of standard errors. Specification ( 3) indicates that off Œgrid electrification estimates are larger (0.7%) compared to grid electrification (0.3%). These estimates are in line with results from test scores. However , while off Œgrid electrification seems to have larger estimates than grid electrification, qualitatively, the estimates are similar for both types of electrification. Part 3: Completion Electrification can also affect completion. For instance, increased and better lighting hours from electrification can create conducive study environment for students. While we do not find any effect on test Œscores, the results could be heterogeneous at the individual level. Thus if electrification increases test Œscores for some individuals, it could also encourage staying at school. It is also important to note that student may stay longer in school if they have strong beliefs that electrification will positively affect thei r future performances. In the current context, the national exam (KCPE) is the ultimate exam that students study for, and if they believe that more study hours will translate to better final grade, they are likely to stay in school longer. In the spirit o f the findings of Kho, Lakdawala and Nakasone (2018), students are also likely to have the same perspective about time invested studying leading to eventual positive results in the long Œrun. Finally, students may prefer co Œstudying and electrification incr eases opportunities for co Œstudying. This creates an attractive environment for students to learn and incentives to stay in school. On the other hand, grid electrification outside of school may also have adverse effects on completion. For instance, jobs cr eated from electrification can attract students leading to drop outs. Dinkleman (2011) find positive labor impacts of electrification for women in South Africa. 165 Table 3.4: Effects of School Electrification on 8th Grade Completion (Dependent: Log Completion) Errors clustered at school level. Standard errors in parenthesis. * p<0.10, ** p<0.05, *** p<0.01. (1) (2) (3) Off-Grid Electricity 0.0378***0.0270***0.0106** (0.0058) (0.0053) (0.0053) Grid Electricity 0.0286***0.0145***-0.0045 (0.0033) (0.0026) (0.0029) Enrolment Boys (8th Grade) 0.0094***0.0145***(0.0029) (0.0014) Enrolment Girls (8th Grade) 0.0164***0.0144***(0.0009) (0.0009) Enrolment Boys (1st -7th Grade) -0.0002** -0.0006*** (0.0001) (0.0001) Enrolment Girls (1st -7th Grade) -0.0004*** -0.0001 (0.0001) (0.0001) Books 4-8th Grade (,00s) 0.00000.00003(0.0000) (0.0000) Total Classrooms (0.0001) 0.0005(0.0003) (0.0004) Rain Water (0.0025) -0.00954 (0.0053) (0.0071) River Water (0.0031) -0.0186** (0.0055) (0.0074) Tap Water (0.0034) -0.0219*** (0.0060) (0.0078) Borehole Water 0.0021-0.0143* (0.0057) (0.0078) Toilets - Boys (0.0000) 0.0005(0.0000) (0.0006) Toilet - Girls 0.0005*0.001**(0.0003) (0.0005) Toilet - Male Teachers (0.0000) -0.00325 (0.0017) (0.0023) Toilet - Female Teachers 0.0000-0.0001 (0.0001) (0.0002) Constant 3.490***3.066***3.045***(0.0024) (0.0388) (0.0311) N524925249252492R-Squared 0.000.410.44School Fixed Effects YYY Year Fixed Effects YYY Controls YY Initial controls x year Y School Cluster Y Y Y 166 Table 3. 4 repeats the analysis of Table 3. 3 but now with log of school completion as the dependent variab le. Completion is defined as completing the 8 th grade national exit exam (KCPE). Omitting school level controls, specification (1) shows positive and statistically significant impacts of electrification on completion. The off Œgrid estimates are slightly l arger (3.8%) than grid estimates (2.9%). Specification (2) shows that estimates are biased from omitted school Œlevel variables. Adding school Œlevel controls decreases coefficients to 2.7% and 1.5% for off Œgrid and gri d electrification respectively, but the estimates remain statistically significant. Specification (3) adds controls to remedy estimates from time Œvarying confounders. This results in estimates shrinking further. The preferred specification (3 ) indicates that off Œgrid electrification increases enrollment by 1% and this estimate is statistically significant. However, grid electrification has a small but negative coefficient. The absence of positive effects for grid electrification on enrollment is surprising given that off Œgrid electrification has positive effects. One would expect that grid has stronger effects particularly since it is perhaps more reliable, might have better lighting quality, likely provides additional lighting and income opportunities outside of the school. One explanation of th ese results is that the completion estimates could be picking up some of the potentially negative effects of electrification outside of school. Presence of grid electricity at school implies that electricity is likely to be available in the areas near the school. If grid electrification encourages students to drop out of school to pursue jobs that come with electrification or distracts students (say through too much time spent on watching television), then electrification may result in more students droppin g out of school. Alternatively, grid electrification may induce students at the margin of dropping out to stay in school longer but only temporarily Œ i.e. students may remain in school longer following electrification but not long enough to complete the n ational exit exam. 167 Since off Œgrid electrification is mainly benefiting students at school only, particularly through lighting, this paper argues that the off Œgrid coefficients provide lower bound estimates of effects of lighting from electrification. This paper argues that most of the off Œgrid electrification (1%) is coming through lighting. Part 4: Test for Common Trend To complete the identification strategy, I attempt to show that treated and control schools followed a common trend prior to electrifica tion program. One of the key identification assumptions is that absence treatment treated schools and control schools would have had similar trends in outcomes. If this assumption is violated, then some of the observed differences in outcomes would be driv en by pre Œtrends. Since the treatment occurs at different times, I restrict the analysis to schools that received treatment in 2014 Œ which is the first year since the government rolled out the electrification program. Controls schools are defined as schoo ls that had no electricity as of 2016. To formally test the common trends assumption, I specify the following model: =+ x( ==) +++ (4) is the outcome of interest in school s in county c in zone z at time t. Treat is a vector of treatment indicators. Some schools are treated with grid electricity in 2014 while others are treated with off Œgrid electricity. captures school fixed effects while absorbs the year fixed effects. A set of dummies ( ==) indicate the i th year. Since treatment fir st occurs in 2014, a test of parallel trends requires that = =0. In the regression, the year 2012 is the omitted year dummy and thus parallel trends are satisfied if =0. Due to data limitations, I can only test for parallel trends in test scores and school completion. In addition, I have a smaller sample size consisting of a panel of about 5,000 schools yielding approximately 25000 observations from 168 2012 to 2016. The data collected was limited by time and travel constraints during my fieldwork. I sampled a few counties 3 while attempting to ensure that my sample was representative. Future fieldwork will attempt to collect more data from the counties not covered. Table 3.5: Test for the Assumption of a Common Trend Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01 3 I gathered data from 18 out of 47 counties in Kenya. Baringo, Bomet, Elgeyo Marakwet, Homa Bay, Kajiado, Kiambu, Kilifi, Kisumu, Kwale, Machakos, Mombasa, Nairobi, Nakuru, Nyamira, Uasin Gishu, Vihiga, West Pokot School Mean Scores (Logs) Completion (Logs) Grid x YEAR=2013 -0.0008 -0.0358 (0.004) (0.025) Grid x YEAR=2014 0.00148-0.0348 (0.005) (0.024) Grid x YEAR=2015 0.00998*-0.0411* (0.006) (0.024) Grid x YEAR=2016 0.0116*-0.0707*** (0.006) (0.026) Off-grid x YEAR=2013 -0.0042 -0.001 (0.007) (0.034) Off-grid x YEAR=2014 -0.0037 0.0233(0.008) (0.035) Off-grid x YEAR=2015 0.01110.0064(0.009) (0.037) Off-grid x YEAR=2016 0.00440.0164(0.009) (0.038) YEAR=2013 0.00220.0591***(0.004) (0.022) YEAR=2014 -0.0032 0.0691***(0.005) (0.022) YEAR=2015 -0.0148*** 0.104***(0.005) (0.022) YEAR=2016 -0.0105* 0.127***(0.006) (0.023) Constant 5.550***3.439***(0.001) (0.007) Observations 2526821125R-squared 0.830.91School Fixed Effects YYYear Fixed Effects YYSchool Cluster YY169 The results reported in Table 3. 5 find no differences in pre Œtrends between treated school (grid and off Œgrid) and controls schools. While the coefficients tend to be negative, they are small and statistically insignificant. Part 5 : Test for Complementarity between School Inputs and Electrification This section concludes by investigating whether electrification is a complement to school inputs. I focus on number of books, classrooms, and toilets. Number of teacher toilets is a proxy for number of teachers Œ Using 2015 and 2016 data shows that the correlation between number of teachers and toilets is approximately 0.3. The results are reported in Table A 3.1 in th e appendix. I find little evidence of electricity acting as a complement to school inputs. Specifically, for test scores, grid electricity acts as a complement to the number of teacher toilets (a proxy for number of teachers). Electricity and books appear to be substitutes but the magnitude of the coefficients are too small for all outcomes. Grid electricity is a complement for the number of classrooms Œ a one percent increase in number of classrooms leads to an additional 0.3 percent increase in enrollment in schools with grid electricity while the magnitude for off Œgrid electricity is 0.6. Overall, electricity does not appear to be a strong complement to school inputs . This, however, may be a short Œterm result and in the long Œterm electricity may become a complement to school inputs. For example, in the long Œrun teachers may adapt their teaching techniques to maximize the benefits of electrification. VI. HETEROGENEITY BY SUBJECT AND GENDER In this section, the paper explores possibility of heterogeneous impact s of electrification on outcomes. For conciseness, the paper limits the analysis to test scores by subject and subsequently look at outcomes by gender. Studies in different countries have shown that treatment effects can vary by subject (Dasso et al ., 2015; Furukawa, 2014; Hassan and Lucchino, 2017). One potential 170 explanation for these findings is that students may choose to specialize on a few subjects when faced with time constraints. Lighting provides more study hours and this can allow students to incre ase study time dedicated to subjects that previously receiving less time. As a consequence, student performances may vary by subject. Gender has been shown to play an important role in different contexts. For instance, women generally have few economic opp ortunities globally in many sectors of the economy. In SSA, girls tend to have fewer education opportunities compared to boys due to cultural preferences for boys over girls. Studies on electrification and education have also documented gender differences in outcomes (Khandker et al ., 2009a; Khandker et al ., 2009b; Dasso et al ., 2015). Part 1: Test scores by subject Table 3. 6 reports the coefficient estimates of school test scores by subject. The subject test scores have been standardized to have a mean of 0, and a standard deviation of 1. Each column reports the preferred Specification (3) used in the previous analysis. Each estimates are from a panel fixed effects model with school Œlevel controls, an interaction between initial school characteristics and time, as well as standard errors clustered at the school level. The results show evidence of heterogeneous treatment effects by subject both for grid and off Œgrid electrification. Grid electrification estimates positive for English, Math, and Social and R eligious Studies but negative for Kiswahili and Science. However, these estimates are quantitatively small. Estimates are larger and statistically significant for off Œgrid electrification. Specifically, off Œgrid electrification increases test scores for En glish and Math by 0.03 and 0.05 standard deviations respectively. Kiswahili scores decrease by 0.05 standard deviations following off Œgrid electrification. The off Œgrid coefficient estimates for Science, and Social and Religious Studies are positive but sm all 171 Table 3.6: Effects of School Electrification on School Mean Test Scores by Subject Errors clustered at school level. Standard errors in parenthesis. * p<0.10, ** p<0.05, *** p<0.01. English Math Kiswahili Science Social and Religious Studies Off-Grid Electricity 0.0308** 0.0496***-0.0500*** 0.00390.0025(0.0122) (0.0148) (0.0150) (0.0150) (0.0138) Grid Electricity 0.00570.0082-0.0034 -0.0018 0.0061(0.0072) (0.0081) (0.0086) (0.0087) (0.0078) Enrolment Boys (8th Grade) -0.0100*** -0.0081*** -0.0116*** -0.0103*** -0.0091*** (0.0009) (0.0008) (0.0013) (0.0011) (0.0010) Enrolment Girls (8th Grade) -0.0052*** -0.0097*** -0.0050*** -0.0161*** -0.0138*** (0.0006) (0.0007) (0.0008) (0.0010) (0.0009) Enrolment Boys (1st -7th Grade) 0.0004***0.00020.0007***0.0005***0.0007***(0.0002) (0.0002) (0.0002) (0.0002) (0.0002) Enrolment Girls (1st -7th Grade) 0.0004** 0.0007***-0.0003 0.0008***0.0006***(0.0002) (0.0002) (0.0002) (0.0002) (0.0002) Books 4-8th Grade (,00s) -0.00008* -0.0002* -0.00008 -0.00001 -0.00012* (0.0000) (0.0000) (0.0001) (0.0000) (0.0001) Total Classrooms -0.0004 -0.0009 -0.0005 -0.00228** -0.0009 (0.0009) (0.0010) (0.0012) (0.0011) (0.0010) Rain Water 0.0358* 0.0005-0.0089 0.01980.0160(0.0189) (0.0199) (0.0217) (0.0220) (0.0200) River Water 0.0555***0.01690.02490.0219 0.0254(0.0195) (0.0207) (0.0225) (0.0228) (0.0207) Tap Water 0.0856***0.02670.03690.002480.0172(0.0200) (0.0214) (0.0233) (0.0235) (0.0212) Borehole Water 0.0356* 0.0070-0.0115 0.03310.0221(0.0205) (0.0215) (0.0236) (0.0240) (0.0216) Toilets - Boys 0.000030.000140.00049-0.00030 -0.00215** (0.0008) (0.0009) (0.0010) (0.0011) (0.0010) Toilet - Girls 0.00090.0025** 0.0011-0.0019 -0.0002 (0.0009) (0.0012) (0.0012) (0.0012) (0.0011) Toilet - Male Teachers 0.0118** 0.00640.0044-0.0026 -0.0083 (0.0059) (0.0064) (0.0074) (0.0072) (0.0060) Toilet - Female Teachers 0.00010.0010** 0.0000-0.0004 -0.0006 (0.0010) (0.0004) (0.0004) (0.0007) (0.0005) Constant -0.168*** -0.0218 0.02430.133***0.0258(0.0332) (0.0382) (0.0422) (0.0404) (0.0377) N5249252492523695249252492R-Squared 0.050.060.040.080.07School Fixed Effects Y Y Y Y Y Year Fixed Effects Y Y Y Y Y Controls Y Y Y Y Y Initial controls x year Y Y Y Y Y School Cluster Y Y Y Y Y 172 and statistically in significant. Overall, this paper finds some evidence of heterogeneous treatment effects for off Œgrid electrification. Part 2: Results by Gender Table 3. 7 explores heterogeneity by gender. For conciseness, this paper only reports the preferred full Speci fication (3) which is a fixed effe cts panel with school controls and controls for time Œvarying confounders (the interaction between initial controls and time), in addition to standard errors clustered at the school level. Starting with test scores, electr ification has a positive impacts for both boys and girls but the estimates are larger for girls. In addition, the estimates tend to be larger for off Œgrid than grid electrification. On the other hand, the estimates are not statistically significant. Conseq uently, this paper finds no evidence that the impact of electrification on test scores varies by gender. Similarly, enrollment results shows a positive effect of electrification that are larger for girls than for boys and larger for off Œgrid than grid. How ever, these estimates are small and statistically insignificant. Turning to completion, off Œgrid electrification continues to have larger impacts, though quantitatively small, relative to grid electrification. Off Œgrid has positive effects while grid has n egative effects. The impacts on girls tend to be larger but statistically insignificant. The only statistically significant result in this analysis is that grid electrification decreases enrollment of boys by 0.8%. In summary, while there appear small diff erences in outcomes between boys and girls, the difference tend to be statistically insignificant. However, there is some suggestive evidence that grid electrification may draw boys away from school and hence decreasing enrollment . This can occur if electr ification improves economic outcomes that require low skill s. 173 Table 3.7: Heterogeneous Impacts by Gender Œ Test scores, Enrollment, and Completion Errors clustered at school level. Standard errors in pare nthesis. * p<0.10, ** p<0.05, *** p<0.01. Boys Girls Boys Girls Boys Girls Off-Grid Electricity 0.00600.01660.00500.00690.00860.01(0.0133) (0.0136) (0.0089) (0.0104) (0.0072) (0.0081) Grid Electricity 0.00180.00740.002050.0024-0.0078* -0.0028 (0.0078) (0.0074) (0.0056) (0.0059) (0.0044) (0.0044) Enrolment Boys (8th Grade) -0.0133*** -0.0107*** --0.0291***-0.0001 (0.0013) (0.0011) --(0.0028) (0.0007) Enrolment Girls (8th Grade) -0.0086*** -0.0095*** ---0.0021* 0.0319***(0.0008) (0.0007) --(0.0012) (0.0015) Enrolment Boys (1st -7th Grade 0.0007***0.0004***0.0001380.0007***-0.0014*** 0.0001(0.0002) (0.0002) (0.0001) (0.0001) (0.0003) (0.0001) Enrolment Girls (1st -7th Grade 0.0003* 0.0005***0.0007***-0.00003 0.0006***-0.0008*** (0.0002) (0.0002) (0.0001) (0.0001) (0.0002) (0.0002) Books 4-8th Grade (,00s) -0.00007** -0.0001*** 0.00002-0.00005 0.000020.000028(0.0000) (0.0000) (0.0000) (0.0001) (0.0001) (0.0000) Total Classrooms -0.0015 -0.0004 0.0030** 0.0036***0.00010.0008(0.0010) (0.0008) (0.0012) (0.0012) (0.0006) (0.0005) Rain Water 0.004830.0235-0.0176 -0.00459 -0.0035 -0.0056 (0.0199) (0.0193) (0.0135) (0.0142) (0.0105) (0.0108) River Water 0.02860.0347* -0.0218 -0.00587 -0.0102 -0.0159 (0.0205) (0.0198) (0.0141) (0.0146) (0.0111) (0.0113) Tap Water 0.0353* 0.0394* -0.0076 -0.0146 -0.0171 -0.0197* (0.0213) (0.0205) (0.0144) (0.0153) (0.0115) (0.0116) Borehole Water 0.01650.0195-0.0134 -0.0122 -0.0103 -0.0117 (0.0212) (0.0207) (0.0144) (0.0153) (0.0112) (0.0119) Toilets - Boys -0.0010 0.00080.00030.00080.000660.00026(0.0008) (0.0008) (0.0013) (0.0009) (0.00107) (0.00052) Toilet - Girls 0.00060.00060.0024***0.00174* 0.0012*0.0009(0.0009) (0.0010) (0.0010) (0.0010) (0.0008) (0.0006) Toilet - Male Teachers 0.00230.00310.0108** 0.0122** -0.00179 -0.0023 (0.0065) (0.0060) (0.0050) (0.0048) (0.0040) (0.0036) Toilet - Female Teachers -0.0001 0.0004-0.0003 0.0002-0.00049 0.0004(0.0006) (0.0005) (0.0003) (0.0004) (0.0004) (0.0003) Constant 0.029-0.0314 2.592***2.571***2.387***2.233***(0.0400) (0.0358) (0.0293) (0.0300) (0.0553) (0.0282) N524925249252202522295232352364R-Squared 0.070.060.010.010.310.41School Fixed Effects Y Y Y Y Y Y Year Fixed Effects Y Y Y Y Y Y Controls Y Y Y Y Y Y Initial controls x year Y Y Y Y Y Y School Cluster Y Y Y Y Y Y Test Scores Log Enrolment Log Completion 174 VII. ROBUSTNESS CHECKS Part 1: School type and location In this section , this paper pursues a number of robustness checks. This section considers the effect of sample restrictions in terms of school ownership and sch ool location. Previously, the paper argues that private schools might endogenously select to get electricity or to locate near the grid network. Consequently, the previous analysis restricted the sample to public schools. However, one can argue that this s ample restriction is can introduce other problems . Specifically, schools in urban areas that do not have electricity are likely to be negatively selected. In other words, given the prevalence of infrastructure and household electrification in urban areas, the lack of electricity at a school can signal that these are poor schools such as those in informal settlements. In addition, students attending those schools are also likely to come from poor backgrounds. This can in turn bias the effect of electrificati on upward if large effects are likely to occur among schools that have limited inputs. On the other hand, if grid only functions as a complement to other school and home inputs, some of which are unobservable, inclusion of urban schools is likely to bias e stimates downwards. Finally, the negative selection of schools without electricity in urban areas implies that electrification of a school is less likely to have any impact on students dropping out to pursue jobs from electrification since the lack of elec trification at school is not driven by distance from the grid network. Thus, electrification of urban schools may not change the local labor market conditions. Table 3. 8 column (2) addresses this question. Each column re Œestimates the preferred panel fixed effects specification that includes school controls, an interaction between initial school controls and time, and clusters standard err ors at the school level. Table 3. 8 column (1) reports the original results from the main section for comparison purposes while column (2) restricts analysis 175 to only rural public schools. Moving down, the column (2) and comparing estimates to column (1), the results are quantitatively similar for test scores, enrollment and school completion. However, the coefficient for gr id becomes weakly significant ( at 10% level ) for grid but remains quantitatively small. Overall, the results are qualitatively similar to the main results. While this paper argues for exclusion of certain school categories due to endogeneity issues, a num ber of objections can be made. For instance, the type of students who attend private school may be different from those who attend public schools both in individual and household characteristics. Wealthy household may enroll their children in private schoo ls, which tend to have better inputs than public schools. In addition, parents may also enroll low Œperforming students who need more teacher attention in private schools. This self Œselection can create problems in estimation. For example, if parents enroll bright students in public schools and low Œperforming students in private schools, exclusion of private schools will yield upwardly biased estimates if bright students attend public school and are more responsive to improved lighting from electrification. Table 3. 8 column (3) investigates this question by first, re Œestimating the model with both private and public schools in rural areas. Starting with test scores in the first panel, the estimates of off Œgrid become smaller and turn negative but remain qua ntitatively small and statistically insignificant, while grid estimates remain similar. In the second panel of column (3) enrollment estimates remains similar to the main results in column (1) and the results in column (2) . Similarly, in column (3), off Œgrid estimates remain unchanged for completion. However, while the magnitude of the coefficient decreases marginally, grid coefficient becomes statistically significant at 5%. Overall, while there are some changes in estimates from inclusion of more school 176 categories, the changes are largely quantitatively small and estimates remain qualitatively similar to the main results . Table 3.8: School Sample Restriction Based on Ownership and Location (rural/urban) Errors clustered at school level. Standard errors in parenthesis. * p<0.10, ** p<0.05, *** p<0.01. (1) (2) (3) (4) All Public Rural Public All Rural All Schools Off-Grid Electricity 0.00200.0010-0.0006 -0.0030 (0.0140) (0.0141) (0.0136) (0.0133) Grid Electricity -0.0018 -0.0025 -0.0013 0.0002(0.0079) (0.0080) (0.0078) (0.0077) N52492499805747964187R-Squared 0.060.060.060.06Off-Grid Electricity 0.00710.00700.00860.0079(0.0069) (0.0069) (0.0068) (0.0067) Grid Electricity 0.00260.00330.00370.0036(0.0041) (0.0042) (0.0041) (0.0040) N52366498635724463872R-Squared 0.020.020.020.02Off-Grid Electricity 0.0106**0.0099*0.0109**0.0117** (0.0053) (0.0053) (0.0051) (0.0051) Grid Electricity -0.0045 -0.0050* -0.0059** -0.0051* (0.0029) (0.0029) (0.0029) (0.0029) N52492499805747964187R-Squared 0.440.450.410.39School Fixed Effects YYYY Year Fixed Effects YYYY Controls YYYY Initial Controls x Year YYYY School Cluster YYYY Restriction Public Rural Public Rural None Log of Completion Log of Enrolment School Mean Scores (Standard Deviations) 177 In column (4) of table (6), the results are derived from inclusion of all schools nationwide both private and public in rural and urban areas . While the elect rification coefficients switch signs for test scores in panel one, they remain small and statistically insignificant. Panel two of column (4) show that enrollment results are robust to inclusion of private schools. Finally, the last panel shows that off Œgrid coefficients remain the same while grid electrification becomes statistically significant but only at 10% level . The coefficients however remain quantitatively similar. Part 2: Time Œvarying unobserved regional factors One strategy of dealing with time Œvarying unobserved factors was to include an interaction term between initial school characteristics and time. To the extent that these unobserved factors are correlated with initial school characteristics, the interaction term will address part of the con cern. However, these interactions only capture school Œlevel time Œvarying factors. In this section, the paper attempts to further control time varying regional factors by interacting region and time. Three regional levels are considered, sub Œcounty and coun ty, in order of increasing magnitude. The results are reported in Table 3. 9. The first column reproduces the main results. The second column includes a county time trend while the third column uses a finer regional trend (sub Œcounty trend). Focusing on col umn (3), starting at panel one, the coefficients become smaller in absolute value and turn negative for off Œgrid electrification while the grid coefficient remain negative. In panel two and three, estimates remain consistent except that the off Œgrid coef ficient becomes statistically weakly significant ( at 10% level from 5% level ) for completion. Generally speaking, the results are robust to inclusion of regional time trends. 178 Table 3.9: Inclusion of Regional Time Trends Errors clustered at school level. Standard errors in parenthesis. * p<0.10, ** p<0.05, *** p<0.01. Part 3: Clustering of standard errors Standard errors should be clustered if there are concerns about correlation between observations within clusters. Abadie et al. (2017) discuss at length the common misconceptions and confusions (1) (2) (3) Off-Grid Electricity 0.0020-0.0099 -0.0076 (0.0140) (0.0139) (0.0142) Grid Electricity -0.0018 -0.0081 -0.0057 (0.0079) (0.0079) (0.0080) N524925249252492R-Squared 0.060.090.12Off-Grid Electricity 0.00710.0161**0.0098(0.0069) (0.0070) (0.0072) Grid Electricity 0.00260.00340.0020(0.0041) (0.0041) (0.0042) N523665236652366R-Squared 0.020.040.06Off-Grid Electricity 0.0106**0.0139***0.0104* (0.0053) (0.0054) (0.0056) Grid Electricity -0.0045 -0.0038 -0.0048 (0.0029) (0.0029) (0.0029) N524925249252492R-Squared 0.440.440.46School Fixed Effects YYY Year Fixed Effects YYY Controls YYY Initial Controls x Ye YYY School Cluster YYY Regional Trend County Subcounty School Mean Scores (Standard Deviations) Log of Enrolment Log of Completion 179 that arise in implementing and justifying clustering. They argue that clustering is either a sampling design or an experimental design issue. It is a sampling issue if data is sampled from the population using clustered sampling design and one would like to use the data to make inferences about the population. On the other hand, clustering becomes necessary due to an experimental design issue if the treatment is clust ered. In this paper, given the panel nature, treatment (electrification) is clustered at the school level since all students face the same treatment. It is also possible that electrification was clustered at regional levels. To i nvestigate this concern, Table 3. 10 reports main estimates using various clusters standard errors Œ school, zone, sub Œcounty, and county Œ in order of increasing regional size. The results are consistent across all cluster Œlevels for test scores and enrollment . Clustering at larger region levels, however, decreases statistical significance from 5% to 10% level for completion. Overall, these are small changes that do not qualitatively affect results. Other concerns The muted effects of electrification on outcomes could be partly dri ven by measurement error in electrification that lead to attenuation bias. Kenya, just like many countries, sometimes experiences power outages. If these outages occur regularly or for extended periods, they can explain the results above. I do not have acc ess to data on outages but future versions of this paper will attempt to gather more data and control for outages. In the meantime, the results should be interpreted as the average treatment effects of being connected to a power source. Teachers are an imp ortant school input that affects educational outcomes. Given that data was missing for 2014, I omitted it as a control variable. This could possibly introduce an omitted variable bias. As robustness check, I restricted my analysis to 2015 and 2016 when dat a was available and found that inclusion or exclusion of number of teachers had no effect on estimates. 180 The estimates are not included for conciseness. In addition, I observe a positive correlation of 0.3 between the number of teachers and the number of te acher toilets in 2015 and 2016. Any bias resulting omitting the number of teachers will be mitigated by the inclusion of teacher toilets as a control. Finally, teacher quality may play a role in outcomes but given the panel structure of the model, the chan ges in composition of teachers at a school based on academic qualification is likely to be minimal. Table 3.10: Results by Cluster Level Œ School, Zone, Sub Œcounty, and County Errors clustered at school le vel. Standard errors in parenthesis. * p<0.10, ** p<0.05, *** p<0.01. Cluster level School Zone Sub-County County Off-Grid Electricity 0.00200.00200.00200.0020(0.0140) (0.0151) (0.0152) (0.0149) Grid Electricity -0.0018 -0.0018 -0.0018 -0.0018 (0.0079) (0.0083) (0.0083) (0.0092) N52492524925249252492R-Squared 0.060.060.060.06Off-Grid Electricity 0.007050.007050.007050.00705(0.0069) (0.0071) (0.0070) (0.0063) Grid Electricity 0.00260.00260.00260.0026(0.0041) (0.0043) (0.0042) (0.0037) N52366523665236652366R-Squared 0.020.020.020.02Off-Grid Electricity 0.0106**0.0106**0.0106*0.0106* (0.0053) (0.0050) (0.0054) (0.0053) Grid Electricity -0.0045 -0.0045 -0.0045 -0.0045 (0.0029) (0.0029) (0.0030) (0.0032) N52492524925249252492R-Squared 0.440.440.440.44School Fixed Effects YYYY Year Fixed Effects YYYY Controls YYYY Initial Controls x Year YYYY Cluster level School Zone Sub-County County Log of Completion Log of Enrolment School Mean Scores (Standard Deviations) 181 The results found in this paper could be biased if households selectively migrated toward electrified areas. Such an outcome would bias the grid results if the families that moved towar ds electrified areas valued education more and perhaps invested more in education. The off Œgrid estimat es are unaffected since the solar panels was only installed at school and only supplied power to the specific school . While I cannot directly observe the migration patterns, I believe the resulting migration was minimal. The electrification project was national and occurred rapidly, and consequently most households would have had little incentive to migrate if they anticipated electricity expansion. In ad dition, within this short time frame, it is unlikely that new attractive employments would have been created following electrification. Another concern is that results for test scores and enrollment have very low R 2. This implies that most of the variation in outcomes are not ready explained by the variables included. Finally, since the analysis uses the within school variation for identification, it is extremely hard to pin down the factors that drive these yearly variations because most of the important s chool inputs do not change significantly within the short time frame. On the other hand, the R 2 for completion is quite high because of the high correlation between enrollment and completion. The high R 2 is a result of including enrollment as a control in the regressions for completion. VIII. CONCLUSION This paper sought to quantify the effects of electrification on primary school test scores, enrollment , and completion for students in Kenya. Using the national examination data and school administrative data, th e paper relied on panel fixed effects models . The estimates showed that grid and off Œgrid electrification have no statistically significant effects on the outcomes of interest Œ test scores and enrollment . However, off Œgrid electricity was found to increas e completion by approximately 1%. In addition , there was no evidence that grid and off Œgrid estimates differ in 182 magnitudes except for the positive impact of off Œgrid electricity on completion (1%) . Taken together these estimates show that, in a short Œterm period, electrification may not have any significant impacts on academic outcomes. Since this paper relies on the off Œgrid estimates to identify the mechanism of interest Œlighting Œthe findings above suggest that lighting alone may not be sufficient to induce improved test scores and enrollment both in the short Œterm . This is consistent with previous empirical work s such as Kho, Lakdawala, and Nakasone (2018) and Dasso et al . (2015). On the other hand, the panel estimates suggesting positive and statisti cally significant impact on completion is encouraging and warrants more scrutiny. This paper finds that, relying on the off Œgrid estimates, lighting only has a statistically significant positive impact on completion, which increases by 1% following electri fication. This study documents heterogeneity in results by subject indicating that provision of electricity may affect student or teacher behavior. As such, measures have to be taken to ensure that students do not skew their studies in favor of particular subjects at the expense of others. However, t here is no evidence of difference by in impacts by gender. The location of a school in urban or rural area has little effect on the impact of electrificati on. Finally , inclusion of private schools in the analysi s does not qualitatively affect the results. The policy implication for these findings is that while electrification may not improve academic outcomes in the short run, positive changes can be experienced in t he long run and thus investment in electrificat ion is encouraged. However, to reap the benefits on the electrification, additional short Œterm and long run investments in complementary academic inputs such as books, teachers, and infrastructure should be made. Providing additional lighting at school may not be sufficient. 183 APPENDIX 184 Table A 3.1: Complementarity between School Inputs and Electricity Standard errors in parenthesis. * p<0.10, ** p<0.05, *** p<0.01. School Mean Scores (Standard Deviations) Enrollment (Logs) Completion (Logs) Off-Grid Electricity -0.0274 -0.0362 0.0332**(0.0457) (0.0223) (0.0168) Grid Electricity -0.0055 -0.0067 0.0124*(0.0191) (0.0118) (0.0073) Enrolment Boys (8th Grade) -0.0120*** -0.0145***(0.0012) (0.0014) Enrolment Girls 8th Grade) -0.0092*** -0.0144***(0.0008) (0.0009) Enrolment Boys (1-7th Grade) 0.0006***0.0005***-0.0006*** (0.0002) (0.0001) (0.0001) Enrolment Girls (1-7th Grade) 0.0004*0.0003***-0.0001 (0.0002) (0.0001) (0.0001) Books 4-8th Grade (,00s) -0.0001 0.0005***0.0003**(0.0002) (0.0002) (0.0001) Total Classrooms 0.00070.00140.0007(0.0014) (0.0013) (0.0005) Rain Water 0.0249-0.0159 -0.01 (0.0202) (0.0098) (0.0071) River Water 0.0351*-0.0192* -0.0190** (0.0207) (0.0101) (0.0074) Tap Water 0.0321-0.0149 -0.0220*** (0.0216) (0.0106) (0.0078) Borehole Water 0.0242-0.0177* -0.0147* (0.0218) (0.0106) (0.0078) Student Toilets -0.0008 0.00228***0.001***(0.0008) (0.0005) (0.0003) Teacher Toilets -0.0006** 0.0002-0.0001 (0.0003) (0.0003) (0.0001) Off-grid x Books 4-8th Grade (,00s) -0.00006 -0.0006*** -0.0004*** (0.0003) (0.0002) (0.0001) Grid x Books 4-8th Grade (,00s) 0.0001-0.0006*** -0.0003** (0.0002) (0.0002) (0.0001) Off-grid x Classrooms 0.00250.0062***-0.0005 (0.0044) (0.0023) (0.0016) Grid x Classrooms -0.0024 0.003***-0.0002 (0.0015) (0.0012) (0.0006) Off-grid x Student Toilets 0.00140.0002-0.0002 (0.0018) (0.0009) (0.0007) Grid x Student Toilets 0.0012-0.0015*** -0.0005* (0.0008) (0.0005) (0.0003) Off-grid x Teacher Toilets -0.0047 -0.0062 -0.0059* (0.0099) (0.0043) (0.0033) Grid x Teacher Toilets 0.0051*0.0007-0.0023*** (0.0028) (0.0018) (0.0009) N524925236652492R-Squared 0.060.020.44School Fixed Effects YY YYear Fixed Effects YY YControls YY YInitial controls x year YY YSchool Cluster YY Y185 BIBLIOGRAPHY 186 BIBLIOGRAPHY Abadie, A., Athey, S., Imbens, G. W., & Wooldridge, J. (2017). When should you adjust standard errors for clustering? (No. w24003). National Bureau of Economic Research. Acemoglu, D., Autor, D. H., and Lyle, D. (2004). Women, war and wages: The effect of female labor supply on the wage structure at midcentury. Journal of Political Economy, 112 (3), 497 Œ551. Asian Development Bank (2010). Asian Development Bank™s Assistance for Rural Electrification in Bhutan: Does Electrification Improve the Quality of Rural Life? Imp act Evaluation Study. Manila: Asian Development Bank. Aevarsdottir, A. M., Barton, N., and Bold, T. (2017). The impacts of rural electrification on labor supply, income and health: experimental evidence with solar lamps in Tanzania. Unpublished Manuscript , June . Almond, D, Hoynes, H. W. and Schanzenbach, D. W. (2011). Inside the war on poverty: The impact of food stamps on birth outcomes. The Review of Economics and Statistitcs, 93 (2), 387 Œ403. Arráiz, I., and Calero, C. (2015). From candles to light: th e impact of rural electrification . Inter ŒAmerican Development Bank. Barnes, D. F., Peskin, H., & Fitzgerald, K. (2003). The benefits of rural electrification in India: Implications for education, household lighting, and irrigation. Draft paper prepared fo r South Asia Energy and Infrastructure, World Bank, Washington, DC . Barron, M., and Torero, M. (2014). Short term effects of household electrification: experimental evidence from northern El Salvador. Job Market Paper . Bates, M. N., Chandyo, R. K., Valen tiner ŒBranth, P., Pokhrel, A. K., Mathisen, M., Basnet, S., ... & Smith, K. R. (2013). Acute lower respiratory infection in childhood and household fuel use in Bhaktapur, Nepal. Environmental health perspectives , 121(5), 637 Œ642. Bernard, T. (2012). Impac t analysis of rural electrification projects in sub ŒSaharan Africa. The World Bank Research Observer , 27(1), 33 Œ51. Bernard, T., and Torero, M. (2013). Bandwagon Effects in Poor Communities Experimental Evidence from a Rural Electrification Program in Eth iopia . mimeo. Cockburn, J. C. (2005). Social Impact Evaluation Project'Fund for the Promotion of Micro ŒHydro Power Stations (MHSP)'. Dasso, R., Fernandez, F. and Ñopo, H.(2015). Electrification and Educational Outcomes in Rural Peru. IZA , Discussion Pape r No. 8928. 187 Dinkelman, T. (2011). The effects of rural electrification on employment: New evidence from South Africa. American Economic Review , 101(7), 3078 Œ3108. Furukawa, C. (2014). Do solar lamps help children study? Contrary evidence from a pilot study in Uganda. Journal of Development Studies , 50(2), 319 Œ341. Hassan, F., and Lucchino, P. ( 2016). Powering education. CEP Discussion Paper No. 1438. Hoynes, H. W., and Schanzenback, D. W. (2009). Consumption responses to in Œkind transfers: Evidence from the introduction of the food stamp program. American Economic Journal: Applied Economics , 1(4), 109 Œ139. Kenya Electricity Transmission Company Limited. https://www.ketraco.co.ke/ Khandker, Shahidur R (1996). Education Achievements and School Efficiency in Rural Bangladesh . World Bank Discussion Paper No. 319. Washington, DC: World Bank. Khandker, S., Barnes, D., Samad, H. and Huu Minh, N. (2009a). Welfare impacts of Rural Electrification: Evidence from Vietnam. World Bank Policy Research Working Paper, N. 5057 . Khandker, S. R., Barnes, D. F., & Samad, H. A. (2009b). Welfare impacts of rural electrification: a case study from Bangladesh. World Bank Policy Research Working Paper Series, Vol . Khandker, S. R., Sam ad, H. A., Ali, R., & Barnes, D. F. (2014). Who benefits most from rural electrification? Evidence in India. The Energy Journal , 35(2). Kho, K., Lakdawala, L. K., & Nakasone, E. (2018). Impact of Internet Access on Student Learning in Peruvian Schools (No . 2018 Œ3). Kudo, Y., Shonchoy, A. S., and Takahashi, K. (2017). Can solar lanterns improve youth academic performance? Experimental evidence from Bangladesh. The World Bank Economic Review . Lam, N. L., Smith, K. R., Gauthier, A., & Bates, M. N. (2012). K erosene: a review of household uses and their hazards in low Œand middle Œincome countries. Journal of Toxicology and Environmental Health, Part B , 15(6), 396 Œ432. Libscomb, M.; Mobarak, A. M. and Barham, T. (2013). Development Effects of Electrification: Evidence from Topographic Placement of Hydropower Plants in Brazil. American Economic Journal: Applied Economics , N. 5, pp. 200 Œ231. Lim, S. S., Vos, T., Flaxman, A. D., Danaei, G., Shibuya, K., Adair ŒRohani, H., ... & Aryee, M. (2012). A comparative risk assessment of burden of disease and injury attributable to 67 risk factors and risk factor clusters in 21 regions, 1990 Œ2010: a systematic analysis for the Global Burden of Disease Study 2010. The lancet , 380(9859), 2224 Œ2260. Martins, Johan. 2005. fiThe Impact Use of Energy Sources on the Quality of Life of Poor Communities.fl Social Indicators Research 72(3): 373 Œ402. 188 World Bank. 2008. The Welfare Impact of Rural Electrification: A Reassessment of the Costs and Benefits . Independent Evaluation Group (IEG) Impact Evaluation. Washington, DC: World Bank.