SUSTAINABLE INTENSIFICATION OF MAIZE PRODUCTION IN TANZANIA: EFFECTS ON CHILD NUTRITION, FOOD SECURITY, AND THE ROLE OF INPUT SUBSIDIES By Jongwoo Kim A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Agricultural, Food and Resource Economics Doctor of Philosophy 201 9 ABSTRACT SUSTAINABLE INTENSIFICATION OF MAIZE PRODUCTION IN TANZANIA: EFFECTS ON CHILD NUTRITION, FOOD SECURITY, AND THE ROLE OF INPUT SUBSIDIES By Jongwoo Kim Degraded and infertile soil, low agricultural productivity, and food and nutrition insecurity are persistent and major challenges facing many countries in sub - Saharan Africa (SSA) up to this day. Agr icultural sustainable intensification (SI) has been proposed as a possible solution to simultaneously address these challenges. Yet, there is little empirical evidence on whether SI incomes, nutrition, and food security. The thr ee essays in this dissertation take various quasi - experimental approaches to investigate child nutrition and household food security effects of SI and examine the role of input subsidies in promoting SI using nationally - representative household panel surve y data from Tanzania. In the empirical analysis, I focus on three important soil fertility management (SFM) practices in Tanzanian maize - based production systems: the use of inorganic fertilizer, the use of organic fertilizer, and maize - legume intercroppin g. I group the eight possible combinations of these technologies into - - legum e maize - legume intercropping). This categorization is used in all three essays. In essay 1, results from a multinomial endogenous treatment effects model suggest that - for - age z - score and weight - for - age z - score, particularly for children beyond breastfeeding age (i.e., those age 25 - 59 months). I also find evidence that these effects come through both productivity and income pathways, and that the combined use of inorganic fertilizer and maize - legume intercropping is a key driver o f these effects on child nutrition. Essay 2 investigates the extent to which the use of practices in each SI category influences household net crop income (per acre and per adult equivalent) and crop productivity as well as household food access (modified household dietary diversity score (HDDS), food expenditure per adult equivalent, and food consumption score (FCS)). Results from a - of practices in each of the ot her SI categories has a positive and significant effect on a - related outcomes and crop productivity. Importantly, for these - in - magnitude effects compared to increases in all three food access outcomes, with the size of these effe cts similar to or greater National Agricultural Input Voucher Schem e (NAIVS), encouraged or of practices in the various SI categories on their maize plots using a multinomial logit model combined with the control function approach. I find statistically significant positive effects of household rec eipt of a NAIVS voucher for inorganic fertilizer on maize - fertilizer with organic fertilizer and/or maize - ). On the other Copyright by JONGWOO KIM 2019 v ACKNOWLEDGEMENTS First and foremost, I would like to express my deepest appreciation to my major professor, Dr. Nicole Mason, for being a wonderful mentor and for her constant support and encouragement throughout my PhD program. I am forever indebted to her . I would also l ike to thank Dr. Sieglinde Snapp, my external committee member, for her support and willingness to provide clear and excellent feedback on my dissertation. I am also grateful for detailed comments and contributions from my other committee members; Dr. Robe rt Myers and Dr. Felicia Wu. It was a tremendous privilege to meet and work with them. I would also like to thank Dr. David Mather who shared critical data for my dissertation and provided detailed suggestions regarding my dissertation . My appreciation also extends to other members in our department; Dr. Scott Swinton, a former graduate program director, as well as Ashleigh Booth and Nancy Creed . I owe much gratitude to them for all their kindness and help whenever I visit ed room 202. I would also like to give special thanks to my wife, Jungmin Lim. During the entire graduate program in AFRE at MSU, she was my be st friend and a lovely colleague who always believed in me, while at the same time has been the best mom to our young son, Kyle. Without her unwavering love, support, and encouragement, I could not have completed my graduate studies. And l ast, but not leas t, I would also like to thank my family in Korea my parents, Eunbae Kim and Boksun Song , as well as my mother - in - law, Hyunsook Ma , for their constant love and support. For without them, none of this would have been possible. I also grateful acknowledge funding support for this research from the Feed the Future (FtF) Innovation Lab for Sustainable Intensification [grant number AID - OAA - L - 14 - 00006] and vi the Feed the Future Innovation Lab for Food Security Policy [grant number AID - OAA - L - 13 - 00001] (Tanzania ASPIRES). vii TABLE OF CONTENTS LIST OF TABLES ix LIST OF FIGURES xi i i KEY TO ABBREVIATIONS xi v INTRODUCTION REFERENCES 5 CHAPTER 1 DOES SUSTAINABLE INTENSIFICATION OF MAIZE PRODUCTION ENHANCE CHILD NUTRITION? EVIDENCE FROM RURAL TANZNIA 7 1.1 Introduction 7 1.2 Sustainable intensification of maize production in Tanzania 10 1.3 Conceptual and econometric framework 13 1.3.1 Conceptual framework 13 1.3.2 Multinomial endogenous tr eatment effects model 15 1.4 19 1.5 22 1.6 29 31 45 CHAPTER 2 THE IMPACTS OF SUSTAINABLE INTENSIFICATION OF MAIZE PRODUCTION ON HOUSEHOLD CROP INCOME, PRODUCTIVITY, AND FOOD ACCESS IN RURAL TANZANIA 52 2.1 Introduction 52 2.2 SI of maize production in Tanzania 58 2.3 Empirical strategy 61 2.3.1 CRE - MNLS model 62 2.3.2 CRE - MESR model 63 2.3.3 Estimation of average treatment effects on the treated (ATT) 66 2.4 Data and key outcome variables 68 2.4.1 Data 68 2.4.2 Outcome variables and explanatory variables 69 2.5 Results and discussion 76 2.5.1 Impacts of using practices in each SI category on household income and productivity 77 2.5.2 Impacts of using practices in each SI category on household food access outcomes 80 2.6 Conclusions and policy implications 84 86 120 viii CHAPTER 3 THE EFFECTS OF THE NATIONAL AGRICULTURAL INPUT VOUCHER SCHEME (NAIVS) ON SUSTAINABLE INTENSIFICATION OF MAIZE PRODUCTION IN TANZANIA 125 3.1 Introduction 125 3.2 B ackground: SI of maize production & the NAIVS program in Tanzania 130 3.2.1 SI of maize production in Tanzania 130 3.2.2 The National Agricultural Input Voucher Scheme 132 3.3 Methodology 139 3.3.1 Conceptual framework 139 3.3.2 Estimation strategy 140 3.4 Data and description of variables 144 3.4.1 Data 144 3.4.2 Explanatory variables 146 3.5 Results 153 3.5.1 Test for endogeneity of household receipt of NAIVS voucher 153 3.5.2 APEs of NAIVS voucher receipt on househol d use of practices in the various SI categories 153 3.6 Conclusions and policy implications 159 162 174 ix LIST OF TABLES Table 1.1 : SI of maize production categories and prevalence on maize plots and among maize - growing households in Tanzania 13 Table 1.2: FE, CRE - POLS, and CRE - METE estimates: Impacts on nutritional outcomes of children aged 6 - 59 months (full sample) 23 Table 1.3: CRE - METE estimates: Impacts on child nutritional outcomes with sub - sample analysis 26 Table 1.4: CRE - METE estimates: Impacts on crop income and productivity 28 Table 1A.1: Falsification test results (parameter estimates from CRE - POLS regressions for - 33 Table 1A.2: Summary statistics for child nutritional status and child characteristics used in the analysis 35 Table 1A.3: Summary statistics for other control variables used in the analysis 36 Table 1A.4: CRE mixed multinomial logit estimates of the determinants of adoption of each SI on - 38 Table 1A.5: Second stage estimates for child nutritional outcomes 40 Table 1A.6: CRE - 42 Table 1A.7: CRE - METE model estimates: Impacts of the adoption of inorganic fertilizer, maize - legume intercropping, and their joint use on child nutritional outcomes full and sub - sample analysis 43 Table 1A.8: Percentage of sample households producing legumes by household - level SI category 44 Table 2.1: SI categories and prevalence on maize plots and among maize - growing households in Tanzania 60 Table 2.2: Descriptive statistics by survey round 74 Table 2.3: ATTs of using practices in each SI category on household net crop income and productivity 80 Table 2.4: ATTs of using practices in each SI category on household food access outcomes 83 x Ta ble 2A.1: Summary statistics by SI category 87 Table 2A.2: CRE - MNLS estimates for net crop income (1,000 TZS) 90 Table 2A.3: CRE - MNLS estimates for net crop income (1,000 TZS) per acre 91 Table 2A.4: CRE - MNLS estimates for net crop income (1,000 TZS) per adult equivalent 92 Table 2A.5: CRE - MNLS estimates for crop productivity 93 Table 2A.6: CRE - MNLS estimates for modified HDDS 94 Table 2A.7: CRE - MNLS estimates for food expenditure per adult equivalent 95 Table 2A.8: CRE - MNLS estimates for FCS 96 Table 2A.9: CRE - MESR second stage estimation results for net crop income (1,000 TZS) 97 Table 2A.10: CRE - MESR second stage e stimation results for net crop income (1,000 TZS) per acre 99 Table 2A.11: CRE - MESR second stage estimation results for net crop income (1,000 TZS) per adult equivalent 101 Table 2A.12: CRE - MESR second stage estimation results for crop productivity 103 Table 2A.13: CRE - MESR second stage estimation results for modified HDDS 105 Table 2A.14: CRE - MESR second stage estimation results for food expenditure (1,000 TZS) per adult equivalent 107 Table 2A.15: CRE - MESR second stage estimation results for FCS 109 Table 2A.16: Average treatment effects on the untreated (ATU) of using practices in each S I category on household net crop income, productivity and food access outcomes 111 Table 2A.17: Average treatment effects of using practices in each SI category on household income, productivity, and food access outcomes (Unconditional average effect s) 112 Table 2A.18: Marginal effects of use of practices in each SI category on net crop income (1,000 TZS) 113 Table 2A.19: Marginal effects of use of practices in each SI category on net crop income (1,000 TZS) per acre 114 xi Table 2A.20: Marginal effects of use of practices in each SI category on net crop income (1,000 TZS) per adult equivalent 115 Table 2A.21: Marginal effects of use of practices in each SI category on crop productivity 116 Table 2A.22: Marginal effects of use of practices in each SI category on modified HDDS 117 Table 2A.23: Marginal effects of use of practices in each SI category on food expenditure per adult equivalent 118 Table 2A.24: Marginal effects of use of practices in each SI category on FCS 119 Table 3.1: SI of maize production categories and prevalence on maize plots in the sample 132 Table 3.2: Household beneficiaries for the NAIVS 134 Table 3.3: Number and percentage of ru ral maize - growing households that received versus redeemed a NAIVS voucher by input voucher type received 137 Table 3.4: Number and percentage of maize plots owned by NAIVS voucher recipients vs. non - recipients under SI category 139 Table 3.5: Summary statistics for the variables used in the analysis 148 Table 3.6: APEs of NAIVS voucher receipt and redemption on household use of practices in the various SI categories 154 Table 3.7: APEs of NAIVS voucher receipt and redemption on household sole or joint use of inorganic fertilizer and maize - legume intercropping 157 Table 3.8: APEs of NAIVS voucher receipt and redemption on household sole or joint use of in organic fertilizer and organic fertilizer 158 Table 3A.1: Reduced form CRE logit regression estimates of factors affecting household NAIVS voucher receipt 163 Table 3A.2: Reduced form CRE logit regression estimates of factors affecting household NAIVS voucher redemption 165 Table 3A.3: CRE - MNL with CF regression results (relative log odds) 167 Table 3A.4: CRE - MNL without CF regression results (relative log odds) 169 xii Table 3A.5: APEs of other (non - NAIVS - related) factors affecting household use of practices in the various SI categories 172 xiii LIST OF FIGURES Figure 1.1: Conceptual pathways between SI of maize production and child nutrition 15 Figure 1A.1: 32 Figure 2.1: Conceptual pathways between SI of maize production and household food access 56 xiv KEY TO ABBRE VIATIONS AMIS Agricultural Market Information System APE Average partial effect ATT A verage treatment effects on the treated ATU A verage treatment effects on the un treated BMI Body mass index CA Conservation Agriculture CF Control function CRE Correlated random effects FCS Food consumption score FE Fixed effects HAZ Height - for - age z - score HDDS Household dietary diversity score IIA Independence of irrelevant alternatives IMR Inverse Mills ratio ISFM Integrated Soil Fe rtility Management ISP Input subsidy program IV Instrumental variable LSMS - ISA Living Standards Measurement Study - Integrated Surveys on Agriculture METE Multinomial endogenous treatment effects MESR Multinomial endogenous switching regression MIT Ministry of Industry and Trade xv MNL Multinomial logit MNLS M ultinomial logit selection NAIVS National Agricultural Input Voucher Scheme OLS Ordinary least squares SACCOS Savings and Credits Cooperatives Societies SFM Soil fertility management SI Sustainable intensification SOC Soil organic carbon SOM Soil organic matter SSA Sub - Saharan Africa TNBS Tanzania National Bureau of Statistics TNPS Tanzania National Panel Surveys TZS Tanzania Shillings WAZ Weight - for - age z - score WHO W orld Health Organization 1 INTRODUCTION Degraded and infertile soil, low agricultural productivity, and food and nutrition insecurity are persistent and major challenges facing many countries in sub - Saharan Africa (SSA) up to this day. In 2017, about 236 million people (23.2% of the population) in SSA were undernourished and approximately 40% of globally stunted children under age five (151 million children) lived in Africa (FAO, IFAD, UNICEF, WFP, and WHO, 2 018). Agriculture in this region is particularly important to alleviate food and nutrition insecurity since most undernourished people reside in rural areas and many of them are small - scale farm households ( Sibhatu et al., 2015 ). This implies that househol could play a crucial role in addressing these challenges, including low agricultural productivity gh - yielding crop varieties could improve the food security and nutritional status of household members by to access diverse and nutritious foods (Jones et al. , 2014). However, there is an emerging consensus that the sole use of these practices, or conventional intensification of agricultural systems, may not be sufficient to sustainably intensify agricultural production and thus it may not be a solution to achi eving food and nutrition security (The Montpellier Panel, 2013; Kassie et al., 2015). Nonetheless, many African governments have still concentrated on encouraging - scale input subsidy prog rams (ISPs) (Jayne et al., 2018). In this context, agricultural sustainable intensification (SI) has received a great deal of attention as a possible solution to address these challenges (The Montpellier Panel, 2013; Petersen and Snapp, 2015). At the core of SI is the goal of improving not only agricultural yields but also nutrition and food security without bringing new land under 2 cultivation, while preserving or enhancing the natural resource base (Pretty et al., 2011; Godfray et al., 2010; Loos et al., 2 014). However, there is little empirical evidence on whether SI indeed The three essays in this dissertation take various quasi - experimental approaches to investigate child nutrition and househol d food security effects of SI and examine the role of input subsidies in promoting SI using nationally - representative household panel survey data from Tanzania. For the empirical analysis, I focus on three important soil fertility management (SFM) practice s in Tanzanian maize - based production systems: the use of inorganic fertilizer, the use of organic fertilizer, and maize - legume intercropping. Given eight possible combinations of these - (use of none of the fertilizer, maize - with organic fertilizer and/or ma ize - legume intercropping). This categorization is used in all three essays. Results from this dissertation will help policymakers understand potential impacts of each SI category on nutrition and food security as well as design agricultural policies for pr omoting SI in maize - based systems. affects the nutritional status (height - for - age z - score (HAZ) and weight - for - age z - score (WAZ)) of household members under age five using a multinomial endogenous treatment effects model. and WAZ, particularly for children beyond breast - feeding age (i.e., those age 25 - 59 months). I also find evidence that these effects come through both productivity and income pathways, and 3 that the combined use of maize - legume intercropping and inorganic ferti lizer is a key driver of the The second essay ( Chapter 2) investigates the exten t to which the use of practices in each SI category influences household net crop income - related variables (net crop income, net crop income per acre, and net crop income per adult equivalent) and crop productivity. Furthermore, I examine whether the use o household dietary diversity score (HDDS), food expenditure per adult equivalent, and food consumption score (FCS)). Results in a multinomial endogenous switching regression model suggest t - crop income - related outcomes and crop productivity. For these outcom es, use of practices in the - associated with increases in all three food access outcomes. These effects are consistently larger ac income pathways. The results also shed light on the findings in Essay 1 and suggest that the driven by improvements in both the quantity and diversity of food items consumed (including legumes produced via maize - legume intercropping). 4 National Agricultural Input Vou cher Scheme (NAIVS), encourages or of individual SFM practices and/or a combination thereof on their maize plots. Understanding implications as i t relates to agricultural productivity, food security, poverty, long - run soil health, and the returns to government spending on ISPs. Using a multinomial logit model combined with the control function approach, I find s tatistically significant positive eff ects of household receipt of a NAIVS voucher for inorganic fertilizer on maize - average 10.0 percentage points higher than for households who do not receive a NAIVS fertilizer voucher. Results further suggest that NAIVS voucher receipt encourages farmers to adopt multiple SFM practices that could contribute to SI. More specifically, NAIVS voucher receipt for inorganic fertiliz er is associated with a 9.6 percentage point increase in the probability of has no The positive effects of NAIVS on joint use of inorganic fertilizer with organic SFM practices is encouraging, as it suggests that the program may have helped promote not just short - run increases in maize yields but also longer - term improvements in soil health. 5 REFERENCES 6 REFERENCES FAO, IFAD, UNICEF, WFP, and WHO. 2018. The State of Food Security and Nutrition in the World 2018. Building climate resilience for food security and nutrition. Rome, FAO. Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., Pretty, J., Robinson, S., Thomas, S.M., and Toulmin, C. 2010. Food security: the challenge of feeding 9 billion people. Science 327(5967):812 - 818. Jayne, T.S., Mason, N. - generation agricultural input subsidy programs. Food Policy 75:1 - 14. Jones, A.D., Shrinivas, A., and Bezner - Kerr, R. 2014. Farm production diversity is associated with greater household dietary diversity in Malawi: Findings from nationally representative data. Food Policy 46:1 - 12. Kassie, M., Teklewold, H., Jaleta, M., Marenya, P., and Erenstein, O. 2015. Understanding the adoption of a portfolio of sustainable intensification practices in eastern and southern Africa. Land Use Policy 42:400 - 411. Loos, J., Abson, D.J., Chappell, M.J., Hanspach, J., Mikulcak, F., and Tichit, M. 2014. Putting meaning back into sustainable intensification. Frontiers in Ecology and the Environment 12:356 - 361. Petersen, B., and Snapp, S.S. 2015. What is sustainable intensification? Views from expert. Land Use Policy 46:1 - 10. Pretty, J., Toulmin, C., and Williams, S. 2011. Sustainable intensification in African agriculture. International Journal of Agricultura l Sustainability 9(1):5 - 24. Sibhatu, K.T., Krishna, V.V., and Qaim, M. 2015. Production diversity and dietary diversity in Proceedings of the National Academy of Sciences 112(34):10657 - 10662. The Montpellier Panel, 2013. Sustainable Intensification: A N ew Paradigm for African Agriculture. Agriculture for Impact, London. 7 CHAPTER 1 DOES SUSTAINABLE INTENSIFICATION OF MAIZE PRODUCTION ENHANCE CHILD NUTRITION? EVIDENCE FROM RURAL TANZNIA This essay has been accepted for publication in Agricultural Economic with Jongwoo Kim as the lead and corresponding author. The citation is as follows: Kim, J., Mason, N.M., Snapp, S., and Wu, F. (in press). Does sustainable intensification of maize production enhance child nutrition? Evidence from rural Tanzania. Agricultural Economics. 1. 1 Introduction Food insecurity and malnutrition continue to be urgent global problems. Although increases in agricultural productivity ha ve dramatically improved food and nutrition security in many parts of the world over the past five decades , approximately 795 million people worldwide remain undernourished and most of them liv e in developing countries (Godfray et al. , 2010; FAO , 2015; Koppmair et al. , 2016 ) . H unger and child malnutrition are especially serious problems in sub - Saharan Africa (SSA) . For example, i n 2017, g lobally about 15 1 million children under age five were stunted and more than one third of the se children live d in Africa (UNICEF , WHO , and World Bank Group , 201 8 ) . Moreover, a pproximately 45% of g l obal deaths of children under age five are linked to malnutrition and the mortality rate of children in SSA is the highest in the world ( Black et al. , 2013 ) . Agriculture and n utrition are closely linked because the majority of undernourished people live in rural areas and many of them are smallholder farmers (Sibhatu et al. , 2015 ; Pinstrup - Andersen , 2007) . This linkage suggests that agricultural intensification via adoption of improved inputs and management practices may improve the nutritional status of nutritionally vulnerable household members including young children by enhancing the 8 agricultural production, productivity , and/or income , as well as by providing better access to more diverse or nutritious foods (Jones et al., 2014; Hawkes and Ruel, 2006) . However, there is an emerging consensus that conventional agricultural intensification via high - yielding crop varieties and inorganic fertilizer may be in sufficient to sustainably raise agricultural productivity and could have negative environmental conseq uences (Pingali , 2012; Montpellier Panel, 2013 ). Moreover, in many parts of SSA, rapidly growing population s and a lack of new land to farm has led to continuous cultivat i on of plots and r educed fallowing, thereby degrading soils and adversely affecting crop yields and yield response to inorganic fertilizer (Kassie et al. , 2013 ; Tully et al. , 2015 ; Jayne et al., 2018 ). Agricultural sustainable intensification (SI) ha s been proposed as a possible solution to address these challenges ( Montpellier Panel, 2013; Petersen and Snapp , 2015 ) . At the core of SI is the goal of producing more food from the same area of land while reducing the environmental impacts ( Godfray et al. , 201 0 , p . 813 ) . B roader definitions of SI also encompass the complex social dimensions of sustainability , including nutrition and food security ( Loos et al. , 2014; Musumba et al. , 201 7 ) . It is an open question, however, whether the use of agricultural inputs and management practices that contribute to SI from an environmental standpoint do indeed improve nutrition and food security. In this study, we contribute to the thin evidence base on this topic by estimating the effects of SI of maize production on the child nutrition outcomes of maize - growing households in Tanzania . We focus on maize due to its importance as a staple food in Tanzania and because it accounts for approximately 75% of total cropped area in the country (Tanzania National Bureau of Statistic s (TNBS) , 2014). To our knowledge, only two previous studies have examined the relationship between SI of maize production and child nutrition ( Manda et al. (2016 a ) and Zeng et al. (2017) ), and both 9 focus on adoption of improved maize varieties . Yet there are numerous other agricultural practices that can contribute to the SI of maize production and potentially affect child nutrition. In this study, we extend the existing literature and focus on three soil fertility management (SFM) practices: th e use of inorganic fertilizer, the use of organic fertilizer, and maize - legume intercropping. ased on their use of these practices on their maize plots - (use of none of the practices) - legume maize - legume intercropping , which is a form of integrate d soil fertility management (ISFM ; Place et al., 2003 ) ). Using nationally representative household panel survey data from Tanzania, we then estimate how the adoption of these SI categories by maize - growing households affects the nutrition outcomes (height - for - age z - score (HAZ) and weight - for - age z - score (WAZ)) of household members under age five . 1 This study further contributes to the literature in several ways . First, to our knowledge , it is the first empirical investigation of how combinations of agricultural practices in general (as opposed to single technologies) and ISFM in particular affect child nutrition . Second, we explore whether these effects operate through the crop productivity and/or income pathways. Third , we use household - level panel data , whereas Manda et al. ( 2016a ) and Zeng et al. ( 2017 ) use cross - sectional data. This enables us to control for time - constant unobserved heterogeneity , which should improve the internal validity of our estimates. And fourth , we contribute to the production diversity - dietary diversity / nutrition literature (see, for example, Jones et al., 2014; Kumar et al., 1 Several recent studies in Agricultural Economics have examined the determinants of adoption (and/or impacts on outcomes other than child nutrition) of some of these practices or other land management practices in SSA (e.g., Wossen et al., 2015; Abdulai, 2016; Manda et al., 2016 b ; Wainaina et al., 2016; Amare and Shiferaw, 2017; and Schmidt et al., 2017). 10 2015; Sibhatu et al., 2015; Hirvonen and Hoddinott, 2017 ; Parvathi, 2018 ) by studying whether production diversity (proxied in this study by maize - legume intercropping), intensification (proxied by inorganic fertilizer use on maize), or a combination of the two is most beneficial for child nutrit ion outcomes. 2 Our r esults suggest that - adoption of SI is consistently associated with improve ments in child WAZ , particularly f or children beyond breast - feeding age ( i.e., those age 25 - 59 months) . We find evidence that these effects come through both the productivity and income pathways, and that t he combined use of maize - legume intercropping and inorganic fertilizer is a key driver of the effects o n child nutrition. 1.2 Sustainable intensification of maize production in Tanzania This study focus es on inorganic fertilizer, organic fertilizer, and maize - legume intercropping on their maize plots . As mentioned above, we define use of - legume fertilizer, maize - legume int I norganic fertilizer is a key input associated with conventional agricultural intensification and it has been a major reason for the dramatic increase in food production globally over the past 50 y ears ( Crews and Peoples , 2005; Pingali , 2012 ) . However, overuse of inorganic fertilizer can result in pollution of ground and surface water ( Byrnes, 1990; Hart et al., 2004) , and chemical fertilizer application without the use of complementary soil building practices ( e.g., maize - 2 We thank an anonymous reviewer for highlighting this. 11 legume intercropping and organic fertilizer ) may lead to a decrease in soil pH, soil organic carbon, soil aggregation, and microbial communities (Bronick and Lal, 2005). Maize - legume intercropping and the use of o rganic fertilizer in the form of manure or compost are widely recognized as S ustainable agricultural practices by agronomists and soil scientists ( Ollenburger and Snapp, 2014; Droppelmann et al., 2017; Mpeketula and Snapp, 2018 ). 3 Organic fertilizer can be produced in a renewable manner, locally, and can enhance soil structure and water retention capacity , encourage the growth of beneficial mic r o - organisms and earthworms , and decrease bulk density (Chen , 2006; Bronick and Lal , 20 05). However, there are often limitation in terms of locally sourcing large quantities, it has a long - time horizon for observed benefits, and it is often not sufficient to substantially raise productivity. M aize - legume intercropping is another local and renewable source of soil fertility . Moreover, compared to continuous sole - cropped maize , it can improv e soil properties for nutrient and moisture - holding capacity , and reduc e weeds, pests, and diseases ( Snapp et al. , 2010; Tilman et al. , 2002; Woodfine , 2009) . L egumes can also benefit household nutrition, providing needed protein and micronutrients such as iron, zinc, or vitamin A (Messina , 1999). Because of these benefit s , some authors consider maize - legume intercropping to be an SI practice (Rusinamhodzi et al. , 2012) . H owever, maize yields in certain contexts may be negatively affected by intercropping (Agboola and Fayemi , 1971 ; Waddington et al. , 2007) , and intercrop systems generally require complementary investments in order to support high crop yields . For the se reasons, we categorize organic fertilizer and maize - legume intercropping as s but not sufficient to sustain ably intensify maize production without joint use with inorganic fertilizer. 3 We recognize that this designation may not be universally accepted. 12 Table 1 .1 shows the prevalence of each of the eight possible combination s of the three SFM practices and each of the four SI categories . Out of 6,383 maize plots pooled acr oss three rounds of survey data (the Tanzania National Panel Surveys (TNPS) of 2008/09, 2010/11, and 2012/13 , described below ), 38% fall in the - For the empirical approa ch used in this study and described below ( a multinomial endogenous treatment effects ( METE ) model ) , we need to define a household - level SI category variable based on the plot - level SI category information . ( This is because the METE model require s that the ) To do so , we calculat e the total area of a maize plots in each SI category and then choose the SI category that has the largest area . The prevalence of these household - level SI categories is summarized in Table 1 .1 and is very similar to the plot - level results. This is because 64% of households in the sample have only one maize plot , and those with multiple maize plots tend to use the same SFM practic es on a ll maize plots . Overall, 87% of the maize plots in the sample have the same SI category at the plot - and household - level . 4 4 There is considerable variation in a time, which is important for the panel data methods used here . O f sample households that appear in only two survey rounds, 43% changed categories between rounds ; of sample households in all three rounds, 56% changed categories at least once. 13 Table 1.1 : SI of maize production categories and prevalence on maize plots and among maize - growing households in Tanzania Case Inorganic fertilizer Organic fertilizer Maize - legume intercropping % of maize plots SI category % % Plot level HH level 1 46. 5 Non - adopt ion 46. 5 44.3 2 7.3 Intensification 7. 3 6. 1 3 6.3 Sustainable 38. 1 40. 8 4 26.8 5 5.0 6 1.7 SI 8. 1 8.8 7 5.2 8 1.2 Use of inorganic fertilizer 15.4 16.1 Use of organic fertilizer 14.2 18.1 Use of maize - legume intercropping 38.2 46.6 Notes: Figures in the plot level column are based on all maize plots (N=6,383) cultivated by rural households pooled across the three waves of the TNPS (2008/09, 2010/11, and 2012/13). Figures in the HH level column are based on the total number of maize growers (N=4,269) in rural areas across these surveys. Legume crops for maize - legume intercropping are beans, soybeans, groundnuts, cowpeas, pigeon peas, chickpeas, field peas, green grams, bambara nuts, and fiwi. 1. 3 Conceptual and econometric framework 1. 3.1 Conceptual framework Tanzania is the third worst affected country in SSA based on the prevalence of stunting (UNICEF , 2009) . A s of 2012/13, 37.4% of children under age five were stunted (i.e., HAZ < - 2) and 12.5% were underweight (i.e., WAZ < - 2), with the prevalence of malnutrition markedly higher in rural than in urban areas ( TNBS , 201 4 ) . 5 HAZ and WAZ reflect long - term factors such 5 HAZ and WAZ measure nutritional status in the form of z - - for - age and weight - for - age, respectively, with that of a reference population of well - nourished children. The World Health Organization (WHO) Child Growth Standards an d WHO Reference 2007 composite data files are used as the reference data. See Heady et al. (2018) for an analysis of differences in child nutrition between rural and urban areas throughout SSA. 14 as deficiencies in nutrition, frequent in fections, and inappropriate feeding practi ces ( Alderman et al. , 2006; TNBS , 2014). R ecent studies suggest that agricultur al interventions or technologies can affect child nutrition through two main pathways : ( 1) food production /productivity ; and ( 2) agricultural income (Herforth and Harris , 2014; Kumar et al. , 2015 ) . F igure 1 .1 depicts the se pathways in the context of this study . First, - adoption of practices in the other SI categories may directly increase production and/or productivity of maize, a key food staple . A dopting maize - legume intercropping and could directly by provid ing leguminous crops with a range of essential nutrients. More diverse and l arger quantities of produced foods could also mean less needs to be items. Practices in may also increase household crop income through generating larger marketable surplus es of maize and/or legume crops, which , in turn , could raise expenditure s on high calorie and protein - rich foods as well as non - food expenditure s on health service s , sanitation, and access to clean water. A doption of the var i ous SFM practices may also labor burden and time allocatio n, which could affect child nutrition outcomes directly or indir nutrition . As described categories on: (i) the HAZ and WAZ of children under age five in the household, and (ii) crop income from an d productivity on their maize plots. The purpose of (ii) is to explore the pathways through which (i) occur s . 15 Figure 1.1 : Conceptual pathways between SI of maize production and child nutrition. Source: Adapted from Herforth and Harris (2014). 1. 3.2 Multinomial endogenous treatment effects model Because farmers often self - select into agricultural technology adopter group s or some technologies are targeted to certain group s of farmers, selection bias and endogeneity may arise ( Manda et al. , 2016a; Kassie et al. , 2015 b ). In the context of this paper, t h ese problems occur if unobserved factors affecting a SI category adoption decision are correlated with . For example, suppose the head of household is highly motivated and curious, and as a result of these traits, actively seeks out information not only on the benefits of various SFM practices but the adoption of certain SI categories is associated with child nutrition outcomes even if there is no causal relationship. To address these concerns , we use a n METE model ( Deb and Trivedi , 2006a, b) because it allows us to evaluate alternative combinations of practices (SI categories) and corrects for both Adoption of practices in a given maize SI category (Intensification, Sustainable, or SI) Ag. Production/ Productivity Health Status and Nutritional Outcomes Diet Com position of Household Food Expenditure Food Access Agricultural Income Non - food Expenditure Health Care Health Status Labor burden Child Nutritional Outcomes Devoted to Child Care 16 self - selection and the potential interdependence of adoption decisions over SFM practices (Wu and Babcock , 1998; Manda et al. , 2016b ). We combine the METE model with Mundlak - Chamberlain correlated random effects (CRE) techniques to further control for time - invariant unobserved household - level heterogeneity that may be correlated with observed covariates (e.g., motivation in the example above) , where the household means of time - varying household - level explanatory variables are included as additional reg ressors (Wooldridge , 2010) . As a benchmark to the CRE - METE models, we also report household fixed effect s (FE) and CRE - pooled ordinary least squares (POLS) results for the main model below. 6 The METE model involves two stages . In the first stage, household i chooses one of the four SI categor ies . Following Deb and Trivedi (2006a, b), let denote the indirect utility obtained by household i from selecting the j th SI category , : Without loss of generality, let j - is a vector of ex ogenou s covariates (described below) with associated parameters ; are independently and identically distributed error terms ; and is unobserved characteristics common to household i j th alternative and the outcome variables ( HAZ and WAZ ) . is not directly observed but we do observe a vector of binary variables, representing whether a household adopted a given SI category . T he probability of treatment can be expressed as : 6 Note that if all explanatory variables are time - varying, FE and POLS - CRE are algebraically equivalent in linear models. However, several household - level regressors in our models are time - invariant for almost all households (e.g., education of the household head, distance to the nearest market, and a binary variable for livestock ownership); per guidance from J. Wooldridge (personal communication, 2017), we exclude the time averages of these variables from models that use CRE. 17 Following Deb and Trivedi (2006 a ), we assume that has a mixed multinomial logit structure , i.e. : In the second stage, we estimate the impact of the adoption of the various SI categories on HAZ and WAZ using OLS with a selectivity correction term from the first stage . 7 The expected outcome equation is written as : where is the nutrition outcome of interest for child n in household i. is a vector of exogenous covariates including two sub - vectors: household i characteristics and child n characteristics . The associated parameter vector is . Parameters (for ) denote the treatment effects relative to the control group ( Non - adopt ion ). is a function of each of the latent factors ; that is, the outcome variable may be influenced by unobserved characteristics that also affect selection into treatment. If is positive (negative), treatment and outcome are positively (negatively) associated with unobserved variables i.e., there is positive (negative) selection. We assume that the outcome variables (z - scores) follow a normal distribution . The model is estimated using a m aximum s imulated l ikelihood approach and 700 Halton sequence - based quasi - random draws . 8 7 We also wanted to estimate models for the probability of being stunted and underweight bu t these models do not converge. 8 500 Halton sequence - based quasi - random draws are used for the WAZ models in the full - sample analys e s in Table A5 because the models do not converge when 700 are used . 18 In principle, the parameters of the semi - structural model are identified through nonlinear functional forms ; h owever, including some variables in that do not enter in is the preferred approach for more robust identification (Deb and Trivedi , 2006a, b). We therefore include the following as excluded instrumental variables (IVs) : the proportion of other households in the (excluding the household itself) that (i) received agricultural production advice , (ii) that used inorganic fertilizer, and (iii) that used maize - legume intercropping ; (iv) electoral threat at the district level ; and (v) the number of National Agricultura l Input Voucher Scheme (NAIVS) subsidized fertilizer region . 9 The first three IVs are related to access to information on and the potential for social learning about SFM practices. 10 We expect these variables to be positively correlated with household i SFM practices but not to directly affect the child nutrition outcomes . Regarding IVs (iv) and (v), a SI category decision could be affected by its receipt of subsidized fertilizer vouchers ; however, this is likely to be endogenous , so we instead use (iv) and (v) such vouchers but are exogenous to an individual household. Electoral threat , as defined by Chang (2005) , i s the proportion of votes for the runner - up divided by the proportion of votes for the presidential winner . Previous studies indicate that the spatial allocation of subsidized input s in some SSA countries, including Tanzania, may be linked to voting patterns during the most recent election ( see, among others , Mason et al. ( 201 7 ) for Zambia ; Mather and Minde ( 2016 ) for Tanzania ; and Mather and Jayne ( 2018 ) for Kenya ). We therefore use Mather and electoral threat variable, which is 9 We also considered the proportion of other households that used organic fertilizer but it did not pass the falsification test described below. 10 Similar variables have been used as selection instruments by Di Falco et al. (2011), Di Falco and Veronesi (20 13), and Manda et al. (2016a, b). 19 based on data from the 2005 and 2010 Tanzania presidential election s . 11 S ubsidized fertilizer vouchers for maize in Tanzania are also targeted based on the suitability of different areas f or maize production . 12 We therefore include as another IV the number of vouchers allocated to the the World Bank (2014 ) . Although there is no formal test for the validity of exclusion restrictions in a nonlinear setting (Deb and Trivedi , 2006a), we follow Di Falco et al. (20 1 1) and perform a simple falsification test where these candidate IVs are included as additional explanatory variables along with in a CRE - POLS regression , while the dependent variable is the HAZ or WAZ of children in household s - group . If the candidate IVs are not statistically significant in this regression , this lends support to the validity of the exclusion restriction s . All IVs used here pass this simple falsification test (see Table 1 A . 1 in the appendix); however, we acknowledge that this is not ironclad evidence that the exclusion restrictions are valid . A useful extension of this study would be a randomized - controlled trial that generates exogenous variation in the adoption of the SI categories (e.g., through different information treatments) and measures the effects on child nutrition. 1. 4 Data With the exception of IVs (iv) and (v) above, t he data come from the TNPS, whic h is a nationally - representative household survey that contains detailed information on socioeconomic characteristics, consumption, agricultural production, and non - farm income generating activities , inter alia . The TNPS is a four - wave panel survey conduct ed in 2008/09, 2010/11, 2012/13 , and 11 The authors thank Dr. David Mather for sharing these data. 12 Recall that we are controlling for time invariant heterogeneity, including suitability for maize production, via CRE. 20 2014/15 but only the data from the first three waves are used here because the sample in the fourth wave was refreshed for future rounds . The TNPS is based on a stratified, multi - stage cluster sample design and the clusters within each stratum are randomly selected as the primary sampling units , where there are four different strata: Dar es Salaam, other urban areas on mainland Tanzania, rural mainland Tanzania, and Zanzibar . The TNPS baseline (2008/09) sample of 3,2 65 households is clustered in 409 e numeration a reas . The se households and their members were tracked and re - interviewed in the second (TNPS 2010/11) and third waves (TNPS 2012/13) with very low attrition rates between waves ( TNBS , 2014) . We start with observations of children under age five ( 0 - 59 months) in rural households that grew maize in the main farming season (i.e., the long - rainy season) in a given wave but drop children age 0 - 5 months because they are typically exclusively breastf ed during that period (Tanzania Food and Nutrition Centre , 2014) and thus less likely to be directly affected by diet changes associated with . There are 1 , 871 total household observations meeting these criteria acros s the three waves of the TNPS ( 532 observations in 2008/09, 560 in 2010/11, and 779 in 2012/13) . These households contain a total of 2, 486 children age 6 - 59 months ( 693 observations in 2008/09, 727 in 2010/11, and 1, 066 in 2012/13 ). Per Table 1 A . 2 in the appendix, among children in our sample, the mean values of HAZ and WAZ are - 1. 82 and - 0.9 8 , respectively ; 4 7 % are stunted and 15% are underweight . (This table and Table 1 A . 3 also show descriptive statistics by SI category .) Anthropometric data to calculate nutritional status were collected from children an average (and median) of 10 months after the household began harvesting the maize (and this timing is controlled for in th e econometric models). This implies that WAZ and HAZ in our sample are mainly 21 influenced by the SFM adoption decision s captured in the data and not by such decision s in the following year . Table s 1 A . 2 and 1 A . 3 in the appendix further provide s ummary statistics for the control variables used in the analysis. These variables were selected based on careful reviews of the technology adoption and child nutrition literatures and include child - level variables (age and gender, whether or not the child had diarrhea in the past two weeks difference between maize harvest and collection of anthropometric data, and dummy variables for number of times the child appears across survey rounds ); household characteristics (age and gender of the household head, educatio n level of the household head and spouse , family labor (as defined in T able 1 A . 3 ) , number of female adults/elderly/child ren /siblings in the household, marital status of the household head, off - farm income, access to a safe drinking water source , use of safe drinking water, basic sanitation (toilet)); agricultural characteristics (total cultivated land ; maize plot , farm equipment, and livestock ownership ; distance to the nearest market ); input and output prices; and community characteristics (wheth er or not a government health center/hospital is available within the community). biological height and weight could also affect his/her nutritional status. roximately 36% (15%) of th e observations in our sample because the individual is no longer a household member or was otherwise not present when measurements were taken . Many models fail to converge with these reduction s in sample size; however, we do report, as a robustness check, estimates that body mass index (BMI) (and age). 13 An important caveat is that BMI could be affected by if the woman is pregnant or not, but the TNPS data do not capture 13 BMI is equal to weight (in kilograms), divided by height (in meters) squared. 22 information on current pregnancy ; thus there is likely to be measurement error in the BMI variable for at least some observations. Our inability to fully control for these biological parent characteristics is a limitation of this study. However, note that if height (of adults) is reasonably assumed to be constant over the survey waves , then our use of CRE indirectly controls for the 1. 5 R esults Table 1. 2 presents the CRE - METE estimates of the local average treatment effects of the various SI categor ies on child for the full sample of children aged 6 - 59 months . See Tables 1 A . 4 and 1 A . 5 in the appendix for the full first - and second - stage results for this model. ( Also note in Table 1 A . 4 that two of the IVs associated with an increased probability of adoption of practices in the SI category by a given household are increase s in the proportion of other households in the or that practice maize - legume intercropping. We return to this point in the final section of the paper on policy implications.) For comparison purpose s , we also report FE and CRE - P OLS results that are estimated under the assumption exogenous after controlling for the observed covariates and time - invariant heterogeneity. The results of both the FE and CRE - POLS models suggest that there are no statistically significant nutritional effects for any of the SI treatment groups. However, we reject the null hypothesis of joint exogeneity of the SI category variables in all CRE - METE models estimated here, which suggests that e ndogeneity is indeed an 23 issue. 14 In subsequent parts of this section we therefore focus on the CRE - METE results , which correct for self - selection . Table 1.2 : FE, CRE - P OLS , and CRE - METE estimates: I mpacts on nutritional outcomes of children aged 6 - 59 months (full sample) Variables HAZ WAZ FE Intensification 0. 069 - 0. 128 (0. 303 ) (0. 235 ) Sustainable 0.0 43 0.0 30 (0. 118 ) (0.0 98 ) SI - 0. 194 - 0. 271 (0. 291 ) (0. 215 ) CRE - P OLS Intensification 0.052 0.093 (0.132) (0.114) Sustainable 0.039 0.020 (0.069) (0.055) SI - 0.070 0.007 (0.106) (0.093) CRE - METE Intensification - 0.463*** - 0.266 (0.176) (0.170) Sustainable 0.116 0.200 (0.160) (0.133) SI 0.355** 0.453*** (0.155) (0.125) Selection terms Intensification 0.443** 0.647*** (0.177) (0.125) Sustainable - 0.232 - 0.103 (0.151) (0.188) SI - 0.592*** - 0.557*** (0.125) (0.155) Notes: N=2,486. Base category is on - adoption . ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the household level in parentheses. 14 This test, f ollowing Deb and Trivedi (2006b), is a likelihood - ratio test where the null hypothesis is that the (selection terms) are jointly equal to zero (exogeneity of treatment). We reject the null in all c ases (p< 0.01), which suggests that treatment is endogenous. To conserve space, we do not report the estimated in subsequent tables. 24 The CRE - METE results in Table 1. 2 suggest that , on average, use of practices in category is associated with increases in of 0. 36 units and 0.4 5 units, respectively, compared to those in non - adopting households. These are sizeable increases relative to the sample mean HAZ and WAZ of - 1. 82 and - 0.9 8 , respectively. 15 Moreover , the estimated increase in HAZ (WAZ) would lift 26% of stunted children (53% of underweight children) in our sample to the - 2 cutoff. In contrast , use of inorganic fertilizer only is associated with a decrease 46 units , and there are no In addition to estimating the CRE - METE model s for the full sample of children age d 6 - 59 months, we also estimate models for : ( i ) children aged 6 - 59 months with interaction terms between the SI treatment groups and an indicator variable for children age d 6 - 24 months ; and ( ii ) children aged 25 - 59 months only. T he major rationale behind the se additional a nalys es is that the growth faltering patterns of children under age five differ across ages ( see Figure 1 A . 1 in the appendix) . Victora et al. (2010) find that rapid growth faltering of HAZ was observed until 24 months of age, then plateauing from 25 - 59 months, while WAZ showed progressive and slow faltering through months 0 - 59, with the most rapid declines from 0 - 24 months. A s a result, the child nutritional impacts of SI adoption decisions may also vary. In particular, the inclusion of the 6 - 24 mo nths interaction terms allow s us to test for different ial effects of the SI treatment groups on the nutritional outcomes of children who are in the cr itical window of opportunity for the promotion of optimal growth, health, and development, which is the 1,000 days from conception through the first two years of li f e . 16 15 Zeng et al. (2017) find that a 0.25 - hectare increase in improved maize variety area is associated with average HAZ and WAZ increases of 0.25 and 0.18 units, respectively, relative to sample means of - 1.51 and - 0.63. 16 We also attempted to estimate models for children age d 6 - 24 months; however, these models do not converge. 25 In Table 1. 3, t he results including the interaction terms are presented in the upper panel and the results for chi ldren aged 25 - 59 months only are in the bottom panel. Together these 25 - 59 months. We continue to find no evidence of statistically significant effects for the re not robust to the model specification, as they cease to be statistically significant when we limit the sample to children aged 25 - 59 months. The lack of statistically significant effects of any SI categories on the HAZ and WAZ of children age d 6 - 24 months may be because these children are still be ing breastfed and largely dependent on compleme ntary/weaning foods instead of consuming adult foods (Zeng et al. , 2017; Tanzania Food and Nutrition Centre , 2014 ; Stephenson et al. , 2017 ). Consistent with our findings, a recent study ( Jain , 2018) f inds that nutrient intake has no association with th e HAZ of children aged 6 - 23 months in rural Bangladesh. 26 Table 1.3 : CRE - METE estimates: Impacts on child nutritional outcomes with sub - sample analysis Variables HAZ WAZ Full - sample (N=2,486) with interaction terms Intensification - 0.400** - 0.238 (0.192) (0.176) Sustainable 0.038 0.191 (0.174) (0.139) SI 0.314* 0.423*** (0.170) (0.134) Intensification×6 - 24 months - 0.129 - 0.083 (0.227) (0.168) Sustainable×6 - 24 months 0.182 0.026 (0.119) (0.090) SI×6 - 24 months 0.077 0.062 (0.172) (0.146) Sub - sample (N=1,411): children aged 25 - 59 months Intensification - 0.162 - 0.104 (0.207) (0.158) Sustainable 0.004 0.235 (0.187) (0.168) SI 0.365** 0.439*** (0.184) (0.145) Notes: - ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the household level in parentheses. Selection terms excluded to conserve space. A ppendix Table 1 A . 6 shows the results for models that include the BMI and age . The se results is positively correlated both child nutrition outcomes. is positively correlated with both HAZ and WAZ. 17 Overall, the robust finding across model specifications is that enhances both HAZ and WAZ . This could be for the following reasons . First, maize plots in Tanzania involve maize - legume intercropping (Table 1 .1 ) and based on the results in Table 1 A . 7, which exclude organic fertilizer, the combined use of maize - legume intercropping 17 Table 1 A . 6 a lso suggests that is negatively associated with HAZ and WAZ for children aged 6 - 59 months. However, we could not confirm that this hold s for children aged 25 - 59 months because the model does not converge . 27 and inorganic fertilizer is a key driv 18 The legume crops produced as a result may directly affect the diet composition of adopting households by providing needed protein and micronutrients (Messina 1999 ) ; this, in turn, may positively affect child nutrition . Indeed , as shown in appendix Table 1 A . 8, 90% of sample ho s , while only 1 9 % and 3 1% of households in the - groups, respectively, produce legumes . 19 The table also indicates that maize - legume intercropping is the dominant way in which maize - growing households in Tanzania produce legumes . In addition, Stahley et al. (2012) report that the mean quantity of legumes consumed by producing households in Ta nzania is double that consumed by purchasing households. Furthermore , these legume crops could help farmers to increase their crop income since per - kilogram prices for legumes are higher than maize prices (see Table 1 A . 3 ) . Second, relative to farmers in the other treatment groups , households treatment group may have higher crop productivity or income s due to synergistic effects when . Indeed , a review by Place et al. (2003) indicates that there is considerable evidence demonstrating positive effects on overall yields and net financial returns of combined use of inorganic fertilizer and organic soil fertility practices including animal manure an d intercropping with legume s . To pathways , we estimate CRE - METE models for two additional outcome variables: (1) gross value of crop production plots as a proxy for crop income; and (2) an 18 We tr ied to estimate a similar model with only organic fertilizer, inorganic fertilizer, and their combined use (excluding maize - legume intercropping) but it do es not converge. 19 The correlation between u se of maize - legume intercropping and production of legumes in other ways is extremely low ( - 0.02 ) . 28 index of crop output per acre on those plots as a proxy for productivity. 20 The associated CRE - METE results are shown in Table 1. 4 and is indeed associated with increases in crop income and productivity on households is as well but the crop income effects are considerably and statistically In contrast associated with negative effec ts on crop income and no significant effects on productivity. These results are consistent with the findings above of WAZ effects . Our results overall also suggest that not al l income and productivity increases are created equal. Simply producing more maize via Intensification without involving legume crops may be insufficient to enhance child nutrition. Table 1.4 : CRE - METE estimates: I mpacts on crop income and productivity Variables Crop income (Tanzanian Shillings) Output index per acre Intensification 350,835.572*** 487.756*** (114,258.251) (131.930) Sustainable - 114,241.755*** 19.272 (41,691.292) (37.026) SI 720,637.260*** 531.401*** (163,209.116) (134.278) Notes: N=1,871. - ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the household level in parenthese s. 20 ) is calculated following Liu and Myers (2009) as , where is the kilograms of crop j produced on farmer i is the regional market price of crop j , and crop 1 is maize. 29 1. 6 Conclusion s and implications In this study, we empirically estimate d various SFM practices on their maize plots on the nutrition outcomes of young children in the household . The r esults consistently suggest that of maize production (joint use of inorganic fertilizer with maize - legume intercropping and/or organic fertilizer) is associated with increases in and WAZ compared to households that adopt none of the practices . These effects are mainly among children age d 25 - 59 months who , compared to younger children, are less likely to be breastfed and may be more directly affected by household diet changes associated with changes in agricultural practices . J oint use of maize - legume intercropping and inorganic fertilizer is a key driver of these results, and the effects appear to come through both crop income and productivity pathways. inorganic fert i maize - legume intercropping but no inorganic fertilizer) improve child nutrition outcomes. These results also link to the production diversity - dietary diversity/nutrition literature and suggest that crop diversification (proxied here by maize - legume intercropping) combined with intensification produces the most favorable child nutrition outcome s. Our results have two main implications for agricultural policy and future research. First, given the potential benefits of joint use of inorganic fertilizer with maize - legume intercropping (and possibly organic fertilizer) for soil fertility , crop inco me, productivity, and child nutrition outcomes, it is important for policy makers to identify ways to promote use of such practices by Tanzanian maize farmers. (At present, Tanzania has much lower adoption rates of these practices than other countries in e astern and southern Africa such as Kenya, Malawi, and Ethiopia (Kassie et al. , 2015a).) Further research is needed to identify cost - effective SI promotion strategies and 30 o ur results do not speak directly to this question in a major way . H owever, based on our results, one general approach that may warrant further investigatio n (among others) is leveraging social learning to encourage SI of maize production . ( Recall that the first stage results in appendix Table 1 A . 4 suggest that increase s using inorganic fertilizer and maize - legume intercropping are associated with an increased probability of by the household itself .) A sec ond area in need of further research is if and how SI of agricultural systems more broadly (i.e., beyond maize) contributes to food security and child nutrition outcomes. 31 APPENDI X 32 Figure 1A . 1 : Mean WAZ and HAZ by age in months, relative to the WHO standard Source s : across the 2008/09, 2010/11, and 2012/13 waves of the TNPS . -3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 Weight-for-age z-score (WAZ) Height-for-age z-score (HAZ) Age in month z - scores 33 Table 1A.1 : Falsification test results (parameter estimates from CRE - POLS regressions for - Variables HAZ WAZ Child characteristics Child age (months) 0.013*** - 0.005* (0.004) (0.003) Child gender (1=male) - 0.162* - 0.003 (0.090) (0.071) Diarrhea (1=yes) - 0.113 - 0.071 (0.138) (0.117) education - 0.021 - 0.018 (0.021) (0.018) Monthly difference 0.009 0.030*** (0.013) (0.010) T2 dummy 0.163* - 0.001 (0.092) (0.074) T3 dummy 1.112*** 0.433 (0.209) (0.440) Household characteristics Head gender (1=male) - 0.066 - 0.028 (0.145) (0.118) Head age (years) - 0.033* - 0.032** (0.017) (0.014) Head education (years) - 0.007 0.005 (0.018) (0.014) Spouse education (years) 0.008 0.009 (0.025) (0.020) Family labor - 0.014 - 0.018 (0.074) (0.027) No. of female adults 0.033 - 0.091 (0.137) (0.115) No. of elderly 0.380 0.412 (0.341) (0.268) No. of child - 0.057 0.023 (0.092) (0.065) No. of siblings - 0.051 0.153 (0.230) (0.214) Head marital status (1=yes) 0.081 0.050 (0.120) (0.097) Off - farm income (1=yes) - 0.019 0.015 (0.159) (0.125) Access to safe drinking water source (1=yes) 0.053 - 0.042 (0.120) (0.091) Safe drinking water 0.106 0.005 (0.104) (0.083) Sanitation (toilet) (1=yes) 0.028 0.008 (0.111) (0.084) 34 Table 1 A .1 ( ) Variables HAZ WAZ Agricultural characteristics Total cultivated land (acres) - 0.019 - 0.024** (0.019) (0.012) Own plot (1=yes) - 0.271* - 0.055 (0.155) (0.140) Market distance (km) - 0.000 0.002 (0.003) (0.002) Farm assets (1,000 TZS ) 0.000 0.000** (0.000) (0.000) Livestock (1=yes) 0.037 0.050 (0.103) (0.082) Input and output prices Maize price ( TZS /kg) 0.000 - 0.000 (0.000) (0.000) Bean price ( TZS /kg) 0.000 0.001** (0.000) (0.000) Groundnut price ( TZS /kg) 0.000 0.000 (0.000) (0.000) Inorganic fertilizer price (TZS /kg) - 0.001* - 0.000 (0.000) (0.000) Community characteristics Gov t . health/hospital (1=yes) 0.045 0.017 (0.094) (0.077) Instrumental variables Electoral threat 0.002 - 0.099 (0.126) (0.073) Number of subsidized fertilizer vouchers - 0.000 - 0.000 (0.000) (0.000) Proportion receiving agricultural advice 0.001 - 0.001 (0.002) (0.002) Proportion adopting inorganic fertilizer 0.001 0.001 (0.003) (0.002) Proportion adopting maize - legume IC - 0.000 - 0.002 (0.002) (0.001) Constant - 2.952*** - 1.572*** (0.460) (0.355) Notes: Sample size is 1 , 084 individuals - - averages of household level variables to control for time - constant unobserved heterogeneity were included in the model but not reported in Table 1 A . 1. ***, **, and * denote statistically significance at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the household level in parentheses. 35 Table 1 A .2 : Summary statistics for child nutritional status and child characteristics used in the analysis Variable s Variable description Mean valu es for each SI category Mean of all SD of all N I S SI Child nutritional status HAZ Height - for - age z - score - 1.86 - 1.82 - 1.76* - 1.98 - 1.82 1.31 Stunted children 1 = yes if HAZ < - 2 0.47 0.46 0.46 0.48 0.47 0.50 WAZ Weight - for - age z - score - 1.02 - 0.90 - 0.94* - 1.02 - 0.98 1.03 Underweighting children 1 = yes if WAZ < - 2 0.15 0.13 0.15 0.15 0.15 0.36 Control variables Child characteristics Child age Age of children under age 5 (months) 37.72 36.75 37.24 37.48 37.44 12.79 Child gender Gender of the child (1 = male) 0.48 0.48 0.50 0.45 0.49 0.50 Diarrhea 1 = yes if the child had diarrhea in the past 2 weeks 0.09 0.08 0.10 0.09 0.09 0.29 Highest grade (years) 4.49 5.63*** 4.55 5.59*** 4.66 3.44 Monthly difference Time difference between maize harvest and 10.10 9.88 9.86 10.10 9.98 4.15 T1 dummy (excluded) 1 = yes if the child is observed once in any of the three waves 0.64 0.60 0.68** 0.52*** 0.65 0.48 T2 dummy 1 = yes if the child is observed twice in any of the three waves 0.35 0.40 0.31* 0.48*** 0.35 0.48 T3 dummy 1 = yes if the child is observed in all three waves 0.01 0.00*** 0.00 0.00*** 0.00 0.07 Number of observations 1084 149 1071 182 2486 Notes: N, I, S, and SI indicate Non - adoption, Intensification, Sustainable, and SI, respectively. A two - sample t - test was used to compare the means of variables between each SI treatment group (I, S, and SI) and the base category (N) under the assumption of uneq ual variance. SD is standard deviation. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 36 Table 1 A .3 : Summary statistics for other control variables used in the analysis Variable s Variable description Mean values for each SI category Mean of all SD of all N I S SI Control variables Household characteristics Head gender Gender of the household head (1 = male) 0.83 0.91*** 0.82 0.89* 0.84 0.37 Head age Age of the household head (years) 43.44 41.02* 44.81* 40.30*** 43.59 14.06 Head education Highest grade completed by the household head (years) 4.57 6.60*** 4.52 6.32*** 4.82 3.31 Spouse education Highest grade completed by the spouse 3.78 5.47*** 3.91 5.57*** 4.09 3.39 Family labor Number of adults (15 - 64 years old) per acre 1.02 0.96 1.09 1.08 1.05 1.41 No. of female adults Number of female adults in the household 1.54 1.52 1.75*** 1.50 1.62 1.05 No. of e lderly Number of household members above 65 years 0.22 0.13* 0.23 0.10*** 0.21 0.49 No. of child Number of household members below 15 years 3.69 3.32** 4.11*** 3.76 3.84 2.17 No. of siblings Number of siblings of children under age 5 0.06 0.02*** 0.07 0.01*** 0.06 0.33 Head marital status 1 = yes if the HH head got married 0.70 0.74 0.70 0.75 0.71 0.46 Off - f arm income 1 = yes if the HH earns other income 0.43 0.52** 0.50*** 0.58*** 0.48 0.50 Access to safe drinking water source 1 = yes if the HH has safe drinking water source (e.g., piped or protected water) 0.22 0.40*** 0.22 0.23 0.23 0.42 Safe drinking water 1 = yes if the HH does drink boiled/bottled/treated water 0.21 0.33** 0.26** 0.25 0.24 0.43 Sanitation (toilet) 1 = yes if the HH has a private toilet 0.80 0.94*** 0.80 0.97*** 0.82 0.38 Agricultural characteristics Total cultivated land Total land area (acres) cultivated 6.99 6.15 7.99 7.51 7.38 19.01 Own plot 1 = yes if the HH owns at least one maize plot 0.92 0.90 0.92 0.97*** 0.92 0.27 Market distance Distance to the nearest market (km) 12.16 10.95 11.47 11.68 11.76 13.05 Farm assets Total value of farm implements and machinery (100,000 TZS) 16.06 8.79 29.12*** 5.16*** 20.04 89.84 Livestock 1 = yes if the HH has livestock (cattle, goats/sheep, pigs, or donkeys) 0.41 0.45 0.58*** 0.54*** 0.49 0.50 37 Table 1 A .3 Variable s Variable description Mean values for each SI category Mean of all SD of all N I S SI Input and output prices Maize price Maize (grain) market price at district level (TZS/kg) 469.81 440.56 474.76 419.93*** 465.99 202.33 Bean price Bean market price at district level (TZS/kg) 1296.55 1267.65 1295.97 1290.15 1293.87 325.53 Groundnut price Groundnut market price at district level (TZS/kg) 1675.66 1690.78 1678.77 1656.75 1676.48 561.07 Inorganic fertilizer price Inorganic fertilizer price at district level (TZS/kg) 1131.46 996.04*** 1178.80** 940.40*** 1126.81 414.54 Community characteristics Govt. health/hospital 1 = yes if there is a governmental health center/hospital in the community 0.45 0.49 0.43 0.45 0.44 0.50 Instrumental variables Electoral threat Proportion of votes for the runner - up divided by the proportion of votes for the presidential winner 0.28 0.17*** 0.28 0.20** 0.27 0.54 Number of subsidized fertilizer vouchers Number of subsidized fertilizer vouchers 58.33 107.03*** 48.84*** 120.23*** 62.54 68.16 Proportion receiving agricultural advice Proportion of other households in the ward that got advice on agricultural production 9.85 20.52*** 10.20 22.71*** 11.71 18.38 Proportion using inorganic fertilizer Proportion of other households in the ward that use inorganic fertilizer 6.98 54.43*** 9.22** 59.43*** 15.15 27.81 Proportion using maize - legume IC Proportion of other households in the ward that use maize - legume intercropping 33.70 44.66*** 48.48*** 58.85*** 42.41 33.36 Number of observations 838 126 762 145 1871 Notes: N , I, S, and SI indicate Non - adoption, Intensification, Sustainable, and SI, respectively. A two - sample t - test was used to compare the means of variables between each SI treatment group (I, S, and SI) and the base category (N) under the assumption of unequa l variance. SD is standard deviation. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. TZS = Tanzania Shillin gs. 38 Table 1A.4 : CRE mixed multinomial logit estimates of the determinants of adoption of each SI - Variables Intensification Sustainable SI Child characteristics Child age (months) - 0.020* 0.005 0.000 (0.010) (0.004) (0.009) Child gender (1=male) - 0.001 0.066 - 0.113 (0.272) (0.119) (0.247) Diarrhea (1=yes) - 0.051 0.248 0.586 (0.455) (0.210) (0.408) 0.003 - 0.003 0.050 (0.062) (0.033) (0.062) Monthly difference - 0.034 - 0.002 0.020 (0.044) (0.019) (0.033) T2 dummy - 0.296 - 0.101 0.069 (0.323) (0.137) (0.283) T3 dummy - 49.806*** - 1.296 - 50.018*** (0.736) (0.802) (0.813) Household characteristics Head gender (1=male) 0.719 - 0.243 0.011 (0.511) (0.230) (0.495) Head age (years) - 0.024 - 0.010 0.034 (0.059) (0.038) (0.056) Head education (years) 0.091 - 0.021 0.076 (0.066) (0.028) (0.053) Spouse education (years) - 0.010 - 0.000 0.010 (0.077) (0.038) (0.068) Family labor - 0.207 0.019 0.232 (0.321) (0.098) (0.183) No. of female adults - 0.480 - 0.491* - 0.703* (0.377) (0.279) (0.382) No. of elderly - 0.164 0.013 - 0.040 (0.637) (0.788) (0.870) No. of child 0.482** 0.022 0.354* (0.206) (0.112) (0.182) No. of siblings - 0.101 - 0.138 - 4.256 (0.659) (0.538) (2.800) Head marital status (1=yes) 0.283 0.090 0.427 (0.334) (0.187) (0.342) Off - farm income (1=yes) 0.422 0.521* 1.441*** (0.614) (0.277) (0.520) Access to safe drinking water source (1=yes) 0.826** 0.184 - 0.331 (0.330) (0.184) (0.339) Safe drinking water 0.805** 0.255 0.351 (0.336) (0.178) (0.328) Sanitation (toilet) (1=yes) 1.633*** - 0.103 1.417** (0.629) (0.195) (0.718) 39 Table 1 A . 4 Variables Intensification Sustainable SI Agricultural characteristics Total cultivated land (acres) 0.037 0.046 - 0.043 (0.054) (0.033) (0.058) Own plot (1=yes) 0.128 - 0.223 1.305** (0.469) (0.265) (0.607) Market distance (km) - 0.020* - 0.004 0.001 (0.011) (0.006) (0.012) Farm assets (1,000 TZS ) 0.000 0.000 - 0.000 (0.000) (0.000) (0.000) Livestock (1=yes) 0.161 0.874*** 0.883*** (0.350) (0.174) (0.325) Input and output prices Maize price ( TZS /kg) 0.002 - 0.002* - 0.002 (0.002) (0.001) (0.002) Bean price ( TZS /kg) - 0.000 - 0.000 - 0.000 (0.001) (0.001) (0.001) Groundnut price ( TZS /kg) 0.000 0.000 - 0.000 (0.001) (0.000) (0.000) Inorganic fertilizer price (TZS /kg) 0.001 0.000 0.001 (0.001) (0.000) (0.001) Community characteristics Gov t . health/hospital (1=yes) 0.007 - 0.030 - 0.542* (0.290) (0.159) (0.324) Instrumental variables Electoral threat - 0.711 0.141 0.804*** (0.727) (0.185) (0.298) Number of subsidized fertilizer vouchers 0.000 - 0.000*** 0.000 (0.000) (0.000) (0.000) Proportion receiving agricultural advice - 0.002 - 0.007 - 0.002 (0.007) (0.005) (0.008) Proportion adopting inorganic fertilizer 0.063*** 0.010** 0.066*** (0.007) (0.004) (0.006) Proportion adopting maize - legume IC - 0.001 0.016*** 0.017*** (0.005) (0.002) (0.005) Constant - 5.277*** - 1.468** - 8.840*** (1.451) (0.719) (1.553) Notes: Sample size is 2 , 486 individuals (1 , 873 households ) . Time - averages of household level variables to control for time - constant unobserved heterogeneity were included in the model but not reported in Table 1 A . 4. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standa rd errors clustered at the household level in parentheses. 40 Table 1A.5 : Second stage estimates for child nutritional outcomes Variables Full - sample (6 - 59 mo.) Full - sample with interactions Sub - sample (25 - 59 mo.) HAZ WAZ HAZ WAZ HAZ WAZ Child characteristics Child age (months) 0.009*** - 0.004** 0.011** - 0.006* 0.014*** - 0.009*** (0.003) (0.002) (0.004) (0.003) (0.004) (0.003) Child gender (1=male) - 0.168*** - 0.018 - 0.167*** - 0.017 - 0.017 0.080 (0.060) (0.048) (0.060) (0.048) (0.068) (0.056) Diarrhea (1=yes) - 0.203** - 0.095 - 0.204** - 0.097 - 0.192 - 0.058 (0.093) (0.076) (0.093) (0.076) (0.135) (0.123) 0.001 - 0.012 0.000 - 0.012 - 0.001 - 0.019 (0.015) (0.012) (0.015) (0.012) (0.016) (0.014) Monthly difference 0.009 0.017** 0.008 0.018** 0.013 0.020*** (0.009) (0.007) (0.009) (0.007) (0.010) (0.008) T2 dummy 0.055 - 0.011 0.053 - 0.008 - 0.241 - 0.026 (0.068) (0.053) (0.069) (0.053) (0.184) (0.138) T3 dummy 1.140*** 0.733 1.124*** 0.731 (0.250) (0.464) (0.241) (0.458) Household characteristics Head gender (1=male) 0.117 0.059 0.113 0.057 0.047 0.025 (0.108) (0.083) (0.107) (0.083) (0.122) (0.096) Head age (years) - 0.002 0.005 - 0.003 0.005 0.008** 0.006** (0.011) (0.010) (0.011) (0.010) (0.004) (0.003) Head education (years) 0.005 0.008 0.004 0.008 0.006 0.006 (0.013) (0.010) (0.013) (0.010) (0.014) (0.011) Spouse education (years) 0.003 0.014 0.003 0.013 0.012 0.026* (0.018) (0.014) (0.018) (0.014) (0.019) (0.015) Family labor - 0.026 - 0.026 - 0.028 - 0.025 0.025 0.025 (0.052) (0.028) (0.052) (0.028) (0.024) (0.019) No. of female adults - 0.066 - 0.046 - 0.067 - 0.045 - 0.002 - 0.002 (0.074) (0.063) (0.073) (0.064) (0.036) (0.029) No. of elderly - 0.008 - 0.024 - 0.009 - 0.025 - 0.022 0.041 (0.315) (0.209) (0.314) (0.211) (0.091) (0.075) No. of child - 0.054 0.021 - 0.051 0.020 - 0.012 - 0.021 (0.037) (0.036) (0.037) (0.037) (0.017) (0.016) No. of siblings 0.098 0.241** 0.097 0.240** 0.087 0.049 (0.160) (0.110) (0.160) (0.108) (0.099) (0.064) Head marital status (1=yes) 0.092 0.031 0.090 0.033 0.139 0.067 (0.086) (0.066) (0.086) (0.065) (0.100) (0.075) Off - farm income (1=yes) - 0.086 - 0.041 - 0.077 - 0.039 0.178** 0.166*** (0.114) (0.088) (0.115) (0.088) (0.078) (0.064) Access to safe drinking water source (1=yes) 0.081 - 0.004 0.078 - 0.005 - 0.067 - 0.057 (0.083) (0.064) (0.083) (0.064) (0.087) (0.066) Safe drinking water 0.099 0.026 0.101 0.026 0.047 0.078 (0.072) (0.058) (0.072) (0.058) (0.078) (0.065) Sanitation (toilet) (1=yes) - 0.234*** - 0.080 - 0.232*** - 0.079 - 0.171* - 0.096 (0.088) (0.071) (0.088) (0.072) (0.096) (0.076) Agricultural characteristics Total cultivated land (acres) - 0.010 - 0.006 - 0.011 - 0.006 0.001 0.001** (0.009) (0.007) (0.009) (0.007) (0.001) (0.001) Own plot (1=yes) - 0.030 0.046 - 0.028 0.047 - 0.008 0.026 (0.119) (0.108) (0.119) (0.108) (0.137) (0.132) 41 Table 1 A .5 Variables Full - sample (6 - 59 mo.) Full - sample with interactions Sub - sample (25 - 59 mo.) HAZ WAZ HAZ W AZ H AZ W AZ Market distance (km) - 0.001 0.001 - 0.001 0.001 0.004 0.003 (0.002) (0.002) (0.002) (0.002) (0.003) (0.002) Farm assets (1,000 TZS ) 0.000 0.000 0.000 0.000 0.000*** 0.000*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Livestock (1=yes) 0.041 0.033 0.042 0.033 - 0.024 - 0.036 (0.073) (0.061) (0.073) (0.061) (0.085) (0.067) Input and output prices Maize price ( TZS /kg) 0.001** 0.000 0.001** 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Bean price ( TZS /kg) 0.000 0.000 0.000 0.000 0.000** 0.000* (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Groundnut price ( TZS /kg) 0.000** 0.000 0.000** 0.000 - 0.000 - 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Inorganic fertilizer price ( TZS /kg) - 0.000 0.000 - 0.000 0.000 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Community characteristics Gov t . health/hospital (1=yes) - 0.008 0.017 - 0.006 0.016 0.049 0.028 (0.067) (0.055) (0.067) (0.054) (0.073) (0.061) SI category Intensification - 0.463*** - 0.266 - 0.400** - 0.238 - 0.162 - 0.104 (0.176) (0.170) (0.192) (0.176) (0.207) (0.158) Sustainable 0.116 0.200 0.038 0.191 0.004 0.235 (0.160) (0.133) (0.174) (0.139) (0.187) (0.168) SI 0.355** 0.453*** 0.314* 0.423*** 0.365** 0.439*** (0.155) (0.125) (0.170) (0.134) (0.184) (0.145) Selection terms Intensification 0.443** 0.647*** 0.450** 0.638*** 0.328** 0.340*** (0.177) (0.125) (0.176) (0.135) (0.160) (0.123) Sustainable - 0.232 - 0.103 - 0.235 - 0.105 - 0.061 - 0.286 (0.151) (0.188) (0.154) (0.195) (0.214) (0.194) SI - 0.592*** - 0.557*** - 0.589*** - 0.545*** - 0.715*** - 0.596*** (0.125) (0.155) (0.129) (0.167) (0.227) (0.134) Age dummy and Interaction terms 6 - 24 months of age dummy - 0.038 - 0.079 (0.115) (0.090) Intensification×6 - 24 months of age - 0.129 - 0.083 (0.227) (0.168) Sustainable ×6 - 24 months of age 0.182 0.026 (0.119) (0.090) SI ×6 - 24 months of age 0.077 0.062 (0.172) (0.146) Constant - 3.285*** - 1.771*** - 3.301*** - 1.679*** - 3.584*** - 1.650*** (0.318) (0.247) (0.346) (0.264) (0.343) (0.287) Joint test for selection terms ( ) 13,265*** 10,952*** 11,849*** 10,894*** 15,141*** 13,524*** Observations 2,486 2,486 2,486 2,486 1,411 1,411 Notes: Time - averages of household level variables to control for time - constant unobserved heterogeneity were included in the full - sample models but not reported in Table 1 A . - ***, **, and * denote statistically significanc e at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the household level in parentheses. 42 Table 1A.6 : CRE - Variables HAZ WAZ Full sample ( children aged 6 - 59 months ) Intensification - 0.630*** - 0.371*** (0.160) (0.136) Sustainable 0.091 0.034 (0.125) (0.105) SI 0.388*** 0.330** (0.140) (0.150) 0.008 - 0.001 (0.005) (0.004) 0.029*** 0.054*** (0.010) (0.008) Selection terms Intensification 0.789*** 0.502*** (0.088) (0.122) Sustainable - 0.107 - 0.051 (0.138) (0.121) SI - 0.642*** - 0.547*** (0.103) (0.162) Full - sample with interaction terms Intensification - 0.599*** - 0.313** (0.175) (0.157) Sustainable - 0.018 0.036 (0.138) (0.114) SI 0.346** 0.302* (0.148) (0.164) Intensification×6 - 24 months of age - 0.085 - 0.164 (0.245) (0.180) Sustainable×6 - 24 months of age 0.240* - 0.006 (0.128) (0.095) SI×6 - 24 months of age 0.084 0.047 (0.179) (0.148) 0.008* - 0.001 (0.005) (0.004) 0.029*** 0.054*** (0.010) (0.008) Notes: - ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the household level in parentheses. The selection terms for the full - sample with intera ction terms are excluded to conserve space. The corresponding model for the sub - sample of children aged 25 - 59 months does not converge. 43 Table 1A.7 : CRE - METE model estimates : I mpacts of the adoption of inorganic fertilizer, maize - legume intercropping, and their joint use on child nutritional outcomes full and sub - sample analysis Variables HAZ WAZ Full sample ( N =2, 486 ): children aged 6 - 59 months Inorganic fertilizer - 0.34 9 ** - 0.2 87 (0.154) (0.3 28 ) Maize - legume intercropping 0.27 3 * 0.08 4 (0.14 6 ) (0.2 91 ) Joint use 0.4 04 *** 0.4 0 6** (0.16 5 ) (0.17 6 ) Selection terms Inorganic fertilizer 0.5 3 2*** 0. 501 (0.13 8 ) (0.4 48 ) Maize - legume intercropping - 0.3 60 ** - 0.12 5 (0.15 4 ) (0.3 68 ) Joint use - 0.6 37 *** - 0.5 63 ** (0.14 8 ) (0.23 5 ) Full - sample ( N =2, 486 ) with interaction terms Inorganic fertilizer - 0. 301 * - 0.2 70 (0.16 6 ) (0. 622 ) Maize - legume intercropping 0.2 47 0. 082 (0.15 2 ) (0. 537 ) Joint use 0.4 36 ** 0.3 52 (0.17 5 ) (0. 245 ) Inorganic fertilizer×6 - 24 months of age - 0.1 19 - 0.1 14 (0.212) (0.1 68 ) Maize - legume intercropping×6 - 24 months of age 0.0 54 - 0.0 43 (0.121) (0.09 4 ) Joint use × 6 - 24 months of age - 0.079 0.0 92 (0.17 6 ) (0.16 1 ) Sub - sample ( N =1,4 11 ): children aged 25 - 59 months Inorganic fertilizer - 0. 182 - 0.0 78 (0.1 96 ) (0.1 39 ) Maize - legume intercropping 0. 025 0.2 16 (0.1 77 ) (0.1 37 ) Joint use 0. 403 *** 0.4 16 *** (0. 200 ) (0.14 3 ) Notes: Base category is on - adoption . ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. Standard errors clustered at the household level in parentheses. The selection terms for the full - sample with interaction terms and sub - sample an alyses are excluded to conserve space. 44 Table 1A.8 : Percentage of sample households producing legumes by household - level SI category Household - level SI category Type of legume production Non - adoption Intensification Sustainable SI Total (a) Only v ia maize - legume intercropping 8.2 10.3 76.0 77.2 41.3 (b) Only o n pure - stand legume plots or via intercropping legumes with non - maize crops 9.4 17.5 2.4 4.1 6.7 (c) Both maize - legume intercropping and pure - stand legumes or intercropping legumes with non - maize crops 1.7 3.2 9.3 8.3 5.4 ( d ) Any legume production (a , b, or c ) 19.3 31.0 87.7 89.6 53.4 Note s : N=1,871 sample households. Values in row (a) for Non - adoption and Intensification are not zero because household - level SI categories are used for this table and recall that these categories are based on the SI category that accounts for the largest share of the h 45 REFERENCES 46 REFERENCES Abdulai, A.N. 2016. Impact of conservation agriculture technology on household welfare in Zambia. Agricultural Economics 47(6) : 729 - 741. Agboola, A.A., and Fayemi, A.A. 1971. Preliminary trials on the intercropping of maize with different tropical legumes in Western Nigeria. The Journal of Agricultural Science 77 (2) : 219 - 225. Alderman, H., Hoogeveen, H., and Rossi, M. 2006. Reducing child malnutrition in Tanzania: Combined effects of income growth and program interventions. Economics and Human Biology 4 (1) : 1 - 23. Amare, M., and Shiferaw, B. 2017. Nonfarm employment, agricultural intensification, and productivity change: Empirical findings from Uganda. Agricultural Economics 48(S1) : 59 - 72. Black, R.E., Victora, C.G., Walker, S.P., Bhutta, Z.A., C hristian, P., De Onis, M., Ezzati, M., Grantham - McGregor, S., Katz, J., Martorell, R., and Uauy, R. 2013. Maternal and child undernutrition and overweight in low - income and middle - income countries. Lancet 382 (9890) : 427 - 451. Bronick, C.J., and Lal, R. 2005. Soil structure and management: a review. Geoderma 124 (1) : 3 - 22. Byrnes, B.H. 1990. Environmental effects of N fertilizer use: an overview. Fertilizer Research 26 : 209 - 215. Chang, E. 2005 . Electoral incentives for political corruption under open - list proportional systems. The Journal of Politics 67(3) : 716 - 730. Chen, J.H. 2006. The combined use of chemical and organic fertilizers and/or biofertilizer for crop growth and soil fertility. Inte rnational workshop on sustained management of the soil - rhizosphere system for efficient crop production and fertilizer use (Vol. 16, p. 20). Land Development Department, Bangkok, Thailand, 16 - 20 October 2006. Crews, T.E., and Peoples, M.B. 2005. Can the synchrony of nitrogen supply and crop demand be improved in legume and fertilizer - based agroecosystems? A review. Nutrient Cycling in Agroecosystems 72 (2) : 101 - 120. Deb, P., and Trivedi, P.K. 2006a. Specification and simulated l The Econometrics Journal 9 (2) : 307 - 331. 47 Deb, P., and Trivedi, P.K. 2006b. Maximum simulated likelihood estimation of a negative binomial re gression model with multinomial endogenous treatment. The Stata Journal 6 (2) : 246 - 255. Di Falco, S., and Veronesi, M. 2013. How can African agriculture adapt to climate change? A counterfactual analysis from Ethiopia. Land Economics 89(4) : 743 - 766. Di Falc o, S., Veronesi, M., and Yesuf, M. 2011. Does adaptation to climate change provide food security? A micro - perspective from Ethiopia. American Journal of Agricultural Economics 93 : 829 - 846. Droppelmann, K.J., Snapp, S.S., and Waddington, S.R. 2017. Sustainable intensification options for smallholder maize - based farming systems in sub - Saharan Africa. Food Security 9(1) : 133 - 150. Food and Agriculture Organization of the United Nations (FAO), 2015. The State of Food Insecurity in the World 2015. Meeting the 2015 international hunger targets: taking stock of uneven progress. Rome, FAO. Accessed April 2017. Available at http://www.fao.org/3/a - i4646e.pdf . Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., Pretty, J., Robinson, S., Thomas, S.M., and Toulmin, C. 2010. Food security: the challenge of feeding 9 billion people. Science 327(5967) : 812 - 818. Hawkes, C., and Ruel, M.T. 2006. Understanding the links between agriculture and health. 2020 Vision Focus 13. International Food Policy Research Institute (IFPRI), Washington, DC, 2006. Accessed August 2017. Available at https://www.researchgate.net/publication/5055724_Understanding_the_Links_Between _Agriculture_and_Health . Hart, M.R., Quin, B.F., and Nguyen, M.L. 2004. Phosphorus runoff from agricultural land and direct fertilizer effects: a review. Journal of Environmental Quality 33(6) : 1954 - 1972. Heady, D., Stifel, D., You, L., and Guo, Z. 2018. Remoteness, urbanization, and child nutrition in sub - S aharan Africa. Agricultural Economics 49(6) : 765 - 775. Herforth, A., and Harris, J. 2014. Understanding and Applying Primary Pathways and Principles. Brief #1. Improving Nutrition through Agriculture Technical Brief Series. Arlington, VA: USAID/Strengthenin g Partnerships, Results, and Innovations in Nutrition Globally (SPRING) Project. Accessed March 2017. Available at https://www .spring - nutrition.org/sites/default/files/publications/briefs/spring_understandingpathways_brief _1.pdf . Hirvonen, K., and from rural Ethiopia. Agricultural Economics 48(4), 469 - 480. 48 intake: evidence from rur al Bangladesh. Economic Development and Cultural Change 66(3) : 555 - 583. Jayne, T.S., Mason, N.M., Burke, W.B., and - generation agricultural input subsidy programs. Food Policy 75 : 1 - 14. Jones, A.D., Shrinivas, A., and Bezner - Kerr, R. 2014. Farm production diversity is associated with greater household dietary diversity in Malawi: Findings from nationally representative data. Food Policy 46 : 1 - 12. Kassie, M., Jaleta, M., Shiferaw, B., Mmbando, F., and Mekuria, M. 2013. Adoption of interrelated sustainable agricultural practices in smallholder systems: Evidence from rural Tan zania. Technological Forecasting and Social Change 80 : 525 - 540. Kassie, M., Teklewold, H., Jaleta, M., Marenya, P., and Erenstein, O. 2015a. Understanding the adoption of a portfolio of sustainable intensification practices in eastern and southern Africa. Land Use Policy 42:400 - 411. Kassie, M., Teklewold, H., Marenya, P., Jaleta, M., and Erenstein, O. 2015b. Production risks and food security under alternative technology choices in Malawi: Application of a multinomial endogenous switching regression. Journ al of Agricultural Economics 66(3) : 640 - 659. Koppmair, S., Kassie, M., and Qaim, M. 2016. Farm production, market access and dietary diversity in Malawi. Public Health Nutrition 20(2) : 325 - 335. Kumar, N., Harris, J., and Rawat, R. 2015. If they grow it, will they eat and grow? Evidence from Zambia on agricultural diversity and child undernutrition. The Journal of Development Studies 51 (8) : 1060 - 1077. Liu, Y., and Myers, R. 2009. Model selection in stochastic frontier analy sis with an application to maize production in Kenya. Journal of Productivity Analysis 33(1) : 33 - 46. Loos, J., Abson, D.J., Chappell, M.J., Hanspach, J., Mikulcak, F., and Tichit, M. 2014. Putting meaning back into sustainable intensification. Frontiers in Ecology and the Environment 12 : 356 - 361. Manda, J., Gardebroek, C., Khonje, M.G., Alene, A.D., Mutenje, M., and Kassie, M. 2016a. Determinants of child nutritional status in the eastern province of Zambia: the role of improved maize varieties. Food Securi ty 8(1) : 239 - 253. Manda, J., Alene, A.D., Gardebroek, C., Kassie, M., and Tembo, G. 2016b. Adoption and impacts of sustainable agricultural practices on maize yields and incomes: evidence from rural Zambia. Journal of Agricultural Economics 67(1) : 130 - 153. 49 Mason, N.M., Jayne, T.S., and Van de Walle, N. 2017. The political economy of fertilizer subsidy programs in Africa: evidence from Zambia. American Journal of Agricultural Economics 99(3) : 705 - 731. Mather, D.L., and Jayne, T.S. 2018. Fertilizer subsidies and the role of targeting in crowding out: evidence from Kenya. Food Security 10(2) : 397 - 417. Mather, D.L., and Minde, I. 2016. Fertilizer subsidies and how targeting conditions crowding in/out: An assessment of small holder fertilizer demand in Tanzania. GISAIA/Tanzania Working Paper #5, Michigan State University, East Lansing, MI. Messina, M.J. 1999. Legumes and soybeans: overview of their nutritional profiles and health effects. American Journal of Clinical Nutriti on 70(3) : 439 - 450. Montpellier Panel, 2013. Sustainable Intensification: A New Paradigm for African Agriculture . Agriculture for Impact, London. Mpeketula P.M., and Snapp, S.S. 2018. Structural stability conditions soil carbon gains from compost management and rotational diversity. Soil Science Society of America Journal 83:203 - 211. Musumba, M., Grabowski, P., Palm, C., and Snapp. S. S. 2017. Guide for the sustainable intensification assessment framework. University of Florida and Michigan State University. Accessed December 2017. Available at https://www.k - state.edu/siil/documents/docs_siframework/Guide%20for%20SI%20Assessment%20Fr amework%20 - %2010.24.17.pdf . Ollenburger, M., and Snapp, S.S. 2014. Model applications for sustainable intensification of maize - based smallholder cropping in a changing in world. In: Ahuja, L.R., Ma, L., Lascano, R.J., (Ed.), Practical Applications of Agricultural System Models to Optimize the Use of Limited Water. Madison, WI, 2014, pp. 375 - 398. Petersen, B., and Snapp, S. S. 2015. What is sustainable intensification? Views from expert. Land Use Policy 46 : 1 - 10. Pingali, P.L. 2012. Green revolution: impacts, limits, and the path ahead. Proceedings of the National Academy of Sciences 109(31) : 12302 - 12308. Pinstrup - Andersen, P. 2007 Agricultural research and policy for better health and nutrition in developing countries: a food systems approach. Agricultural Economics 37 : 187 - 198. Place, F., Barrett, C.B., Freeman, H.A., Ramisch, J.J., and Vanlauwe, B. 2003. Prospects for integra ted soil fertility management using organic and inorganic inputs: evidence from smallholder African agricultural systems. Food Policy 28 : 365 - 378. 50 Parvathi, P. 2018. Does mixed crop - livestock farming lead to less diversified diets among smallholders? Evide nce from Laos. Agricultural Economics 49(4) : 497 - 509. Rusinamhodzi, L., Corbeels, M., Nyamangara, J., and Giller, K.E. 2012. Maize - grain legume intercropping is an attractive option for ecological intensification that reduces climatic risk for smallholder farmers in central Mozambique. Field Crops Research 136 : 12 - 22. Schmidt, E., Chinowsky, P., Robinson, S., and Strzepek, K. 2017. Determinants and impact of sustainable land management (SLM) investments: A systems evaluation in the Blue Nile Basin, Ethiopia. Agricultural Economics 48(5) : 613 - 627. Sibhatu, K.T., Krishna, V.V., and Qaim, M. 2015. Production diversity and dietary diversity in Proceedings of the National Academy of Sciences 112(34) : 10657 - 10662. Snapp, S.S., Blackie, M.J., Gilbert, R.A., Bezner - Kerr, R., and Kanyama - Phiri, G.Y. 2010. Biodiversity can support a greener revolution in Africa. Proceedings of the National Academy of Sciences 107(48) : 20840 - 20845. Stahley, K., Slakie, E., Derksen - Schrock, K., Gugerty, M.K., and Anderson, C.L. 2012. Tanzania National Panel Survey LSMS - ISA: Legumes. Evans School Policy Analysis and Research (EPAR) No. 189. Accessed May 2018. Available at https://evans.uw.edu/sites/default/files/EPAR_UW_Request%23189_LSMSLegumes_0 7.11_0.pdf . Stephenson, K.B., Agapova, S.E., Divala, O., Kaimila, Y., Maleta, K.M., Thakwalakwa, C., Ordiz M.I., Trehan, I., and Manary, M.J. 2017. Complementary feeding w ith cowpea reduces growth faltering in rural Malawian infants: a blind, randomized, controlled clinical trial. American Journal of Clinical Nutrition 106(6):1500 - 1507. Tanzania Food and Nutrition Centre . 2014. Tanzania National Nutrition Survey 2014. Acce ssed August 2017. Available at https://www.unicef.org/esaro/Tanzania_National_Nutrition_Survey_2014_Final_Repor t_18012015.pdf . Tanzania National Bureau of Statistics (TNBS) . 2014. Tanzania National Panel Survey Wave 3, 2012 - 2013, Dar es Salaam, Tanzania, NBS. Accessed June 2016. Available at http://www.nbs.go.tz/nbs/takwimu/Statistical_Methods_and_Standards/NPS_Wave_3% 20_Final%20_Report.pdf . Tilman, D., Cassman, K.G., Matson, P.A., Naylor, R., and Polasky, S. 2002. Agricultural sustainability and intensive production practices. Natu re 418(6898) : 671 - 677. Tully, K., Sullivan C., Weil, R., and Sanchez, P. 2015. The state of soil degradation in sub - Saharan Africa: Baselines, trajectories, and solutions. Sustainability 7 : 6523 - 6552. 51 UNICEF . 2009. Tracking progress on child and maternal n utrition: a survival and development priority. New York, NY. Accessed July 2017. Available at https://www.unicef.org/publications/index_51656.html . UNICEF, WHO, World Bank Group. 2018. Levels and trends in child malnutrition Joint child malnutrition estimates 2018 edition. Accessed May 2018. Available at http://www.who.int/nut growthdb/2018 - jme - brochure.pdf?ua=1 . Victora, C.G., de Onis, M., Hallal, P.C., Blössner, M., and Shrimpton, R. 2010. Worldwide timing of growth faltering: revisiting implications for interventions. Pediatrics 125 : e473 - 480. Waddington, S.R., Mekuria, M., Siziba, S., and Karigwindi, J. 2007. Long - term yield sustainability and financial returns from grain legume - maize intercrops on a sandy soil in subhumid north central Zimbabwe. Experimental Agriculture 43(3) : 489 - 503. Wainaina, P., Tongruksawattana, S., an d Qaim, M. 2016. Tradeoffs and complementarities in the adoption of improved seeds, fertilizer, and natural resource management technologies in Kenya. Agricultural Economics 47(3) : 351 - 362. Woodfine, A. 2009. Using sustainable land management practices to adopt to and mitigate climate change in sub - Saharan Africa. Resource Guide Version 1. Accessed June 2017. Available at http://www.ipcinfo.org/fileadmin/user_upload/terrafrica/docs/SLM_SUB - SAHARAN_AFRICA.pdf . Wooldridge, J.M. 2010. Econometric Analysis of Cross Section and Panel Data . MIT Press, Cambridge, MA. Wossen, T., Berger, T., and Di Falco, S. 2015. Social capital, risk preference and adoption of improved farm land management practices in Ethiopia. Agricultural Economics 46(1) : 81 - 97. World Bank. 2014. Tanzania public expenditure review: National Agricultural Input Voucher Schem e. Washington, DC. Accessed November 2018. Available at https://openknowledge.worldbank.org/handle/10986/18247?locale - attribute=en . Wu, J.J., and Babcock, B.A. 1998 . The choice of tillage, rotation, and soil testing practices: economic and environmental implications. American Journal of Agricultural Economics 80(3) : 494 - 511. Zeng, D., Alwang, J., Norton, G.W., Shiferaw, B., Jaleta, M., and Yirga, C. 2017. Agricultura l technology adoption and child nutrition enhancement: improved maize varieties in rural Ethiopia. Agricultural Economics 48 : 573 - 586. 52 CHAPTER 2 THE IMPACTS OF SUSTAINABLE INTENSIFICATION OF MAIZE PRODUCTION ON HOUSEHOLD CROP INCOME, PRODUCTIVITY, AND FOOD ACCESS IN RURAL TANZANIA 2.1 Introduction A key challenge in sub - Saharan Africa (SSA) is how to simultaneously raise agricultural productivity and household incomes while achieving food security and environmental sustainability goals . In many SSA countries, low crop yields are closely linked to degraded and infertile soils, which are caused by a variety of factors including continuous monocropping, inadequate investment in organic matter recycling, and climatic variability (Ngwira et al., 2012; Manda et al., 2016a). Given that agriculture is the main source of livelihood for the majority of rural small - scale farm households in SSA, the use of more efficient farming practices or technologies is crucial for alleviating food insecurity (D i Falco and Veronesi, 2013; Khonje et al., 2018). For example, conventional agricultural intensification via improved seed and inorganic fertilizer can improve crop yields in the short - term. This could, in turn, increase the quantity of food available for home consumption or increase household income, which could then be used to purchase more or better quality food, thereby contributing to improved food access, an important dimension of food security. 21 However, there is growing agreement that increased use of these inputs alone is insufficient to intensify agricultural production over the long - term (The Montpellier Panel, 2013; Kassie et al. 2015a). In addition, continuous use of inorganic fertilizer without complementary organic inputs and management practi ces could result 21 Food security is defined as when safe and nutritious food that meets their dietary needs and food prefe 1996). F ood security consists of four dimensions: food availability, access, utilization, and stability. 53 in negative environmental externalities (Pingali, 2012). Thus, conventional agricultural intensification may not be a viable solution for achieving and maintaining food security and environmental sustainability in the long - run. Researcher s and policymakers are therefore increasingly interested in how to achieve sustainable agricultural intensification (SI) and to leverage it to food security and environmental objectives (Godfray 2015). At the core of SI is the goal of increasing agricultur al yields without bringing new land under cultivation, while minimizing adverse environmental impacts (Godfray et al. 2010; Pretty et al. 2011). Holden (2018) and Jayne et al. (2019) suggest that the combined use of inorganic fertilizer and organic soil fe rtility practices (which is a form of Integrated Soil Fertility Management (ISFM)) is an approach to SI. 22 Given the potential of SI of crop production to address low crop yields and food insecurity issues, the main objective of this study is to estimate the impacts of SI on smallholder farm household productivity, incomes, and food security something that has not been rigorously exa mined in the previous literature. Instead, most previous studies on the household welfare and food security effects of various farming practices have focused on individual practices like minimum or zero tillage, improved maize/wheat varieties, inorganic fe rtilizer, or cereal - legume rotation or intercropping (e.g., Jaleta et al., 2016; Zeng et al. 2015; Shiferaw et al. 2014; Magrini and Vigani, 2016; Sauer et al., 2018). However, the use of any one of these practices individually is unlikely to contribute to SI. Moreover, while there are a handful of combinations of agricultural practices that could contribute to SI (e.g., Manda et al. (2016a) and Khonje et al. (2018) for Zamb ia; Kassie et al. (2015b) for Malawi; and Teklewold et al. (2013) and Kassie et 22 Holden (2018) also lists Conservation Agriculture (CA) as an approach to SI. CA is based on three princi ples: crop rotation / intercropping with legumes , permanent soil cover, and minimum or zero tillage . 54 al. (2018) for Ethiopia), none of these studies estimate the effects of the practices on household food security. To fill these gaps, this study uses nationally representative household panel survey data from Tanzania (the Tanzania National Panel Surveys (TNPS) of 2008/09, 2010/11, and 2012/13 described below) to estimate the effects on rural maize - crop income, and food access of the use of various combinations of three important soil fertility management (SFM) practices that could contribute to SI of maize - based production systems in Tanzania. 23 The three focal SFM practices are inorganic fertilizer, organic fertilizer, and maize - legume i ntercropping . 24 Understandi ng SI in the context of maize - based production systems is particularly important in Tanzania because maize provides about half the household calories for smallholder farms across Tanzania (Cochrane and D ' Souza 2015) and the area p lanted with maize accounts for 75% of the total cultivated area in the country (Tanzania National Bureau of Statistics ( TNBS ) 2014 ). We follow Kim et al. (in press) and group the eight possible combinations of these three SFM practices on a given maize plot into four SI categories: (i) - (animal manure or compost) , maize - legume intercropping, or both groupings is discussed further below and at length in Kim et al. (in press) but, b riefly, joint use of inorganic fertilizer with maize - legume intercropping and/or organic fertilizer on a given maize 23 We focus on food access due to data constraints that limit our ability to examine impacts on the other three dimensions of food security. 24 There are other practices for example, maize - legume rotation and minimum tillage that have the potential to contribute to SI, but they are not widely used in Tanzania as of yet and are not captured in the TNPS data. The three SFM practices on which we focus are the most common ones used in maize - based systems in rural Tanzania. 55 productivity while preserving or enhancing soil hea lth due to synergistic or complementary effects among the practices. We estimate the effects of use of practices in the various SI categories on net crop income and producti vity in addition to food access because crop income and food production /productivi ty are considered the two main potential impact pathways through which changes in cropping practices including the SFM practices studied here are likely to affect household food access as well as food secu rity and nutrition more broadly (Herforth and Harri s, 2014; Kumar et al., 2015). 25 For example , the practice(s) in each SI category - crop production or productivity in terms of the quality and/or quantity of crops produced on their maize plot , which household members could consume directly . In addition, it could increase a household s crop income through generating larger quantities of the crops that can be sold at the market, which, in turn, allows farmers to purchase more and/ or better quality food (see Figure 2.1). We consider several measures of food access (described further below): household food expenditure per adult equivalent, a mod ified version consumption score (FCS). 25 contain information that would enable empirical analysis of this pathway. 56 Figure 2.1: Conceptual pathways between S I of maize production and household food access Source s : Modified from Herforth and Ha rris (2014) and Kim et al. (in press) This study makes several contributions to the previous literature. First, to our knowledge, it is the first empirical examination of the impacts of households' use of combinations of SFM practices (as opposed to individual practices) on household food access . Second, we g o beyond previous studies on the impacts of combined use of agricultural practices by considering joint use of maize - legume intercropping with inorganic fertilizer and rigorously examin ing the effects of such joint use on household crop income, productivity, and food access. Some previous studies (Kassie et al., 2015b; Kassie et al., 2018) consider maize - legume intercropping in their analyses but group it with maize - rop diver sification . Third , we use nationally representative household panel survey data, whereas most of the previous studies that are closely related to this study use either cross - sectional or panel data but not nationally representative panel data (Manda et al . 2016a; Khonje et al., 2018; Kassie et al. 2015b, 2018; Teklewold et al. 2013) . The data used here should improve both the external validity of our findings (because the data are nationally representative) as well as the internal validity thereof (because we use panel data methods namely, the Mundlak - Chamberlain correlated random effects (CRE) approach combined with multinomial endogenous switching regression (MESR) methods to control for selection bias ). Finally, the study complements and extends Use of practices in a given maize SI category (Intensification, Sustainable, or SI) Crop P roduction/ P roductivity Crop Income Household Food Access Food Expenditure 57 Kim et al. (in press), which estimates the effects of the use of the same set of practices we consider here on child nutrition outcomes among rural maize - growing households in Tanzania but does not estimate the effects on food access and makes only a cursory examination of the effects of the practices on productivity and incomes. Kim et al. find positive effects of use of this is also the case for household food access - related outcomes. Moreover, understanding if there are such effects on household food access could help to further explain the pathways through which SI of maize production affects child nutrition. Our CRE - MESR results suggest that relative to Non - adoption, the use of practices in each of the other three SI categor ies s a - related outcomes an d crop productivit y. Of these group was the most effective, providing the largest effects on net crop income and net crop income per adult equivalent. On the other hand, increases in all three food acce ss outcomes, while the effects of using practices in the statistical significance and the extent of the effects. The rest of this study is organized as follows. Section 2.2 provides background information on the use of the focal SFM practices in Tanzania. Section 2.3 outlines the econometric approaches. Section 2.4 describes the data and food security outcome variables used in this study. The results are prese nted and discussed in Section 2.5 and the last section draws conclusions and policy implications. 58 2.2 SI of maize production in Tanzania We begin this section by briefly describing the rationale for the SI categories used here. The reader is referred to Kim et al. (in press) for a much more detailed discussion of the rationale, i ncluding extensive references to empirical evidence supporting the categorizations. We then describe the prevalence of use of practices in the various SI categories in rural Tanzania. ) is not considered use of inorganic fertilizer without complementary SFM practices ( Matson et al., 1997; Pingali, 2012; Petersen and Snapp, 2015; Bronick and L al, 2005 ). For this reason, although use of inorganic fertilizer can raise maize yields in the short - run, these yield increases are unlikely to be sustained in the long - maize - legume i ntercropping) can improve soil fertility in the longer - run and use locally available resources but in the absence of inorganic fertilizer, they are unlikely to appreciably increase crop productivity, particularly in the short - run. However, when inorganic f ertilizer is used jointly with maize - legume intercropping or organic fertilizer, there are several potential synergistic or complementary effects which can result in higher productivity while maintaining or improving soil fertility. For instance, improving soil organic matter ( SOM ) levels through the application of organic fertilizer or maize - legume intercropping could increase maize yield response to inorganic fertilizer (Marenya and Barrett 2009; Jayne et al. 2018). Moreover, there is empirical evidence t returns. For instance , Waddington et al. (2007) observed during the years from 1993 to 2006 in Zimbabwe that maize yields were about two times larger on average with a joint application of maize - legume intercropping and inorganic fertilizer than with maize - legume intercropping alone. 59 Moreover, work by Mekuria an d Waddington (2002) in Zimbabwe suggests that maize gross margins per hectare increased by about 7.5 times when inorganic fertilizer was jointly used with animal manure compared to when the same quantity of inorganic fertilizer was used without manure . Tab le 2.1 summarizes the prevalence of use of the three focal SFM practices and combinations thereof on maize plots in Tanzania. Out of 5,419 maize plots in the sample (TNPS 2008/09, 2010/11, and 2012/13, described below), 46.5% (case 1) of them have none of the SFM practices applied while 39.9% have only one of the three practices applied: 7.8% for the use of inorganic fertilizer only (case 2), 6.6% for the use of organic fertilizer only (case 3), and 25.5% for intercropping maize with legumes (case 4). On th e contrary , relatively few maize plots (13.6%) have two or more SFM practices applied (cases 5, 6, 7, and 8). Table 2.1 also shows the four SI groups at the plot - accounts for 37.1% of all maize plots prevalent. Among the and at least maize - legume intercropping is the dominant case (6.8% of all maize plot s), while combined use of inorganic fertilizer and at least organic fertilizer is less common. Since some households have multiple maize plots that might be managed in different ways, we generate a household - level SI category variable and use it to estima te the effects of the SFM strategy on household - level food access and other outcome variables. For the household level SI category variable , we com p ute the maize areas cultivated by the household under each SI category and then select the category with the largest area. As shown in Table 2.1, the plot - and HH - level prevalences of the various SI categories are very similar. This is due to the following reasons: (i) about 65% of sample 60 households in this study have only one maize plot , and (2) households with multiple maize plots hav e a tendency to use the same set of practices on all maize plots. Overall, 87% of the maize plots owned by our sample households fall in the same SI group at the plot and household level s . Table 2.1: SI categories and prevalence on maize plots and among maize - growing households in Tanzania Cas e Inorganic fertilizer Organic fertilizer Maize - legume intercropping Number of maize plots (%) SI category Plot level (%) HH level (%) 1 2,519 (46.5) Non - adoption 46.5 44.4 2 422 (7.8) Intensification 7.8 6.6 3 358 (6.6) Sustainable 37.1 39.5 4 1,384 (25.5) 5 267 (4.9) 6 102 (1.9) SI 8.7 9.4 7 296 (5.5) 8 71 (1.3) Use of inorganic fertilizer 15.4 16.1 Use of organic fertilizer 14.2 18.1 Use of maize - legume intercropping 38.2 46.6 Notes: Figures in the plot level column are based on all maize plots (N=5,419) completed harvesting by rural households pooled across the three waves of the TNPS (2008/09, 2010/11, and 2012/13). Figures in the HH level column are based on the total number of mai ze growers (N=3,641) in rural areas across these surveys. Legume crops for maize - legume intercropping are beans, soybeans, groundnuts, cowpeas, pigeon peas, chickpeas, field peas, green grams, bambara nuts, and fiwi. 61 2.3 Empirical strategy In this section, we outline the econometric approaches used in this study. To empirically estimate the impacts of a household use of a given set of agricultural practices based on observational data, a key challenge is to con trol for potential selection bias, where farmers often self - select into use or non - use of a given technology or combination of technologies . In the context of this study , selection bias occurs if unobserved characteristics influencing a decision on which set of SFM practic es to use are correlated with the outcome variables considered here . If the selection bias is not adequately addressed, then econometric estimates are biased and inconsistent. One frequently used method to control for selection bias is propensity score matching; however, this approach only controls for selection on observable characteristics (Smith and Todd, 2005). Selection may also be related to unobservable factors. In order to address selection bias issues originati ng from observed and unobserved heterogeneity, we use an MESR approach following Kassie et al. (2018) and Khonje et al. (2018). The MESR framework involves a two - of which set of SFM practi ces to use (i.e., their SI category) is estimated in a multinomial logit selection (MNLS) model accounting for unobserved heterogeneity, and an inverse Mills ratio (IMR) is generated for each SI category. These are referred to as s election correction terms . In the second stage, the impacts of using each set of practices on a given outcome variable are estimated using ordinary least squares (OLS) with the IMRs included as additional covariates to capt ure selection bias arising from time - varying unobserved he terogeneity (Kassie et al. 2018). Other empirical studies that have applied the MESR model include Di Falco and Veronesi (2013), Teklewold et al. (2013), Kassie et al. (2015, 2018), and Khonje et al . (2018), among others. 62 In addition, we combine the MESR m odel with CRE techniques to further control for time - invariant unobserved household - level heterogeneity . To implement this approach, the means of time - varying covariates are included as additional regressors in both the first and the second stages (Wooldridge, 2010). 2.3.1 CRE - MNLS model In the first stage, a farmer s decision of which SI category to be in is modeled in a random utility framework. Following Kassie et al. (2018) and K honje et al. (2018), consider the following latent variable ( ) below that specifies a maize grower i choosing strategy j (i.e., - ); and ) in this study) at time t over all other alternative strategies, m : , (1) where is a vector of observed exogenous covariates that represent household head characteristics, household e ndowments of physical, human, and social capital, agricultural extension and access to information and market services, shocks and other constraints, an d input and expected output prices (described in 2.4.2 and Table 2A.1); are the time - average s of these covariates to control for time - invariant household - level unobserved heterogeneity; is time - varying unobserved characteristics; and and are vectors of parameters to be estimated, respectively. Farmer i t directly observed but we do observe their SI category decision (strategy) . It is assumed that farmer i will ch oose strategy j if strategy j provides greater utility than any other strategy m j (equation (2)) : 63 (2) where . U nder the assumption that the are independently and identically Gumbel - distributed, the probability that farmer i at time t will choose SI category j can be specified by a CRE - MNLS model as follows (McFadden, 1973): , (3) 2.3.2 CRE - MESR model In the second stage of the CRE - MESR model, we investigate the impacts of each strategy on a on and net crop income from its maize plots as we ll as its food access decision. The model in our study implies that households face a total of four regimes (i.e., ). The outcome equation for each regime is specified as: (4) where is the value of a given outcome variables for household i in regime j at time t ; and are vectors of explanatory variables and their household time - averages, respectively; and the error terms . The outcome equation for each regime is estimated separately v ia OLS. However, if the error terms of the CRE - MNLS model ( in equation (1)) are correlated with the error terms of the outcome 64 equation, the expected values of conditional on the sample selection are non - zero, and then OLS estimates of equation (4) will be inconsistent. To address this potential inconsistency, selection correction terms for the alternative choices are included in equation (4), which takes into account the correlation between the he Bourguignon et al. (2007), consistent estimates of and in the outcome equations (equation (4)) can be obtained via estimation of the following CRE - MESR models: (5) In equation (5), is the error term, the expected value of which is zero; denotes the covariance between the is the estimated IMR, computed as follows: , (6) where is the correlation coefficient between the selection correction terms are included in the outcome equation , one for each adoption regime. Standard errors in equation (5) are bootstrapped to account for the two - stage estimation procedure (Di Falco and Veronesi, 2013). For the model to be identified, it is critical to use selection instruments as exclusion rest rictions in addition to selection terms automatically generated by the selection model of adoption (Di Falco and Veronesi, 2013). Following Kim et al. (in press), this study considers six candida te instrumental variables (IVs) that may directly influence a decision of which set of SFM practices to use but not their food access and other outcome variables . The candidate IVs are: household itself) (i) that received adv ice on agricultural production, (ii) that used inorganic fertilizer, (iii) that used organic fertilizer, and (iv) that used maize - legume intercropping; (v) 65 electoral threat at the district level; and (vi) the number of National Agricultural In put Voucher S 26 The first four IVs are associated with access to info rmation on and the potential for social learning about the three SFM practices considered in this study. 27 For I Vs strategies, especially the sole use and/or combinations of inorganic fertilizer and other practices, coul d be influenced by subsidized fertilizer vouchers obtained from the Tanzanian government input subsidy pr ogram. Ho vouchers but are exogenous to an individual household. The district - level electoral thr eat IV, d efined as in Chang (2005), is the proportion of votes won in the most recent presidential election by the runner - up candidate divided by the proportion of votes won by the ultimately winning candidate. Several recent studies have used measures of electoral outcomes as an IV for household receipt of subsidized fertilizer because the spatial allocation of subsidized inputs in SSA countries, including Tanzania, may be connected to voting patterns during the most recent election (Mather and Minde, 2016 ; Mason e t al., 2017; Mather and Jayne, 2018). Using constituency - level data from the 2005 and 2010 Tanzania presidential elections , this study generates the district - level electoral threat variable by aggregating up the vote totals to the district - concentrated on the areas that are the most suitable for maize and rice production. We thus 26 TNPS includes total 26 regions, where each region is subdivided into districts. The districts are further divided into wards, where a ward is an administrative structure for one single town or portion of a bigger town. 27 In recent studies on agricultural technology adoption decisions and their impacts (Kassie et al. 2015; Di Falco and Veroneisi 2013; Khonje et al. 2018), similar variables representing better access to information on modern twork, and government extension, etc.) have been used as selection instruments. 66 include the number of vouchers for in organic fertilizer allocated to th the World Bank (2014). Of these six candidate IVs, we only include in a given first - stage regression the IVs that pass a simple falsification test . To be a valid selection instrument, it affects t he category decision, but does not directly affect the outcome variable (Di Falco et al. 2011; Kassie et al. 2018). We conduct the falsification test fo llowing Khonje et al. (2018) and Kassie et al. (2018) : the six candidate IVs are tested to determine if the y are statistically significant in the CRE - pooled OLS model of each SI category ; we then drop the IVs that are significantly correlated with a given outcome variable. 2.3.3 Estimation of average treatment effects on the treated (ATT) The CRE - MESR framework above can be used to compute the average treatment effects on the treated (ATT) by comparing the expected outcomes of and non - users - of each SI strateg y in actual and counterfactual sce narios. These actual and counterfactual scenarios can b e specified as follow: Adopters with adoption (actual), . (7) Nonadopter without adoption (actual), . (8) Adopters had they decided not to adopt (counterfactual), . (9) Nonadopter had they decided to adopt (counterfactual), . (10) 67 Equations (7) and (8) denote for adopters and non - adopters, respectively, the expected values of a given outcome variable that are actually revealed in the sample while equations (9) and (10) refer to their counterfactual s (Kassie et al. 2018) . For example, the counterfactual scenario described in equation (9) is defined as the outcome of adopters tha t would have been obtained if the coefficients on their explanatory variables ( ) had been the same as the coefficients on the explanatory variables of the nonadopters, and vice versa ( Ibid. ). After estimating the CRE - MESR model, these conditional expectations are used to der ive the ATT, which is defined as the difference between equa tions (7) and (9): 28 . (11) The first term in equation (11) indicates the expected change in the mean of the outcome variable if the characteristics of adopters had been the same as nonadopters. The second term ( ) in equation (9) along with the CRE approach ( ) corrects for selection bia s and endogeneity caused by unobserved heterogeneity. 28 We also estimate the average treatment effects on the untreated (ATU), calculated as the difference between equations (8) and (10). These results are presented in Table 2A.16 of the Appendix. 68 2.4 Data and key outcome variables 2.4.1 Data This study primarily use s the 2008 / 09, 2010 / 11, an d 2012 /13 TNPS data . 29 The TNPS is part of - Integrated Surveys on Agriculture (LSMS - ISA) project, which was implemented by the TNBS with support from the World Bank. The topics covered in the survey include agricultural produc tion, off - farm activitie s, consumption expenditure, and socioeconomic characteristics, among others . The TNPS was based on a stratified, multi - stage cluster sample design ; the strata were Dar es Salaam, other urban areas and rural areas in mainland Tanzania, and Zanzibar . Clusters, the primary sampling units, were randomly selected from within each stratum with the probability of selection proportional to their population size. Then, eight households were randomly selected from each cluster. 30 The TNP S baseline sample (2008/09 TNPS) comprises 409 clusters and 3,265 households . 97% of households in the first round were re - interviewed in the second round (2010/11 TNPS) , and 96% of the households in the second round were re - interviewed in the third round (2012/13 TNPS) , which gives very low attrition rates between survey rounds (TNBS 2014 ). The analytical sample used for the empirical analysis consists of the unbalanced panel of maize - growing households who have completed harvesting on their maize plots : 3,641 total 29 Data for TNPS 2014/15 (i.e., the fourth wave of the survey) is now publicly available. However, the sample in the fourth wave of the survey was entirely refreshed for all future rounds, where only 860 households corresponding to 68 clus ters were re - interviewed from the TNPS 2012/13. Thus, this study uses only the first three rounds of the survey for analysis. 30 The unit for clusters is census enumeration areas (EAs) in urban areas (as defined in the 2002 Population and Housing Census) an d villages in rural areas. 69 household observations ( 967 observations in 2008/0 9 , 1,176 in 2010/11, and 1,498 in 2012/13) . A slightly different analytical sample is used for the FCS outcome variable . 31 T he TNPS data also include various geospatial variables from other sources such as rainfall data and soil nutrient availability data , which were merged at the household level. 32 Among these, we use household distance to the nearest main road, town, and market . 33 In addition to the TNPS, there are three additional sets of variables d erived from other data sources that are used for the empirical analysis: (i) monthly wholesale pri ce data for maize and rice from the Agricultural Market Information System (AMIS) of the Tanzania Ministry of Industry and Trade (MIT) ; 34 (ii) the number of subsidized inorganic fertilizer vouchers distributed to regions from World Bank (2014); and (iii) co nstituency - level data for the 2005 and 2010 presidential elections from the Electoral Commission of Tanzania. 35 2.4.2 Outcome variables and explanatory vari ables access , we use seven outcome va riables: (i) net crop income from maize plots (henceforth simply i) net crop income per adult equivalent; (iv) crop productivity ( per unit of land ) on maize plots (he crop 31 In the TNPS data used in this study, consumption frequency is captured in the second and third waves (TNPS 2010/11 and 2012/13) but not in the first (2008/09). Therefore, for the FCS outcome variable, the analytical sample invo lves the balanced maize - growing households that were interviewed in both the second and third waves : 1,622 total household observations (811 observations in each wave) . 32 The source of the rainfall data is the National Oceanic and Atmospheric Administration - Climate Prediction Center and that of the soil nutrient availability data is the Harmonized World Soil Database . 33 Each distance variable is from a different source. Distance to main road is from OpenStreetMaps, distance to town is from City Population, and distance to main market is from Famine Early Warning Systems Network. 34 These data are collected weekly from twenty wholesale markets in Tanzania. There are six regions (out of 26 in the TNPS) that are not covered by these data. For these markets is used for the empirical analysis. 35 The author thank s Dr. David Mather for sharing these data. 70 adult equivalent ; (vi) modified household dietary diversity score (HDDS); and (vii) food consumption score (FCS). The first four outcome variables ((i) to (iv)) ar e used as measures for the access and the other three outcome variables ((v) to (vii)) are used as measures of access. We discuss the construction of each of these outcome variables in turn. include intercrops with legumes and/or other crops. To take into account income from all crops on a given maize plot, we compute net crop income at the household level as follows: , (12) where is the quantity (kg) of crop j harvested by household i , is the regional median market price of crop j in TZS/kg, is the cost of input h (i.e., land rental, purchased inorganic fertilizer and organic fertilizer, and hired labor) used by household i , is the cost of seed purchased to produce crop j , and m indexes the maize plots cultivated by household i . Using net crop income per e quation (12), we then generate net crop income per acre of maize plots and net crop income per adult equivalent by dividing net crop income by the total acreage of maize plots cultivated by household i and by the number of adult equivalents in household i , respectively. (Acreage of maize plots refers to the acreage of plots that contain at least some maize.) Given that food prices increase over time and inflation is considerably higher in rural areas than urban areas (TNBS, 2014), real 2013 prices for , x , and s are used to generate the net crop income - related variables. For the crop productivity outcome variable, we calculate an output index following Liu ) including th e area of intercropped plots as follows: 71 , (11) where is the regional median price of maize and the other variables are as defined above. For the food/beverage consumption expenditure, modified HDDS, and FCS outcome variables, we draw on the household food consumption data that were collected in the TNPS. Th ese data are based on a seven - day recall period prior to the survey and cover over 50 food/beverage items. HDDS and FCS are both indicators of the food access component of household food security (Jones et al., 2013; Leroy et al., 2015). The modified HDDS is calculated as a count over 12 food groups (cereals, roots and tubers, vegetables, fruits, meat and poultry , eggs, fish and seafood, pulses/legumes/nuts, milk and milk products, oils and fats, sugar and honey, and miscellaneous) consumed during the seven - day reference period; this variable is thus a count variable with values ranging from zero to 12. 36 The FCS takes on values ranging from zero to 112 as it is calculated as the consumption frequency of nine food groups (main staples, pulses, vegetables, fru it, meat and fish, milk, sugar, oil, and condiments) during the last seven days multiplied by a group - specific weight and then summed up (World Food Programme, 2008). Our third indicator of household food access, consumption expenditure on food and beverag es per adult equivalent , is provided in each round of the TNPS. All sources of consumption are included (purchases, own production, gifts received, and goods bartered in) and the variable only includes the actual consumption of the household over the previ ous seven days. 37 36 The standard HDDS is calculated based on food consumption during the previous 24 hours (Swin dale and Bilinsky, 2006). However, such data are not available in the TNPS so we calculate a modified HDDS based on food consumption during the previous 7 days. 37 For all of these outcome variables except for the modified HDDS and the FCS, a one percent wi nsorization in each tail was used to prevent the results from being heavily influenced by outliers. 72 Descriptive statistics for the seven outcome variables and control variables used in the analysis are presented in Table 2.2. 38 Summary statistics on the six candidate instrumental variables are also included in this table. The control variables were selected based on a careful review of the literature associated with technology adoption and its impacts on household income, productivity, and food security in African countries (e.g., Khonje et al., 2018; Kassie et al., 2015a, b; Kassie et al., 2018; Teklewold et al., 2013; Manda et al., 2016). These variables include characteristics of the household head (age, gender, and education); household endow ments of physical, human, and social capital (family labor defined as the number of adults (15 - 64 years old) per acre of cultivated land, total cultivated land, off - farm income, real value of farm assets (1,000 TZS), livestock ownership, access to credit, membership in a Savings and Credits Cooperatives Society (SACCOS); agricultural extension and access to information, markets, and services (household - level receipt of extension advice from government/NGO, household distance to main road/town/market, presen ce of cooperatives/input supplier within the village); shocks and other constraints (drought/flood and crop disease/pest shocks in the past two years, total rainfall, soil nutrient constraint); 39 and input and proxies for expected output prices (inorganic fertilizer price 38 A detailed description of the variables and summary statistics by SI category are presented in Appendix Table 2A.1. In addition, note that some of the con trol variables in our models are time - invariant for almost all households (e.g., education of the household head, distance to the nearest market, and a binary variable for livestock ownership). Thus, we excluded the time - averages of these variables from bo th stages of the CRE - MESR model. 39 According to the Harmonized World Soil Database, soil nutrient s are estimated based on soil texture, soil organic carbon, soil pH, and total exchangeable bases of the topsoil (0 - 30 cm) and the subsoil (30 - 100 cm). In gene ral, the moderate constraint of the soil nutrient availability is rated between 60% and 80% of the plant growth potential while the severe and very severe constraints are rated between 40% and 60%, and less than 40% of the growth plant potential, respectiv ely. A challenge associated with this database is the coarse resolution relative to the variability of these properties so although this is the only available soil data that could be associated with the TNPS, the mismatch in scale must be acknowledged. 73 at district level, lagged prices of maize and rice at the region level, bean and groundnut price at region level). 40 40 T he average price of inorganic fertilizer per kilogram at district level is included as the major relevant input price in this study. Using data from AMIS - MIT, this study includes the average wholesale prices of maize and rice during the post - harvest period of maize and rice. However, such data are not available for beans and groundnuts, so we instead use the average producer prices of these crops at region level in each TNPS survey round as a proxy for the expected legume prices. All of these input and output prices are deflated by the CPI (2013=100). 74 Table 2.2: Descriptive statistics by survey round TNPS 2008/09 TNPS 2010/11 TNPS 2012/13 Full sample Variables Mean SD Mean SD Mean SD Mean SD Outcome variables Net crop income (1000 TZS) 315.41 544.52 317.29 549.28 382.09 614.87 343.45 576.72 Net crop income per acre 98.32 113.69 93.54 106.01 106.34 119.09 100.08 113.67 Net crop income per adult equivalent 79.85 126.15 74.83 119.58 84.44 118.76 80.12 121.07 Crop productivity 408.55 434.83 394.25 417.22 389.39 414.54 396.05 420.84 Food expenditure per adult equivalent 388.43 218.54 423.40 255.95 578.99 334.10 478.12 294.99 Modified HDDS 7.69 2.03 8.08 1.91 7.90 2.00 7.90 1.98 FCS - - 50.87 16.50 50.68 17.73 50.78 17.12 Explanatory variables Male - headed HH (yes = 1) 0.78 0.42 0.77 0.42 0.78 0.41 0.78 0.42 Age of HH head (years) 47.27 15.95 48.13 15.62 48.57 16.12 48.08 15.92 Education of HH head (years) 4.52 3.32 4.52 3.46 4.72 3.46 4.60 3.42 Family labor (number of adults per acre) 0.99 1.22 1.15 2.00 1.15 1.79 1.11 1.73 Total cultivated land (acres) 6.32 22.75 5.48 8.27 6.93 13.78 6.30 15.42 Off - farm income (yes = 1) 0.54 0.50 0.62 0.49 0.64 0.48 0.61 0.49 Farm assets (1,000 TZS) 1,296.49 7,144.48 1,296.32 7,802.01 1,738.75 7,407.92 1,478.39 7,470.75 Livestock ownership (yes = 1) 0.45 0.50 0.45 0.50 0.42 0.49 0.44 0.50 Access to credit (yes = 1) 0.06 0.23 0.07 0.25 0.10 0.30 0.08 0.27 Membership (SACCOS) (yes = 1) 0.04 0.19 0.05 0.22 0.04 0.19 0.04 0.20 0.16 0.36 0.08 0.28 0.06 0.24 0.09 0.29 Distance to main road (km) 21.45 22.17 23.18 23.55 22.04 22.29 22.25 22.68 Distance to town (km) 56.25 37.14 56.71 38.83 57.66 38.87 56.98 38.40 Distance to main market (km) 85.13 52.53 85.06 54.00 87.09 54.14 85.91 53.66 75 Table 2.2 TNPS 2008/09 TNPS 2010/11 TNPS 2012/13 Full sample Variables Mean SD Mean SD Mean SD Mean SD Explanatory variables Cooperatives (yes = 1) 0.45 0.50 0.42 0.49 0.37 0.48 0.41 0.49 Input supplier (yes = 1) 0.33 0.47 0.36 0.48 0.40 0.49 0.37 0.48 Drought/Flood (yes = 1) 0.10 0.31 0.13 0.34 0.11 0.31 0.11 0.32 Crop disease/Pests (yes = 1) 0.10 0.31 0.10 0.30 0.06 0.24 0.09 0.28 Total rainfall (mm) 756.16 307.13 817.15 296.25 825.24 246.11 804.28 281.29 Soil nutrient constraint (yes = 1) 0.60 0.49 0.63 0.48 0.64 0.48 0.63 0.48 Inorganic fertilizer price (TZS/kg) 1,765.72 712.48 1,426.31 758.01 1,504.83 889.20 1,548.76 814.56 Lagged price of maize (TZS/kg) 353.56 75.44 517.61 89.34 527.35 114.54 478.05 122.91 Lagged price of rice (TZS/kg) 1,089.67 162.80 1,484.96 182.41 1,524.18 146.03 1,396.11 246.56 Bean price (TZS/kg) 1,579.56 241.94 1,615.55 125.51 1,523.58 126.28 1,568.15 169.58 Groundnut price (TZS/kg) 1,685.18 275.80 2,368.40 344.96 1,986.39 301.09 2,029.78 406.51 Year dummy (2010/11) 0.00 0.00 1.00 0.00 0.00 0.00 0.32 0.47 Year dummy (2012/13) 0.00 0.00 0.00 0.00 1.00 0.00 0.41 0.49 T2 dummy 0.28 0.45 0.38 0.49 0.27 0.44 0.31 0.46 T3 dummy 0.54 0.50 0.44 0.50 0.35 0.48 0.43 0.49 Instrumental variables Electoral threat 0.18 0.71 0.20 0.69 0.36 0.34 0.26 0.59 Number of subsidized fertilizer vouchers 47,718 51,537 113,335 106,153 49,854 30,967 69,791 75,116 Proportion receiving agricultural advice 17.25 20.58 10.87 18.15 8.10 14.77 11.43 17.95 Proportion adopting inorganic fertilizer 17.14 28.39 20.22 31.61 17.15 29.86 18.14 30.08 Proportion adopting organic fertilizer 19.59 24.51 18.22 27.75 19.49 27.38 19.11 26.77 Proportion adopting maize - legume IC 45.06 31.89 40.62 31.12 40.96 34.34 41.94 32.73 Notes: TZS = Tanzanian Shillings. SD = standard deviation. IC = intercropping. T2 and T3 dummies are variables for frequency of the household across survey rounds. 76 2.5 Results and discussion The primary objective of this study is to analyze the impacts of the use of practices in each SI crop income and productivity which could be the primary pathways to improve smallholder farmers food access . At t he same time, we examine whether the use of these practices indeed enhance s the first stage regression results in detail beyond those related to the effects of the IVs on the choice of SI strategy . The first stage results are presented in Appendix Table s 2A.2 - 2A.8. 41 The results from a joint significance test of the excl uded IVs in these t ables confirm that the IVs are jointly significant at the 1% level . Moreover, the IVs used in each first stage regression pass the simple falsification test, suggesting that they income - , crop productivity - , or food access - related outcome variables . (See Tables 2A.9 - 2A.15 for the simple falsification test results) . The full CRE - MESR regression results for the secon d stage are reported in Appendix Table s 2A.9 - 2A.15. In some of the outcome equations, the IMRs ( s ) and the mean of time varying variables are statistically significant, implying the presence of sample selection in SI category choice (Kassie et al. 2018). The p redicted outcomes from the CRE - MESR models are used to estimate adoption effects on hou sehold income, productivity, and food a cc ess . Unconditional average effects of various SI category choices on each outcome variable are reported in Appendix Table 2A.17, which are calculated based on the actual and counterfactual distributions. The results show that for all SI categories except for use of practices in each SI category is positively 41 Since the coefficients reported in Appendix Table 2A.2 - 2A.8 are the log - odds of each respective SI category , we need to calculate marginal effects to make inferences based on actual probabilities. The marginal effects for each outcome variable are reported in Appendix Table 2A.18 - 2A.24. 77 income - , productivit y - , and food access - related outcomes relative to non - adoption, o n average. However, these results could be misleading because selection bias from both observed and unobserved factors that may affect the outcome variables has not been addressed in these results (Khonje et al., 2018). Below, we therefore focus on the ave rage effects of use of practices in the various SI categories after controlling for selection bias. 2.5.1 Im pact s of using practices in each SI category on household income and productivity Table 2.3 presents the ATT of the use of practices in the various SI categories crop income and crop productivity , which is calculated as th e difference between column (1) and (2) : f or example maize plots based on the actual combination of SFM practices they used , and the counterfactual that they used none of the practices (i.e., columns (1) and (2), respectively). In all cases, households who use a given set of SFM practice s would have obtained less desirable outcomes if they had not done s o ; all ATTs are positive and s tatistically significant at the 1% level. Of the three SI categories (i.e., s . g roup are hand, for both net crop income and net crop income per adult equivalent, the greatest effects on the outcome variables are observed for the y and these effects are statistically . More specifically, the use of practices in category increases net crop income by 153.2% on average (ATT divided by 78 average counterfactual net cro p income) and net crop income per adult equivalent by 41.5% . For not statistically different from each other. practices in terms of all net crop income - related and prod This is consistent with evidence in the agronomic literature cited above and cited extensively in Kim et al. (in press) that th ere are synergistic or complementary effects when inorganic fertilizer and organic - based SFM practices are used together . For example, the use of organic fertilizer and/or maize - legume intercropping could improve soil quality through increases in SOM and s oil pH level and then enhance crop yield response of applied ino rganic fertilizer use, which could lead to increases in crop income and productivity. In addition, legume crops produced through the use of maize - legume intercropping could help these farmers to further increase their crop income du e to relatively higher market price per kilogram than maize price (if this higher price offsets potentially higher costs of production and lower legume yields per unit of land relat ive to maize) . - legume intercropping and organic fertilizer without inorganic fertilizer being unlikely to signif icantly increase crop yields in the short run. T he results in Table 2.3 are difficult to directly compare with findings in previous studies because each study considered different combinations of agricultural practices. Howev er, our household crop income and productivity effects are consistent with the main findings that the 79 combined use of practices potentially associated with SI p rovides higher maize yields and maize income relative to the use of other practices in Ethiopia (Kassie et al., 2018; Teklewold et al., 2013), and Zambia (Khonje et al. 2018) . More specifically, Kassie et al. (2018) considered the combinations of inorganic fertilizer, an improved maize variety, and legume diversification (maize - legume in tercropping or rotation) and found that the use of legume diversification jointly with at least one of the other two technologies substantially improve household maize yi elds compared to sole or combined use of the other practice s . Similarly, the other two studies used combinations of an improve d maize variety and at least one other practice (minimum tillage in Khonje et al. (2018 ) and maize - legume rotation and /or conservation tillage in Teklewold et al. (2013)) . These latter studi that the combined use of an improved maize variety and at least one of the other practices considered could deliver higher returns on household maize yields and/or maize income compared to any practice on its own . 80 Table 2.3: ATTs of us ing practices in each SI cat egory on household net crop income and productivity Adoption status Outcome variables SI category Adopting Nonadopting ATT (1) (2) (3) = (1) (2) Net crop income (1,000 TZS) Intensification 441.40 (30.62) 232.45 (11.91) 208.95*** (28.99) (N=3,641) Sustainable 365.17 (8.97) 283.45 (5.80) 81.72*** (5.08) SI 549.53 (30.63) 217.02 (8.15) 332.51*** (28.27) Net crop income (1,000 TZS) Intensification 116.52 (4.91) 76.36 (2.00) 40.16*** (4.65) per acre Sustainable 108.15 (0.97) 80.89 (0.85) 27.26*** (0.79) (N=3,641) SI 114.75 (3.86) 81.09 (1.65) 33.66*** (3.97) Net crop income (1,000 TZS) Intensification 102.96 (6.30) 60.55 (2.00) 42.41*** (5.68) per adult equivalent Sustainable 81.77 (1.16) 63.12 (0.86) 18.65*** (0.93) (N=3,641) SI 120.62 (6.08) 56.99 (1.55) 63.63*** (5.53) Crop productivity Intensification 633.22 (19.89) 346.86 (6.97) 286.36*** (18.31) (N=3,641) Sustainable 384.47 (3.59) 286.10 (3.18) 98.37*** (2.90) SI 652.78 (16.78) 346.82 (6.08) 305.96*** (14.91) Notes: Standard errors in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. For each outcome variable, indicates that the ATT is statistically different from at or below the 5% level, while statistically different f ATT . 2.5.2 Impacts of using practices in each SI category on household food access outcomes T he ATTs f or outcomes representing household food access (modified HDDS, food expenditure per adult equivalent, and FCS) are reported in Table 2.4. One observation is that use of practices - HDDS outcome. 42 Below, we discuss the results for each food access indicator in more detail. For the modified HHDS outcome, we find that the use of practices in the 42 Relatedly, Snapp and Fisher (2015) find that a one - crop incre ase in the number of crops intercropped raises the HDDS and FCS. 81 HDDS of 6.5% and 9.2%, respectively s also statistically different from zero but at 0.6%, this effect is very small in magnitude relative to the effects of the other two categories (Table 2.4). Thes e results are consistent with the positive effects of all three of these SI categories on net crop income and productivity in Table 2.3. They are also consistent with the two of the three net crop income - related outcomes in Table 2.3. Thus, improvements in HDDS as a result of and SI appear to be coming through both the crop income and there is no statistically significant difference in the productivity ATTs for - legume intercropping in some of the sets odified HDDS. For the food expenditure per adult equivalent and FCS outcomes, we find that , relative to with increase s of 4.1% and 3.0% on average, respectively . However, in contrast to the HDDS results , the food expenditure per adult equivalent and FCS results suggest that use of practices also substantially increases these outcomes (by 4.0% and 3.6% has no 82 equivalent and FCS . 43 Further research is needed to investigate what is driving the food expenditure resu the inclusion of maize - tegories but not in groups involve maize - legume intercropping (Table 2.1). If households consume some or all of the legumes they produce through maize - legume interc ropping, this could considerably increase their FCS because pulses are highly weighted in the FC S . 44 Moreover, per Kim et al. (in press), m aize - legume intercropping is the ma in way in which maize - growing households in rural Tanzania produce legumes (as oppo sed to growing legumes separately from maize). Furthermore, l egume consumption among legume - producing households is two times greater than legume consumption among those who only purchase legumes (Stahley et al., 2012). For focusing on maize production through the sole use of inorganic fertilizer may not be enough to substantially raise FCS . Finally, the positive and i.e., households may not just be producing more and more diverse foods which they then consume, they may be purchasing them as well. nutrition outcomes found by Kim et al. (in press) could be linked to increases in HDDS, FCS, and f ood expenditure. 43 44 The weight of pulses is thre e which is the second highest among the nine food groups used to calculate the FCS. The food groups with the highest weight, four, are meat and fish, and milk items. 83 Tab le 2.4: ATTs of using practices in each SI category on household food access outcomes Adoption status Outcome variables SI category Adopting Nonadopting ATT (1) (2) (3) = (1) (2) Modified HDDS Intensification 8.39 (0.07) 7.88 (0.05) 0.51*** (0.07) (N=3,641) Sustainable 7.87 (0.02) 7.82 (0.02) 0.05*** (0.01) SI 8.67 (0.05) 7.89 (0.04) 0.78*** (0.04) Food expenditure (1,000 TZS) Intensification 516.00 (11.94) 503.66 (8.05) 12.34 (10.33) per adult equivalent Sustainable 486.35 (2.71) 467.65 (3.22) 18.70*** (1.65) (N=3,641) SI 526.89 (9.80) 506.23 (6.08) 20.66*** (7.07) FCS Intensification 49.62 (0.88) 49.56 (0.57) 0.06 (0.80) (N=1,622) Sustainable 52.28 (0.32) 50.47 (0.35) 1.81*** (0.28) SI 53.08 (0.99) 51.52 (0.50) 1.56** (0.81) Notes: Standard errors in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. For each outcome variable, indicates that the ATT is statistically different from at or below the 5% level, while ATT . Before concluding, it is important to note the limitations of the study. First, the data used here do not in the farming seasons between survey rounds or in years prior to the first survey. We therefore only capture the short - run effects of the various SFM practices studied here but the long - run effects could also be import ant. In addition, we - level SI category decisions using dummy variables that denoted whether a given SFM practice was applied or not without considering the intensity of application (e.g., the amount of inorganic fertilizer or organic fertilizer applied or the proportion of area covered by legume crops) . T well as their food access outcomes. Future research with richer data (if available) to address these short comings would be worthwhile. We also used observational data and have relied on econometric methods (each with their own assumptions) to try to estimate causal effects; 84 however, we may not have fully addressed selection bias issues. Exploring how to examin e similar research questions using a randomized - controlled trial would thus also be useful. 2.6 Conclusion s and policy implications Low agricultural productivity and food insecurity are major challenges in SSA , where agriculture is central to rural liveli hood s . S ustainable intensification ha s received considerable attention as a possible means to address these challenges but there have been very few studies that have attempted to evaluate the relationship between SI and practices that could contribute to SI of maize production in Tanzania on rural maize - growing ortant dimension of food security. We also estimate the effects on their net crop income and productivity the two primary pathways through which changes in To deal with potential selection bias originating from both observed and unobserved heterogeneity, we use CRE - MESR models to estimate these effects. Our findings suggest that - legume intercropping but no practices ) on maize plots all have - related outcomes and crop productivity relative to use of none of the three SFM practices considered here . effects than 85 DS, with the largest effects occurring and by a similar magnitude as oss all the outcomes considered here, use of practices in the us - that there may be major food security and nutrition benefits (not to m ention soil fertility and productivity benefits) to promoting joint use of inorganic fertilizer with complementary organic soil fertility practices. While further research is needed to determine how best to do this, our first stage regression results (Tab les 2A.18 - 2A.24) suggest that improving education, access to agricultural extension service s , and access to credit are key drivers decisions to jointly use these practices. 86 APPENDIX 87 Table 2A.1: Summary statistics by SI category Variables Variable description Mean value of each SI category Mean of all N I S SI Outcome variables Net crop income Real net crop income on maize plots (2013 = 100) 265.73 441.40 365.17 549.53 343.45 Net crop income per acre Real net crop income per acre of maize plots (2013 = 100) 87.33 116.52 108.15 114.75 100.08 Net crop income per adult equivalent Real net crop income per adult equivalent (2013 = 100) 66.64 102.96 81.77 120.62 80.12 Crop productivity Crop productivity based on output index following Liu and Myers (2009) 316.54 633.22 384.47 652.78 396.05 Food expenditure per adult equivalent Real food and beverage consumption expenditure 454.80 516.01 486.35 526.89 478.12 Modified HDDS Modified household dietary diversity score (0 - 12) 7.68 8.40 7.88 8.67 7.90 FCS Food consumption score (0 112) 49.14 49.62 52.28 53.08 50.78 Explanatory variables Male - headed HH 1 = yes if the household head is male 0.77 0.84 0.77 0.82 0.78 Age of HH head Age of the household head (years) 47.38 44.81 49.45 47.91 48.08 Education of HH head Highest grade completed by the household head (years) 4.30 6.13 4.38 5.91 4.60 Family labor Number of adults (15 - 64 years old) per acre of cultivated land 1.10 0.99 1.15 1.06 1.11 Total cultivated land Total land area cultivated (acres) 6.09 5.99 6.53 6.58 6.30 Off - farm income 1 = yes if the HH earned off - income in the past 12 months 0.59 0.69 0.62 0.62 0.61 Farm assets Real total value of farm implements and machinery (1,000 TZS) owned in the past 12 months (2013=100) 1,085.97 1,094.78 2,183.51 639.02 1,478.39 Livestock ownership 1 = yes if the HH has livestock (cattle, goats, sheep, pigs, or donkeys) 0.34 0.44 0.51 0.58 0.44 Access to credit 1 = yes if the HH borrowed cash, goods, or services in the past 12 months 0.07 0.11 0.08 0.14 0.08 Membership (SACCOS) 1 = yes if the HH has a member of SACCOS 0.03 0.08 0.04 0.08 0.04 1 = yes if the HH received agricultural advice from government/NGO in the past 12 months 0.08 0.17 0.07 0.20 0.09 88 Table 2A.1 Variables Variable description Mean value of each SI category Mean of all N I S SI Explanatory variables Distance to main road Household distance to main road (km) 24.57 17.06 22.05 15.85 22.25 Distance to town Household distance to nearest town of > 20,000 population (km) 58.99 48.81 57.69 50.27 56.98 Distance to main market Household distance to major market (km) 84.73 92.25 84.90 91.29 85.91 Cooperatives 0.40 0.58 0.37 0.47 0.41 Input supplier 1 = yes if improved maize seed supplier present within the village 0.33 0.56 0.33 0.55 0.37 Drought/Flood 1 = yes if the HH was negatively affected by drought or flood in the past two years 0.12 0.07 0.12 0.05 0.11 Crop disease/Pests 1 = yes if the HH was negatively affected by crop diseases or pests for the past two years 0.08 0.07 0.09 0.08 0.09 Rainfall 12 - month total rainfall (mm) in July - June 798.99 863.75 791.33 841.86 804.28 Soil nutrient constraint 1 = yes if soil nutrient availability constraint is moderate or (very) severe 0.61 0.70 0.61 0.73 0.63 Inorganic fertilizer price Real inorganic fertilizer price at district level (TZS/kg) (2013=100) 1,549.92 1,419.75 1,615.28 1,354.87 1,548.76 Lagged price of maize Real average price of maize from Jul. to Sep. in prior year (TZS/kg) (2013=100) 480.17 423.73 495.05 434.82 478.05 Lagged price of rice Real average price of maize from Jul. to Sep. in prior year (TZS/kg) (2013=100) 1,402.55 1,365.56 1,398.24 1,378.28 1,396.11 Bean price Real bean market price at region level (TZS/kg) (2013=100) 1,578.39 1,527.97 1,571.06 1,535.90 1,568.15 Groundnut price Real groundnut market price at region level (TZS/kg) (2013=100) 2,008.68 2,057.88 2,037.31 2,077.84 2,029.78 Year dummy (2010/11) 1 = yes if the household is in TNPS 2010/11 sample 0.34 0.31 0.29 0.38 0.32 Year dummy (2012/13) 1 = yes if the household is in TNPS 2012/13 sample 0.38 0.46 0.44 0.39 0.41 T2 dummy 1 = yes if the household is observed twice in any of the three waves 0.32 0.27 0.30 0.29 0.31 T3 dummy 1 = yes if the household is observed in all three waves 0.41 0.58 0.38 0.57 0.43 89 Table 2A.1 Variables Variable description Mean value of each SI category Mean of all N I S SI Instrumental variables Electoral threat Proportion of votes for the runner - up divided by the proportion of votes for the presidential winner 0.30 0.18 0.25 0.18 0.26 Number of subsidized fertilizer vouchers Number of inorganic fertilizer (nitrogen) vouchers distributed to region 62,622 113,500 57,338 125,118 69,791 Proportion receiving agricultural advice Proportion of other households in the ward that got advice on agricultural production 9.39 20.42 9.91 21.09 11.43 Proportion using inorganic fertilizer Proportion of other households in the ward that use inorganic fertilizer 8.76 60.13 11.79 59.51 18.14 Proportion using organic fertilizer Proportion of other households in the ward that use organic fertilizer 13.99 26.23 21.41 28.57 19.11 Proportion using maize - legume IC Proportion of other households in the ward that use maize - legume intercropping 32.97 45.63 48.77 52.98 41.94 Notes: TZS = Tanzanian Shillings. SD = standard deviation. IC = intercropping. N, I, S, and SI indicate Non - adoption, Intensification, Sustainable, and SI, respectively. 90 Table 2A.2: CRE - MNLS estimates for net crop income (1,000 TZS) Variables Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 0.134 0.227 0.020 0.097 - 0.092 0.195 Age of HH head (years) - 0.022 0.039 0.028 0.019 0.041 0.038 Education of HH head (years) 0.142*** 0.030 0.013 0.013 0.145*** 0.026 Family labor (number of adults per acre) - 0.159* 0.091 0.005 0.044 - 0.021 0.095 Total cultivated land (acres) - 0.022 0.024 - 0.001 0.011 - 0.030 0.020 Off - farm income (yes = 1) 0.268 0.279 - 0.018 0.140 - 0.008 0.250 Farm assets (1,000 TZS) 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) 0.372** 0.185 0.592*** 0.088 1.047*** 0.166 Access to credit (yes = 1) 0.573** 0.284 0.008 0.154 0.871*** 0.247 Membership (SACCOS) (yes = 1) 0.807** 0.347 0.038 0.211 0.671** 0.317 0.662*** 0.242 - 0.092 0.149 0.804*** 0.215 Distance to main road (km) - 0.009* 0.005 - 0.003 0.002 - 0.012*** 0.004 Distance to town (km) - 0.009*** 0.003 - 0.002 0.001 - 0.008*** 0.002 Distance to main market (km) 0.007*** 0.002 0.003*** 0.001 0.008*** 0.002 Cooperatives (yes = 1) 0.203 0.180 - 0.045 0.084 - 0.248 0.163 Input supplier (yes = 1) 0.077 0.179 - 0.111 0.088 0.114 0.160 Drought/Flood (yes = 1) - 0.169 0.318 - 0.043 0.120 - 0.546* 0.301 Crop disease/Pests (yes = 1) 0.020 0.318 0.035 0.139 - 0.027 0.275 Total rainfall (mm) 0.001 0.001 - 0.001 0.000 0.000 0.001 Soil nutrient constraint (yes = 1) - 0.102 0.197 0.081 0.087 0.234 0.181 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 0.000 0.000 Lagged price of maize (TZS/kg) - 0.003 0.003 0.001 0.001 0.000 0.003 Lagged price of rice (TZS/kg) - 0.001 0.001 - 0.001 0.001 - 0.001 0.001 Bean price (TZS/kg) - 0.001 0.001 0.000 0.001 0.001 0.001 Groundnut price (TZS/kg) 0.000 0.001 0.000 0.000 0.000 0.000 Year dummy (2010/11) 0.859* 0.511 - 0.101 0.204 0.511 0.441 Year dummy (2012/13) 1.434*** 0.428 0.240 0.182 0.715* 0.377 T2 dummy 0.098 0.261 - 0.236** 0.104 0.372 0.237 T3 dummy 0.044 0.253 - 0.211** 0.104 0.204 0.235 Constant - 6.620 1.799 - 2.976 0.660 - 4.613 1.508 Joint significance of excluded IVs: 276.49*** 165.54*** 359.18*** Joint significance of time - varying covariates: 8.41 12.76 10.06 Wald 1141.23*** Number of observations 3,641 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 91 Table 2A.3: CRE - MNLS estimates for net crop income (1,000 TZS) per acre Variables Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 0.135 0.227 0.007 0.096 - 0.092 0.196 Age of HH head (years) - 0.022 0.039 0.027 0.019 0.041 0.038 Education of HH head (years) 0.142*** 0.030 0.014 0.013 0.145*** 0.026 Family labor (number of adults per acre) - 0.157* 0.090 0.006 0.044 - 0.021 0.095 Total cultivated land (acres) - 0.022 0.024 - 0.002 0.011 - 0.031 0.020 Off - farm income (yes = 1) 0.269 0.279 - 0.016 0.140 - 0.011 0.250 Farm assets (1,000 TZS) 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) 0.394** 0.182 0.645*** 0.086 1.063*** 0.163 Access to credit (yes = 1) 0.574** 0.284 0.010 0.154 0.871*** 0.247 Membership (SACCOS) (yes = 1) 0.811** 0.348 0.025 0.210 0.675** 0.318 0.656*** 0.243 - 0.071 0.149 0.800*** 0.216 Distance to main road (km) - 0.009* 0.005 - 0.004* 0.002 - 0.012*** 0.004 Distance to town (km) - 0.010*** 0.003 - 0.002 0.001 - 0.008*** 0.003 Distance to main market (km) 0.007*** 0.002 0.003*** 0.001 0.008*** 0.002 Cooperatives (yes = 1) 0.199 0.180 - 0.041 0.084 - 0.245 0.163 Input supplier (yes = 1) 0.079 0.177 - 0.100 0.088 0.108 0.159 Drought/Flood (yes = 1) - 0.170 0.317 - 0.055 0.119 - 0.545* 0.301 Crop disease/Pests (yes = 1) 0.009 0.319 0.033 0.139 - 0.029 0.276 Total rainfall (mm) 0.001 0.001 0.000 0.000 0.000 0.001 Soil nutrient constraint (yes = 1) - 0.116 0.197 0.065 0.087 0.225 0.181 Inorganic fertilizer price (TZS/kg) 0.000 0.000 0.000 0.000 0.000 0.000 Lagged price of maize (TZS/kg) - 0.003 0.003 0.001 0.001 0.000 0.003 Lagged price of rice (TZS/kg) - 0.001 0.001 - 0.001 0.001 - 0.001 0.001 Bean price (TZS/kg) - 0.001 0.001 0.000 0.001 0.001 0.001 Groundnut price (TZS/kg) 0.000 0.001 0.000 0.000 0.000 0.000 Year dummy (2010/11) 0.887* 0.516 - 0.167 0.204 0.558 0.445 Year dummy (2012/13) 1.480*** 0.434 0.200 0.183 0.774** 0.382 T2 dummy 0.104 0.261 - 0.220** 0.104 0.379 0.237 T3 dummy 0.053 0.253 - 0.192* 0.103 0.214 0.235 Electoral threat - 0.685 0.501 - 0.225** 0.105 - 0.267 0.371 Proportion receiving agricultural advice 0.002 0.005 - 0.002 0.003 0.002 0.004 Proportion adopting inorganic fertilizer 0.046*** 0.003 0.006*** 0.002 0.046*** 0.003 Proportion adopting maize - legume IC 0.003 0.003 0.014*** 0.001 0.011*** 0.003 Constant - 6.654*** 1.797 - 3.053*** 0.661 - 4.569*** 1.502 Joint significance of excluded IVs: 279.42*** 156.60*** 362.36*** Joint significance of time - varying covariates: 8.64 12.57 10.18 Wald 1130.97*** Number of observations 3,641 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 92 Table 2A.4: CRE - MNLS estimates for net crop income (1,000 TZS) per adult equivalent Variables Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 0.161 0.228 0.017 0.097 - 0.044 0.197 Age of HH head (years) - 0.021 0.039 0.028 0.019 0.043 0.038 Education of HH head (years) 0.142*** 0.030 0.013 0.013 0.146*** 0.026 Family labor (number of adults per acre) - 0.153 0.093 0.004 0.044 - 0.006 0.096 Total cultivated land (acres) - 0.021 0.024 - 0.002 0.011 - 0.029 0.020 Off - farm income (yes = 1) 0.266 0.280 - 0.022 0.140 0.003 0.252 Farm assets (1,000 TZS) 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) 0.355* 0.185 0.598*** 0.088 1.034*** 0.167 Access to credit (yes = 1) 0.602** 0.284 0.005 0.154 0.903*** 0.247 Membership (SACCOS) (yes = 1) 0.782** 0.348 0.040 0.211 0.653** 0.319 0.642*** 0.242 - 0.089 0.149 0.783*** 0.216 Distance to main road (km) - 0.008 0.005 - 0.003 0.002 - 0.010** 0.005 Distance to town (km) - 0.009*** 0.003 - 0.002 0.001 - 0.008*** 0.002 Distance to main market (km) 0.007*** 0.002 0.003*** 0.001 0.008*** 0.002 Cooperatives (yes = 1) 0.256 0.180 - 0.053 0.085 - 0.198 0.164 Input supplier (yes = 1) 0.006 0.180 - 0.103 0.089 0.011 0.163 Drought/Flood (yes = 1) - 0.135 0.318 - 0.045 0.120 - 0.497 0.302 Crop disease/Pests (yes = 1) 0.028 0.318 0.033 0.139 - 0.015 0.276 Total rainfall (mm) 0.001 0.001 0.000 0.000 0.000 0.001 Soil nutrient constraint (yes = 1) 0.041 0.204 0.066 0.090 0.429** 0.188 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.001 0.003 0.001 0.001 0.003 0.003 Lagged price of rice (TZS/kg) - 0.002 0.001 - 0.001 0.001 - 0.003** 0.001 Bean price (TZS/kg) - 0.001 0.001 - 0.000 0.001 0.001 0.001 Groundnut price (TZS/kg) - 0.000 0.001 - 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 0.501 0.538 - 0.061 0.210 0.016 0.466 Year dummy (2012/13) 1.617*** 0.440 0.244 0.182 0.980** 0.389 T2 dummy 0.103 0.261 - 0.233** 0.104 0.393* 0.238 T3 dummy 0.030 0.254 - 0.202* 0.104 0.181 0.237 Electoral threat - 0.639 0.492 - 0.225 0.107 - 0.210 0.356 Number of subsidized fertilizer vouchers 0.000** 0.000 0.000 0.000 0.000*** 0.000 Proportion adopting inorganic fertilizer 0.045*** 0.003 0.004** 0.002 0.045*** 0.003 Proportion adopting organic fertilizer 0.001 0.004 0.006*** 0.002 0.000 0.003 Proportion adopting maize - legume IC 0.002 0.003 0.014*** 0.001 0.010*** 0.003 Constant - 7.930*** 1.906 - 2.820*** 0.684 - 5.880*** 1.593 Joint significance of excluded IVs: 277.64*** 165.96*** 365.46*** Joint significance of time - varying covariates: 10.61 12.27 10.89 Wald 1141.35*** Number of observations 3,641 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 93 Table 2A.5: CRE - MNLS estimates for crop productivity Variables Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 0.188 0.211 0.019 0.097 - 0.006 0.179 Age of HH head (years) - 0.014 0.035 0.029 0.019 0.041 0.034 Education of HH head (years) 0.148*** 0.028 0.013 0.013 0.151*** 0.024 Family labor (number of adults per acre) - 0.123 0.081 0.005 0.045 0.000 0.083 Total cultivated land (acres) - 0.021 0.024 - 0.001 0.010 - 0.029 0.020 Off - farm income (yes = 1) 0.232 0.257 - 0.018 0.140 - 0.077 0.225 Farm assets (1,000 TZS) 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) 0.272 0.170 0.575*** 0.087 0.991*** 0.151 Access to credit (yes = 1) 0.576** 0.261 0.016 0.154 0.847*** 0.226 Membership (SACCOS) (yes = 1) 1.040*** 0.317 0.065 0.210 0.788*** 0.292 Extension from 0.654*** 0.223 - 0.070 0.148 0.820*** 0.195 Distance to main road (km) - 0.009* 0.005 - 0.003* 0.002 - 0.016*** 0.004 Distance to town (km) - 0.014*** 0.003 - 0.002 0.001 - 0.011*** 0.002 Distance to main market (km) 0.010*** 0.002 0.003*** 0.001 0.011*** 0.002 Cooperatives (yes = 1) 0.439*** 0.164 - 0.017 0.084 - 0.004 0.146 Input supplier (yes = 1) 0.401** 0.162 - 0.095 0.087 0.469*** 0.143 Drought/Flood (yes = 1) - 0.156 0.292 - 0.057 0.119 - 0.553** 0.276 Crop disease/Pests (yes = 1) 0.015 0.297 0.035 0.139 0.002 0.253 Total rainfall (mm) 0.001 0.001 - 0.001 0.000 - 0.000 0.001 Soil nutrient constraint (yes = 1) - 0.215 0.180 0.084 0.087 0.089 0.163 Inorganic fertilizer price (TZS/kg) 0.000** 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.004 0.003 0.001 0.001 - 0.001 0.002 Lagged price of rice (TZS/kg) 0.000 0.001 - 0.001 0.001 - 0.000 0.001 Bean price (TZS/kg) - 0.001 0.001 - 0.000 0.001 0.000 0.001 Groundnut price (TZS/kg) - 0.001 0.000 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 1.331*** 0.480 - 0.103 0.205 1.029*** 0.409 Year dummy (2012/13) 1.784*** 0.394 0.237 0.184 1.112*** 0.345 T2 dummy 0.087 0.242 - 0.236** 0.104 0.247 0.214 T3 dummy 0.177 0.231 - 0.210** 0.104 0.247 0.210 Electoral threat - 0.572 0.420 - 0.231** 0.107 - 0.192 0.279 Proportion receiving agricultural advice 0.027*** 0.004 - 0.001 0.003 0.026*** 0.003 Proportion adopting organic fertilizer 0.012*** 0.003 0.007*** 0.002 0.011*** 0.003 Proportion adopting maize - legume IC 0.009*** 0.003 0.014*** 0.001 0.017*** 0.002 Constant - 5.622*** 1.523 - 2.997*** 0.668 - 4.630*** 1.285 Joint significance of excluded IVs: 92.67*** 166.16*** 156.38*** Joint significance of time - varying covariates: 18.25* 12.74 15.05 Wald 942.33*** Number of observations 3,641 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 94 Table 2A.6: CRE - MNLS estimates for modified HDDS Variables Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 0.127 0.226 0.008 0.096 - 0.100 0.195 Age of HH head (years) - 0.023 0.039 0.027 0.019 0.040 0.038 Education of HH head (years) 0.143*** 0.030 0.014 0.013 0.145*** 0.026 Family labor (number of adults per acre) - 0.160* 0.091 0.006 0.044 - 0.022 0.096 Total cultivated land (acres) - 0.022 0.024 - 0.002 0.011 - 0.030 0.020 Off - farm income (yes = 1) 0.268 0.279 - 0.015 0.140 - 0.009 0.250 Farm assets (1,000 TZS) 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) 0.389** 0.181 0.643*** 0.086 1.058*** 0.163 Access to credit (yes = 1) 0.571** 0.284 0.009 0.154 0.868*** 0.247 Membership (SACCOS) (yes = 1) 0.801** 0.347 0.028 0.210 0.665** 0.317 0.669*** 0.242 - 0.077 0.149 0.812*** 0.215 Distance to main road (km) - 0.009* 0.005 - 0.004* 0.002 - 0.012*** 0.004 Distance to town (km) - 0.009*** 0.003 - 0.002 0.001 - 0.008*** 0.002 Distance to main market (km) 0.007*** 0.002 0.003*** 0.001 0.008*** 0.002 Cooperatives (yes = 1) 0.207 0.179 - 0.046 0.084 - 0.242 0.163 Input supplier (yes = 1) 0.077 0.177 - 0.100 0.088 0.107 0.159 Drought/Flood (yes = 1) - 0.172 0.317 - 0.054 0.119 - 0.545* 0.301 Crop disease/Pests (yes = 1) 0.021 0.317 0.032 0.139 - 0.023 0.275 Total rainfall (mm) 0.001 0.001 - 0.000 0.000 - 0.000 0.001 Soil nutrient constraint (yes = 1) - 0.109 0.196 0.065 0.087 0.230 0.181 Inorganic fertilizer price (TZS/kg) 0.000 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.003 0.003 0.001 0.001 0.000 0.003 Lagged price of rice (TZS/kg) - 0.001 0.001 - 0.001 0.001 - 0.001 0.001 Bean price (TZS/kg) - 0.001 0.001 - 0.000 0.001 0.001 0.001 Groundnut price (TZS/kg) - 0.000 0.001 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 0.843* 0.509 - 0.155 0.203 0.514 0.439 Year dummy (2012/13) 1.434*** 0.425 0.215 0.182 0.732* 0.375 T2 dummy 0.104 0.261 - 0.220** 0.104 0.378 0.237 T3 dummy 0.049 0.252 - 0.192* 0.103 0.210 0.235 Electoral threat - 0.710 0.501 - 0.223 0.105 - 0.284 0.373 Proportion adopting inorganic fertilizer 0.047 0.003 0.006 0.002 0.047 0.003 Proportion adopting maize - legume IC 0.003 0.003 0.014 0.001 0.011 0.003 Constant - 6.652*** 1.795 - 3.077*** 0.660 - 4.569*** 1.501 Joint significance of excluded IVs: 279.00*** 156.42*** 362.64*** Joint significance of time - varying covariates: 8.55 12.67 10.08 Wald 1,131.36*** Number of observations 3,641 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 95 Table 2A.7: CRE - MNLS estimates for food expenditure per adult equivalent Variables Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 0.113 0.226 0.005 0.096 - 0.107 0.195 Age of HH head (years) - 0.023 0.039 0.028 0.019 0.041 0.038 Education of HH head (years) 0.142*** 0.030 0.015 0.013 0.146*** 0.026 Family labor (number of adults per acre) - 0.161* 0.091 0.004 0.044 - 0.023 0.095 Total cultivated land (acres) - 0.022 0.025 - 0.001 0.011 - 0.030 0.020 Off - farm income (yes = 1) 0.252 0.279 - 0.025 0.140 - 0.015 0.250 Farm assets (1,000 TZS) 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) 0.375** 0.185 0.597*** 0.088 1.051*** 0.166 Access to credit (yes = 1) 0.587** 0.284 0.015 0.154 0.880*** 0.247 Membership (SACCOS) (yes = 1) 0.781** 0.347 0.028 0.210 0.662** 0.317 0.677*** 0.241 - 0.086 0.149 0.809*** 0.215 Distance to main road (km) - 0.009* 0.005 - 0.004* 0.002 - 0.013*** 0.004 Distance to town (km) - 0.009*** 0.003 - 0.001 0.001 - 0.008*** 0.002 Distance to main market (km) 0.007*** 0.002 0.003*** 0.001 0.008*** 0.002 Cooperatives (yes = 1) 0.194 0.180 - 0.054 0.084 - 0.257 0.163 Input supplier (yes = 1) 0.074 0.178 - 0.127 0.088 0.107 0.160 Drought/Flood (yes = 1) - 0.159 0.318 - 0.033 0.120 - 0.537* 0.301 Crop disease/Pests (yes = 1) 0.033 0.317 0.045 0.139 - 0.023 0.275 Total rainfall (mm) 0.001 0.001 - 0.001 0.000 0.000 0.001 Soil nutrient constraint (yes = 1) - 0.080 0.196 0.095 0.087 0.243 0.180 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 0.000 0.000 Lagged price of maize (TZS/kg) - 0.004 0.003 0.001 0.001 0.000 0.003 Lagged price of rice (TZS/kg) 0.000 0.001 - 0.001 0.001 - 0.001 0.001 Bean price (TZS/kg) - 0.001 0.001 0.000 0.001 0.001 0.001 Groundnut price (TZS/kg) 0.000 0.001 0.000 0.000 0.000 0.000 Year dummy (2010/11) 0.904* 0.504 - 0.029 0.201 0.567 0.440 Year dummy (2012/13) 1.387*** 0.419 0.264 0.181 0.717* 0.375 T2 dummy 0.099 0.261 - 0.238** 0.104 0.371 0.237 T3 dummy 0.055 0.252 - 0.207** 0.104 0.209 0.235 Proportion adopting inorganic fertilizer 0.046*** 0.003 0.004** 0.002 0.046*** 0.003 Proportion adopting organic fertilizer 0.002 0.003 0.006*** 0.002 0.002 0.003 Proportion adopting maize - legume IC 0.003 0.003 0.014*** 0.001 0.011*** 0.003 Constant - 7.013*** 1.771 - 2.964*** 0.660 - 4.667*** 1.497 Joint significance of excluded IVs: 277.16*** 165.08*** 360.75*** Joint significance of time - varying covariates: 8.58 13.55 10.04 Wald 1143.15*** Number of observations 3,641 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 96 Table 2A.8: CRE - MNLS estimates for FCS Variables Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) - 0.317 0.312 - 0.009 0.156 - 0.082 0.296 Age of HH head (years) - 0.020 0.061 0.067** 0.031 0.081 0.054 Education of HH head (years) 0.113*** 0.044 0.028 0.021 0.152*** 0.039 Family labor (number of adults per acre) - 0.078 0.209 - 0.040 0.052 0.068 0.114 Total cultivated land (acres) - 0.014 0.037 0.004 0.018 - 0.008 0.025 Off - farm income (yes = 1) 0.385 0.421 0.100 0.205 0.223 0.379 Farm assets (1,000 TZS) 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) 0.446* 0.257 0.663*** 0.136 1.040*** 0.238 Access to credit (yes = 1) 0.472 0.419 - 0.115 0.254 0.933** 0.376 Membership (SACCOS) (yes = 1) 0.817* 0.476 - 0.164 0.299 0.464 0.460 0.797** 0.357 - 0.251 0.266 0.668** 0.330 Distance to main road (km) - 0.009 0.007 - 0.007** 0.003 - 0.008 0.006 Distance to town (km) - 0.007* 0.004 0.000 0.002 - 0.005 0.004 Distance to main market (km) 0.009*** 0.003 0.003** 0.002 0.008*** 0.002 Cooperatives (yes = 1) 0.218 0.261 - 0.207 0.134 - 0.336 0.243 Input supplier (yes = 1) 0.008 0.258 - 0.097 0.140 - 0.137 0.240 Drought/Flood (yes = 1) 0.393 0.414 0.139 0.191 - 0.351 0.458 Crop disease/Pests (yes = 1) - 0.527 0.491 - 0.319 0.229 - 0.477 0.423 Total rainfall (mm) 0.001 0.001 - 0.001** 0.001 - 0.002 0.001 Soil nutrient constraint (yes = 1) 0.203 0.301 0.149 0.147 0.498* 0.283 Inorganic fertilizer price (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) 0.001 0.005 0.007*** 0.003 0.012** 0.005 Lagged price of rice (TZS/kg) - 0.005** 0.002 - 0.003*** 0.001 - 0.007*** 0.002 Bean price (TZS/kg) - 0.000 0.003 0.002 0.002 0.004 0.003 Groundnut price (TZS/kg) - 0.002* 0.001 - 0.000 0.001 - 0.001 0.001 Year dummy (2012/13) 0.768 0.634 0.408 0.299 1.331** 0.587 Number of subsidized fertilizer vouchers 0.000 0.000 0.000 0.000 0.000 0.000 Proportion adopting inorganic fertilizer 0.046 0.004 0.005 0.003 0.046 0.004 Proportion adopting maize - legume IC 0.005 0.005 0.014 0.002 0.011 0.004 Constant - 11.951*** 3.950 - 6.630*** 1.409 - 11.805*** 3.682 Joint significance of excluded IVs: 145.21*** 61.36*** 181.58*** Joint significance of time - varying covariates: 13.17 19.25* 14.39 Wald 559.24*** Number of observations 1,622 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 97 Table 2A.9: CRE - MESR second stage estimation results for net crop income (1,000 TZS) Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 110.22*** 20.08 101.37 141.72 103.36*** 23.22 - 6.93 130.39 Age of HH head (years) 3.34 5.74 5.56 27.53 4.77 5.80 - 1.31 31.31 Education of HH head (years) 4.44 3.52 4.89 24.54 2.17 3.54 14.48 27.62 Family labor (number of adults per acre) 8.95 14.34 14.85 73.57 - 16.81 16.02 - 42.64 59.22 Total cultivated land (acres) 6.76 4.39 18.88 28.23 14.61* 8.77 12.94 22.36 Off - farm income (yes = 1) 36.48 39.88 130.85 159.64 46.57 37.74 - 4.97 197.86 Farm assets (1,000 TZS) 0.01 0.01 - 0.07 0.07 0.01 0.01 0.02 0.07 Livestock ownership (yes = 1) 147.51*** 30.46 95.33 166.31 149.43*** 30.11 - 96.24 121.91 Access to credit (yes = 1) - 21.99 38.27 - 166.97 208.99 - 20.16 45.22 342.82* 181.98 Membership (SACCOS) (yes = 1) 54.91 53.84 - 195.33 240.43 - 59.53 53.75 - 319.55** 158.02 - 50.75 36.47 123.99 179.33 51.54 52.97 158.09 151.60 Distance to main road (km) 0.55 0.63 0.18 4.10 - 1.81*** 0.70 - 6.84** 3.19 Distance to town (km) 1.54*** 0.35 3.23 2.36 2.23*** 0.47 7.03*** 2.15 Distance to main market (km) - 0.49* 0.28 - 0.39 1.19 0.11 0.32 - 0.47 1.12 Cooperatives (yes = 1) - 22.25 25.42 55.75 133.03 - 55.23** 25.90 - 137.55 106.54 Input supplier (yes = 1) 32.49 22.61 - 161.85 121.23 5.95 28.62 97.83 96.88 Drought/Flood (yes = 1) - 66.53** 29.87 439.58 353.31 17.25 37.37 46.49 261.62 Crop disease/Pests (yes = 1) - 40.47 31.76 - 101.01 182.28 - 71.65* 36.87 167.17 220.25 Total rainfall (mm) 0.01 0.10 0.52 0.90 0.02 0.14 - 0.07 0.59 Soil nutrient constraint (yes = 1) 7.26 19.32 250.13 159.57 - 24.65 25.52 - 23.12 109.18 Inorganic fertilizer price (TZS/kg) 0.00 0.02 0.07 0.13 0.01 0.02 - 0.07 0.10 Lagged price of maize (TZS/kg) 0.56* 0.34 1.41 3.06 - 0.14 0.42 0.81 1.93 Lagged price of rice (TZS/kg) - 0.26* 0.15 - 0.81 0.92 - 0.03 0.18 0.40 0.76 Bean price (TZS/kg) 0.32** 0.14 1.95 1.43 0.14 0.18 0.41 0.75 Groundnut price (TZS/kg) 0.03 0.07 - 0.43 0.42 0.06 0.08 0.23 0.34 98 Table 2A.9 Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Year dummy (2010/11) - 23.13 54.09 250.18 606.91 23.34 58.29 - 390.81 347.30 Year dummy (2012/13) 55.65 45.32 270.52 451.63 57.81 63.35 - 347.00 297.53 T2 dummy - 5.24 26.20 204.19 227.12 - 46.29 30.60 - 14.35 190.38 T3 dummy 51.60** 25.12 210.56 231.35 35.10 31.60 - 22.87 200.53 Constant 643.18*** 171.69 288.33 2012.07 34.78 212.59 2100.94 1347.78 Joint significance of excluded IVs F(4,1572)=0.85 F(4,195)=1.51 F(4,1395)=0.90 F(4,299)=1.16 Joint significance of time - varying covariates 22.99** 4.18 12.39 13.95 Ancillary 261,497*** 60,155 615,060 883,982 199,524*** 44,462 1,629,908 1,118,932 - 0.15 0.53 - 0.24 0.16 0.81* 0.47 - 1.10** 0.49 - 0.18 0.52 - 0.89** 0.39 0.45* 0.24 0.61 0.53 - 0.08 0.42 0.78 0.55 - 0.46 0.42 0.38 0.48 Number of observations 1,617 240 1,440 344 Notes: Non - adoption is the reference category. Standard errors were bootstrapped with 100 replications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 99 Table 2A.10: CRE - MESR second stage estimation results for net crop income (1,000 TZS) per acre Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 15.60** 6.30 5.93 35.40 10.18 6.34 - 4.44 20.51 Age of HH head (years) - 0.20 0.85 1.71 8.31 2.29 1.73 7.16 5.58 Education of HH head (years) 0.31 0.95 1.64 5.64 0.13 0.89 1.03 4.24 Family labor (number of adults per acre) 11.51*** 3.96 18.38 16.61 17.97** 7.97 - 6.01 14.46 Total cultivated land (acres) - 1.85** 0.81 - 2.64 4.72 - 1.45 1.22 - 2.41 2.11 Off - farm income (yes = 1) 4.34 8.95 45.48 40.42 - 6.47 9.61 10.02 30.19 Farm assets (1,000 TZS) 0.00 0.00 - 0.01 0.01 0.00 0.00 0.00 0.01 Livestock ownership (yes = 1) 14.97* 8.81 - 24.31 38.63 16.74** 8.04 2.74 19.71 Access to credit (yes = 1) - 8.79 10.00 - 17.59 41.95 - 14.04 10.86 108.04*** 32.31 Membership (SACCOS) (yes = 1) 6.39 15.57 - 27.71 59.95 2.52 13.66 - 63.06** 31.83 - 10.33 10.43 37.70 42.47 14.66 15.44 26.20 23.76 Distance to main road (km) 0.09 0.16 1.45 0.92 - 0.04 0.17 - 0.15 0.52 Distance to town (km) 0.05 0.11 0.36 0.44 0.31*** 0.09 0.60** 0.30 Distance to main market (km) 0.11 0.08 0.01 0.27 - 0.08 0.06 - 0.08 0.19 Cooperatives (yes = 1) 5.40 5.78 25.80 27.33 0.07 5.75 - 12.01 16.49 Input supplier (yes = 1) 3.67 6.04 - 39.49 24.95 2.84 5.94 11.94 17.00 Drought/Flood (yes = 1) - 6.38 7.27 49.02 57.77 3.63 8.14 12.07 39.55 Crop disease/Pests (yes = 1) - 11.43 8.11 - 0.60 38.90 - 2.91 8.44 27.48 38.54 Total rainfall (mm) 0.05* 0.03 0.03 0.17 0.00 0.03 0.03 0.11 Soil nutrient constraint (yes = 1) - 10.00* 5.61 0.97 31.93 - 16.42*** 6.08 - 2.51 24.25 Inorganic fertilizer price (TZS/kg) 0.01 0.01 0.04 0.03 0.01 0.01 - 0.02 0.01 Lagged price of maize (TZS/kg) 0.07 0.08 - 0.23 0.61 0.02 0.10 0.23 0.37 Lagged price of rice (TZS/kg) - 0.05 0.04 - 0.05 0.18 0.00 0.03 - 0.09 0.12 Bean price (TZS/kg) 0.06* 0.03 0.05 0.29 0.00 0.04 0.09 0.13 Groundnut price (TZS/kg) 0.04 0.02 - 0.05 0.09 0.05*** 0.02 0.04 0.06 100 Table 2A.10. Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Year dummy (2010/11) - 32.96** 16.19 78.10 103.40 - 40.46*** 14.15 - 66.85 68.48 Year dummy (2012/13) 10.44 13.27 73.26 87.44 - 16.91 13.14 - 58.64 53.46 T2 dummy 0.00 8.01 - 28.01 50.97 - 8.33 6.63 21.93 28.70 T3 dummy 4.24 6.58 - 47.41 49.28 10.51 8.23 5.73 27.71 Constant 22.56 46.19 115.58 393.51 12.21 60.12 404.05 218.07 Joint significance of excluded IVs F(4,1572)=0.51 F(4,195)=1.03 F(4,1395)=0.76 F(4,299)=0.82 Joint significance of time - varying covariates 7.01 4.52 24.00** 17.71* Ancillary 21,111*** 6,546 35,214 54,002 10,728*** 1,839 46,240* 25,026 0.08 0.52 - 0.27 0.19 0.11 0.50 0.65 0.42 0.12 0.44 - 0.96*** 0.34 0.69*** 0.16 0.53 0.58 0.72** 0.36 - 1.05*** 0.39 - 0.79* 0.43 0.11 0.39 Number of observations 1,617 240 1,440 344 Notes: Non - adoption is the reference category. Standard errors were bootstrapped with 100 replications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 101 Table 2A.11: CRE - MESR second stage estimation results for net crop income (1,000 TZS) per adult equivalent Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 17.15*** 4.90 26.47 30.38 14.76** 6.53 - 15.28 29.95 Age of HH head (years) 0.37 1.32 1.55 5.81 2.49 1.90 4.27 7.38 Education of HH head (years) 1.31 0.84 - 3.27 4.76 - 0.28 0.86 5.22 5.47 Family labor (number of adults per acre) 2.95 4.35 0.89 14.89 - 8.03* 4.85 - 6.62 14.28 Total cultivated land (acres) 0.84 0.68 8.72 6.19 3.63*** 1.01 3.12 4.83 Off - farm income (yes = 1) 8.06 9.46 22.93 35.29 - 1.48 8.96 - 19.47 42.02 Farm assets (1,000 TZS) 0.00 0.00 - 0.01 0.01 0.00 0.00 0.01 0.01 Livestock ownership (yes = 1) 13.54** 6.82 - 3.35 36.50 9.85 6.50 - 23.95 26.00 Access to credit (yes = 1) 3.79 12.15 - 26.46 43.07 - 10.90 8.08 87.99 36.39 Membership (SACCOS) (yes = 1) - 6.08 9.82 - 54.04 50.62 - 12.81 12.84 - 67.43* 36.58 - 18.12*** 6.91 18.26 39.94 15.30 14.00 24.93 31.76 Distance to main road (km) 0.13 0.16 1.13 1.00 - 0.38** 0.17 - 1.34** 0.67 Distance to town (km) 0.33*** 0.08 0.52 0.45 0.39*** 0.11 1.61*** 0.44 Distance to main market (km) - 0.13* 0.07 - 0.26 0.29 - 0.03 0.07 - 0.32 0.23 Cooperatives (yes = 1) 1.85 5.65 29.85 25.54 - 7.68 5.90 - 35.40 23.36 Input supplier (yes = 1) 6.21 4.73 - 38.32 26.35 3.68 7.55 9.61 19.55 Drought/Flood (yes = 1) - 10.05 7.05 34.71 64.63 9.24 10.61 - 27.96 48.03 Crop disease/Pests (yes = 1) - 10.44 7.46 - 1.26 42.87 - 14.38 9.40 61.96 47.38 Total rainfall (mm) 0.01 0.03 0.06 0.19 0.02 0.04 - 0.06 0.14 Soil nutrient constraint (yes = 1) 1.43 4.68 78.15** 35.04 - 8.88 6.14 6.19 24.52 Inorganic fertilizer price (TZS/kg) 0.00 0.00 0.02 0.03 0.00 0.00 - 0.01 0.02 Lagged price of maize (TZS/kg) 0.13 0.08 0.43 0.61 - 0.02 0.10 0.41 0.39 Lagged price of rice (TZS/kg) - 0.06* 0.03 - 0.17 0.20 - 0.03 0.04 0.09 0.16 Bean price (TZS/kg) 0.07** 0.03 0.35 0.31 0.02 0.05 0.27* 0.16 Groundnut price (TZS/kg) 0.01 0.01 - 0.06 0.09 0.01 0.02 0.02 0.07 102 Table 2A.11. Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Year dummy (2010/11) - 11.83 12.22 5.73 111.75 4.24 15.43 - 118.69* 71.57 Year dummy (2012/13) 8.53 11.69 10.23 84.11 5.12 13.83 - 125.79* 66.28 T2 dummy - 7.02 6.12 16.37 47.28 - 10.74 7.35 - 25.49 45.29 T3 dummy 6.52 5.92 14.82 46.06 6.76 7.12 - 42.56 46.54 Constant 145.24*** 39.99 100.07 457.97 16.45 56.53 647.60** 281.28 Joint significance of excluded IVs F(5,1571)=0.43 F(5,194)=1.72 F(5,1394)=1.35 F(5,298)=0.82 Joint significance of time - varying covariates 26.61*** 4.75 10.80 18.49* Ancillary 10,845*** 2,503 34,651 45,517 12,219*** 3,468 84,063 55,547 - 0.11 0.54 - 0.31** 0.15 0.89** 0.42 - 0.70 0.43 0.55 0.44 - 0.87*** 0.31 0.57*** 0.18 0.71 0.55 - 0.19 0.43 0.28 0.45 - 0.59 0.38 - 0.30 0.42 Number of observations 1,617 240 1,440 344 Notes: Non - adoption is the reference category. Standard errors were bootstrapped with 100 replications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 103 Table 2A.12: CRE - MESR second stage estimation results for crop p roductivity Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 43.83* 22.48 34.81 116.59 30.66 21.75 4.79 94.31 Age of HH head (years) 0.60 4.27 1.33 30.62 11.63* 6.32 33.64 27.77 Education of HH head (years) 0.63 3.30 14.22 21.87 - 1.71 3.45 - 3.15 19.67 Family labor (number of adults per acre) 42.66*** 14.66 81.40 71.60 63.30** 26.20 96.09* 56.85 Total cultivated land (acres) - 6.11** 3.02 - 12.47 16.06 - 5.26 4.12 - 2.47 10.60 Off - farm income (yes = 1) 16.95 35.55 264.71* 141.13 - 21.40 34.08 19.92 138.00 Farm assets (1,000 TZS) 0.00 0.01 - 0.01 0.03 0.00 0.00 0.00 0.05 Livestock ownership (yes = 1) 65.58** 31.06 - 160.92 166.27 76.20*** 26.36 - 58.53 84.11 Access to credit (yes = 1) - 19.10 36.61 - 4.96 201.24 - 20.91 41.27 305.88** 139.49 Membership (SACCOS) (yes = 1) 8.34 51.54 - 127.43 215.71 23.74 43.27 - 156.62 118.28 - 50.04 34.87 217.68 178.91 54.13 47.44 105.82 104.39 Distance to main road (km) 0.26 0.60 3.52 3.57 - 0.12 0.53 - 1.84 2.25 Distance to town (km) 0.44 0.37 1.83 1.86 1.09*** 0.31 3.66*** 1.41 Distance to main market (km) 0.02 0.31 - 0.50 1.06 - 0.41* 0.23 - 1.02 0.90 Cooperatives (yes = 1) 8.01 24.15 56.60 110.36 - 15.05 18.45 - 4.44 72.08 Input supplier (yes = 1) - 3.54 20.73 - 209.28** 106.73 - 2.70 21.49 - 36.94 79.85 Drought/Flood (yes = 1) - 10.47 26.24 162.40 237.49 24.77 33.35 18.89 154.19 Crop disease/Pests (yes = 1) - 35.24 33.20 - 106.79 153.11 - 7.47 32.97 11.44 156.02 Total rainfall (mm) 0.17 0.12 0.07 0.64 - 0.05 0.11 - 0.49 0.48 Soil nutrient constraint (yes = 1) - 44.70* 23.47 - 119.07 122.36 - 56.22*** 20.04 - 50.12 96.96 Inorganic fertilizer price (TZS/kg) 0.02 0.02 0.10 0.12 0.02 0.02 0.00 0.10 Lagged price of maize (TZS/kg) 0.21 0.28 - 0.39 2.16 0.00 0.29 1.92 1.35 Lagged price of rice (TZS/kg) - 0.26** 0.12 - 0.21 0.62 - 0.10 0.12 - 0.70 0.47 Bean price (TZS/kg) 0.09 0.14 0.98 0.88 - 0.13 0.14 0.08 0.54 Groundnut price (TZS/kg) 0.16* 0.09 - 0.35 0.32 0.13** 0.06 0.19 0.20 104 Table 2A.12. Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Year dummy (2010/11) - 91.08 59.19 264.95 419.85 - 60.14 56.17 - 310.31 248.23 Year dummy (2012/13) 20.30 47.86 281.07 340.13 - 70.98 50.45 - 226.17 204.10 T2 dummy 1.39 29.46 - 80.32 191.43 - 26.71 22.56 - 38.48 134.78 T3 dummy 15.48 22.85 - 135.67 187.95 39.30* 23.37 - 56.23 135.99 Constant 557.16*** 167.20 1058.47 1,698.19 587.53*** 163.27 1,741.38** 886.95 Joint significance of excluded IVs F(4,1572)=0.36 F(4,195)=1.18 F(4,1395)=0.46 F(4,299)=1.30 Joint significance of time - varying covariates 5.87 2.86 31.08*** 17.85* Ancillary 309,416** 124,066 578,334 901,736 141,052*** 44,279 515,045 508,778 0.29 0.44 0.01 0.21 - 0.64 0.54 0.23 0.47 - 0.66 0.49 0.23 0.57 0.90*** 0.17 0.48 0.63 0.73* 0.39 - 0.90** 0.39 - 0.90 0.55 0.47 0.44 Number of observations 1,617 240 1,440 344 Notes: Non - adoption is the reference category. Standard errors were bootstrapped with 100 replications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 105 Table 2A.13: CRE - MESR second stage estimation results for modified HDDS Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) - 0.01 0.14 0.55 0.42 - 0.08 0.11 - 0.23 0.32 Age of HH head (years) - 0.02 0.02 - 0.10 0.11 0.03 0.03 - 0.08 0.08 Education of HH head (years) 0.08*** 0.02 0.12** 0.05 0.12*** 0.02 0.08* 0.05 Family labor (number of adults per acre) - 0.07 0.06 0.11 0.25 0.02 0.06 - 0.21 0.19 Total cultivated land (acres) 0.01 0.01 0.07 0.05 0.00 0.01 0.00 0.04 Off - farm income (yes = 1) 0.10 0.19 0.23 0.42 0.25 0.21 0.48 0.43 Farm assets (1,000 TZS) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Livestock ownership (yes = 1) 0.30** 0.13 0.76 0.46 0.47*** 0.16 0.88*** 0.29 Access to credit (yes = 1) 0.44** 0.21 0.10 0.51 0.22 0.18 - 0.16 0.34 Membership (SACCOS) (yes = 1) 0.47** 0.24 0.46 0.57 0.99*** 0.21 - 0.01 0.35 0.54*** 0.18 0.03 0.34 0.05 0.17 0.31 0.30 Distance to main road (km) 0.00 0.00 - 0.02* 0.01 0.00 0.00 - 0.01 0.01 Distance to town (km) 0.00* 0.00 0.00 0.00 - 0.01 0.00*** 0.00 0.00 Distance to main market (km) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Cooperatives (yes = 1) - 0.23** 0.11 0.20 0.30 0.05 0.12 - 0.21 0.28 Input supplier (yes = 1) - 0.04 0.12 - 0.50* 0.27 0.02 0.12 - 0.13 0.26 Drought/Flood (yes = 1) 0.32* 0.17 - 1.52** 0.63 0.12 0.15 - 0.21 0.53 Crop disease/Pests (yes = 1) 0.52*** 0.18 - 0.62 0.49 - 0.08 0.16 - 0.45 0.33 Total rainfall (mm) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Soil nutrient constraint (yes = 1) - 0.03 0.09 0.09 0.36 0.12 0.12 0.16 0.33 Inorganic fertilizer price (TZS/kg) 0.00* 0.00 0.00* 0.00 0.00 0.00 0.00 0.00 Lagged price of maize (TZS/kg) 0.00 0.00 0.00 0.01 0.00* 0.00 0.00 0.00 Lagged price of rice (TZS/kg) 0.00 0.00 0.00 0.00 0.00** 0.00 0.00 0.00 Bean price (TZS/kg) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Groundnut price (TZS/kg) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 106 Table 2A.13 Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Year dummy (2010/11) - 0.23 0.25 0.70 1.10 0.01 0.26 0.52 0.70 Year dummy (2012/13) - 0.31 0.26 0.76 0.80 - 0.50** 0.25 0.62 0.58 T2 dummy 0.03 0.14 - 0.38 0.55 0.01 0.13 0.53 0.49 T3 dummy 0.02 0.13 0.00 0.55 0.00 0.14 0.75* 0.45 Constant 8.06*** 0.77 2.45 3.84 7.54*** 0.93 8.31*** 2.92 Joint significance of excluded IVs F(3,1573)=1.45 F(3,196)=0.73 F(3,1396)=0.92 F(3,300)=1.47 Joint significance of time - varying covariates 22.08** 10.81 12.55 12.52 Ancillary 3.93*** 1.34 3.52 5.43 3.58*** 0.89 3.89 4.45 0.45 0.55 - 0.09 0.18 - 0.46 0.56 - 0.41 0.59 0.45 0.54 - 0.28 0.49 - 0.25 0.24 - 0.72 0.70 0.66* 0.39 0.58 0.55 0.22 0.60 - 0.45 0.50 Number of observations 1,617 240 1,440 344 Notes: Non - adoption is the reference category. Standard errors were bootstrapped with 100 replications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 107 Table 2A.14: CRE - MESR second stage estimation results for food expenditure (1,000 TZS) per adult equivalent Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) - 25.53 16.00 8.22 65.86 - 29.55 19.68 - 103.62* 57.72 Age of HH head (years) - 0.38 2.94 - 31.68 33.20 4.31 3.50 5.79 9.93 Education of HH head (years) 6.13** 2.95 11.77 7.56 8.52*** 2.38 11.46 9.26 Family labor (number of adults per acre) - 3.02 10.89 - 56.75 37.62 - 1.64 11.57 - 38.88 29.68 Total cultivated land (acres) 1.64 1.79 - 0.49 7.17 5.08 3.50 - 0.32 4.18 Off - farm income (yes = 1) 42.72** 21.16 - 39.94 65.14 29.68 29.48 - 10.70 52.49 Farm assets (1,000 TZS) 0.00 0.00 - 0.01 0.02 0.00 0.00 0.00 0.01 Livestock ownership (yes = 1) 30.88 19.43 75.44 63.55 - 32.81* 18.84 57.65 40.04 Access to credit (yes = 1) 66.99* 35.03 - 29.25 80.92 4.98 28.90 69.10 60.14 Membership (SACCOS) (yes = 1) 41.17 45.28 - 70.33 75.89 94.81** 38.55 - 60.22 63.68 15.76 26.00 120.81** 55.77 51.61 32.44 74.75* 43.85 Distance to main road (km) - 0.08 0.37 0.93 1.66 0.12 0.33 - 0.79 1.02 Distance to town (km) - 0.30 0.24 - 0.32 0.77 - 0.57** 0.25 - 0.44 0.60 Distance to main market (km) 0.07 0.20 - 0.54 0.45 0.18 0.18 0.00 0.46 Cooperatives (yes = 1) - 9.43 17.71 - 5.09 52.35 5.53 16.09 - 23.20 39.34 Input supplier (yes = 1) - 27.98* 15.50 - 51.37 39.33 0.85 16.17 22.36 35.11 Drought/Flood (yes = 1) 23.06 22.99 - 174.12* 90.17 10.46 27.18 - 56.34 64.82 Crop disease/Pests (yes = 1) 19.09 26.64 17.24 71.12 10.64 28.63 - 75.35 58.84 Total rainfall (mm) 0.00 0.06 - 0.04 0.27 - 0.02 0.08 0.27 0.19 Soil nutrient constraint (yes = 1) - 27.21* 14.71 81.32 51.16 - 39.90** 18.95 35.12 44.92 Inorganic fertilizer price (TZS/kg) - 0.02 0.02 - 0.12** 0.05 0.00 0.01 0.00 0.04 Lagged price of maize (TZS/kg) - 0.01 0.23 1.24 0.90 0.05 0.21 - 0.63 0.62 Lagged price of rice (TZS/kg) - 0.04 0.09 - 0.10 0.30 - 0.06 0.10 - 0.16 0.25 Bean price (TZS/kg) - 0.12 0.08 0.50 0.52 0.02 0.10 - 0.04 0.26 Groundnut price (TZS/kg) - 0.03 0.04 - 0.18 0.19 0.02 0.04 0.15 0.11 108 Table 2A.14 Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Year dummy (2010/11) 20.22 33.98 118.26 179.40 18.11 32.73 70.66 111.03 Year dummy (2012/13) 175.18*** 29.04 215.43 161.03 153.35*** 35.02 261.88*** 100.10 T2 dummy - 51.14** 22.31 - 20.20 94.65 - 32.07 24.03 - 41.99 87.60 T3 dummy - 72.15*** 17.63 - 139.47 90.01 - 30.64 20.54 - 61.44 90.68 Constant 158.10 120.98 246.39 775.50 186.55 153.97 118.60 491.70 Joint significance of excluded IVs F(3,1573)=1.89 F(3,196)=0.04 F(3,1396)=0.49 F(3,300)=1.12 Joint significance of time - varying covariates 31.34*** 12.17 13.02 29.48*** Ancillary 116,596** 51,518 59,725 114,432 109,321*** 26,249 93,603 77,571 - 0.20 0.58 - 0.18 0.18 0.25 0.64 0.46 0.52 0.86* 0.52 - 0.81 0.50 0.55*** 0.19 0.02 0.74 0.23 0.49 - 0.94** 0.46 0.30 0.57 - 0.76* 0.43 Number of observations 1,617 240 1,440 344 Notes: Non - adoption is the reference category. Standard errors were bootstrapped with 100 replications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 109 Table 2A.15: CRE - MESR second stage estimation results for FCS Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Male - headed HH (yes = 1) 3.17** 1.51 1.43 6.84 - 1.21 1.71 - 3.53 4.86 Age of HH head (years) - 0.18 0.28 0.12 1.80 - 0.16 0.35 0.35 3.73 Education of HH head (years) 0.20 0.22 1.25 0.94 0.95*** 0.25 1.37* 0.73 Family labor (number of adults per acre) - 1.06 0.74 0.04 6.47 - 0.29 0.52 - 3.13 2.22 Total cultivated land (acres) - 0.05 0.16 - 0.58 1.00 - 0.10 0.22 - 0.36 0.59 Off - farm income (yes = 1) 2.72 2.09 2.83 8.21 0.05 2.21 - 1.61 4.88 Farm assets (1,000 TZS) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 Livestock ownership (yes = 1) 4.80*** 1.83 10.51 7.61 3.75* 2.05 6.80* 3.88 Access to credit (yes = 1) 3.58 3.17 2.72 10.13 0.32 3.41 1.46 4.01 Membership (SACCOS) (yes = 1) 1.19 3.19 - 6.50 8.40 3.77 3.08 2.76 6.91 3.57 2.95 - 0.07 6.32 2.34 3.42 5.29 3.71 Distance to main road (km) - 0.02 0.03 - 0.24 0.17 0.04 0.03 - 0.15 0.09 Distance to town (km) - 0.04 0.03 0.00 0.08 - 0.03 0.03 - 0.07 0.06 Distance to main market (km) 0.03* 0.02 0.02 0.05 0.00 0.02 0.02 0.04 Cooperatives (yes = 1) - 2.71* 1.51 - 0.98 5.34 3.32** 1.52 5.07 4.02 Input supplier (yes = 1) 0.67 1.47 - 6.00 6.19 - 1.37 1.73 - 8.79*** 2.91 Drought/Flood (yes = 1) 1.16 2.16 - 11.07 10.16 0.94 2.35 3.69 6.76 Crop disease/Pests (yes = 1) - 0.60 1.99 - 12.72 11.50 - 3.97* 2.41 - 0.44 6.79 Total rainfall (mm) 0.00 0.01 0.06 0.05 0.01 0.01 - 0.01 0.02 Soil nutrient constraint (yes = 1) - 3.38** 1.39 - 2.41 6.37 - 4.61** 1.82 3.35 3.99 Inorganic fertilizer price (TZS/kg) 0.00 0.00 - 0.01 0.01 0.00 0.00 0.00 0.00 Lagged price of maize (TZS/kg) - 0.04 0.03 - 0.12 0.17 - 0.01 0.03 - 0.05 0.08 Lagged price of rice (TZS/kg) 0.02 0.01 0.00 0.03 0.01 0.01 - 0.02 0.02 Bean price (TZS/kg) - 0.02 0.02 - 0.04 0.08 0.01 0.02 - 0.05 0.04 Groundnut price (TZS/kg) 0.00 0.01 0.01 0.03 0.00 0.01 0.01 0.02 110 Table 2A.15 Adoption choice ( j ) Variables Non - adoption Intensification Sustainable SI Coef. SE Coef. SE Coef. SE Coef. SE Year dummy (2012/13) - 1.52 3.36 3.54 12.88 - 2.15 3.07 0.19 12.87 Constant 17.36 12.37 - 47.81 181.98 65.90*** 18.22 - 38.45 77.58 Joint significance of excluded IVs F(3,667)=1.03 F(3,97)=1.10 F(3,550)=1.84 F(3,144)=0.38 Joint significance of time - varying covariates 9.1 5.79 20.53** 12.70 Ancillary 309.01* 163.71 545.01 1440.91 340.46*** 92.94 192.24 469.31 - 0.82 0.63 0.01 0.28 - 0.39 0.75 0.38 0.58 0.89 0.59 0 .21 0.64 0.49 0.33 0.16 0.69 0.14 0.66 - 0.75 0.53 0.85 0.77 - 0.70 0.50 Number of observations 708 138 591 185 Notes: Non - adoption is the reference category. Standard errors were bootstrapped with 100 replications. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 111 Table 2A.16: Average treatment effects on the unt reated (ATU) of using practices in each SI category on household net crop income, productivity and food access outcomes Adoption status Average treatment effects Outcome variables SI category Adopting Nonadopting (1) (2) (3) = (1) (2) Net crop income (1,000 TZS) Intensification 452.74 (22.47) 265.73 (4.60) 187.01*** (19.96) (N=3,641) Sustainable 332.51 (7.47) 265.73 (4.60) 66.78*** (5.00) SI 492.46 (22.18) 265.73 (4.60) 226.73*** (19.71) Net crop income (1,000 TZS) Intensification 159.81 (2.94) 87.33 (0.87) 72.48*** (2.78) per acre Sustainable 107.16 (1.00) 87.33 (0.87) 19.83*** (0.70) (N=3,641) SI 115.85 (2.20) 87.33 (0.87) 28.52*** (2.15) Net crop income (1,000 TZS) Intensification 113.41 (4.29) 66.64 (0.79) 46.77*** (4.08) per adult equivalent Sustainable 85.19 (1.10) 66.64 (0.79) 18.55*** (0.90) (N=3,641) SI 104.12 (3.72) 66.64 (0.79) 37.48*** (3.47) Crop productivity Intensification 716.82 (10.12) 316.54 (3.12) 400.28*** (9.57) (N=3,641) Sustainable 390.70 (3.76) 316.54 (3.12) 74.16*** (2.53) SI 503.06 (8.81) 316.54 (3.12) 186.52*** (7.83) Modified HDDS Intensification 8.36 (0.03) 7.68 (0.02) 0.68*** (0.03) (N=3,641) Sustainable 7.79 (0.02) 7.68 (0.02) 0.11*** (0.01) SI 8.79 (0.03) 7.68 (0.02) 1.11*** (0.03) Food expenditure (1,000 TZS) Intensification 544.01 (6.11) 454.80 (2.92) 89.22*** (5.35) per adult equivalent Sustainable 481.36 (2.44) 454.80 (2.92) 26.56*** (1.57) (N=3,641) SI 628.48 (5.10) 454.80 (2.92) 173.68*** (4.36) FCS Intensification 49.72 (0.72) 49.14 (0.24) 0.58 (0.64) (N=1,622) Sustainable 50.95 (0.27) 49.14 (0.24) 1.81*** (0.21) SI 56.86 (0.59) 49.14 (0.24) 7.72*** (0.51) Notes: Standard errors in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 112 Table 2A.17: Average treatment effects of using practices in each SI category on household income, productivity, and food access outcomes (Unconditional average effects) Adoption status Average treatment effects Outcome variables SI category Adopting Nonadopting (1) (2) (3) = (1) (2) Net crop income (1,000 TZS) Intensification 426.88 (13.09) 265.94 (3.28) 160.94*** (12.42) (N=3,641) Sustainable 342.54 (5.09) 265.94 (3.28) 76.60*** (3.16) SI 542.15 (15.28) 265.94 (3.28) 276.21*** (13.29) Net crop income (1,000 TZS) Intensification 142.77 (1.85) 83.47 (0.56) 59.30*** (1.75) per acre Sustainable 106.63 (0.63) 83.47 (0.56) 23.16*** (0.48) (N=3,641) SI 114.11 (1.58) 83.47 (0.56) 30.64*** (1.56) Net crop income (1,000 TZS) Intensification 103.81 (2.64) 63.93 (0.53) 39.88*** (2.49) per adult equivalent Sustainable 83.11 (0.71) 63.93 (0.53) 19.18*** (0.59) (N=3,641) SI 108.38 (2.39) 63.93 (0.53) 44.45*** (2.20) Crop productivity Intensification 674.46 (6.44) 309.36 (2.04) 365.10*** (6.12) (N=3,641) Sustainable 396.17 (2.40) 309.36 (2.04) 86.81*** (1.74) SI 532.60 (5.91) 309.36 (2.04) 223.24*** (5.25) Modified HDDS Intensification 8.46 (0.02) 7.77 (0.01) 0.69*** (0.02) (N=3,641) Sustainable 7.90 (0.01) 7.77 (0.01) 0.13*** (0.01) SI 8.73 (0.02) 7.77 (0.01) 0.96*** (0.02) Food expenditure (1,000 TZS) Intensification 535.36 (3.97) 467.96 (2.00) 67.40*** (3.54) per adult equivalent Sustainable 486.41 (1.66) 467.96 (2.00) 18.45*** (1.06) (N=3,641) SI 603.78 (3.52) 467.96 (2.00) 135.82*** (3.10) FCS Intensification 49.58 (0.49) 49.93 (0.18) - 0.35 (0.43) (N=1,622) Sustainable 51.19 (0.18) 49.93 (0.18) 1.26*** (0.15) SI 57.23 (0.41) 49.93 (0.18) 7.30*** (0.34) Notes: Standard errors in parentheses. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 113 Table 2A.18: Marginal effect s of use of practices in each SI categor y on net crop income (1,000 TZS) Variables Intensification Sustainable SI dy/dx SE dy/dx SE dy/dx SE Male - headed HH (yes = 1) 0.008 0.011 0.004 0.019 - 0.009 0.011 Age of HH head (years) - 0.002 0.002 0.005 0.004 0.002 0.002 Education of HH head (years) 0.004*** 0.001 - 0.003 0.003 0.006*** 0.002 Family labor (number of adults per acre) - 0.008* 0.004 0.004 0.009 0.001 0.006 Total cultivated land (acres) - 0.001 0.001 0.001 0.002 - 0.001 0.001 Off - farm income (yes = 1) 0.014 0.013 - 0.008 0.028 - 0.005 0.014 Farm assets (1,000 TZS) - 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) - 0.010 0.008 0.091*** 0.017 0.045*** 0.009 Access to credit (yes = 1) 0.012 0.012 - 0.029 0.030 0.044*** 0.014 Membership (SACCOS) (yes = 1) 0.027* 0.015 - 0.021 0.040 0.027 0.017 0.020* 0.010 - 0.049* 0.029 0.041*** 0.012 Distance to main road (km) - 0.000 0.000 - 0.000 0.000 - 0.001** 0.000 Distance to town (km) - 0.000** 0.000 - 0.000 0.000 - 0.000** 0.000 Distance to main market (km) 0.000** 0.000 0.000 0.000 0.000*** 0.000 Cooperatives (yes = 1) 0.015* 0.008 - 0.006 0.016 - 0.018** 0.009 Input supplier (yes = 1) 0.003 0.008 - 0.027 0.017 0.009 0.009 Drought/Flood (yes = 1) 0.002 0.015 0.007 0.024 - 0.030* 0.018 Crop disease/Pests (yes = 1) 0.001 0.015 0.008 0.027 - 0.003 0.016 Total rainfall (mm) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Soil nutrient constraint (yes = 1) - 0.011 0.009 0.013 0.017 0.015 0.011 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.000 0.000 0.000 0.000 0.000 0.000 Lagged price of rice (TZS/kg) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Bean price (TZS/kg) - 0.000 0.000 - 0.000 0.000 0.000 0.000 Groundnut price (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 0.035 0.024 - 0.047 0.041 0.019 0.027 Year dummy (2012/13) 0.054*** 0.020 0.010 0.036 0.013 0.023 T2 dummy 0.002 0.012 - 0.060*** 0.021 0.028* 0.014 T3 dummy 0.002 0.012 - 0.049** 0.021 0.017 0.014 Electoral threat - 0.026 0.024 - 0.028 0.024 0.001 0.023 Proportion adopting inorganic fertilizer 0.001 *** 0.000 - 0.001 *** 0.000 0.002 *** 0.000 Proportion adopting organic fertilizer - 0.000 0.000 0.001 *** 0.000 - 0.000 0.000 Proportion adopting maize - legume IC - 0.000 ** 0.000 0.002 *** 0.000 0.000 ** 0.000 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 114 Table 2A.19: Marginal effects of use of practices in each SI category on net crop income (1,000 TZS) per acre Variables Intensification Sustainable SI dy/dx SE dy/dx SE dy/dx SE Male - headed HH (yes = 1) 0.008 0.011 0.002 0.019 - 0.008 0.012 Age of HH head (years) - 0.002 0.002 0.005 0.004 0.002 0.002 Education of HH head (years) 0.004*** 0.001 - 0.003 0.003 0.006*** 0.002 Family labor (number of adults per acre) - 0.008* 0.004 0.004 0.009 0.001 0.006 Total cultivated land (acres) - 0.001 0.001 0.001 0.002 - 0.001 0.001 Off - farm income (yes = 1) 0.014 0.013 - 0.007 0.028 - 0.005 0.014 Farm assets (1,000 TZS) - 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) - 0.010 0.008 0.102*** 0.016 0.044*** 0.009 Access to credit (yes = 1) 0.012 0.012 - 0.028 0.030 0.044*** 0.014 Membership (SACCOS) (yes = 1) 0.027* 0.015 - 0.024 0.040 0.027 0.017 0.019* 0.011 - 0.045 0.029 0.040*** 0.012 Distance to main road (km) - 0.000 0.000 - 0.000 0.000 - 0.001** 0.000 Distance to town (km) - 0.000** 0.000 - 0.000 0.000 - 0.000** 0.000 Distance to main market (km) 0.000** 0.000 0.000 0.000 0.000*** 0.000 Cooperatives (yes = 1) 0.015* 0.008 - 0.006 0.017 - 0.018* 0.009 Input supplier (yes = 1) 0.004 0.008 - 0.025 0.017 0.008 0.009 Drought/Flood (yes = 1) 0.002 0.015 0.005 0.024 - 0.030* 0.018 Crop disease/Pests (yes = 1) 0.000 0.015 0.008 0.027 - 0.003 0.016 Total rainfall (mm) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Soil nutrient constraint (yes = 1) - 0.011 0.009 0.010 0.017 0.015 0.011 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.000 0.000 0.000 0.000 0.000 0.000 Lagged price of rice (TZS/kg) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Bean price (TZS/kg) - 0.000 0.000 - 0.000 0.000 0.000 0.000 Groundnut price (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 0.036 0.025 - 0.063 0.041 0.023 0.027 Year dummy (2012/13) 0.056*** 0.021 - 0.001 0.036 0.017 0.023 T2 dummy 0.002 0.012 - 0.057*** 0.021 0.028* 0.014 T3 dummy 0.002 0.012 - 0.046** 0.021 0.017 0.014 Electoral threat - 0.026 0.024 - 0.029 0.024 0.001 0.023 Proportion receiving agricultural advice 0.000 0.000 - 0.000 0.001 0.000 0.000 Proportion adopting inorganic fertilizer 0.001*** 0.000 - 0.001* 0.000 0.002*** 0.000 Proportion adopting maize - legume IC 0.000** 0.000 0.003*** 0.000 0.000** 0.000 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 115 Table 2A.20: Marginal effects of use of practices in each SI category on net crop income (1,000 TZS ) per adult equivalent Variables Intensification Sustainable SI dy/dx SE dy/dx SE dy/dx SE Male - headed HH (yes = 1) 0.009 0.011 0.002 0.019 - 0.006 0.011 Age of HH head (years) - 0.002 0.002 0.005 0.004 0.002 0.002 Education of HH head (years) 0.004*** 0.001 - 0.003 0.003 0.006*** 0.002 Family labor (number of adults per acre) - 0.008* 0.004 0.003 0.009 0.002 0.006 Total cultivated land (acres) - 0.000 0.001 0.001 0.002 - 0.001 0.001 Off - farm income (yes = 1) 0.013 0.013 - 0.009 0.028 - 0.004 0.014 Farm assets (1,000 TZS) - 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) - 0.011 0.008 0.093*** 0.017 0.044*** 0.009 Access to credit (yes = 1) 0.013 0.013 - 0.030 0.030 0.046*** 0.013 Membership (SACCOS) (yes = 1) 0.026* 0.015 - 0.020 0.040 0.026 0.017 0.019* 0.010 - 0.047* 0.029 0.039*** 0.012 Distance to main road (km) - 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Distance to town (km) - 0.000** 0.000 - 0.000 0.000 - 0.000* 0.000 Distance to main market (km) 0.000** 0.000 0.000 0.000 0.000*** 0.000 Cooperatives (yes = 1) 0.017** 0.008 - 0.010 0.017 - 0.016* 0.009 Input supplier (yes = 1) 0.002 0.008 - 0.022 0.017 0.003 0.009 Drought/Flood (yes = 1) 0.003 0.015 0.005 0.024 - 0.028 0.018 Crop disease/Pests (yes = 1) 0.001 0.015 0.007 0.027 - 0.002 0.016 Total rainfall (mm) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Soil nutrient constraint (yes = 1) - 0.007 0.009 0.003 0.018 0.025** 0.011 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.000 0.000 0.000 0.000 0.000 0.000 Lagged price of rice (TZS/kg) - 0.000 0.000 - 0.000 0.000 - 0.000* 0.000 Bean price (TZS/kg) - 0.000 0.000 - 0.000 0.000 0.000 0.000 Groundnut price (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 0.026 0.026 - 0.021 0.042 - 0.007 0.028 Year dummy (2012/13) 0.058*** 0.021 0.001 0.036 0.026 0.023 T2 dummy 0.001 0.012 - 0.059*** 0.021 0.028** 0.014 T3 dummy 0.001 0.012 - 0.047** 0.021 0.016 0.014 Electoral threat - 0.024 0.024 - 0.032 0.024 0.004 0.022 Number of subsidized fertilizer vouchers 0.000 0.000 - 0.000** 0.000 0.000*** 0.000 Proportion adopting inorganic fertilizer 0.001*** 0.000 - 0.001** 0.000 0.002*** 0.000 Proportion adopting organic fertilizer - 0.000 0.000 0.001*** 0.000 - 0.000 0.000 Proportion adopting maize - legume IC - 0.000** 0.000 0.003*** 0.000 0.000* 0.000 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 116 Table 2A.21: Marginal effect s of use of practices in each SI category on crop productivity Variables Intensification Sustainable SI dy/dx SE dy/dx SE dy/dx SE Male - headed HH (yes = 1) 0.010 0.011 0.001 0.019 - 0.004 0.012 Age of HH head (years) - 0.002 0.002 0.005 0.004 0.002 0.002 Education of HH head (years) 0.006*** 0.001 - 0.005* 0.003 0.008*** 0.002 Family labor (number of adults per acre) - 0.007 0.004 0.003 0.009 0.002 0.006 Total cultivated land (acres) - 0.001 0.001 0.001 0.002 - 0.002 0.001 Off - farm income (yes = 1) 0.014 0.013 - 0.006 0.028 - 0.008 0.015 Farm assets (1,000 TZS) - 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) - 0.010 0.009 0.086*** 0.017 0.050*** 0.010 Access to credit (yes = 1) 0.020 0.013 - 0.034 0.030 0.052*** 0.014 Membership (SACCOS) (yes = 1) 0.045*** 0.015 - 0.031 0.040 0.041** 0.018 0.026** 0.011 - 0.053* 0.029 0.052*** 0.012 Distance to main road (km) - 0.000 0.000 - 0.000 0.000 - 0.001*** 0.000 Distance to town (km) - 0.001*** 0.000 0.000 0.000 - 0.001*** 0.000 Distance to main market (km) 0.000*** 0.000 0.000 0.000 0.001*** 0.000 Cooperatives (yes = 1) 0.024*** 0.008 - 0.012 0.017 - 0.006 0.010 Input supplier (yes = 1) 0.017** 0.008 - 0.042** 0.017 0.031*** 0.009 Drought/Flood (yes = 1) 0.000 0.015 0.008 0.025 - 0.036* 0.019 Crop disease/Pests (yes = 1) 0.000 0.015 0.007 0.028 - 0.001 0.017 Total rainfall (mm) 0.000 0.000 - 0.000 0.000 0.000 0.000 Soil nutrient constraint (yes = 1) - 0.014 0.009 0.019 0.018 0.007 0.011 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Lagged price of rice (TZS/kg) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Bean price (TZS/kg) - 0.000 0.000 - 0.000 0.000 0.000 0.000 Groundnut price (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 0.061** 0.025 - 0.079* 0.041 0.059** 0.028 Year dummy (2012/13) 0.077*** 0.021 - 0.019 0.037 0.049** 0.023 T2 dummy 0.006 0.013 - 0.059*** 0.021 0.024 0.015 T3 dummy 0.010 0.012 - 0.055*** 0.021 0.022 0.014 Electoral threat - 0.024 0.023 - 0.032 0.024 0.001 0.020 Proportion receiving agricultural advice 0.001*** 0.000 - 0.002*** 0.000 0.002*** 0.000 Proportion adopting organic fertilizer 0.000** 0.000 0.001*** 0.000 0.000** 0.000 Proportion adopting maize - legume IC - 0.000 0.000 0.002*** 0.000 0.001*** 0.000 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 117 Table 2A.22: Marginal effect s of use of practices in each SI category on modified HDDS Variables Intensification Sustainable SI dy/dx SE dy/dx SE dy/dx SE Male - headed HH (yes = 1) 0.008 0.011 0.002 0.019 - 0.009 0.011 Age of HH head (years) - 0.002 0.002 0.005 0.004 0.002 0.002 Education of HH head (years) 0.004*** 0.001 - 0.003 0.003 0.006*** 0.002 Family labor (number of adults per acre) - 0.008* 0.004 0.004 0.009 0.001 0.006 Total cultivated land (acres) - 0.000 0.001 0.001 0.002 - 0.001 0.001 Off - farm income (yes = 1) 0.014 0.013 - 0.007 0.028 - 0.005 0.014 Farm assets (1,000 TZS) - 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) - 0.010 0.008 0.102*** 0.016 0.044*** 0.009 Access to credit (yes = 1) 0.012 0.012 - 0.029 0.030 0.044*** 0.014 Membership (SACCOS) (yes = 1) 0.027* 0.015 - 0.023 0.040 0.027 0.017 0.020* 0.010 - 0.047 0.029 0.041*** 0.012 Distance to main road (km) - 0.000 0.000 - 0.000 0.000 - 0.001** 0.000 Distance to town (km) - 0.000** 0.000 - 0.000 0.000 - 0.000** 0.000 Distance to main market (km) 0.000** 0.000 0.000 0.000 0.000*** 0.000 Cooperatives (yes = 1) 0.015* 0.008 - 0.007 0.017 - 0.018* 0.009 Input supplier (yes = 1) 0.003 0.008 - 0.025 0.017 0.008 0.009 Drought/Flood (yes = 1) 0.002 0.015 0.005 0.024 - 0.030* 0.018 Crop disease/Pests (yes = 1) 0.001 0.015 0.007 0.027 - 0.003 0.016 Total rainfall (mm) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Soil nutrient constraint (yes = 1) - 0.011 0.009 0.010 0.017 0.015 0.011 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.000 0.000 0.000 0.000 0.000 0.000 Lagged price of rice (TZS/kg) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Bean price (TZS/kg) - 0.000 0.000 - 0.000 0.000 0.000 0.000 Groundnut price (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 0.035 0.024 - 0.058 0.041 0.021 0.027 Year dummy (2012/13) 0.054*** 0.020 0.004 0.036 0.015 0.022 T2 dummy 0.002 0.012 - 0.057*** 0.021 0.028* 0.014 T3 dummy 0.002 0.012 - 0.046** 0.021 0.017 0.014 Electoral threat - 0.026 0.024 - 0.028 0.024 0.001 0.023 Proportion adopting inorganic fertilizer 0.001*** 0.000 - 0.001** 0.000 0.002*** 0.000 Proportion adopting maize - legume IC - 0.000** 0.000 0.003*** 0.000 0.000** 0.000 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 118 Table 2A.23: Marginal effect s of use of practices in each SI category on food expenditure per adult equivalent Variables Intensification Sustainable SI dy/dx SE dy/dx SE dy/dx SE Male - headed HH (yes = 1) 0.008 0.011 0.002 0.019 - 0.009 0.011 Age of HH head (years) - 0.002 0.002 0.005 0.004 0.002 0.002 Education of HH head (years) 0.004*** 0.001 - 0.003 0.003 0.006*** 0.002 Family labor (number of adults per acre) - 0.008* 0.004 0.004 0.009 0.001 0.006 Total cultivated land (acres) - 0.000 0.001 0.001 0.002 - 0.001 0.001 Off - farm income (yes = 1) 0.013 0.013 - 0.009 0.028 - 0.005 0.014 Farm assets (1,000 TZS) - 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) - 0.010 0.008 0.092*** 0.017 0.045*** 0.009 Access to credit (yes = 1) 0.013 0.012 - 0.028 0.030 0.045*** 0.014 Membership (SACCOS) (yes = 1) 0.026* 0.015 - 0.023 0.040 0.027 0.017 0.020* 0.010 - 0.048* 0.029 0.041*** 0.012 Distance to main road (km) - 0.000 0.000 - 0.000 0.000 - 0.001** 0.000 Distance to town (km) - 0.000** 0.000 0.000 0.000 - 0.000** 0.000 Distance to main market (km) 0.000** 0.000 0.000 0.000 0.000*** 0.000 Cooperatives (yes = 1) 0.015* 0.008 - 0.008 0.016 - 0.019** 0.009 Input supplier (yes = 1) 0.004 0.008 - 0.030* 0.017 0.009 0.009 Drought/Flood (yes = 1) 0.002 0.015 0.009 0.024 - 0.030* 0.018 Crop disease/Pests (yes = 1) 0.001 0.015 0.009 0.027 - 0.003 0.016 Total rainfall (mm) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Soil nutrient constraint (yes = 1) - 0.010 0.009 0.015 0.017 0.015 0.011 Inorganic fertilizer price (TZS/kg) 0.000* 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.000 0.000 0.000 0.000 0.000 0.000 Lagged price of rice (TZS/kg) 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Bean price (TZS/kg) - 0.000 0.000 - 0.000 0.000 0.000 0.000 Groundnut price (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Year dummy (2010/11) 0.035 0.024 - 0.034 0.040 0.020 0.027 Year dummy (2012/13) 0.052*** 0.020 0.015 0.036 0.013 0.022 T2 dummy 0.002 0.012 - 0.060*** 0.021 0.028* 0.014 T3 dummy 0.002 0.012 - 0.049** 0.021 0.017 0.014 Proportion adopting inorganic fertilizer 0.001*** 0.000 - 0.001*** 0.000 0.002*** 0.000 Proportion adopting organic fertilizer - 0.000 0.000 0.001*** 0.000 - 0.000 0.000 Proportion adopting maize - legume IC - 0.000** 0.000 0.003*** 0.000 0.000** 0.000 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 119 Table 2A.24: Marginal effects of use of practices in each SI category on FCS Variables Intensification Sustainable SI dy/dx SE dy/dx SE dy/dx SE Male - headed HH (yes = 1) - 0.017 0.017 0.006 0.029 0.002 0.019 Age of HH head (years) - 0.004 0.004 0.012** 0.006 0.005 0.004 Education of HH head (years) 0.003 0.002 - 0.000 0.004 0.008*** 0.003 Family labor (number of adults per acre) - 0.006 0.012 - 0.008 0.010 0.008 0.009 Total cultivated land (acres) - 0.001 0.002 0.001 0.003 - 0.000 0.002 Off - farm income (yes = 1) 0.016 0.023 0.007 0.038 0.004 0.025 Farm assets (1,000 TZS) - 0.000 0.000 0.000 0.000 0.000 0.000 Livestock ownership (yes = 1) - 0.010 0.014 0.096*** 0.025 0.047*** 0.015 Access to credit (yes = 1) 0.008 0.022 - 0.055 0.046 0.059*** 0.023 Membership (SACCOS) (yes = 1) 0.041* 0.025 - 0.059 0.055 0.018 0.028 0.037** 0.018 - 0.081* 0.048 0.035* 0.019 Distance to main road (km) - 0.000 0.000 - 0.001* 0.001 - 0.000 0.000 Distance to town (km) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Distance to main market (km) 0.000** 0.000 0.000 0.000 0.000* 0.000 Cooperatives (yes = 1) 0.025* 0.014 - 0.036 0.025 - 0.024 0.016 Input supplier (yes = 1) 0.006 0.014 - 0.016 0.026 - 0.008 0.015 Drought/Flood (yes = 1) 0.030 0.023 0.029 0.036 - 0.039 0.031 Crop disease/Pests (yes = 1) - 0.015 0.028 - 0.041 0.043 - 0.013 0.028 Total rainfall (mm) 0.000* 0.000 - 0.000** 0.000 - 0.000 0.000 Soil nutrient constraint (yes = 1) - 0.003 0.017 0.013 0.028 0.027 0.019 Inorganic fertilizer price (TZS/kg) - 0.000 0.000 0.000 0.000 - 0.000 0.000 Lagged price of maize (TZS/kg) - 0.000 0.000 0.001** 0.000 0.001* 0.000 Lagged price of rice (TZS/kg) - 0.000 0.000 - 0.000* 0.000 - 0.000** 0.000 Bean price (TZS/kg) - 0.000 0.000 0.000 0.000 0.000 0.000 Groundnut price (TZS/kg) - 0.000 0.000 - 0.000 0.000 - 0.000 0.000 Year dummy (2012/13) 0.007 0.036 0.032 0.056 0.067* 0.039 Number of subsidized fertilizer vouchers 0.000 0.000 - 0.000 0.000 0.000*** 0.000 Proportion adopting inorganic fertilizer 0.002*** 0.000 - 0.001** 0.000 0.002*** 0.000 Proportion adopting maize - legume IC - 0.000 0.000 0.002*** 0.000 0.000 0.000 Notes: SE is standard errors. Non - adoption is the reference category. ***, **, and * denote statistical significance at the 1%, 5%, and 10% levels, respectively. 120 REFERENCES 121 REFERENCES Bourguignon, F., Fournier, M., and Gurgand, M. 2007. Selection bias corrections based on the multinomial logit model: Monte Carlo comparisons. Journal of Economic Surveys 21(1):174 - 205. Bronick, C.J., and Lal, R. 2005. Soil structure and management: a rev iew. Geoderma 124:3 - 22. Chang, E.C. 2005. Electoral incentives for political corruption under open - list proportional representation. The Journal of Politics 67(3):716 - 730. food basket approach. EIB - 135, U.S. Department of Agriculture, Economic Research Service. Di Falco, S., and Veronesi, M. 2013. How can African agriculture adapt to climate change? A counterfactual analysis from Ethiopia. Land Economics 89(4):743 - 766. Di Falco, S., Veronesi, M., and Yesuf, M. 2011. Does adaptation to climate change provide food security? A micro - perspective from Ethiopia. American Journal of Agricultural Economics 93(3):829 - 846. FAO. 1996. Declaration on the world food security. World foo d summit, Rome: FAO. Godfray, H.C.J. 2015. The debate over sustainable intensification. Food Security 7(2):199 - 208. Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., Pretty, J., Robinson, S., Thomas, S.M., and Toulmin, C. 2010. Food security: the challenge of feeding 9 billion people. Science 327(5967):812 - 818. Herforth, A., and Harris, J. 2014. Understanding and Applying Primary Pathways and Principles. Brief #1. Improving Nutrition through Agriculture Technical Brief Series. Arlington, VA: USAID/Strengthening Partnerships, Results, and Innovations in Nutrition Globally (SPRING) Project. Holden, S.T. 2018. Fertilizer and sustainable intensification in Sub - Saharan Africa. Global Food Security 18:20 - 26. Jaleta, M., Kas sie, M., Tesfaye, K., Teklewold, T., Jena, P.R., Marenya, P., and Erenstein, O. 2016. Resource saving and productivity enhancing impacts of crop management innovation packages in Ethiopia. Agricultural Economics 47(5):513 - 522. Jayne, T.S., Mason, N.M., Bu - generation agricultural input subsidy programs. Food Policy 75:1 - 14. 122 Jayne, T.S., Snapp, S., Place, F., and Sitko, N. 2019. Sustainable agricultural intensification in an era of rural transfor mation in Africa. Global Food Security 20:105 - 113. Jones, A.D., Ngure, F.M., Pelto, G., and Young, S.L. 2013. What are we assessing when we measure food security? A compendium and review of current metrics. Advances in Nutrition 4(5):481 - 505. Kassie, M., Marenya, P., Tessema, Y., Jaleta, M., Zeng, D., Erenstein, O., and Rahut, D. 2018. Measuring farm and market level economic impacts of improved maize production technologies in Ethiopia: evidence from panel data. Journal of Agricultural Economics 69(1):76 - 95. Kassie, M., Teklewold, H., Jaleta, M., Marenya, P., and Erenstein, O. 2015a. Understanding the adoption of a portfolio of sustainable intensification practices in eastern and southern Africa. Land Use Policy 42:400 - 411. Kassie, M., Teklewold, H., Marenya, P., Jaleta, M., and Erenstein, O. 2015b. Production risks and food security under alternative technology choices in Malawi: Application of a multinomial endogenous switching regression. Journal of Agricultural Economics 66(3):640 - 659. Khonje, M.G., Manda, J., Mkandawire, P., Tufa, A.H., and Alene, A.D. 2018. Adoption and welfare impacts of multiple agricultural technologies: evidence from eastern Zambia. Agricultural Economics 49(5):599 - 609. Kim, J., Mason, N.M., Snapp, S., and Wu, F. (in press). Does Sustainable Intensification of Maize Production Enhance Child Nutrition? Evidence from Rural Tanzania. Accepted for publication in Agricultural Economics on 8 July 2019. Kumar, N., Harris, J., and Rawat, R. 2015. If they g row it, will they eat and grow? Evidence from Zambia on agricultural diversity and child undernutrition. The Journal of Development Studies 51 (8):1060 - 1077. Liu, Y., and Myers, R. 2009. Model selection in stochastic frontier analysis with an application to maize production in Kenya. Journal of Productivity Analysis 31(1):33 - 46. Leroy, J.L., Ruel, M., Frongillo, E.A., Harris, J., and Ballard, T.J. 2015. Measur ing the food access dimension of food security: a critical review and mapping of indicators. Food and Nutrition Bulletin 36(2):167 - 195. Magrini, E., and Vigani, M. 2016. Technology adoption and the multiple dimensions of food security: the case of maize i n Tanzania. Food Security 8(4):707 - 726. Manda, J., Alene, A.D., Gardebroek, C., Kassie, M., and Tembo, G. 2016a. Adoption and impacts of sustainable agricultural practices on maize yields and incomes: Evidence from rural Zambia. Journal of Agricultural Ec onomics 67(1):130 - 153. 123 Manda, J., Gardebroek, C., Khonje, M.G., Alene, A.D., Mutenje, M., and Kassie, M. 2016b. Determinants of child nutritional status in the eastern province of Zambia: the role of improved maize varieties. Food Security 8:239 - 253. Mare nya, P.P., and Barrett, C.B. 2009. State - conditional fertilizer yield response on western Kenyan farms. American Journal of Agricultural Economics 91(4):991 - 1006. Mason, N.M., Jayne, T.S., and Van De Walle, N. 2017. The political economy of fertilizer sub sidy programs in Africa: Evidence from Zambia. American Journal of Agricultural Economics 99(3):705 - 731. Mather, D.L., and Jayne, T.S. 2018. Fertilizer subsidies and the role of targeting in crowding out: evidence from Kenya. Food Security 10(2):397 - 417. Mather, D.L., and Minde, I. 2016. Fertilizer subsidies and how targeting conditions crowding in/out: An assessment of smallholder fertilizer demand in Tanzania. GISAIA/Tanzania Working Paper #5, Michigan State University, East Lansing, MI. Matson, P.A., Parton, W.J., Power, A.G., and Swift, M.J. 1997. Agricultural intensification and ecosystem properties. Science 277(5325):504 - 509. McFadden, D. 1973. Conditional logit analysis of qualitative choice behavior. In: Zarembka, P. (Ed.). Frontiers in Econometrics. Academic Press, New York. Mekuria, M., and Waddington, S.R. 2002. Initiatives to encourage farmer adoption of soil - fertility technologies for maize - based cropping systems in Southern Africa. In C.B. Barrett, F. Place and A.A. Ab oud, eds. Natural resources management in African agriculture: Understanding and improving current practices. New York:219 - 233. Ngwira, A.R., Aune, J.B., and Mkwinda, S. 2012. On - farm evaluation of yield and economic benefit of short term maize legume int ercropping systems under conservation agriculture in Malawi. Field Crops Research 132:149 - 157. Petersen, B., and Snapp, S. 2015. What is sustainable intensification? Views from experts. Land Use Policy 46:1 - 10. Pingali, P.L. 2012. Green revolution: impac ts, limits, and the path ahead. Proceedings of the National Academy of Sciences 109(31):12302 - 12308. Pretty, J., Toulmin, C., and Williams, S. 2011. Sustainable intensification in African agriculture. International Journal of Agricultural Sustainability 9 (1):5 - 24. Sauer, C.M., Mason, N.M., Maredia, M.K., and Mofya - Mukuka, R. 2018. Does adopting legume - based cropping practices improve the food security of small - scale farm households? Panel survey evidence from Zambia. Food Security 10(6):1463 - 1478. 124 Shifer aw, B., Kassie, M., Jaleta, M., and Yirga, C. 2014. Adoption of improved wheat varieties and impacts on household food security in Ethiopia. Food Policy 44:272 - 284. nonexperime ntal estimators? Journal of Econometrics 125(1 - 2):305 - 353. of household diet in Malawi. Food Security 7(1):83 - 96. Stahley, K., Slakie, E., Derksen - Schrock, K ., Gugerty, M.K., and Anderson, C.L. 2012. Tanzania National Panel Survey LSMS - ISA: Legumes. Evans School Policy Analysis and Research (EPAR) No. 189. Swindale, A, and Bilinsky, P. 2006 Household dietary diversity score (HDDS) for measurement of household food access: Indicator guide, Academy for Educational Development, Washington, DC. Tanzania National Bureau of Statistics. 2014. Tanzania National Panel Survey Report (NPS) Wave 3, 2012 - 2013, Dar es Salaam, Tanzania, NBS. Teklewold, H., Kassie, M., Shiferaw, B., and Köhlin, G. 2013. Cropping system diversification, conservation tillage and modern seed adoption in Ethiopia: Impacts on household income, agrochemical use and demand for labor. Ecological Economics 93:85 - 93. The Montpellier Panel, 2013. Sustainable Intensification: A New Paradigm for African Agriculture. Agriculture for Impact, London. World Food Programme, 2008. Food consumption analysis: Calculation and use of the food consumption score in food security ana lysis. Rome: United Nations Vulnerability Analysis and Mapping Branch. Waddington, S.R., Mekuria, M., Siziba, S., and Karigwindi, J. 2007. Long - term yield sustainability and financial returns from grain legume maize intercrops on a sandy soil in subhumid north central Zimbabwe. Experimental Agriculture 43(4):489 - 503. Wooldridge, J.M. 2010. Econometric Analysis of Cross Section and Panel Data . MIT Press, Cambridge, MA. World Bank. 2014. Public expenditure review: national agricultural input voucher scheme (NAIVS). Zeng, D., Alwang, J., Norton, G.W., Shiferaw, B., Jaleta, M., and Yirga, C. 2015. Ex post impacts of improved maize varieties on poverty in rural Ethiopia. Agricultural Economics 46(4):515 - 526. 125 CH APTER 3 THE EFFECTS OF THE NATIONAL AGRICULTURAL INPUT VOUCHER SCHEME (NAIVS) ON SUSTAINABLE INTENSIFICATION OF MAIZE PRODUCTION IN TANZANIA 3.1 Introduction Hunger and food insecurity continue to be major challenges in sub - Saharan Africa (SSA). Currently, SSA is the region with the largest gap between cereal consumption and production and about a quarter of the population suffered from chronic food deprivation in 2017 (van Ittersum et al., 2016; FAO, IFAD, UNICEF, WF P and WHO, 2018). These problems may become more serious in the future because by 2050 the population in SSA is projected to increase 2.5 - fold and its cereal demand is projected to triple, while the region already imports substantial quantities of cereals to meet current demand (van Ittersum et al., 2016). In addition, there is an emerging consensus that conventional intensification of agricultural systems involving the use of inorganic fertilizer and high - yielding crop varieties may be insufficient to sust ainably intensify agricultural production and that conventional intensification can have negative environmental externalities (Petersen and Snapp, 2015; Pingali, 2012). In this context, sustainable intensification (SI) has been identified as a potential me ans to feed an increasing global population and meet rising food demand (Godfray et al., 2010). The main goal of SI is to produce more agricultural output from the same area of land (or less land) on a sustainable basis without adverse environmental impact (Pretty et al., 2011; The Montpellier Panel, 2013). While SI does not refer to a specific set of agricultural inputs or management practices and there are likely to be many pathways to SI, Holden (2018) points to integrated soil fertility management (ISFM ) and conservation agriculture (CA) as two potential approaches to SI. ISFM is defined as 126 the combined use of inorganic fertilizer and locally available soil amendments and organic matter, whereas CA involves crop rotation/intercropping with legumes, perma nent soil coverage, and minimum soil disturbance. productivity have primarily focused on conventional intensification in particular, trying to raise smallholder large scale input subsidy programs (ISPs). In recent years, 10 African countries spent approximately US$0.6 - 1 billion annually on ISPs . B ut despite the heavy spending on the programs, the effects of ISPs on crop production and productivity as well as incomes and poverty have generally been smaller than anticipated ISPs , Jayne et al. (2018) argue that low crop yield response to inorganic fertilizer consistently reduces the productivity effects of ISPs . In particular, poor soil quality (e.g., low soil organic of low crop yield response to inorganic fertilizer application (Marenya and Barrett, 2009; Burke et al., 2017). It is therefore important to address poor soil quality issues (e.g., through an application of complementary soil fertility management (SFM) pra ctices) in order to improve the agronomic 2018). In recognition of the importance of integrated agricultural practices that improve soil health and the effic iency of inorganic fertilizer use , contribute to SI of agricultural systems, and have impl i cations for the effectiveness of ISPs, the main research question of this study is joint use of inorganic fertilizer with other SFM practices ; this joint use can be considered a form of SI . To our knowledge, there have been no 127 previous studies on this relationship. Instead, there have bee n only a few empirical studies on th e individual SFM practices other than inorganic fertilizer in Malawi (Holden and Lunduka, 2012; Kassie et al., 2015a; Koppmair et al., 2017) and Zambia (Morgan et al., 2019). This is in contrast to the larger literature on the effects of ISPs on inorganic fertilizer purchases or use, which does not consider the p SFM practices or joint use of inorganic fertilizer with practices. (See Jayne et al. (2013) and Jayne et al. (2018) for listings and syntheses of these studies.) We focus here on the case of Tanzania and the ISP implemented by the Government of Tanzania from 2008/09 through 2013/14: the National Agricultural Input Voucher Scheme (NAIVS). NAIVS provided targeted beneficiaries with vouchers for inorganic fertilizer and seed for improved varieties of maize or rice two major staple c rops in Tanzania. NAIVS is a - market - overcome the shortcomings of past programs including the ir limited impacts on productivity, high costs (and low benefit - cost ratios), politicization, and sidelining of the private sector (Jayne et al., 2018; Dorward, 2009; Morris et al., 2007 ; Pan and Christiaensen, 2012 ). 45 NAIVS is an important case study on this to pic because it is widely considered to be the most private sector - friendly ISP in SSA to date (Wanzala et al., 2013). NAIVS was implemented through vouchers redeemable at private agro - - mentioned studies on the ef ISPs on individual SFM practices cover periods 45 Most first generation ISPs were phased out in the 1990s , and second generation ISPs beg an being introduced in the early - mid 2000s - ISP and Pan and Christiaensen (2012) briefly define such ISPs as follows: ISPs - o f a broader productivity enhancement program, if they have a clear exit strategy, and most importantly, if they are 128 (Malawi) or farmer cooperatives (Zambia) and not through the private sector (Mason and Ricker - Gilbert, 2013). Th effects in Malawi and Zambia. Furthermore, the design and implementation of ISPs varies across countries and time, so insights from a new country (in this case, Tanzania) can also help deepen our understanding of how ISP effects on SFM practices may vary depending on differences in program design and implementation. This study focuses on SFM practices for maize production because maize is both the main staple fo od cultivated by the majority of Tanzanian smallholders and the main crop promoted through NAIVS (World Bank, 2004). The SFM practices considered here include the use of inorganic fertilizer, organic fertilizer such as animal manure or compost, and maize - l egume intercropping . We focus on these three because they are the main SFM practices used by maize growing households in rural Tanzania . We follow Kim et al. ( in press ) and group the eight possible combinations of use of these three SFM practices into four SI categories - , meaning none of the practices are used ; inorganic fertilizer use only ; meaning use of organic fertilizer , maize - legume intercropping , or both ; meaning joint use of inorganic fertilizer with at least one of the practices category . Using nationally representative household panel survey data from Tanzania (the Tanzania National Panel Survey (TNPS) of 2008/09 and 2012/13), we estimate the impacts of receipt of vouchers for inorganic fertilizer and/or maize estimated using a multinomial logit (M NL) model combined with correlated random effects (CRE) and the control function (CF) approach to control, respectively, for time - invariant and 129 time - decisions and their rec eipt of NAIVS vouchers. This study contributes to the literature in several ways beyond being the first analysis of joint use of inorganic fertilizer and complementary SFM practices. First, unlike several of the previous st udies in the ISP - SFM literature that did not use nationally representative data (Holden and Lunduka, 2012; Kassie et al., 2015a; Koppmair et al., 2017) or used cross - sectional data (Kassie et al., 2015a), this study uses nationally representative household panel survey data. By using panel data method s , the internal validity of our results should be enhanced as we can control for time - invariant unobserved heterogeneity . Also, external validity should be improved by using the nationally - representative data. Second, we use the CF approach to address potential correlation of receipt of subsidized inputs with time - varying unobserved heterogeneity; in contrast, Kassie et al. (2015a) and Koppmair et al. (2017) do not directly addres s this issue, which may result in biased and inconsistent estimates. We find statistically significant positive effects of household receipt of a NAIVS voucher for inorganic fertilizer on maize - growing use of inorganic fertilizer only (i.e., using inorganic fertilizer only is on average 10.0 percentage points higher than for households who do not receive a NAIVS voucher. Our results further suggest that NAIVS vouche r receipt encourages farmers to use inorganic fertilizer jointly with organic fertilizer and/or maize - legume intercropping . More specifically, NAIVS voucher receipt for inorganic fertilizer is associated with a 9.6 percentage point increase in s probability of using practices in the the practices in the has no statistically significant effect on fa r category decision s . 130 The remainder of this study is organized as follows. First, we provide background information on the NAIVS program and SI of maize production in Tanzania. Next, we outline the conceptual framework and empirical strategies for estimating the effects of the NAIVS program on a maize - , including joint use of inorganic fertilizer with other SFM practices . Then, we describe the data and variable specifications. Finally, we present our re sults and conclude by discussing policy implications. 3.2 Background: SI of maize production & the NAIVS program in Tanzania 3.2.1 SI of maize production in Tanzania Per Kim et al. (in press), the main rationale for the categorization of inorganic fertili zer use only use of inorganic fertilizer has substantially contributed to raising agricultural productivity over the last several decades (Godfray et al., 2010; Pingali, 2012) , its sole use can have adverse consequences including over - reliance on fossil fuels ; decrease s in biodiversity ; ground and water pollution ; and reduction s in soil pH, soil organic carbon (SOC), soil aggregation, and m icrobial communities ( Matson et al., 1997; Pingali, 2012; Petersen and Snapp, 2015; Bronick and Lal, 2005). O rganic fertilizer use and maize - legume intercropping are local and renewable ways to raise soil fertility but their use without in organic fertilizer is unlikely to significantly raise maize yields. Finally, the combined use of inorganic fertilizer with either organic fertilizer and/or maize - legume intercropping is expected to result in sustainable increases in maize yields from the same area of the land while preserving or improving soil health due to the synergistic e ffects of joint use of the practices . See 131 Kim et al. (in press) for a much more detailed discussion of the rationale for these categorizations , including extensive references to the agronomy and other related literatures. Table 3.1 shows the prevalence of the various SFM practices and SI categories on maize plots in Tanzania. Out of 2,559 maize plots in the sample (TNPS 2008/09 and 2012/13 , described below ), 41.4% of them are cultivated with only one of the three SFM practices. The maize plots with inorganic fertilizer only and organic fertilizer only account for 8.8% (case 2) and 6.5% (case 3) of all maize plots, respectively; and the maize plots intercropped with legumes but without use of the other two practices account for 26.1% (case 4). On the other hand, the proportion of maize plots cultivated with two or more SFM practices is relatively low, accounting for 13.3% of total maize plots (i.e., cases 5, 6, 7, and 8). Table 3.1 also shows the plot - level SI categories used for the empirical analysis : o ut of 2,559 maize plots, the 37% , while t account for much lower proportion s at approximately 9% of maize plots each. In particular, among the maize plots included in the up , the combined use of inorganic fertilizer and at least maize - legume intercropping accounts for 6.9%, while joint use of inorganic fertilizer and at least - c ategory. Among the three SFM practices, maize - legume intercropping is the most common among maize - growing households in rural Tanzania as it is used on 38% of all maize plots in the sample (alone or in combination with other practices). I norganic fertilizer use and organic fertilizer use are much lower at 18% and 14% of maize plots, respectively (Table 3.1). 132 Table 3.1: SI of maize production categories and prevalence on maize plots in the sample Case Inorganic fertilizer Organic fertilizer Maize - legume intercropping No. of maize plots (%) SI category No. of maize plots (%) 1 1,159 (45.3) Non - adoption 1,159 (45.3) 2 224 (8.8) Intensification 224 (8.8) 3 166 (6.5) Sustainable 948 (37.0) 4 669 (26.1) 5 113 (4.4) 6 50 (2.0) SI 228 (8.9) 7 147 (5.7) 8 31 (1.2) Total number of maize plots 2,559 (100.0) 2,559 (100.0) Use of inorganic fertilizer 452 (17.7) Use of organic fertilizer 360 (14.1) Use of maize - legume intercropping 960 (37.5) Note: Figures in the table are based on maize plots ( n =2,559) cultivated by the balanced panel of rural maize - growin g households across two waves of the TNPS (2008/09, and 2012/13). The eight cases and SI categories are each mutually exclusive, while the number of maize plots for the practices listed at the bottom of the table include maize plots for which the practice wa s applied alone or in combination with other practices. The l egume crops reported as being intercropped with maize in the survey are beans, soybeans, groundnuts, cowpeas, pigeon peas, chickpeas, field peas, green grams, bambara nuts, and fiwi. Source: Auth 3.2.2 The National Agricultural Input Voucher Scheme In Tanzania, there were large - scale , universal subsidy programs between the 1960s and the 1980s, where the government controlled importation and distribution of agricultural inputs and heavily subsidized input prices (World Bank, 2014). With the economic crisis in the mid - 1980s that resulted in an economic reform program, the Tanzanian government greatly reduced subsidy rates on fertilizer from 80% in 1990 to 55% in early 1992, and to no more than 20% by mid - 1992 (Putterman, 1995). These subsidies were ultimately phased out altogether after li beralization of agricultural markets between 1991 and 1994. In 2003, after a decade with no subsidized agricultural inputs, the Government of Tanzania resumed a transport subsidy for companies that 133 were involved in the distribution of fertilizers. However, the transport subsidy was not successful since the distributors and agro - dealers who directly received the subsidy did not pass on the cost savings to smallholder farmers (Mather et al., 2016). Also, there were some constraints frequently reported under t his system: delayed input delivery, inputs not being effective due to quality deterioration, and smuggling to neighboring countries (Aloyce et al., 2014). Eventually, due to concerns regarding the cost effectiveness of the program , targeting , and the distr ibution of subsidy benefits , the program was phase d out and redesign ed in 2007. Following the 2007/2008 food price crisis, the Government of Tanzania decided to launch a voucher - based input subsidy program that was piloted in two districts within the Mbeya and Rukwa regions in 2007/08. T he Tanzanian government with financial support f rom the World Bank in 2008/09 rapidly scaled up the existing input voucher pilot program with the goal of enhancing short and longer - term food security in the country (Mather et al., 2016; Pan and Christiaensen, 2012). The scaled - up program was called the NAIVS and it operated in 58 districts across 11 regions in 2008/09 ; the goal was to eventually reach 2.5 million households for three consecutive years each . 46 The NAIVS wa s initially geographically targeted to areas favorable to maize and rice production in Tanzania. However, the NAIVS program was expanded nationwide by 2011/12 due to political pressure, which allowed other rural regions to receive at least small quanti ties of vouchers while a substantial share of the vouchers was still concentr ated in the originally designated regions (World Bank, 2014). Table 3.2 shows the number of household beneficiaries of the NAIVS program between 2008/09 and 2013/14, where the 730,667 households in the 2008/09 crop season were expected to receive vouchers for three consecutive years. The number of household beneficiaries reached its peak in 2010/11 and then 46 The targeted regions were intially Iringa, Mbeya, Ruvuma, Rukwa , Kilimanjaro, Arusha, Manyara, Kigoma, Tabora, Mara, and Morogoro , with Pwani added in 2009/10 (World Bank, 2014). 134 declined as beneficiaries completed their three years of assistance . NAIVS officially ended during the 2013/14 cropping season. 47 Table 3.2: Household beneficiaries for the NAIVS 2008/09 2009/10 2010/11 2011/12 2012/13 2013/14 Planned 740,000 1,500,000 2,040,000 1,800,000 1,000,000 500,000 Actual 730,667 1,511,900 2,011,000 1,779,867 940,783 932,100 Source: World Bank (2014) The major goals of NAIVS were to: (i) increase the production of maize and rice, the two improved maize and rice varieties; and (iii) strengthen private sector improved seed and inorganic fertilizer value chains and increase agro - dealer activity at village level (World Bank, 2014; Mather et al., 2016). distribution systems for subsidized in puts and have only recently started engaging the private sector in major ways, from its start, NAIVS used a much more private sector - oriented approach whereby the private sector handled importation, distribution, and retailing of the subsidized fertilizer 48 In addition, NAIVS primarily targeted households with limited experience usin g modern inputs 47 External funding through the World Bank was finally terminated in 2014, which was the official closure of NAIVS. However, in subsequent years the government of Tanzania continued providing input subsidies to farmers through different approaches including: (i) credit - based subsidies in 2014/15 through which the government voucher - based system in 2015/16; and (iii) subsidized fertilizer by en tering into contracts with seed and fertilizer companies to supply inputs in 2016/17 (Masinjila and Lewis, 2018). 48 In Malawi, the government parastatal distributed fertilizers from the port to parastatal depots (Mather et al., 2016) and until recently , f ertilizer vouchers for the ISP could only be redeemed at government depots (and not at private agro - (Lunduka et al., 2013). In Zambia, an electronic - voucher pilot program was launched in t use vouchers ; rather, subsidized fertilizers and seeds were distributed through a dedicated system that operated separately from private agro - dealers instead of through them. 135 but that had the farming resources required to use these inputs well (Worl d Bank, 2014). More specifically, to be eligible for the program, beneficiaries had to : (i) have the ability and willingness to co - finance the input purchase (i.e., upon redeeming the vouchers for each subsidized input which had a face value of half of the market price, the recipient needed to pay the remaining 50% of the price ) ; and ii) be full time farmers with one hectare or less of maize or rice under cultivation, where female - headed households and farmers that had not used modern inputs o n maize or rice within the past five years were to be priorit ized . 49 Given these targeting criteria, NAIVS was not intended to help the most vulnerable households among the poor beca use farmers who cannot co - finance the inputs purchased with the voucher are less likely to be able to purchase the inputs at market prices once subsidies are phased out. In addition, the second criterion was designed to prevent the vouchers from reaching households who were already capable of self - financing purchase of the inputs (Mather et al., 2016). Each voucher recipient was to obtain three vouchers for three consecutive years and approximately 80 % of the vouchers were assigned to maize - growing households . 50 The vouchers were for: i) one 50 kg bag of urea, ii) one 50 kg bag of Di - Ammonium Phosphates (DAP) or two 50 kg bags of Minjingu Rock Phosphate (MRP) with nitrogen supplement, and iii) 10 kg of hybrid or open - pollinated maize seed or 16 kg of rice seed, w hich is suitable for planting approximately one acre of land (World Bank, 2014 ; Pan and Christiaensen, 2012 ). The voucher 49 Mather and Minde (2016) provide descriptive evidence based on data from the TNPS and a World Bank household survey that the majority of NAIVS recipients met the major targeting criteria such as voucher distribution to the most suitable regions for maize and rice production and targeted farmers who have one hectare or less of maize or rice area and who had previously not been using modern inputs within the last five years. However, out of 2.5 million voucher recipients between 2008 and 2013, only 14.7% of them were women although female - headed households were supposed to be given pr eference (Masinjila and Lewis, 2018). 50 There may be lagged or enduring effects of the vouchers received for three consecutive years, but this study cannot directly control for this due to lack of data on NAIVS participation in years prior to the years cap tured in the surveys. 136 recipients were to redeem their vouchers at local agro - dealerships participating in the program and pay the 50% top - up fee for the sub sidized inputs at that time. 51 In general, the NAIVS vouchers were geographically allocated each year through a multi - stage targeting process. As the first step , a national voucher committee which consisted of central and regional government officials and representatives from private sector input supply chains would meet to determine how vouchers should be allocated among regions . Then, a similar voucher committee at the district level set the number of vouchers to assign to each district (ward/village). At each level of government, the vouchers were allocated based on the estimated numbers of farmers that to population size (Mather et al., 2016). At the last stage of the distribution, a village voucher committee which consisted of elected village leaders, several resident farmers, an d extension agents generated a list of beneficiary farmers which was then submitted to the village assembly for approval . Finally , the input vouchers were distributed to farmers that were approved by the village assembly and met the eligibility criteria. A mong the 1,624 maize growing households in our sample (which is drawn from the 2008/09 and 2012/13 TNPSs), 6.7% (108 households) of them received vouchers for inorganic fertilizers and/or maize seed through the NAIVS program (Table 3.3). Unlike the planned input subsidy package that three vouchers be allocated to each targeted farmer, Table 3.3 shows that 65.7% of recipient households (pooled across both waves of the TNPS) obtained vouchers only for inorganic fertilizer while 11.1% of them received only a v oucher for improved maize seed; just 23.1% of recipient farmers received vouchers for both inorganic fertilizer and improved 51 Although vouchers were intended to cover 50% of the input costs, increasing fertilizer prices in some years meant that they only covered 40 - 45% of the input cost (World Bank, 2014). 137 maize seed. 52 Given the geographic targeting and eligibility criteria for NAIVS, most of these voucher recipients reside in high pot ential maize production regions e.g., approximately 73.1% of them live in the Southern Highlands (i.e., Ruvum a, Iringa, Mbeya, and Rukwa regions); and 21.3% of them live in the northern part of the country (i.e., Arusha, Kilimanjaro, Tanga, Mara, and Manyara regions). Table 3.3 further shows that 87% of the sample farmers that received vouchers actually redeemed them at local agro - dealerships. According to Mather and Minde (2016), some voucher recipients did not redeem their vouchers because they could not afford the top - up fee; other recipients may have redeemed their vouchers with payment of the top - up fee and then sold one or more of their inputs to another farmer or back to the agro - dealer for cash. We cannot observe resale of inputs acquired with N AIVS vouchers in the TNPS data. Table 3.3: Number and percentage of rural maize - growing households that received versus redeemed a NAIVS voucher by input voucher type received TNPS 2008/09 (%) TNPS 2012/13 (%) Total (%) Voucher receipt Inorganic fertilizer only 14 (50.0) 57 (71.3) 71 (65.7) Improved maize seed only 3 (10.7) 9 (11.3) 12 (11.1) Both 11 (39.3) 14 (17.5) 25 (23.1) Total number of households 28 (100.0) 80 (100.0) 108 (100.0) Voucher receipt and redemption Inorganic fertilizer only 13 (92.9) 50 (87.7) 63 (88.7) Improved maize seed only 2 (66.7) 8 (88.9) 10 (83.3) Both 8 (72.7) 13 (92.9) 21 (84.0) Total number of households 23 (82.1) 71 (88.8) 94 (87.0) 52 In the TNPS, the reasons why farmers may not ha ve received the full set of vouchers are not reported , but Masinjila and Lewis (2018) provide several potential explanations for this. For example, some farmers with limited financial resources may want to take a voucher for a specific input type instead o f the entire package of the vouchers. In other cases, farmers were asked to sign for all the vouchers but did not receive all their inputs when inputs were delayed or local agro - dealers had run out of that input. 138 Table 3.4 shows the number and percentage of sample maize plots in each SI category owned by recipients of a NAIVS voucher (for inorganic fertilizer and/or improved maize seed) versus NAIVS non - recipients. Out of 2,559 maize plots, 8.4% (215 maize plots) a re owned by households who received a NAIVS voucher while 91.6% (2,344 maize plots) are owned by non - recipients. Among the 215 maize plots owned by NAIVS voucher recipients, approximately 36% ctively. Considering the input voucher types, recipients who received a voucher for inorganic fertilizer only or vouchers for both fertilizer and maize seed are more likely to fall compared to those who received imp roved maize seed only. On the other hand, approximately - 53 Unlike the case of the NAIVS voucher recipients, most of the maize plots owned by non - - categories are much less prevalent among NAIVS non - beneficiaries, at approximately 6% an d 7% of maize plots each. This may indicate that maize - producing households have difficulty affording inorganic fertilizers at unsubsidized prices. 53 Note that even if a farmer received an in organic fertilizer voucher, they could fall in the Non - adoption or Sustainable categories if they used the inorganic fertilizer acquired on a crop other than maize and/or if they did not redeem their voucher for inorganic fertilizer. 139 Table 3.4: Number and percentage of maize plots owned by NAIVS voucher recipients vs. non - recipients unde r SI category Non - adoption (row %) Intensification (row %) Sustainable (row %) SI (row %) Total (row %) Voucher recipients 31 (14.4) 77 (35.8) 41 (19.1) 66 (30.7) 215 (100.0) Input voucher type Inorganic fertilizer only 16 (11.2) 57 (39.9) 21 (14.7) 49 (34.3) 143 (100.0) Improved maize seed only 10 (45.5) 1 (4.5) 9 (40.9) 2 (9.1) 22 (100.0) Both 5 (10.0) 19 (38.0) 11 (22.0) 15 (30.0) 50 (100.0) Non - recipients 1,128 (48.1) 147 (6.3) 907 (38.7) 162 (6.9) 2,344 (100.0) Total maize plots 1,159 (45.3) 224 (8.8) 948 (37.0) 228 (8.9) 2,559 (100.0) 3.3 Methodology 3.3.1 Conceptual framework Following previous studies ( e.g., Marenya and Barrett, 2007; Di Falco and Veronesi, 2013; Teklewold et al., 2013), we use a random utility framework to conceptualize the effects of NAIVS voucher receipt on use of SFM practices on a given maize plot. Let d enote a lat ent variable that represents farmer on maize plot , ( in this study given that there are four SI categories). This study specifies the latent variable as: , (1) where t indexes the agricultural year; and , respectively, capture the observed household, plot, and community characteristics and their corresponding parameters (discussed in Section 3.4.2 below); with associated parameter is a dummy variable equal to one if the 140 household receive d a NAIVS voucher for inorganic fertilizer , improved maize seed, or both, and equal to zero otherwise; is household - level time - invariant unobserved heterogeneity; and is the time - varying error term. 54 However, we do not directly observe the expected utility from choosing alternative , only the choice ultimately made by the farmer . I t is assumed that farmer will choose alternative if using provides greater expected utility than any other alter native . This can be expressed as: (2) 3.3.2 Estimation strategy For the empirical analysis, we apply an MNL model, which is widely used in economic applications such as studi es on adoption of multiple agricultural technologies and their impacts (Grabowski et al., 2016; Teklewold et al., 2013; Kassie et al., 2015a; Khonje et al., 2018). The main advantage of using an MNL model (compared to a multivariate probit model, discussed below) is its computational simplicity in calculating choice probabilities without any requirement of multivariate integration (Tse, 1987; Hass a n and Nhemachena, 2008). In addition, the log - likelihood function for the MNL specification is globally concave , which makes the maximization problem straightforward (Hausman and McFadden, 1984). T he main drawback of the MNL model is the assumption of independence of irrelevant alternatives (IIA), which implies 54 The TNPS is a ho usehold - level panel dataset, not a plot - level one; thus we are only able to control for household - level (not plot - level) time - invariant unobserved heterogeneity. 141 that the relative odds between any two alternatives ar e independent of the characteristics of the other alternatives in the choice set (Wooldridge, 2010; Hausman and McFadden, 1984). An alternative approach to the MNL model is the multinomial probit model, which relaxes the IIA property by assuming that from choosing alternative for ) has a multivariate nor mal distribution with arbitrary correlations between and for all . The multinomial probit model is theoretically attractive but it also has some practical challenges: (i) the choice probabilities are very complicated, which makes it diffic ult to obtain partial effects on the choice probabilities; (ii) it requires that multivariate normal integrals be evaluated to estimate the unknown parameters; and (iii) it is not feasible for more than five alternatives, although this latter issue is not a constraint in the current application (Hausman and McFadden, 1984; Wooldridge, 2010). For these reasons, we use an MNL model instead of a multinomial probit model here. Assuming that the in equation (1) are identically and independently Gumbel distri buted, the probability that farmer characterized by , , and in equation (1) will choose alternative can be specified by the MNL model (McFadden, 1973) as: (3) As noted above, relatively few NAIVS beneficiaries received vouchers for both inorganic fertilizer and maize seed, while approximately 66% of the recipients pooling across both waves received only vouchers for inorganic fertilizer. The effects of NAIVS voucher receipt on therefore generate two alternative NAIVS variables based on input types: i) equals one if the household received a voucher for inorganic fert ilizer, and ii) equals one 142 if the household received a voucher for improved maize seed. In addition, when farmers received the vouchers bu t did not redeem them, the actual effects of the NAIVS program on each adoption strategy may b e under - or over - estimated. We thus also estimate a set of models using another set of alternative NAIVS To control for time - constant unobserved household - level heterogeneity ( ) that may be correlate d with the observed explanatory variables, a CRE/Mundlak - Chamberlain device approach is applied . This entails including the household - level time averages of the explanatory variables that change across and as additional regressors in equation (3) (Mu ndlak, 1978; Chamberlain, 1984; Wooldridge, 2010). This approach requires the assumptions of strict exogeneity of the explanatory variables conditional on the unobserved heterogeneity, and that the unobserved effects are linearly correlated with the household - level time averages of the observed explanatory variables. Even though our model controls for time - invariant unobserved heterogeneity via CRE, we still have concerns about potential e ndogeneity related to time - varying unobserved heterogeneity , particularly since NAIVS beneficiaries are not randomly selected. The NAIVS voucher receipt variables (i.e., , , and ) may be systemati cally related to time - varying unobse . To test and control for th is potential endogeneity of the NAIVS variables, we use the CF approach. The CF approach in the context of the current study consists of two steps (Wooldridge 2015). In the first step, we estimate a reduced form model via CRE logit in which the relevant NAIVS variables are the covariates in e quation (3) and at least as many instrumental variables (IVs) as there are potentially endogenous NAIVS variables (J.M. Wooldridge, personal 143 communication, May 2017). The logit generalized residuals obtained from the reduced form serve as the control funct ions. In the second step, the reduced form logit residuals are included as additional regressors in the main MNL model. If the coefficient on a given logit generalized residual variable is statistically significant at the 10% level or lower, then the null hypothesis that that NAIVS variable is exogenous is rejected. However , including the logit residuals in the main MNL model corrects for endogeneity of that NAIVS variable (Rivers and Vuong 1988). Because the logit generalized residuals are generated in a f irst stage estimation, we use bootstrapping to obtain valid standard errors for the parameter estimates in the MNL model (Wooldridge, 2010). To be valid IVs, there are two requirements: i) the IVs must be strongly partially correlated with the NAIVS variab les, and ii) partially uncorrelated with , where condition ii) is a maintained assumption that cannot be tested. This study considers two candidate IVs for the NAIVS variables. The first IV, , is the number of vouchers for inorganic fertilizer (nitrogen) distributed to region r . This variable is expected to be positively correlated with fertilizer and maize seed voucher receipt by a maize - growing household because most of the fertilizer vouche rs are geographically targeted to the most suitable areas for maize. The second IV is , which was us ed by Mather and Minde (2016) as an IV for the quantity of subsidized fertilizer received by a household . It is defined as the district - level ( d ) ratio of the proportion of votes for the runner - up in the most recent presidential election (Ibrahim Lipumba in 2005 and Willibrod Peter Slaa in 2010) over the proportion of votes for the winner (Jakaya Mrisho Kikwete in both 2005 an d 2010). 55 According to previous studies (e.g., Banful, 2011; 55 Both IVs, and , used in this study are ti me - varying in addition to varying across regions and districts, respectively.To construct , we used constituency - level data on electoral results from the 2005 and 2010 presidential elections and then aggregated them to the district level because the TNPS does not provide village names, which prevents us from being able to match households with their constituency. The electoral results in 2005 and 2010 were used to construct the IV for household receipt of NAIVS vouchers in T NPS 2008/09 and 2012/13, respectively. 144 Mason et al., 2017; Mather and Minde, 2016), past election results and voting patterns in a given area (district, constituency, etc.) have been found to affect the targeting of subsidized fertilizer in Ghana, Malawi, Zambia, and Tanzania. In Tanzania in particular, Mather and Minde (2016) found that electoral threat significantly affects the quantity of subsidized fertilizer received by the household. The reduced for m CRE logit results indicate that these IVs are indeed very strongly partially correlated with the potentially endogenous NAIVS variables ; the IVs are jointly significant for all NAIVS variables at the 1% level (see Table 3A.1 and Table 3A.2 in the Appendi x). Regarding requirement (ii) for the validity of the two IVs , and , we argue that after controlling for the rich set of observed covariates described below and time invariant household - level unobserved heterogeneity via CRE, these variables should only affect a household s SI category decisions through their effects on the household s receipt of NAIVS vouchers. Moreover , these IVs are exogenous to an individ ual household because district - level election results reflect the decision s of thousands of voters and the regional allocation of NAIVS vouchers is decided by the central government. 3.4 Data and description of variables 3.4.1 Data Our primary data source is the TNPS, which is a three - wave nationally representative household panel survey conducted in 2008/09, 2010/11, and 2012/13. 56 The TNPS was implemented by the Tanzania National Bureau of Statistic s with technical assistance from the World Bank thr ough 56 Data from the fourth wave of the survey (TNPS 2014/15) are now publicly available. However, only 860 households corresponding to 68 clusters were selected from the TNPS 2012/13 sample as part of the 2014/15 first three rounds of the survey in this study. 145 the Living Standards Measurement Study - Integrated Surveys on Agriculture (LSMS - ISA) program. The survey captures information on agricultural production and input use , off - farm income sources , household consumption, socio - economic characteristics , and other topics . A stratified random sampling procedure was employed to select the households in four analytical strata: Dar es Salaam, other urban areas in mainland Tanzania, rural areas in mainland Tanzania, and Zanzibar. Within each stratum, clusters were randomly chosen as the primary sampling units and eight households from each cluster were randomly selected in the last stage. 57 The 2008/09 TNPS consisted of 3,265 households that were clustered in 409 enumeration areas. This original sample of 3,265 house holds and individual members in these households were tracked and re - interviewed in the second (2010/11 TNPS) and third rounds (2012/13 TNPS). The second round tracked 97% of the first round households and the third round tracked 96% of the second round ho useholds ; thus attrition between rounds was very low (Tanzania National Bureau of Statistics, 2014). For the empirical analysis, we exclude the second (2010/11) wave of the TNPS because the questions on NAIVS participation are not comparable to those on the first and third waves. Specifically, the survey instrument in 2010/11 recorded input voucher receipt at the plot level (only if a given input was used) and has no information on whether recipients indeed redeemed the vouchers, while the voucher receipt and redemption information in the other two rounds (2008/09 and 2012/13 TNPS) was directly collected at the household level ( and so it captured all voucher receip t regardless of whether a given input was used) ; the latter data are used to generate the NAIVS variables described above. Our analytical sample involves the balanced panel of maize - growing households interviewed in both TNPS 2008/09 and 2012/13, and their associated 57 In urban areas, the clusters are census enumeration areas based on the 2002 Population and Housing Census ; in rural areas, the clusters are villages. 146 maize plots: 1,624 total household observations (812 observations in each round) and 2,559 total maize plots cultivated by these households (1,225 maize plots in 2008/09 and 1,334 maize plots in 2012/13). In addition, the TNPS data provided by the World Bank include a range of secondary geospatial variables from other sources. Among these, we use in the empirical analysis the rainfall data from the National Oceanic & Atmospheric Administration - Climate Prediction Center (NOAA - CPC) and the soil nu trient availability data from the Harmonized World Soil Database. Other data used in the analysis are: (i) monthly wholesale price data for maize and rice from the Agricultural Market Information System (AMIS) of the Ministry of Industry and Trade (MIT); 58 and (ii) constituency - level data from the 2005 and 2010 presidential elections from the national election commission of Tanzania. 59 3.4.2 Explanatory variables Table 3.5 provides descriptive statistics for the explanatory variables used in the analysis. These variables were selected based on a careful review of the technology adoption literature and the literature on the impacts of ISPs on SFM in other SSA countries (e.g., Pender and Gebremedhin, 2007; Ndiritu et al., 2014; Doss and Morris, 2001 ; de Janvry et al., 1991; Kassie et al., 2013; Kassie et al., 2015a and 2015b; Amsalu and Graaff, 2007; Morgan et al., 201 9 ; Koppmair et al., 2017 ). 58 These prices were collected on a weekly basis from 20 wholesale markets that are matched to regions in Tanzania. Out of 26 regions in the TNPS, there are six regions that are not covered by AMIS . For the wholesale prices in these regions, we use an average price calculated from wholesale markets in adjacent regions. 59 We thank Dr. David Mather for sharing these data. 147 The key explanatory variables of interest in this study are the NAIVS variables. Out of 1,624 household obs ervations during TNPS 2008/09 and 2012/13, 7% of the sample (3% and 10% of the sample households in 2008/09 and 2012/13, respectively) received a NAIVS fertilizer and/or maize seed voucher . By input type of the voucher received, 6% and 2% of the sample households received a NAIVS fertilizer voucher and a NAIVS seed voucher , respectively. 148 Table 3.5: Summary statistics for the variables used in the analysis Variables Variable description Mean Std. dev. Household characteristics 1=yes if the household received a NAIVS voucher for inorganic fertilizer and/or maize seed 0.07 0.25 1=yes if the household received a NAIVS voucher for inorganic fertilizer 0.06 0.24 1=yes if the household received a NAIVS voucher for maize seed 0.02 0.15 Male - Headed HH 1=yes if the household head is male 0.79 0.41 Age of HH head Age of the household head (years) 48.96 15.15 Education of HH head Highest grade completed by the household head (years) 4.74 3.38 Household endowments of physical, human, and social capital Family labor Number of adults (15 - 64 years old) per acre of cultivated land 0.97 1.33 Total cultivated land Total land area cultivated (acres) 6.23 10.41 Off - farm income 1 = yes if the HH earned off - income in the past 12 months 0.43 0.49 Farm assets Total value of farm implements and machinery (1,000 TZS) owned in the past 12 months 1,131.23 5,761.07 Livestock ownership 1 = yes if the HH has livestock (cattle, goats, sheep, pigs, or donkeys) 0.46 0.50 Access to credit 1 = yes if the HH borrowed cash, goods, or services in the past 12 months 0.07 0.25 Membership (SACCOS) 1 = yes if the HH has a member of SACCOS 0.04 0.19 Agricultural extension and access to information and input suppliers Extension from 1 = yes if the HH received agricultural advice from government/NGO in the past 12 months 0.12 0.32 Extension from cooperative 1 = yes if the HH received agricultural advice from cooperative/large scale farmer in the past 12 months 0.04 0.19 Cooperatives within the village 0.46 0.50 Input supplier 1 = yes if improved maize seed supplier present within the village 0.39 0.49 Shocks and other constraints Drought/Flood 1 = yes if the HH was negatively affected by drought or flood in the past two years 0.11 0.31 Crop disease/Pests 1 = yes if the HH was negatively affected by crop diseases or pests in the past two years 0.08 0.28 Rainfall 12 - month total rainfall (mm) in July - June 766.64 270.51 Soil nutrient constraint 1 = yes if soil nutrient availability constraint is moderate or severe 0.62 0.49 149 Table 3.5 Variables Variable description Mean Std. dev. Input and expected output prices Inorganic fertilizer price Inorganic fertilizer price at district level (TZS/kg) 1,141.35 371.39 Real price of maize Average price of maize from Jul. to Sep. in prior year (TZS/100kg bag) 29,941.11 7879.33 Real price of rice Average price of rice from Jul. to Sep. in prior year (TZS/100kg bag) 91,313.88 17,695.48 Bean price Bean market price at region level (TZS/kg) 1281.17 274.05 Groundnut price Groundnut market price at region level (TZS/kg) 1541.44 499.50 Plot characteristics Plot size Plot size (acres) 2.94 5.68 Plot tenure 1 = yes if the HH has title deed for the plot 0.09 0.28 Distance from home Distance from plot to home (km) 3.66 20.16 Distance from main road Distance from plot to main road (km) 2.05 5.11 Distance from market Distance from plot to major market (km) 10.84 14.18 Good soil quality plot is good 0.50 0.50 Poor soil quality plot is poor 0.05 0.22 Flat plot slope is flat 0.64 0.48 Moderate plot slope is slightly sloped 0.32 0.47 Instrumental variables Proportion of votes for the presidential runner - up divided by the proportion of votes for the winner 0.24 0.47 Number of inorganic fertilizer (nitrogen) vouchers distributed to region 52,373.05 42,070.34 Note: The means and standard deviations for plot characteristics are calculated based on the plot level data (n=2,559), whereas the means and standard deviations for the other control variables are calculated based on the balanced household - level data (n=1,624). This study controls for household - level heterogeneity by including characteristics of the household head such as his/her age, gender, and education level which are relevant variables that may influence decision - making processes within the household. That is, use of modern 150 inputs and management practices may differ across households depending on the characteristics of the household head as a main decision - maker. For example, more educated farmers may be more awar e of the benefits from the use of each SFM practice (or combined use thereof ), and thus they may be more likely to purchase inputs or adopt agricultural practices that could have the potential to improve crop yields (Pender and Gebremedhin, 2007). Moreover, there may exist gender differences in adoption strategies for the SFM prac tices since female farmers often have less access to things like land, labor, credit, education, and information (Ndiritu et al., 2014; Doss and Morris, 2001). In the context of imperfect or missing markets for land and labor, a household endowme nts (physical, human, and social) , represented by total cultivated land, off - farm income, farm assets, livestock ownership, family labor, access to credit, and membership in Savings and Credits Cooperatives Societies ( SACCOS) in this study, may significant decision to use external inputs and SFM practices (de Janvry et al., 1991; Pender and Gebremedhin, 2007). Households with greater physical assets and social capital generally have more savings and better access to credit which would he lp them to finance the purchase of inputs such as inorganic fertilizer and improved seeds (Kassie et al., 2013). Livestock ownership could also facilitate use of organic fertilizer because animal manure is one of the major sources of organic fertilizer and it can rarely be purchased from the market. In addition, family labor availability, defined here as the number of adults aged 15 to 64 within the household per acre of total cultivated land, could be an important determinant of household use choices among the SFM practices. For example, particularly in the context of missing or imperfect labor markets, greater availability of family labor could enable households to choose relatively labor - intensive 151 practices (e.g., maize - legume intercropping o r organic fertilizer, or both) rather than investing in inorganic fertilizer only. Agricultural extension services are a key channel to promote the use of modern inputs and improved management practices (Pender and Gebremedhin, 2007; Kassie et al., 2015a) . We thus include two dummy variables associated with agricultural extension services depending on the organizations: i) one is a variable equal to one if the household received agricultural extension advice from government or an NGO in the past 12 months; and ii) the other equals one or large - scale farmer within the community could provide farmers with better access to information about or better physical access to farm inputs. Thus, this study includes dummy variables for the existence of a es for access to information and agricultural inputs. Given that African farmers are often vulnerable to weather shocks and crop pest/disease outbreaks , which could affect their use o f SFM practices in subsequent seasons, we also control for the following two binary variables (following Kassie et al., 2015b): i) drought/flood which equals one if the household was negatively affected by a drought or flood during the past two years; and ii) crop diseases/pests which equals one if the household was negatively affected by crop diseases or pests in the past two years. We also control for two geospatial variables: i) 12 - month soil nutrient availability constraint which equals one if soil nutrie moderate, severe, or very severe (with the base category being no or slight constraint). 60 60 According to the Harmonized World Soil Database, soil nutrient availability i s one of the key soil qualities for crop production (where maize is used as the reference crop). It is measured based on important characteristics (i.e., 152 Input and expected output prices could also be key factors when the household makes decisions to use inputs and agricultural management practices on their maize plots. In particular, there is a significant gap between the prices of leguminous crops and maize in Tanzania, and thus use of SFM practic e s may vary depending on the (expected) prices of these crops. (Output prices at harvest are not known at planting time.) For the price of maize, we assume naïve price expectations i.e., that the expected harvest price of the crop equals the observed mar ket price in the previous year. Given that the MIT collects wholesale prices throughout the year for maize, we calculate the average real wholesale price per 100 kg bag from the nearest wholesale market during the post - harvest period (i.e., from July throu price. The data available on legume prices are more limited. Due to these data limitations, we utilize the price info rmation available in the TNPS and include the average prices of beans and groundnuts per kilogram at region level as a proxy for the expected prices of legume crops (i.e., we assum e perfect foresight ) . We also control for the average price of inorganic fer tilizer per kilogram at district level as the major relevant input price in this study. 61 Plot - specific attributes such as plot size, plot tenure status, distance from the plot to the soil quality and slope of the plot are also included in our model. Per previous studies (Amsalu and Graaff, 2007; Kassie et al., 2013; Kassie et al., 2015a), these plot characteristics are often important determinants of the use of soil conservation and SFM practices in eastern and southern Africa including Tanzania. soil texture, soil organic carbon, soil pH, total exchangeable bases) of the top soil (0 - 30 cm) and th e subsoil (30 - 100 cm). Moderate, severe, and very severe constraints are generally rated between 60% and 80%, between 40% and 60%, and less than 40% of the growth potential, respectively. 61 No data are available on maize seed prices at district level. 153 3.5 Results 3.5.1 Test for endogeneity of household receipt of NAIVS voucher The parameter estimates from the CRE - MNL regression models with CF approach are reported in Appendix Table 3A. 3 . Two sets of estimated coefficients are presented based on different NAIVS variables: i) NAIVS variable for receipt of any input voucher ( , column 1) ; and ii) two NAIVS variables by input types ( and , column 2). We find that the generalized residuals from the CF first - stage CRE logit models in both model specifications are not statistically significant, implying we fail to reject the exogeneity of the NAIVS variables considered in this study. 62 Similar results hold if NAIVS variables and residuals based on redemption (instead of receipt) are used. Thus, in the remainder of this study, we focus on the results of CRE - MNL models that exclude the CF residuals. Parameter estimates for these models are reported in Table 3A. 4 in the Appendix. These coefficients are the log - control variable relative to the reference SI categor - variables constant. To reach conclusions based on actual probabilities, we need to calculate average partial effects (APEs). We report and discuss these APEs below. 3.5.2 APEs of NAIVS voucher receipt on household us e of practices in the various SI categories Table 3.6 shows the APEs of household receipt of NAIVS vouchers on household practices in various SI categories by input voucher type received (Panel s A and B , column 1). As 62 T he p - values on the generalized residuals for , , and are 0.449, 0.498, and 0.430, respectively. Mather and Minde (2016), who used the electoral threat IV for household quantity of NAIVS fertilize r, also fail to reject exogeneity of NAIVS fertilizer quantity received. 154 noted in Section 3. 3 .2, becau se some farmers did not redeem their vouchers, we also report the s A and B , column 2). Table 3.6: APEs of NAIVS voucher receipt and redemption on household use of practices in the various SI categories NAIVS voucher receipt NAIVS voucher redemption Variables N I S SI N I S SI Panel A NAIVS for any input - 0.214*** 0.096*** 0.027 0.091*** - 0.212*** 0.100*** 0.008 0.104*** (0.048) (0.015) (0.048) (0.016) (0.056) (0.016) (0.052) (0.015) Panel B NAIVS for inorganic fertilizer - 0.251*** 0.100*** 0.055 0.096*** - 0.219*** 0.099*** 0.020 0.101*** (0.059) (0.017) (0.056) (0.016) (0.065) (0.018) (0.060) (0.016) NAIVS for maize seed - 0.034 0.011 0.023 - 0.001 - 0.102 0.039 0.025 0.038 (0.074) (0.026) (0.074) (0.032) (0.097) (0.025) (0.089) (0.030) indicates that the corresponding coefficients are significant at the 10%, 5%, and 1% levels, respectively. There are three main empirical findings drawn from Table 3.6. First, based on the results from Panel A in column 1, we find that receipt of a NAIVS voucher for any input (i.e., inorganic fertilizer with other SF receipt of a NAIVS voucher is associated with a 9.6 percentage point average increase in the crease credit considered as major constraints that farmers in SSA face, this significant positive effect on household inorganic fertilizer use is entirely reaso nable. This is consistent with findings in Mather and Minde (2016) that household receipt of one NAIVS fertilizer voucher (50kg of 155 subsidized fertilizer) increase s 4.0 percentage point s , on average group are as follows. First, for households who originally considered using inorganic fertilizer only on their maize plot, the subsidized NAIVS voucher for inorganic fertilizer and/o r maize seed could free up their resources to invest in other inputs (e.g., legume seeds or organic fertilizers in our study) that facilitate joint use of these practices with inorganic fertilizer. Second, for households who initially planned use of organi c fertilizer and/or maize - legume intercropping but not inorganic fertilizer, a NAIVS voucher, especially a voucher for inorganic fertilizer, could be a great incentive to or make it possible for the household to jointly use these SFM practices. The positiv e effect of receipt of a NAIVS voucher on the use of practices in the maize plots is an encouraging result, as it could suggest that NAIVS stimulated ISFM and could improve soil health of the associated maize plots as well as maize yields and yield response to inorganic fertilizer in the long term. On the other hand, we find no statistically significant effects of NAIVS voucher receipt on the use The second main finding based on Table 3.6 is that the statistically significant positive appear to be mainly driven (as expected) by receipt of a voucher for inorganic fertilizer as opposed to receipt of a vouch er for maize seed. In particular, note that based on the results in Panel B, the APEs of the NAIVS inorganic fertilizer voucher are positive and statistically e NAIVS maize seed voucher is not statistically different from zero. However, no significant effects of the NAIVS maize seed voucher may be explained by the very small proportion of sample households that received it. That is, there may indeed be an impact of maize seed voucher 156 receipt, but such an impact may not be detected unless it is very large due to low statistical power. The third main finding based on Table 3.6 is that the estimated effects of NAIVS on the use of practices in various SI categor ies are very similar in sign, significance, and magnitude in the results with voucher receipt (column 1) versus voucher redemption (column 2). This finding is perhaps not that surprising given that overall 87% of household beneficiaries who received at least one NAIVS voucher indeed redeem ed it (Table 3.3). Nevertheless, it shows that our results are robust to alternative definitions To further explore the above findings, we conduct additional analyses to unpack how NAIVS voucher receipt affects the use group. To do this, we consider two sets of c ategorizations focusing on the use of at inorganic fertilizer and at least one of the Sustainable practices on a given maize plot, respectively: i) four categories based on the combinations of inorganic fertilizer and maize - legume intercropping irrespectiv e of the use of organic fertilizer (Table 3.7), and ii) four categories based on the combinations of inorganic fertilizer and organic fertilizer irrespective of the use of maize - legume intercropping (Table 3.8). The APEs of these categorizations in CRE - MNL models are reported in Tables 3.7 and 3.8, respectively. The results suggest that household receipt of a NAIVS using each of the inorganic fertilizer plus Sustainable practice combinations included in the , although the effect on the probability of joint use of inorganic fertilizer and at least maize - legume intercropping is larger in magnitude . More specifically, household receipt of a NAIVS voucher for any input is associated with an 7.5 percentage point average increase in the probability of joint use of inorganic fertilizer and at least maize - legume intercropping on a given maize plot and a 2.9 157 percentage point average increase in the probability of joint use of inorganic fertili zer and at least organic fertilizer. Tables 3.7 and 3.8 also show that the three main findings drawn in Table 3.6 are largely upheld. 63 Table 3.7: APEs of NAIVS voucher receipt and redemption on household sole or joint use of inorganic fertilizer and maize - legume intercropping NAIVS voucher receipt NAIVS voucher redemption Variables None Inorganic fertilizer only Maize - legume IC only Both None Inorganic fertilizer only Maize - legume IC only Both Panel A NAIVS for any - 0.180*** 0.113*** - 0.009 0.075*** - 0.208*** 0.122*** 0.001 0.084*** input (0.049) (0.016) (0.048) (0.014) (0.058) (0.017) (0.055) (0.014) Panel B NAIVS for - 0.188*** 0.120*** - 0.007 0.075*** - 0.196*** 0.123*** - 0.005 0.078*** inorganic fertilizer (0.060) (0.018) (0.057) (0.015) (0.066) (0.018) (0.063) (0.015) NAIVS for maize - 0.016 0.001 0.006 0.009 - 0.065 0.036 - 0.012 0.040 seed (0.077) (0.030) (0.075) (0.028) (0.105) (0.028) (0.099) (0.027) Notes: *, **, and *** indicates that the corresponding coefficients are significant at the 10%, 5%, and 1% levels, respectively. 63 In addition to these main findings, APEs of other factors influencing the use of practices in various SI categories are presented in Appendix Table 3A. 5. 158 Table 3.8: APEs of NAIVS voucher receipt and redemption on household sole or joint use of inorganic fertilizer and organic fertilizer NAIVS voucher receipt NAIVS voucher redemption Variables None Inorganic fertilizer only Organic fertilizer only Both None Inorganic fertilizer only Organic fertilizer only Both Panel A NAIVS for any - 0.196*** 0.157*** 0.009 0.029*** - 0.201*** 0.169*** - 0.002 0.034*** input (0.035) (0.021) (0.028) (0.010) (0.039) (0.022) (0.030) (0.009) Panel B NAIVS for - 0.225*** 0.160*** 0.032 0.033*** - 0.212*** 0.162*** 0.015 0.035*** inorganic fertilizer (0.041) (0.023) (0.033) (0.011) (0.043) (0.024) (0.033) (0.011) NAIVS for maize - 0.014 0.018 - 0.004 - 0.000 - 0.101 0.073* 0.012 0.015 seed (0.063) (0.040) (0.042) (0.018) (0.105) (0.028) (0.099) (0.027) Notes: *, **, and *** indicates that the corresponding coefficients are significant at the 10%, 5%, and 1% levels, respectively. These findings are new and important considering that previous studies (Holden and Lunduka, 2012; Koppmair et al., 2017; Morgan et al., 2019; Kassie et al., 2015a) hav e typically found evidence of no significant effects or negative effects of fertilizer subsidies in Malawi and Zambia on the use of SFM practices specifically organic manure, intercropping maize with other crops, ridges, terraces and stone bunds, and fallowing when considered individually . This may be similar with our findings of no significant effects of NAIVS voucher receipt on use of the evidence in our study suggests significant positive subsidy program effects on inorganic fertilizer use only as well as joint use of inorganic fertilizer with other SFM practices someth ing that is not explicitly investigated in previous studies. a key limitation of the study is that although NAIVS beneficiaries were to receive input vouchers for three consecutive years , our data only capture 159 one year of participation in the NAIVS program. Hence, our findings should be considered as the immediate or short - rather than the long - run effects of their full participation in the program. F uture research using alternative data sources (if available) could seek to address this limitation. 3.6 Conclusions and policy implications In many African countries, government policies through large - scale ISPs have primarily focused on conventional intensification of agricultural systems involving the use of inorganic fertili zer and high - yielding crop varieties. Yet there is an emerging consensus that these conventional means are unlikely to be sufficient to sustainably intensify agricultural production. Despite heavy spending on ISPs in SSA, the productivity and welfare effec ts of these programs have , in many cases, been considerably smaller than expected (Jayne et al., 2018) . One of the major reasons for soil quality ( Ibid. ). Given this limited effect of ISPs, it is increasingly apparent that use of complementary SFM practices along with inorganic fertilizer is needed to improve the agronomic efficiency of inorganic fertilizer use as well as the effectiveness of ISPs (Holden , 2018; Jayne et al., 2018). However, no previous studies have investigated the effects of an African ISP on joint use of inorganic fertilizer with other SFM practices. Using nationally representative household panel survey data from Tanzania, this study e stimates the effects of household receipt of vouchers for inorganic fertilizer and/or maize seed - MNL models suggest that receipt of a NAIVS voucher for any input (i.e. , inorganic fertilizer, improved maize seed, or both) is associated with increases in maize - 160 probability of using inorganic fertilizer only ( referred to as well as joint use of inorganic fertilizer with organic fert ilizer and/or maize - legume intercropping ( referred to as a voucher for inorganic fertilizer as opposed to receipt of a voucher for improved maize seed. No statistically significant NAIVS effects are found for the practices in the organic fertilizer use only, maize - legume intercropping use only, or both). These findings are also robust to a thermore, we find that household fertilizer and inorganic fertilizer with maize - legume intercropping , with the latter effect found to be larger in magnitude sole use of inorganic fertilizer , b ut more importantly, that the program also incentivized which could raise inorganic fertilizer use efficiency as well as contribute to SI goals. The results have seve ISPs. 64 First, our main findings demonstrate that NAIVS increased fertilizer only as well as joint use of inorganic fertilizer with organic fertilizer and/or maize - legume intercropping as sustainable forms of agricultural intensification . Although further research is needed, these positive effects could be explained by its more private sector - friendly design and more effective targeting criteria and impl ementation . Compared to other SSA program was designed to target relatively resource poor households who have limited experience in using modern inputs and the majority of voucher recipients met 64 Although the NAIVS program officially ended in 2014 , a similar ISPs was implemented in 2015/16 and it is possible that a similar program will be re - introduced in Tanzania in the future. 161 these criteria (Mather and Minde, 2016) . 65 In addition, our data also show that most of voucher recipients redeemed their voucher(s) at local agro - dealerships . positive effects on the sole use of inorganic fertilizer and joint use of it with other SFM practices may imply that subsidy achiev ing the goals of ISPs and stimulating SI . In addition, because most NAIVS beneficiaries prior to NAIVS had very limited experience with using inorganic fertilizer (unlike many subsidized fertilizer recipients in Malawi and Zambia see Ricker - Gilbert et al. (2011) and Jayne et al. (2013)) and relied mainly on organic sources of soil fertility, they may consider inorganic fertilizer to be a complement to rather than a substitute for practices like use of organic fertilizer and maize - legume intercropping. Therefore, the receipt of a NAIVS voucher for inorganic fertilizer may have encourage d households combined use of i norganic fertilizer with these other practices . The second policy implication is related to the fact that approximately 38% of the maize plots in rural Tanzania involve d maize - legume intercropping (Table 3.1) but this use rate is still far from universal and much lower relative to other countr ies in the reg ion such as Kenya (Kassie et al., 2015a). Given this, promoting wider adoption of legume intercropping with maize through including a legume seed subsidy in the ISP may be a country - specific strategy to incentivize joint use of inorganic fertilizer with maize - legume intercropping as an SI strategy . However, further research is needed to identify if this policy shift would be a cost - effective means of promoting SI of maize production in Tanzania. 65 In contrast, i n Malawi and Zambia, households with greater land and asset wealth receiv ed more subsidized already using commercially - priced inorganic fertilizer on maize a few years before the programs started (Mather and Mind e, 2016). 162 APPENDIX 163 Table 3A.1: Reduced form CRE logit regression estimates of factors affecting househol d NAIVS voucher receipt CRE logit (1) CRE logit (2) CRE logit (3) Variables =1 if the household received a NAIVS voucher for inorganic fertilizer and/or maize seed =1 if the household received a NAIVS voucher for inorganic fertilizer =1 if the household received a NAIVS voucher for maize seed 0.734*** 0.357 1.042*** (0.220) (0.632) (0.398) 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Male - Headed HH 0.422 0.380 - 0.861 (0.411) (0.443) (0.546) Age of HH head 0.029 0.097** - 0.032 (0.057) (0.043) (0.080) Education of HH head 0.131** 0.160*** 0.133* (0.052) (0.053) (0.074) Family labor 0.482* 0.506* - 0.058 (0.249) (0.270) (0.368) Total cultivated land 0.382*** 0.374*** 0.277*** (0.077) (0.082) (0.076) Off - farm income 0.317 - 0.011 1.223* (0.399) (0.411) (0.634) Farm assets 0.000 0.000 - 0.000 (0.000) (0.000) (0.000) Livestock ownership 0.977*** 1.137*** 0.225 (0.354) (0.376) (0.495) Access to credit - 1.514* - 1.341 - 1.596 (0.831) (0.895) (1.056) Extension from 0.472 0.153 - 0.317 (0.519) (0.531) (0.737) Extension from 0.349 0.218 0.302 cooperative (0.751) (0.806) (1.107) Cooperative 0.102 0.178 - 1.257** (0.335) (0.362) (0.623) Input supplier 0.890*** 1.031*** 1.636*** (0.323) (0.344) (0.557) Drought/Flood - 1.262* - 1.073 - 1.223 (0.665) (0.793) (0.951) Crop disease/Pests 0.299 0.705 - 0.508 (0.802) (0.974) (1.330) Rainfall - 0.003* - 0.004* - 0.001 (0.002) (0.002) (0.003) Soil nutrient constraint 3.973*** 3.581** 1.881 (1.508) (1.762) (1.381) 164 Table 3A.1 CRE logit (1) CRE logit (2) CRE logit (3) Variables =1 if the household received a NAIVS voucher for inorganic fertilizer and/or maize seed =1 if the household received a NAIVS voucher for inorganic fertilizer =1 if the household received a NAIVS voucher for maize seed Inorganic fertilizer price - 0.001* - 0.002* - 0.001 (0.001) (0.001) (0.001) Real price of maize 0.000** 0.000* - 0.000 (0.000) (0.000) (0.000) Real price of rice - 0.000 - 0.000 - 0.000 (0.000) (0.000) (0.000) Bean price 0.001 0.001 0.002 (0.001) (0.001) (0.004) Groundnut price 0.001 0.001 0.002 (0.001) (0.001) (0.003) Plot size - 0.045 - 0.036 - 0.053 (0.031) (0.034) (0.035) Plot tenure - 0.209 - 0.147 - 1.370** (0.412) (0.443) (0.638) Distance from home 0.009* 0.012** - 0.005*** (0.005) (0.005) (0.002) Distance from main road - 0.076 - 0.072 - 0.269 (0.059) (0.060) (0.165) Distance from market - 0.005 - 0.010 0.011 (0.009) (0.010) (0.011) Good soil quality 0.215 0.328 - 0.034 (0.236) (0.251) (0.302) Poor soil quality 0.025 0.235 - 0.861 (0.463) (0.430) (0.939) Flat plot slope 0.120 - 0.010 - 0.483 (0.432) (0.444) (0.599) Moderate plot slope 0.046 - 0.011 - 0.743 (0.440) (0.456) (0.676) Constant - 21.132*** - 17.583*** - 32.942*** (4.230) (4.016) (9.691) Joint significance of IVs 41.52*** 43.72*** 16.20*** Pseudo R - squared 0.426 0.445 0.442 Observations 2,599 2,599 2,599 Notes: *, **, and *** indicates that the corresponding coefficients are significant at the 10%, 5%, and 1% levels, respectively. Time - averages of household characteristics to control for time invariant unobserved heterogeneity were included in the model but not reported in Table 3A.1. Robust standard errors clustered at the household level are in parentheses. 165 Table 3A. 2 : Reduced form CRE logit regression estimates of factors affecting household NAIVS voucher redemption CRE logit (1) CRE logit (2) CRE logit (3) Variables =1 if the household redeemed a NAIVS voucher for inorganic fertilizer and/or maize seed =1 if the household redeemed a NAIVS voucher for inorganic fertilizer =1 if the household redeemed a NAIVS voucher for maize seed 0.778*** 0.210 1.125** (0.220) (0.747) (0.441) 0.000*** 0.000*** 0.000*** (0.000) (0.000) (0.000) Male - Headed HH 0.572 0.600 - 0.529 (0.421) (0.463) (0.613) Age of HH head 0.039 0.121*** 0.018 (0.063) (0.040) (0.077) Education of HH head 0.133*** 0.154*** 0.162** (0.049) (0.051) (0.074) Family labor 0.459** 0.477** 0.019 (0.209) (0.214) (0.294) Total cultivated land 0.336*** 0.345*** 0.191** (0.072) (0.078) (0.087) Off - farm income 0.286 0.196 1.285* (0.409) (0.435) (0.718) Farm assets 0.000 0.000 - 0.000* (0.000) (0.000) (0.000) Livestock ownership 1.144*** 1.215*** 0.398 (0.389) (0.414) (0.611) Access to credit - 1.987* - 1.916 - 3.331*** (1.076) (1.254) (1.287) Extension from 0.167 0.043 - 1.003 (0.552) (0.572) (0.741) Extension from 0.363 0.265 0.814 cooperative (0.751) (0.824) (0.939) Cooperative 0.247 0.277 - 0.809 (0.340) (0.371) (0.688) Input supplier 0.899*** 1.048*** 1.518*** (0.328) (0.358) (0.580) Drought/Flood - 0.474 - 0.601 0.424 (0.638) (0.785) (0.881) Crop disease/Pests 0.593 1.148 0.149 (0.826) (1.040) (1.311) Rainfall - 0.003 - 0.003 0.002 (0.002) (0.002) (0.004) Soil nutrient constraint 3.693** 3.626* 3.396* (1.575) (1.964) (1.760) 166 Table 3A. 2 CRE logit (1) CRE logit (2) CRE logit (3) Variables =1 if the household redeemed a NAIVS voucher for inorganic fertilizer and/or maize seed =1 if the household redeemed a NAIVS voucher for inorganic fertilizer =1 if the household redeemed a NAIVS voucher for maize seed Inorganic fertilizer price - 0.001 - 0.002 - 0.001 (0.001) (0.001) (0.002) Real price of maize 0.000** 0.000* 0.000 (0.000) (0.000) (0.000) Real price of rice - 0.000 0.000 - 0.000 (0.000) (0.000) (0.000) Bean price 0.001 0.001 0.004 (0.002) (0.002) (0.005) Groundnut price 0.001 0.000 0.003 (0.001) (0.001) (0.004) Plot size - 0.030 - 0.031 - 0.022 (0.029) (0.033) (0.035) Plot tenure - 0.295 - 0.113 - 2.482** (0.461) (0.461) (1.045) Distance from home 0.010** 0.014*** - 0.003 (0.005) (0.005) (0.003) Distance from main road - 0.124** - 0.136** - 0.176 (0.060) (0.061) (0.150) Distance from market - 0.013 - 0.012 - 0.020 (0.012) (0.012) (0.014) Good soil quality 0.033 0.162 - 0.430 (0.255) (0.273) (0.355) Poor soil quality 0.136 0.338 - 0.265 (0.433) (0.419) (0.897) Flat plot slope 0.334 0.239 - 0.087 (0.469) (0.462) (0.663) Moderate plot slope 0.251 0.211 - 0.628 (0.487) (0.491) (0.698) Constant - 21.672*** - 18.865*** - 35.252*** (4.836) (4.623) (12.584) Joint significance of IVs 38. 74*** 36.46*** 10.19*** Pseudo R - squared 0.436 0.462 0.484 Observations 2,559 2,559 2,559 Notes: *, **, and *** indicates that the corresponding coefficients are significant at the 10%, 5%, and 1% levels, respectively. Time - averages of household characteristics to control for time invariant unobserved heterogeneity were included in the model but not reported in Tabl e 3A. 2 . Robust standard errors clustered at the household level are in parentheses. 167 Table 3A. 3 : CRE - MNL with CF regression results (relative log odds) CRE - MNL with CF (1) CRE - MNL with CF (2) Variables I S SI I S SI Male - Headed HH - 0.208 - 0.039 - 0.480 - 0.167 - 0.040 - 0.432 (0.270) (0.149) (0.339) (0.267) (0.150) (0.342) Age of HH head - 0.028 - 0.001 - 0.014 - 0.034 - 0.003 - 0.025 (0.033) (0.016) (0.025) (0.034) (0.016) (0.024) Education of HH head 0.141*** 0.019 0.112** 0.141*** 0.017 0.109** (0.048) (0.018) (0.049) (0.047) (0.018) (0.047) Family labor - 0.163 0.117** - 0.011 - 0.146 0.110** - 0.013 (0.149) (0.051) (0.178) (0.131) (0.051) (0.179) Total cultivated land - 0.104** - 0.035** - 0.112* - 0.099** - 0.035** - 0.110* (0.048) (0.017) (0.060) (0.049) (0.017) (0.059) Off - farm income - 0.100 0.111 - 0.533* - 0.086 0.133 - 0.495 (0.321) (0.167) (0.317) (0.327) (0.170) (0.303) Farm assets - 0.000 - 0.000 0.000 - 0.000 - 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Livestock ownership 0.333 0.355*** 0.802*** 0.360 0.345*** 0.791*** (0.274) (0.125) (0.300) (0.283) (0.126) (0.295) Access to credit 0.670 0.042 1.130** 0.670 0.038 1.096** (0.457) (0.214) (0.452) (0.435) (0.209) (0.445) Extension from 0.585 - 0.059 0.517 0.664 - 0.054 0.574 (0.411) (0.186) (0.399) (0.425) (0.187) (0.392) Extension from 0.380 0.408 0.943** 0.422 0.432 0.994** cooperative (0.624) (0.499) (0.435) (0.633) (0.511) (0.462) Cooperative 0.408 0.038 0.200 0.432 0.030 0.208 (0.420) (0.149) (0.403) (0.427) (0.146) (0.405) Input supplier 0.147 - 0.042 0.300 0.136 - 0.053 0.269 (0.347) (0.130) (0.354) (0.352) (0.132) (0.359) Drought/Flood 0.762 - 0.221 0.568 0.843 - 0.238 0.540 (0.682) (0.210) (0.443) (0.731) (0.213) (0.463) Crop disease/Pests 0.248 - 0.071 0.167 0.237 - 0.101 0.118 (0.584) (0.266) (0.756) (0.586) (0.266) (0.766) Rainfall - 0.001 0.000 - 0.001 - 0.001 0.000 - 0.001 (0.002) (0.001) (0.001) (0.002) (0.001) (0.001) Soil nutrient constraint - 0.925 - 2.236** - 1.843 - 0.937 - 2.222** - 1.792 (1.547) (1.078) (1.297) (1.597) (1.090) (1.320) Inorganic fertilizer price 0.002*** 0.000 0.001* 0.002*** 0.000 0.001* (0.001) (0.000) (0.001) (0.001) (0.000) (0.001) Real price of maize - 0.000 - 0.000** - 0.000 - 0.000 - 0.000** - 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Real price of rice 0.000 - 0.000 0.000 0.000 - 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Bean price - 0.000 - 0.001 - 0.001 0.000 - 0.001 - 0.001 (0.001) (0.001) (0.001) (0.001) (0.000) (0.001) Groundnut price - 0.000 0.001** 0.000 - 0.001 0.001** 0.000 (0.001) (0.000) (0.001) (0.001) (0.000) (0.001) 168 Table 3A. 3 CRE - MNL with CF (1) CRE - MNL with CF (2) Variables I S SI I S SI Plot size 0.095** 0.043** 0.125** 0.095** 0.043** 0.124** (0.046) (0.018) (0.054) (0.046) (0.018) (0.054) Plot tenure - 0.067 0.188 0.524 - 0.037 0.183 0.542 (0.579) (0.211) (0.495) (0.591) (0.209) (0.509) Distance from home - 0.016 - 0.046*** - 0.007 - 0.016 - 0.047*** - 0.007 (0.012) (0.015) (0.016) (0.012) (0.015) (0.015) Distance from main road - 0.041 0.021 - 0.048 - 0.038 0.022 - 0.048 (0.035) (0.018) (0.044) (0.035) (0.018) (0.044) Distance from market 0.001 0.004 - 0.010 - 0.000 0.005 - 0.010 (0.010) (0.005) (0.011) (0.010) (0.006) (0.011) Good soil quality 0.048 - 0.241** - 0.323 0.037 - 0.242** - 0.346 (0.228) (0.120) (0.216) (0.233) (0.121) (0.214) Poor soil quality - 0.685 - 0.167 0.252 - 0.678 - 0.180 0.216 (2.737) (0.241) (0.464) (2.644) (0.242) (0.471) Flat plot slope 1.797 0.065 0.189 1.844 0.074 0.241 (4.815) (0.310) (0.688) (4.816) (0.311) (0.676) Moderate plot slope 1.714 0.241 0.286 1.755 0.246 0.341 (4.727) (0.313) (0.623) (4.728) (0.314) (0.617) NAIVS for any input 2.912** 0.846 3.056*** (1.316) (0.729) (1.183) CRE logit residuals - 0.581 - 0.324 - 0.883 (Any input) (1.343) (0.777) (1.166) NAIVS for inorganic 2.595* 1.394 3.147*** fertilizer (1.405) (0.860) (1.165) NAIVS for maize seed 2.103 - 0.162 1.328 (1.654) (1.113) (1.632) CRE logit residuals 0.004 - 0.760 - 0.778 (Inorganic fertilizer) (1.395) (1.014) (1.148) CRE logit residuals - 2.248 0.351 - 1.440 (Improved maize seed) (1.747) (1.116) (1.826) Notes: Bootstrapped standard are in parentheses. To control for time - invariant unobserved household heterogeneity, time - averages of household characteristics were included in the model but not reported in Table 3A. 3 - ** indicates that the corresponding coefficients are significant at the 10%, 5%, and 1% levels, respectively. 169 Table 3A. 4 : CRE - MNL with out CF regression results (relative log odds) Voucher receipt (1) Voucher receipt (2) Voucher redemption (1) Voucher redemption (2) Variables I S SI I S SI I S SI I S SI Male - Headed HH - 0.164 0.048 - 0.407 - 0.158 0.051 - 0.399 - 0.191 0.047 - 0.444 - 0.190 0.044 - 0.435 (0.335) (0.143) (0.300) (0.338) (0.143) (0.302) (0.336) (0.143) (0.297) (0.338) (0.143) (0.299) Age of HH head - 0.037 0.004 - 0.017 - 0.048* 0.003 - 0.027 - 0.037 0.003 - 0.018 - 0.047 0.004 - 0.028 (0.032) (0.015) (0.030) (0.029) (0.015) (0.027) (0.033) (0.015) (0.031) (0.031) (0.015) (0.028) Education of HH head 0.126*** 0.002 0.099** 0.121*** 0.001 0.098** 0.129*** 0.002 0.103** 0.124*** 0.001 0.102** (0.040) (0.019) (0.041) (0.040) (0.019) (0.041) (0.039) (0.019) (0.040) (0.040) (0.019) (0.040) Family labor - 0.123 0.147*** 0.024 - 0.134 0.146*** 0.017 - 0.138 0.146*** 0.000 - 0.145 0.146*** - 0.000 (0.146) (0.055) (0.131) (0.146) (0.055) (0.132) (0.150) (0.054) (0.134) (0.149) (0.055) (0.133) Total cultivated land - 0.101** - 0.028* - 0.124*** - 0.098** - 0.028* - 0.121*** - 0.104*** - 0.027* - 0.130*** - 0.107*** - 0.027* - 0.131*** (0.040) (0.015) (0.045) (0.040) (0.015) (0.045) (0.040) (0.014) (0.045) (0.041) (0.014) (0.044) Off - farm income - 0.071 0.100 - 0.466* - 0.029 0.112 - 0.428 - 0.099 0.105 - 0.493* - 0.074 0.107 - 0.465* (0.291) (0.143) (0.263) (0.294) (0.142) (0.261) (0.291) (0.142) (0.265) (0.291) (0.142) (0.262) Farm assets - 0.000 - 0.000* 0.000 - 0.000 - 0.000* 0.000 - 0.000 - 0.000* 0.000 - 0.000 - 0.000* 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Livestock ownership 0.429* 0.462*** 0.925*** 0.423* 0.461*** 0.927*** 0.432* 0.462*** 0.910*** 0.430* 0.463*** 0.908*** (0.255) (0.123) (0.234) (0.254) (0.123) (0.233) (0.252) (0.123) (0.233) (0.249) (0.123) (0.232) Access to credit 0.740* 0.007 1.142*** 0.713* 0.004 1.097*** 0.761* - 0.003 1.203*** 0.783** - 0.001 1.201*** (0.411) (0.234) (0.423) (0.413) (0.234) (0.421) (0.398) (0.233) (0.392) (0.395) (0.233) (0.390) Extension from 0.657* - 0.091 0.653* 0.697** - 0.084 0.679** 0.695** - 0.070 0.673** 0.749** - 0.073 0.719** (0.336) (0.196) (0.337) (0.342) (0.197) (0.343) (0.334) (0.194) (0.337) (0.335) (0.194) (0.342) Extension from 0.508 0.561 0.999** 0.551 0.572 1.040** 0.515 0.561 1.009** 0.539 0.559 1.042** cooperative (0.534) (0.422) (0.417) (0.546) (0.425) (0.419) (0.546) (0.422) (0.425) (0.555) (0.423) (0.436) Cooperative 0.467* 0.144 0.324 0.465* 0.144 0.315 0.467* 0.142 0.331 0.465* 0.143 0.331 (0.253) (0.114) (0.216) (0.251) (0.115) (0.218) (0.254) (0.114) (0.219) (0.249) (0.114) (0.220) Input supplier 0.189 - 0.051 0.336 0.163 - 0.057 0.321 0.190 - 0.046 0.321 0.156 - 0.048 0.300 (0.223) (0.107) (0.226) (0.227) (0.107) (0.226) (0.223) (0.107) (0.225) (0.227) (0.107) (0.226) Drought/Flood 0.643 - 0.234 0.473 0.628 - 0.239 0.439 0.543 - 0.260 0.396 0.526 - 0.261 0.360 (0.458) (0.207) (0.413) (0.466) (0.208) (0.418) (0.450) (0.205) (0.396) (0.455) (0.205) (0.396) Crop disease/Pests 0.204 - 0.166 0.235 0.105 - 0.180 0.163 0.165 - 0.161 0.177 0.058 - 0.167 0.076 (0.510) (0.261) (0.511) (0.502) (0.261) (0.504) (0.509) (0.261) (0.515) (0.505) (0.261) (0.504) Rainfall - 0.001 0.000 - 0.001 - 0.001 0.000 - 0.001 - 0.001 0.000 - 0.001 - 0.001 0.000 - 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) 170 Table 3A. 4 Voucher receipt (1) Voucher receipt (2) Voucher redemption (1) Voucher redemption (2) Variables I S SI I S SI I S SI I S SI Soil nutrient constraint - 1.081 - 2.258*** - 1.733** - 1.095 - 2.256*** - 1.729** - 1.096 - 2.264*** - 1.734** - 1.080 - 2.266*** - 1.713** (0.926) (0.708) (0.791) (0.918) (0.711) (0.788) (0.928) (0.707) (0.786) (0.918) (0.708) (0.782) Inorganic fertilizer price 0.002*** 0.000 0.001** 0.002*** 0.000 0.001** 0.002*** 0.000 0.001** 0.002*** 0.000 0.001** (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) Real price of maize - 0.000 - 0.000** - 0.000 - 0.000 - 0.000** - 0.000 - 0.000 - 0.000** - 0.000 - 0.000 - 0.000** - 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Real price of rice 0.000** - 0.000 0.000 0.000** - 0.000 0.000 0.000** - 0.000 0.000 0.000** - 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Bean price - 0.000 - 0.000 - 0.001 - 0.000 - 0.000 - 0.001 - 0.000 - 0.000 - 0.001 - 0.000 - 0.000 - 0.001 (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Groundnut price - 0.000 0.001** 0.000 - 0.000 0.001** 0.000 - 0.000 0.001** 0.000 - 0.000 0.001** 0.000 (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) (0.001) (0.000) (0.000) Plot size 0.101*** 0.036** 0.139*** 0.100*** 0.037** 0.137*** 0.104*** 0.036** 0.143*** 0.107*** 0.036** 0.145*** (0.034) (0.016) (0.038) (0.034) (0.016) (0.038) (0.034) (0.015) (0.038) (0.034) (0.015) (0.038) Plot tenure 0.047 0.237 0.468 0.038 0.241 0.442 0.087 0.244 0.494 0.097 0.246 0.494 (0.400) (0.188) (0.393) (0.396) (0.188) (0.393) (0.403) (0.188) (0.400) (0.396) (0.188) (0.400) Distance from home - 0.014* - 0.053*** - 0.007** - 0.015* - 0.053*** - 0.007** - 0.015* - 0.052*** - 0.007*** - 0.016* - 0.052*** - 0.007*** (0.007) (0.014) (0.003) (0.008) (0.014) (0.003) (0.008) (0.014) (0.003) (0.009) (0.014) (0.003) Distance from main road - 0.029 0.023 - 0.042 - 0.027 0.023 - 0.041 - 0.022 0.023 - 0.038 - 0.019 0.023 - 0.035 (0.031) (0.015) (0.035) (0.031) (0.015) (0.035) (0.031) (0.015) (0.036) (0.031) (0.015) (0.035) Distance from market - 0.004 0.002 - 0.012 - 0.004 0.002 - 0.011 - 0.004 0.002 - 0.011 - 0.004 0.002 - 0.011 (0.006) (0.005) (0.008) (0.006) (0.005) (0.008) (0.006) (0.005) (0.008) (0.006) (0.005) (0.008) Good soil quality 0.022 - 0.248** - 0.271 0.007 - 0.249** - 0.287 0.056 - 0.242** - 0.248 0.046 - 0.242** - 0.260 (0.214) (0.109) (0.211) (0.214) (0.109) (0.211) (0.214) (0.109) (0.210) (0.214) (0.109) (0.211) Poor soil quality - 0.556 - 0.155 0.398 - 0.583 - 0.162 0.357 - 0.576 - 0.159 0.367 - 0.597 - 0.162 0.322 (0.518) (0.249) (0.405) (0.520) (0.249) (0.409) (0.518) (0.249) (0.411) (0.523) (0.249) (0.414) Flat plot slope 1.653** 0.011 0.150 1.674** 0.016 0.179 1.587** 0.012 0.078 1.598** 0.013 0.115 (0.813) (0.257) (0.502) (0.819) (0.258) (0.498) (0.794) (0.257) (0.478) (0.793) (0.257) (0.470) Moderate plot slope 1.486* 0.160 0.117 1.512* 0.162 0.151 1.413* 0.161 0.038 1.439* 0.163 0.086 (0.795) (0.260) (0.513) (0.800) (0.261) (0.509) (0.771) (0.260) (0.488) (0.771) (0.261) (0.481) 171 Table 3A. 4 Voucher receipt (1) Voucher receipt (2) Voucher redemption (1) Voucher redemption (2) Variables I S SI I S SI I S SI I S SI NAIVS for any input 2.382*** 0.610** 2.229*** 2.507*** 0.552* 2.449*** (0.323) (0.249) (0.321) (0.356) (0.284) (0.335) NAIVS for inorganic 2.580*** 0.786*** 2.426*** 2.510*** 0.603* 2.430*** fertilizer (0.376) (0.301) (0.350) (0.398) (0.325) (0.363) NAIVS for maize seed 0.267 0.148 0.108 1.028* 0.324 0.979 (0.535) (0.372) (0.578) (0.551) (0.478) (0.603) Notes: Robust standard are in parentheses. To control for time - invariant unobserved household heterogeneity, time - averages of household characteristics were included in the model but not reported in Table 3A. 4 . I, S, and SI - and 1% levels, respectively. 172 Table 3A. 5 : APEs of other (non - NAIVS - related) factors affecting household use of practices in the various SI categories CRE - MNL with voucher receipt CRE - MNL with voucher redemption Variables I S SI I S SI Male - Headed HH - 0.004 0.023 - 0.025 - 0.005 0.024 - 0.027 (0.019) (0.029) (0.017) (0.019) (0.029) (0.017) Age of HH head - 0.002 0.002 - 0.001 - 0.002 0.002 - 0.001 (0.002) (0.003) (0.002) (0.002) (0.003) (0.002) Education of HH head 0.006*** - 0.004 0.004* 0.006*** - 0.005 0.004* (0.002) (0.004) (0.002) (0.002) (0.004) (0.002) Family labor - 0.011 0.033*** 0.000 - 0.011 0.034*** - 0.001 (0.008) (0.011) (0.007) (0.008) (0.011) (0.007) Total cultivated land - 0.003 - 0.001 - 0.006** - 0.004 - 0.001 - 0.006** (0.002) (0.003) (0.003) (0.003) (0.003) (0.003) Off - farm income 0.002 0.034 - 0.032** 0.000 0.036 - 0.033** (0.016) (0.029) (0.015) (0.016) (0.028) (0.015) Farm assets - 0.000 - 0.000* 0.000 - 0.000 - 0.000* 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Livestock ownership 0.001 0.066*** 0.042*** 0.002 0.067*** 0.041*** (0.014) (0.025) (0.013) (0.014) (0.025) (0.013) Access to credit 0.026 - 0.041 0.062*** 0.026 - 0.045 0.066*** (0.021) (0.047) (0.024) (0.021) (0.046) (0.022) Extension from 0.031 - 0.048 0.034* 0.032* - 0.045 0.034* (0.019) (0.040) (0.021) (0.019) (0.039) (0.021) Extension from 0.003 0.084 0.043* 0.003 0.084 0.044* cooperative (0.028) (0.080) (0.023) (0.028) (0.080) (0.022) Cooperative 0.020 0.013 0.010 0.020 0.013 0.010 (0.014) (0.023) (0.013) (0.014) (0.023) (0.013) Input supplier 0.007 - 0.023 0.020 0.007 - 0.021 0.019 (0.012) (0.022) (0.014) (0.012) (0.021) (0.014) Drought/Flood 0.036 - 0.074* 0.026 0.032 - 0.075* 0.023 (0.026) (0.042) (0.025) (0.026) (0.042) (0.024) Crop disease/Pests 0.012 - 0.045 0.016 0.010 - 0.041 0.013 (0.029) (0.052) (0.030) (0.028) (0.052) (0.030) Rainfall - 0.000 0.000 - 0.000 - 0.000 0.000 - 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Soil nutrient constraint 0.008 - 0.413*** - 0.041 0.007 - 0.414*** - 0.040 (0.042) (0.127) (0.031) (0.043) (0.127) (0.031) Inorganic fertilizer price 0.000*** - 0.000 0.000 0.000*** - 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Real price of maize - 0.000 - 0.000* - 0.000 - 0.000 - 0.000* - 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Real price of rice 0.000** - 0.000 0.000 0.000** - 0.000 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Bean price 0.000 - 0.000 - 0.000 0.000 - 0.000 - 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Groundnut price - 0.000 0.000** 0.000 - 0.000 0.000** 0.000 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) 173 Table 3A. 5 CRE - MNL with voucher receipt CRE - MNL with voucher redemption Variables I S SI I S SI Plot size 0.003* 0.002 0.007*** 0.003* 0.002 0.007*** (0.002) (0.003) (0.002) (0.002) (0.003) (0.002) Plot tenure - 0.010 0.038 0.024 - 0.008 0.038 0.025 (0.021) (0.038) (0.022) (0.021) (0.038) (0.022) Distance from home 0.000 - 0.011*** 0.001** 0.000 - 0.011*** 0.001** (0.001) (0.003) (0.000) (0.001) (0.003) (0.000) Distance from main road - 0.002 0.006** - 0.003 - 0.001 0.006* - 0.003 (0.002) (0.003) (0.002) (0.002) (0.003) (0.002) Distance from market - 0.000 0.001 - 0.001 - 0.000 0.001 - 0.001 (0.000) (0.001) (0.001) (0.000) (0.001) (0.001) Good soil quality 0.011 - 0.046** - 0.012 0.012 - 0.046** - 0.011 (0.012) (0.022) (0.012) (0.012) (0.022) (0.012) Poor soil quality - 0.038 - 0.031 0.039 - 0.038 - 0.031 0.038 (0.029) (0.049) (0.024) (0.029) (0.049) (0.025) Flat plot slope 0.098** - 0.034 - 0.018 0.095** - 0.031 - 0.022 (0.048) (0.056) (0.031) (0.047) (0.055) (0.030) Moderate plot slope 0.086* 0.001 - 0.021 0.083* 0.005 - 0.026 (0.047) (0.055) (0.032) (0.045) (0.055) (0.031) Notes: To control for time - invariant unobserved household heterogeneity, time - averages of household characteristics were included in the model but not reported in Table 3A. 5 . I, S, and SI denote coefficients are significant at the 10%, 5%, and 1% levels, respectively. 174 REFERENCES 175 REFERENCES Aloyce, G.M., Gabagambi, D.M., and Hella, J.P. 2014. Assessment of operational aspects of the input supply chain under national agriculture input voucher scheme (NAIVS) in Tanzania. Journal of Development and Agricultural Economics 6(3): 94 - 104. Amsalu, A ., and de Graaff, J. 2007. Determinants of adoption and continued use of stone terraces for soil and water conservation in an Ethiopian highland watershed. Ecological E conomics 61:294 - 302. Banful, A.B. 2011. Old problems in the new solutions? Politically motivated allocation of World Development 39(7):1166 - 1176. Bronick, C.J., and Lal, R. 2005. Soil structure and management: a review. Geoderma 124(1):3 - 22. Burke, W.J., Jayne, T.S., and Black, J.R. 2017. Factors explaining the low and variable profitability of fertilizer application to maize in Zambia. Agricultura l Economics 48(1):115 - 126. Chamberlain, G. 1984. Panel data. Handbook of Econometrics , vol. 2, 1247 - 1318. De Janvry, A., Fatchamps, M., and Sadoulet, E. 1991. Peasant household behavior with missing markets: some paradoxes explained. The Economic Journal 101:1400 - 1417. Di Falco, S., and Veronesi, M. 2013. How can African agriculture adapt to climate change? A counterfactual analysis from Ethiopia. Land Economics 89(4):743 - 766. Dorward, A. 2009. Rethinking agricultural input subsidy programmes in develop ing countries. In A. Elbehri, & A. Sarris (Eds.), Non - distorting farm support to enhance global food production. Rome: FAO. Doss, C.R., and Morris, M.L. 2001. How does gender affect the adoption of agricultural innovations? The case of improved maize tech nology in Ghana. Agricultural Economics 25(1):27 - 39. FAO, IFAD, UNICEF, WFP and WHO, 2018. The state of food security and nutrition in the world 2018. Building climate resilience for food security and nutrition. Rome, FAO. Godfray, H.C.J., Beddington, J. R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F., Pretty, J., Robinson, S., Thomas, S.M., and Toulmin, C. 2010. Food security: the challenge of feeding 9 billion people. Science 327(5967):812 - 818. 176 Grabowski, P.P., Kerr, J.M., Haggblade, S., and Kabwe, S. 2016. Determinants of adoption and disadoption of minimum tillage by cotton farmers in eastern Zambia. Agriculture, Ecosystems and Environment 231:54 - 67. Hassan, R., and Nhemachena, C. 2008. Determ to climate change: Multinomial choice analysis. African Journal of Agricultural and Resource Economics 2(1):83 - 104. Hausman, J., and McFadden, D. 1984. Specification tests for the multinomial logit model. Econometrica: Journal of the Econometric Society : 1219 - 1240. Holden, S.T. 2018. Fertilizer and sustainable intensification in Sub - Saharan Africa. Global Food Security 18:20 - 26. Holden, S., and Lunduka, R. 2012. Do fertilizer subsidies crowd out organic manure? The case of Malawi. Agricultural Economics 43(3):303 - 314. Jayne, T.S., Mather, D., Mason, N., and Ricker - Gilbert, J. 2013. How do fertilizer subsidy programs affect total fertilizer use in sub - Saharan Africa? Crowding out, diversion, and benefit/cost assessments. Agricultural Economics 44: 687 - 703. second - generation agricultural inp ut subsidy programs. Food Policy 75:1 - 14. Jayne, T.S. and Rashid, S. 2013. Input subsidy programs in sub - Saharan Africa: a synthesis of recent evidence. Agricultural Economics 44:547 - 562. Kassie, M., Teklewold, H., Jaleta, M., Marenya, P., and Erenstein, O. 2015a. Understanding the adoption of a portfolio of sustainable intensification practices in eastern and southern Africa. Land Use Policy 42:400 - 411. Kassie, M., Teklewold, H., Marenya, P., Jaleta, M., and Erenstein, O. 2015b. Production risks and foo d security under alternative technology choices in Malawi: Application of a multinomial endogenous switching regression. Journal of Agricultural Economics 66(3):640 - 659. Kassie, M., Jaleta, M., Shiferaw, B., Mmbando, F., and Mekuria, M. 2013. Adoption of interrelated sustainable agricultural practices in smallholder systems: Evidence from rural Tanzania. Technological Forecasting & Social Change 80:525 - 540. Khonje, M.G., Manda, J., Mkandawire, P., Tufa, A.H., and Alene, A.D. 2018. Adoption and welfare imp acts of multiple agricultural technologies: evidence from eastern Zambia. Agricultural Economics 49(5):599 - 609. 177 Kim, J., Mason, N.M., Snapp, S., and Wu, F. (in press). Does Sustainable Intensification of Maize Production Enhance Child Nutrition? Evidence from Rural Tanzania. Accepted for publication in Agricultural Economics on 8 July 2019. Koppmair, S., Kassie, M., and Qaim, M. 2017. The influence of farm input subsidies on the adoption of natural resource management technologies. Australian Journal of A gricultural and Resource Economics 61(4):539 - 556. Lunduka, R., Ricker - Gilbert, J., and Fisher, M. 2013. What are the farm - level impacts of Agricultural Economics . 44:563 - 579. Masinjila, S. and Lewis, L. 2018. The future of smallholder farmer support in Tanzania: where to after the National Agricultural Input Voucher System (NAIVS)?. https://acbio.org.za/sites/default/files/documents/The%20Future%20of%20smallholder %20farmer%20support%20in%20Tanzania%20Where%20to%20after%20the%20NAI VS%20system.pdf Mas on, N.M. and Ricker - Gilbert, J. 2013. Disrupting demand for commercial seed: input subsidies in Malawi and Zambia. World Development 45:75 - 91. Mason, N.M., Jayne, T.S., and van de Walle, N. 2017. The Political Economy of Fertilizer Subsidy Programs in Africa: Evidence from Zambia. American Journal of Agricultural Economics 99 (3):705 - 731. Mather, D. and Minde, I. 2016. Fertilizer subsidies and h ow targeting conditions crowding in/out: an assessment of smallholder fertilizer demand in Tanzania. GISAIA/Tanzania working paper No. 5. Department of Agricultural, Food, and Resource Economics, Michigan State University, East Lansing, MI. Mather, D., Wa ized, B., Ndyetabula, D., Temu, A., Minde, I., and Nyange, D. 2016. The effects of NAIVS on private sector fertilizer and seed supply chains in Tanzania. GISAIA/Tanzania working paper No. 3. Department of Agricultural, Food, and Resource Economics, Michiga n State University, East Lansing, MI. Matson, P.A., Parton, W.J., Power, A.G., and Swift, M.J. 1997. Agricultural intensification and ecosystem properties. Science 277(5325):504 - 509. Marenya, P.P., and Barrett, C.B. 2007. Household - level determinants of adoption of improved natural resources management practices among smallholder farmers in western Kenya. Food policy 32(4):515 - 536. Marenya, P.P., and Barrett, C.B. 2009. State - conditional fertilizer yield response on Western Kenyan farms. American Journal of Agricultural Economics 91(4): 991 - 1006. 178 McFadden, D. 1973. Conditional logit analysis of qualitative choice behavior. In: Zarem bka, P. (Ed.). Frontiers in Econometrics. Academic Press, New York. Morgan, S., Mason, N.M., Levine, N.K., and Mbata - Zulu, O. 2019. Dis - incentivizing sustainable - fertilizer subsidy program. World Development 122 :54 - 69. Morris, M., Kelly, V.A., Kopicki, R.J., and Byerlee, D. 2007. Fertilizer use in African agriculture: lessons learned and good practice guidelines. Washington , DC: World Bank. Mundlak, Y. 1978. On the pooling of time series and cross section data. Econometrica 64:69 - 85. Ndiritu, S.W., Kassie, M., and Shiferaw, B. 2014. Are there systematic gender differences in the adoption of sustainable agricultural intensification practices? Evidence from Kenya. Food Policy 49:117 - 127. Pan, L., and Christiaensen, L. 2012. Who is vouching for the input voucher? Decentralized targeting and elite capture in Tanzania. World Development 40(8):1619 - 1633. Pender, J., and Gebremedhin, B. 2007. Dete rminants of agricultural and land management practices and impacts on crop production and household income in the highlands of Tigray, Ethiopia. Journal of African Economies 17 (3):395 - 450. Petersen, B., and Snapp, S. 2015. What is sustainable intensification? Views from expert. Land Use Policy 46:1 - 10. Pingali, P.L. 2012. Green revolution: impacts, limits, and the path ahead. Proceedings of the National Academy of Sciences 109(31):12302 - 123 08. Pretty, J., Toulmin, C. and Williams, S. 2011. Sustainable intensification in African agriculture. International journal of agricultural sustainability 9(1):5 - 24. Putterman, L. 1995. Economic reform and smallholder agriculture in Tanzania: A discussi on of recent market liberalization, road rehabilitation, and technology dissemination efforts. World Development 23(2):311 - 326. Ricker - Gilbert, J., Jayne, T.S., and Chirwa, E. 2011. Subsidies and crowding out: a double - hurdle model of fertilizer demand in Malawi. American Journal of Agricultural Economics 93(1): 26 - 42. Rivers, D., and Vuong, Q.H. 1988. Limited information estimators and exogeneity tests for simultaneous probit models. Journal of econometrics 39(3):347 - 366. Tanzania National Bureau of Sta tistics. 2014. Tanzania National Panel Survey Report (NPS) Wave 3, 2012 - 2013, Dar es Salaam, Tanzania, NBS. 179 Teklewold, H., Kassie, M., Shiferaw, B., and K ö hlin, G. 2013. Cropping system diversification, conservation tillage and modern seed adoption in Et hiopia: Impacts on household income, agrochemical use and demand for labor. Ecological Economics 93:85 - 93. The Montpellier Panel. 2013. Sustainable Intensification: A New Paradigm for African Agriculture. Imperial College London, London Tse, Y.K. 1987. A diagnostic test for the multinomial logit model. Journal of Business & Economic Statistics 5(2):283 - 286. Van Ittersum, M.K., Van Bussel, L.G., Wolf, J., Grassini, P., Van Wart, J., Guilpart, N., ... and Yang, H. 2016. Can sub - Saharan Africa feed itself?. Proceedings of the National Academy of Sciences 113(52):14964 - 14969. Wanzala - Mlobela, M., Fuentes, P., an d Mkumbwa, S. 2013. Practices and policy options for the improved design and implementation of fertilizer subsidy programs in sub - Saharan Africa. NEPAD policy document. IFDC, Alabama. Wooldridge, J. M. Econometric Analysis of Cross Section and Panel Data . MIT press, 2010 Wooldridge, J. M. 2015. Control function methods in applied econometrics. Journal of Human Resources 50(2): 420 - 445. World Bank. 2014. Public expenditure review: national agricultural input voucher scheme (NAIVS).