Report No. 32971-UG Uganda Policy Options for Increasing Crop Productivity and Reducing Soil Nutrient Depletion and Poverty July 18, 2005 AFTS2 Environmentally and Socially Sustainable Development Africa Region Document of the World Bank Acknowledgements The World Bank i s gratefid to the Trust Fund for Environmentally and Socially Sustainable Development for providing financial support to this research, as well as to Makerere University Institute o f Statistics and Applied Economics, the Uganda Bureau o f Statistics, the National Agricultural Research Organization o f Uganda, the Agricultural University o f Norway for partnership with IFPRI in this project and to the many farmers and community leaders who participated inthe survey uponwhich this study i s based. .. 11 Tableof contents Summary iv 1 Introduction . ..................................................................................................................................... 1 2 Theory on LinkagesbetweenPovertyandNaturalResourceManagement .................................................................................................. .................... 4 3 EmpiricalModels andData .. ............................................................................................................................ 6 Dataanalysis .................................................................................................................... 17 17 4. Discussionof Econometricmodels: ......................................................................................................... Results 17 DescriptiveResults ............................................................................................................. ............................................................................................................ 17 22 5 ConclusionsandPolicyImplications . EconometricResults ........................................................................................................... ................................................................................... 36 References .................................................................................................................................. 41 Listof Fimres Figure1 Agro-climatic Zones inUganda .............................................................................. 15 Figure2 Classificationof Market Access inUganda .. ........................................................... 16 Listof Tables Table 1:Major sources of nitrogeninflowsandchannels of outflowsat plotlevel ............18 Table 2: Major sources of phosphorusinflowsandchannelsof outflowsat plotlevel ......18 Table3: Major sources ofpotassiuminflowsandchannelsof outflowsat plotlevel Table4: Severity of soil nutrientdepletionanditseconomic magnitude .....................................19 19 Listof Amendices Appendix 1:SelectedDistricts.Communitiesandhouseholds ............................................. 48 Appendix 2 :Descriptivestatisticsof plot andhouseholdlevelvariables ........................... 49 Appendix 3: Determinantsof landmanagementpractices(Probit models) ....................... 51 Appendix4: Determinantsof inputuse (purchasedseed. inorganicfertilizer and organic residuesapplied) (Probit models) ............................................................................................ 52 Appendix5: Factorsaffectingplotproductivity .................................................................... 54 Appendix 6: Factorsaffectingper capitahouseholdincome ................................................ 56 Appendix 7: Pre-NAADSvalue of crop productionper acre inNAADSvs non-NAADS sub-countiesof sample districts ............................................................................................... 57 Appendix8: Pre-NAADSincomeper capitainNAADS vs .non-NAADSsub-countiesof sample districts .......................................................................................................................... 57 Appendix 9: Determinantsof soil erosion [ln(Soilerosion)] 58 Appendix 10: Determinantsof nutrientbalances .................................................................. ................................................. 59 Appendix 11:Qualitative results(summary) 62 Appendix 12 TheoreticalDynamicHouseholdModel . ......................................................................... ............................................................... ......................................................... 64 Appendix 13 EconometricModelsandApproach . 69 ... 111 Summary This study was conducted with the main objective o f determining the linkages between poverty and landmanagement inUganda. The study used the 2002/03 Uganda National Household Survey (UNHS) and more focused data collected from a sub-sample o f 851 households o f the 2002/03 UNHS sample households in eight districts representing six major agro-ecological zones and farming systems. Farmers in these districts deplete an average o f 179 kg/ha o f nitrogen, phosphorus and potassium, which i s about 1.2% o f the nutrient stock stored in the topsoil. The value o f replacing the depleted nutrients using the cheapest inorganic fertilizers is equivalent to about 20% o f household income obtained from agricultural production. This underscores the reliance o f smallholder farmers on soil nutrient mining for their livelihoods and the highcosts that would be required to avoid this problem. The findings o f this study also underscore the great concern that soil nutrient depletion poses since it contributes to declining agricultural productionin the near term as well as the longer term. For example, data shows that a 1% decrease in the nitrogen stock inthe topsoil leads to a predicted 0.26% reduction incrop productivity. This loss in agricultural productivity may contribute to food insecurity. Furthermore, soil nutrient depletion may also contribute to deforestation and loss o f biodiversity since farmers may be forced to abandon nutrient-depleted soils and cultivate more marginal areas such as hillsides and rainforests. Hence, there is an urgent need to design strategies to address the soil nutrient depletion and other forms o f land degradation. Such strategies include, but are not limited to, reducing the cost o f inorganic fertilizer and developing and promoting organic soil fertility technologies that are cost effective andrelevant to local farming systems. Econometric analysis o f the survey results provides evidence o f linkages between poverty and land management. Some land investments contribute to better current land management practices, higher productivity and income, reduced erosion and, in the case o f soil and water conservation (SWC) structures, improve soil nutrient balances. These findings imply that land investments can lead to win-win-win outcomes since they increase agricultural productivity and income and conserve naturalresources. Many inputs and land management practices, including labor, purchased seed, organic fertilizer, and incorporation o f crop residues increase crop production per acre. However, inorganic fertilizer does not have a statistically significant impact on crop productivity, and the estimated marginal value cost ratio o f fertilizer is much less than 1, suggesting that adoption o f fertilizer is likely to remain low unless its price is reduced substantially or crop prices improve substantially. An inverse farm size - crop productivity relationship was observed, due to lower farming intensity by larger farms. Although smaller farms obtain higher value o f production per acre, this does not fully compensate for the fact that they have less land, and they earn lower per capita incomes as a result. These findings are consistent with those o f previous studies in Uganda and with numerous studies from other developing countries. Despite these differences related to farm size, no significant differences insoil erosion or soil nutrient depletion due to farm size were found. Thus, improving access of small farmers to land, for example by improvingthe functioning o f land markets, can increase aggregate agricultural productivity and small fanners' incomes in Uganda, with no apparent tradeoff interms o f landdegradation. iv Non-land assets, including livestock and value o f equipment, have mixed impacts on land management and outcomes. Value o f equipment increases labor intensity and but decreases crop productivity. Livestock ownership decreases the probability to fallow and the level o f nitrogen balances but increases crop productivity andper capita income. These results suggest that livestock poor farmers are likely to remain inpoverty with low productivity. Education o f female household members has generally a limited impact on land management, while male education i s associated with greater use o f inorganic fertilizer. Both female post-secondary and male primary and secondary education are associated with higher crop productivity, but female education also i s associated with soil erosion and nutrient depletion, while male education has less impact on land degradation. These results imply that simply investing in education will not solve the problem o f land degradation in Uganda, even though education i s critical to the long-term success o f poverty reduction efforts, and contributes to improved productivity and fertilizer adoption (in the case o f male education). Improvements in the educational curriculum, incorporating teaching on principles o f sustainable landmanagement, may help to address the land degradation problem. Larger families use more erosive practices but realize higher value o f crop production per acre but have lower per capita income, suggesting that population pressure at the household level contributes to poverty and landdegradation inUganda. Access to financial capital has limited effect on land management but increases crop productivity and per capita household income. The results suggest that rural financing could significantly contribute to poverty reductionbut it has limitedimpact on land management. Likewise, access to markets and all-weather roads has limited influence on most landmanagement practices. However, roads contribute to higher per capita household income and less soil nutrient depletion. These results support the Ugandan government's efforts to build rural roads as investments that can reduce poverty, as well as potentially helping to reduce landdegradation. Both the traditional and the new agricultural extension program increase use of fertilizer and crop productivity, suggesting that investment in extension services could significantly contribute to agricultural modernization and poverty reduction. The results show that participation inthe new National Agricultural Advisory Services (NAADS) is associated with 15% higher crop productivity, while a 10% increase in the number o f contact hours with the traditional extension agent increases predicted crop productivity by 1%. These results therefore provide support to the emphasis in the Uganda Plan for Modernization o f Agriculture (PMA) on increasing the availability of agricultural technical assistance through expansion o f NAADS. Nevertheless, more focused research on the impacts o f NAADS would be very useful to better understand how the NAADS program is having such favorable impacts inthe initial districts and sub-counties where it has been implemented, and whether these beneficial impacts are likely to scale out to other sub- counties and districts as the program expands. The results suggest the need to give incentives for technical assistance programs to operate inremote areas, where accessto extension services is limited. Access to extension services didnot have significant impact on organic land management practices or on land degradation. The results V suggest the need for the new extension program to give greater attention to promoting organic land soil fertility practices to help address the soil fertility depletion problem. Households pursuing n o n - f m activities are more able to fallow their landand less likely to use slash and bum, and obtain higher value o f production per acre and per capita income. These results imply that non-farm activities can be complementary to crop production. Hence, efforts to increase rural households' access to non-farm activities can help increase agricultural productivity as well as helping to reduce poverty. Perennial crop producers deplete soil nutrients more rapidly, implyingthe need to promote measures to restore soil nutrients inperennial (especially banana) production areas. The landtenure system also i s associated with some differences in land management practices and land degradation; e.g., land degradation i s somewhat worse on land under mailo than freehold tenure, while nutrient depletion is less on customary than freehold tenure. Nevertheless, no significant differences were found incrop productivity or income per capita associatedwith differences inland tenure systems. Findings suggest that customary land tenure, which i s the most common form o f tenure, is not a constraint to improvements in land productivity or use o f sustainable land management. Overall, the results provide general support for the hypothesis that promotion o f poverty reduction and agricultural modernization through technical assistance programs and investments in infrastructure and education can improve agricultural productivity and help reduce poverty. However, they also show that some o f these investments do not necessarily reduce land degradation, and may contribute to worsening land degradation inthe near term. Thus, investing in poverty reduction and agricultural modernization i s not sufficient to address the problem o f land degradation in Uganda, and must be complemented by greater efforts to address this problem. vi 1. Introduction The world celebrated the beginningof the new millenniumwith more than one billionpeople living on less than one US$ a day. This has posed an enormous challenge to poor countries and their development partners. At a global scale, the UnitedNations has set millenniumdevelopment goals (MDG) to halve the proportion of people living on less than one US$ a day by 2015. Most poor countries, including Uganda, have ratifiedthe MDG's and committed to achieve them. Eventhough Uganda has reduced absolute poverty from 56% o f the population in 1992 to 35% in 1999/00 (Appleton, 2001), poverty reduction remains the primary goal of the country's policies and strategies. To achieve this goal, the government has laid out an ambitious strategy for addressing poverty through the Poverty Eradication Action Plan (PEAP), which sets a target o f reducing the proportion of the population living in absolute poverty to below 10% by 2017 (MFPED, 2001). However, there is concern over whether this goal can be achieved and whether poverty reduction statistics reflect an improvement inthe living standards o f the majority o fthe people, particularlyin rural areas, where 96% o f the poor live. Agricultural productivity in general has stagnated or declined for most farmers (Deininger and Okidi, 2001). Recent data also show an increase in the incidence o fpoverty to 38% (8.9 million people) in2002/03 (UBOS, 2003). Since poor householdsdependon natural resources more than wealthier households, one of PEAP's objectives i s to ensure that poverty reduction efforts do not compromise natural resources (NEMA, 2002; MFPED, 2001). There is concern in Uganda, as elsewhere in Africa that poor households in Africa face a downward spiral of land degradation and poverty (NEMA, 2002; Cleaver and Schreiber, 1994). Most communities in Uganda perceive that natural resources are degrading and that food insecurity is worsening (Pender, et al. 2001; APSEC 2001). However, scientific studies to verify these perceptions and to quantify land degradation in Uganda are limited. Available estimatesindicate that the rate of soil fertility depletion inUgandais among the highest in sub-Saharan Africa (Stoorvogel and Smaling, 1990; Wortmann and Kaizzi 1998). A recent study of maize producinghouseholds ineasternUganda estimated that the average value o f soil nutrient depletion i s equal to about one-fifth o f average household income (Nkonya, et al. 2004b). Soil fertility depletion thus represents a substantial loss in Uganda's natural capital, as well as reducing agricultural productivity and income. Soil erosion is also a serious problem in the highlands (hid.; Magunda and Tenywa 1999; NEAP 1992). Soil nutrient depletion and erosion pose a serious concern since they contributes to declining agricultural production (Bekunda, 1999; Deininger and Okidi 2001; Pender, et al. 2001), which intum contribute to food and nutrition insecurity. Soil nutrient depletion and erosion could also lead to deforestation and loss of biodiversity since farmers are forced to abandon nutrient-starved soils and cultivate more marginal areas such as hillsides and rainforests. The overall implication o f these impacts is increasedpoverty, which pose an enormous development challenge inSSA. Poverty may also contribute to landdegradation, ifpoor people lack the ability or incentive to invest in conserving and improving their land. However, little empirical evidence is available concerning the relationships between land degradation and poverty in Uganda and other African countries, or about the policy, institutional or technological responses that could most effectively address these problems. This study seeks to address this information gap, using analysis o f data from a survey conducted in 2003 at the community, household and plot level by the International Food Policy Research Institute (IFPRI), in collaboration with the Uganda Bureau o f Statistics (UBOS), the National Agricultural Research Organization (NARO) and Makerere University - 1 hereafter referred to as IFPRI/UBOS survey. These data were collected from a sub-sample of households participating in the 2002/2003 Uganda NationalHousehold Survey (UNHS), and some o fthe 2002/2003 data were also usedinthe analysis. There is a strong desire by the government to understand the nature o f poverty and what can be done to address it. Particularly policy makers and other stakeholders would like to know the policies and strategies that effectively alleviate poverty and conserve the environment and natural resources. For example, one of the deficiencies of the PEAP is a weak framework on strategies for conserving the environment and natural resources (NEMA, 2004). The PEAP i s beingrevised to address this and other deficiencies. This paper contributes to better understanding o f policies and strategies that would increase agricultural productivity and conserve the environment. The main focus of this study is on investigating how poverty, broadly defined to include limitations in physical, human, natural and financial capital as well as limited access to infrastructure and services, influences farmers' land management practices, land degradation in the form o f soil erosion and depletion of soil nutrients, crop productivity, and household incomes inUganda. Contributionsof ThisStudy This paper is the fourth in a series of papers produced by the IFPRI-UBOS-NARO- Makerere UniversityProject "Poverty and Natural Resource Management inUganda", which was supported by the World Bank Trust Fund for Environmentally and Socially Sustainable Development (Pender, et al. 2004; Nkonya, et al. 2004a; Kaizzi, et al., 2004). It builds on that work (especially the studies by Pender, et al. 2004 and Kaizzi, et al., 2004) and on earlier research inUganda (Nkonya, et al. 2004b) to identify the impacts ofpoverty-broadly definedto include limitations in communities' and households' endowments of physical, human, natural, and financial capital, as well as access to infrastructure and key services, such as agricultural technical assistance-on their landmanagement decisions and land degradation; andto identify the impacts o f landmanagement andlanddegradation on agricultural productivity andpoverty. The study by Nkonya, et al. (2004a) sought to understand the determinants o f natural resource management (NRM) at the community level. The results showed that greater awareness of regulations contributes to more sustainable NRM. Awareness i s greater in areas closer to all- weather roads, probably due to better access to information in such areas. Development o f roads andcommunication can thus facilitate better communityNRM. Other low cost options to increase awareness could include use of radio programs, environmental education inschools, resource user seminars, brochures, and district level training workshops. Nkonya, et al. (2004a) also found that compliance with bylaws that influence NRMi s greater for bylaws enacted by local LC1 councils than for bylaws enacted at a higher level. These results suggest that involvement of locally accountable and representative authorities in enacting and enforcing NRM requirements appears critical for the legitimacy and success of such regulation. The results also showed that involvement of extemal programs and organizations focusing on agriculture and environment issues can help to promote local enactment of such bylaws (Ibid.). Several dimensions o f poverty, including greater income poverty, poor education, and poor access to credit were found to be associated with lower compliance with NRM requirements (Ibid.). This supports the hypothesis of a poverty-natural resource degradation trap, and suggests that measures to reduce poverty can 2 have "win-win" benefits helping to improve NRM as well. Land tenure had mixed relationships with enactment and compliancewithNRMrequirements (hid.). The study by Pender, et al. (2004) assessedthe household-level linkages between poverty and land management to the extent possible by analyzing available survey data from the 1999/2000 UNHS, which collected information on use o f inputs in crop production (e.g., use o f seeds, inorganic and organic fertilizer) and crop production and income at the household level. This analysis provided mixed support for the hypothesis that poverty causes poor land management and low productivity. For example, the results showed that smaller (land poor) farms compensate for land constraints by using some inputs more intensively, and thus obtain higher land productivity. However, smaller farmers' incomes are lower as they are unable to fully compensate for land constraints. Households with lower value land use less o f most inputs and obtain lower land productivity and income. To the extent that land degradation contributes to future declines in land quality and value, these results support the land degradation - poverty spiral hypothesis, though longitudinal data are needed to verify this. Lack o f ownership o f physical assets such as livestock and equipment was found to be associated with less use o f fertilizer (inorganic or organic) and other inputs, and for some assets, lower productivity and income. These results also support the poverty spiral hypothesis, though the role o f land degradation inreducing productivity and income i s not clear. Lack o f human capital was found to have mixed impacts on land management, with male education associated with adoption o f fertilizer and some other land management practices, while female education had less impact on landmanagement. Limited access to markets androads was found to reduce adoptiono f fertilizer and some other inputs, though impacts on productivity and incomes were more region-specific. Limited access to credit, agricultural extension and market information were also associated with less use o f fertilizer and, in the case o f credit, lower productivity. Lower wage rates were associated with lower adoption o f fertilizer and some other inputs, as well as lower productivity and incomes. This study also found low marginal retums to investments ininorganic or organic fertilizer, suggesting that it will be difficult for farmers to increase investment in these inputs in the present market environment. Many, but not all, o f the results in Pender, et al. (2004) support the idea that poverty, broadly defined, contributes to less intensive land management and lower productivity and income. However, several limitations o f that study limited its ability to draw definitive conclusions about the linkages between poverty and land degradation. No land quality indicators were measured in the 1999/2000 UNHS, so estimated land value was used as a proxy; but land values may be poorly estimated andmay reflect many factors other than landquality. The level o f use o f inputs and crop production were measured only at the household level, limiting the ability to take into account plot-specific characteristics that affect these responses. More importantly, no indicators o f land degradation were measured, so that the relationships o f poverty with land degradation could not be directly assessed. Assessing some o f the linkages between poverty and land degradation requires longitudinal data on both poverty and land degradation, as well as on intervening factors such as landmanagement decisions. The present study addresses most o f these shortcomings. Information on land quality indicators, land management and land degradation were collected at the plot level so that plot specific characteristics and responses could be taken into account. Soil samples were taken and use to quantify measures o f soil fertility and as an input into estimation o f soil erosion and soil nutrient losses based on the survey data. These soil analyses were led by a soil scientist from NARO, and the methodological approach and results are reported in Kaizzi, et al. (2004). The main conclusions o f Kaizzi, et al. (2004) are presented in this study. The assessment o f determinants o f soil nutrient losses inthis study builds on an approach pioneered ina small study o f determinants o f household soil nutrient balances in eastern Uganda reported in Nkonya, et al. (2004b). Inthis study, the assessment o f nutrient depletion is at the plot rather thanthe household level (which i s the more relevant level to consider land degradation impacts), and has broader coverage with a much larger sample size, so that more robust conclusions are possible. Although the present study i s still limited by the cross-sectional nature o f the results, it has laid the foundation for future longitudinal studies o f poverty-land degradation relationships by being linked to the 2002/2003 UNHS sample. The remainder o f the paper i s organized as follows: In the next section, the theory o f poverty-NRh4 linkages is discussed, followed by discussion o f the empirical approach, key variables, hypotheses and data sources. The empirical results are presented in the fourth section, andconclusions andimplications are discussed inthe last section. 2. Theory on Linkages between Poverty and NaturalResourceManagement Interest inresearch on poverty and its linkage with natural resource management has grown enormously inthe past few decades (Grepperud, 1997). There i s yet no consensus on the impact o f poverty on natural resource management (NRM). One view posits that in a perfect market setting, there i s no linkage between poverty and NRM since households (and firms) would allocate natural resources such that they yield the highest returns to investment (Singh, et al., 1986). Under this unrealistic perfect market assumption, household endowments would therefore not determine allocation and management o f natural resources since such decisions are dictated by local biophysical factors and market prices that determine the returns to investment. An alternative view assumes that poor households have high discount rate, hence have short-term planning horizons (Griffin and Stoll, 1984; Rausser, 1980; Hammer, 1986). Many studies have argued that lack o f resources and alternative opportunities and their short-term perspective force poor farmers to degrade natural capital in order to meet their short-term needs (WCED, 1987; Leonard, 1989; Cleaver and Schreiber, 1994). Empirical evidence has shown that poverty-NRM linkages are more complex than these simplified views. The first view rests on an unrealistic assumption o f perfect markets that i s not easy to observe inreal life. Imperfect markets o f different types are a rule rather than an exception in most areas - especially in low income countries like Uganda - and contribute to failure of farmers to efficiently use their scarce resources. As will be discussed later, market imperfections greatly influence NRM. Some studies have also challenged the second view on both theoretical and empirical grounds. For example, Pender (1998) noted that farmers optimally choose to invest inhigher-returninvestments inorder to obtain higher income andbetter welfare, but suchdecisions may lead to natural resource degradation in the near term if the returns to investment in natural capital are lower than returns to investing in other forms o f capital (until returns are relatively equalized across different investments). This suggests that natural resource degradation inthe near term may be part of the process o f poverty reduction, rather than a cause of increasing poverty. Neither does poverty necessarily lead to natural resource degradation. Poor households may invest 4 more than wealthier ones in labor-intensive NRMpractices because they depend more on natural resources for their livelihood or becausethey have lower labor opportunity costs. A large number o f factors may influence the direction and severity o f impact o fpoverty on NRM.Empirical evidence suggeststhat marketfailure isone ofthemostimportant factors that give credence to the poverty-NRM linkage view. The impact o fmarket failure on the linkage o fpoverty- NRMdepends uponthe nature ofthe market failure, the nature ofpoverty, andthe type ofresource management and resource degradation considered. For example, if there is no land or credit market, but all other markets function perfectly, households with less wealth or income will be less able to invest in soil and water conservation measures than wealthier households, other factors being equal, and thus may suffer greater land degradation (Pender and Kerr 1998). On the other hand, wealthier households are more able to invest in livestock, mechanical equipment, or other assets that may contribute to soil erosion or other forms o f land degradation. Furthermore, the land management practices pursued by wealthier households may increase some forms o f resource degradation (e.g., more soil erosion due to use o f mechanical equipment, or more damage to water resources and biodiversity due to greater use o f agro-chemicals), while reducing other forms o f resource degradation (e.g., less soil nutrient depletion as a result o f greater ability to purchase fertilizers or greater ownership o f livestock andrecycling o f manure) (Swinton, et al., 2003). Ifthere are imperfect labor and landmarkets, households with access to more family labor relative to their land are likely to use more labor-intensive and less land-intensive farming practices, such as shorter fallow periods or no fallowing, farming on steep slopes, and tilling more frequently, all o f which could contribute to land degradation. On the other hand, households with surplus labor (relative to land) may adopt labor intensive practices that lead to better NRM. Example o f these practices are applying manure or mulch, investing in soil and water conservation measures, etc. (Scherr and Hazel11994; Tiffen, et al. 1994). In an imperfect market setting, the nature of poverty is also important in determining its impact on natural resource management and degradation. Households that are not poor by welfare criteria such as minimum levels o f consumption may still face "investment poverty" that prevents them from makingprofitable investments inresource conservation and improvement (Reardon and Vosti 1995). Households that lack access to roads and markets, or that own little land may deplete soil nutrients less rapidly since they are subsistence-oriented and thus export less soil nutrients in the form of crop harvest and sales. On the other hand, households that are livestock poor may deplete soil nutrients more rapidly because they lack access to manure. A recent study o f determinants o f soil nutrient depletion in eastern Uganda found support for these hypotheses o f divergent effects o f different types o f assets (Nkonya, et al. 2004b). Inthis research, the linkages betweenpoverty andNRMare investigatedby examining the impact o f various types o f poverty on private land management, soil erosion and soil nutrient depletion, agricultural productivity, and income. The focus is on private land management because private land i s the most important natural resource to most rural households in Uganda. The problem o f landdegradation on private land inUganda is severe, and the linkages between poverty and private landmanagement may be very direct. This is not to say that linkages between poverty and management of other natural resources are not important; some o f these linkages are investigated by Nkonya, et al. (2004a), which investigated the impacts o f poverty on community 5 level natural resource management decisions, and found some support for the hypothesis that poverty contributes to poor NRMat the community level. Poverty can be defined in many ways, and has many dimensions. Typically, economists study income or consumption poverty, but poverty may also be measuredby lack o f assets, lack o f access to infrastructure and services, lack of education, or other factors that determine a household's livelihood status. Among the poor, the meaning o f poverty also differs widely, depending on their livelihoods and endowments of physical, human, natural and financial capital. The Uganda Participatory Poverty Assessment Process (MFPED, 2002) defines poverty as lack of basic needs and services (food, clothing, and shelter), basic health care, education and productive assets. Poverty may also include lack o f democracy or power to make decisions that affect the livelihoods o f the poor and social exclusion. For the case of farmers in northern Uganda, poverty also includes insecurity and internal displacement. Inthis study, a broad definition o f poverty has been considered, focusing on the impacts o f limited endowments o f physical, natural, human and financial capital, as well as poor access to infrastructure and services. Investigation o f the impacts o f other more political or social components of poverty such as lack o f democracy and power, social exclusion, insecurity and internal displacement, was beyondthe scope o fthe study. 3. EmpiricalModelsandData The objective i s to analyze the impacts of different aspects o f poverty on landmanagement practices and the impacts on crop productivity, household income, and measures o f land degradation. This is done by using an empirical model based on the sustainable livelihoods framework (Carney 1998) and literature on agricultural household models (Singh, et al. 1986; de Janvry, et al. 1991). Inthe theoretical framework, it is assumedthat ruralhouseholds make choices about labor allocation, land management, input use, and savings and investmentto maximize their discounted expected lifetime welfare, subject to the factors that determine their income opportunities, constraints and risks, including their endowments o f physical, human, natural, and financial capital, land tenure, agro-climatic potential, population pressure, commodity and factor prices, and access to markets, extension and other services. Under standard assumptions used inthe dynamic programming literature (e.g., Stokey and Lucas 1989), this life-cycle decision problem reduces to a series of decision problems in each year, in which the household decides what i s best to do inthe current year baseduponthe endowments and information that it has at the beginningo f the year and its expectations about how the decisions it makes will affect current consumption and the value o f endowments that it will carry over to the next year.' These decision problems imply that current decisions about labor allocation, land management, input use and investments will dependuponthe endowments of different types ofcapital that the householdhas at the beginningof the year, and other factors influencing the household's income potentials and risks in the present and future. The empirical models that are estimated inthis paper are based upon such a dynamic householdmodel.' 'Thisis a verbal statement o fthe Bellmanprinciple o f dynamic programming (Stokey and Lucas, 1989). See Appendix 12 for the specification o f the theoretical dynamic householdmodel and derivation of the empirical models usedinthis paper. 6 Response and outcome variables The particular interest o f this study is to know how different types o f capital and access constraints (as measures o f different types of poverty) influence household decisions on labor use, landmanagement practices anduse of agricultural inputs and implications for productivity, income and land degradation. The major landmanagement practices and inputs that we analyze are those that are sufficiently common among survey respondents to be investigated empirically. These include application o f organic matter (plant residues and animal manure) and inorganic fertilizer, use o f short term soil and water conservation (SWC) practices, crop rotation, slash and burn, and use o f purchased seeds. The short-term SWC practices include trash lines, deep tillage, zero tillage, andplowing andplanting along contour lines. The impacts of land management decisions on the value o f crop production per acre (and hence indirectly on income) are investigated thus quantifying some o f the linkages from land management to poverty. Total value o f crop production was measured as the product o f quantity produced and village level price, aggregated over the two seasons, per acre cultivated. Area cultivated was derived as the weighted average for both seasons. The impacts o f endowments on crop production per acre and household income per capita in reduced form are also investigated, throughwhich the total effects o f asset holdings on income poverty (via impacts on labor use, land management and input use) canbe assessed. As indicators o f land degradation, the focus is on soil erosion and soil nutrient depletion, which are among the most severe forms o f land degradation inUganda. The severity o f estimated soil erosion using the revised universal soil loss equation (RUSLE) (Renard, et al. 1991), and soil nutrient depletion by computing the soil nutrient inflows, outflows and balances are also investigated (Smaling, et al. 1993). Soil nutrient flow is defined as the amount o f plant nutrients that flow in and out o f a system or area during a specified time period (one year inthis case). The difference between soil nutrient inflow and outflow i s referred to as "nutrient balance." Nutrient flows and balances may be measured at different scales, such as at the plant, plot, household, water catchment, village, district, national, or higher level (Ibid.). The study measures soil nutrient flows and balances at the plot level since there are wide variations across plots in soil nutrient balances, and it i s at this level that actual impacts on sustainability o f landuse will be most evident3 Determinants of responses and outcomes The analysis is centered on land management since land is the major resource for the livelihoods o f the poor. A large body o f past research shows that the major determinants o f land management include households' endowments o f different types o f capital, land tenure, and the biophysical and socio-economic environment inwhich rural households live (e.g., see Reardon and Vosti, 1995; Barrett, et al., 2002; Nkonya, et al., 2004b). The capital endowments are the constraints in the welfare maximization model presented in Appendix 12. As noted earlier, due to imperfect or missing markets o f these capital goods and services, household land management decisions may differ depending on the levels o f their capital endowments. For instance, holding all else constant, households with abundant labor but with land scarcity are likely to invest more labor For details on estimationof household andplot level soil nutrient flow andbalances, see Kaizzi et al., 2004. 7 on their small land parcel thanthe case o fhouseholds with large farms ifland and labor markets do not function perfectly (e.g. see Feder 1985). Specifically, the capital endowments that may influence land management practices (depending also on nature o f markets) include: Natural capital: The natural capital endowment that i s considered inthis research is mainly land, which includes the amount o f land owned, the quality o f the land - measured as topsoil depth, the stock o f macronutrients (nitrogen, phosphorus, and potassium) and average slope, and the presence o f prior land investments on plot. Most past studies consider land endowment as only farm size since it is difficult and expensive to measure quality o f land. As noted earlier, one way inwhich this study contributes to the literature i s its use o fbetter data on landquality indicators. The topsoil is a storehouse o fplant nutrients (Sanchez, et al., 1997). Hence, infarming systems where farmers apply a limited amount o f inorganic fertilizer as i s the case in Uganda, topsoil depth largely determines soil quality (Ssali, 2002). The measure o f land quality i s enriched by including the stock o f macronutrients, which i s a more specific measure o f soil fertility. Also included is the slope o f plot since it measures the potential for soil erosion, which accounts for a large share o f nutrient loss (Wortmann and Kaizzi, 1998). Land investments - such as soil and water conservation structures, agroforestry, etc -- also can improve soil moisture holding capacity and fertility (Sanchez, et al., 1997), hence canincrease landquality. The impacts o f natural capital on land management decisions may be mixed. As noted earlier, farmers who own more landmay farm the land that they own less intensively iffactor markets are imperfect, andhence be less prone to invest inlabor andinput intensive land management practices. On the other hand, greater land ownership may increase households' ability to hire labor or purchase inputs by increasing their access to credit (Pender and Kerr 1998). The need to invest in intensive SWC practices will be greater on steeper soils, but the costs o f such investmentsmay be higher or the returns lower ifslopesareverysteep. Thebenefitsofinvestinginfertilizermaybeloweronmorefertile soils, unless there are complementarities between different types o f nutrients or between organic practices and use of inorganic fertilizer (Palm, et al. 1997). The presence o f land investments such as SWC structures maypromote greater use o f inputs such as fertilizer, by increasing the retum to such inputs (e.g., because they conserve soil moisture which may be complementary to fertilizer, seeds or other inputs). On the other hand, such structures may reduce the need for inputs (since less may be lost through erosion). Furthermore other types o f land investments may be oriented more to livestock or other production (e.g., paddocks, fishponds) andthus may tendto reduce farmers' use o fcrop inputs. Clearly, the theoretical impacts o f natural capital endowments on land management practices are ambiguous, and empirical research is needed to identify the actual impacts ina particular context. Since the impacts o fnatural capital on land management are theoretically ambiguous, impacts on land degradation will also be ambiguous. The same can be said regarding the impacts of most other endowments as well. (ii) Physical capital includes the value o f farm buildings, equipment andother durable goods, number of livestock, etc. As with naturalcapital, these assetsmayhavemixedimpacts on 8 land management. Ownership o fmarketable assets ingeneral increases the household's ability to finance investments andpurchase o f inputs, whichmay favor use o fpurchased inputs such as inorganic fertilizer. Onthe other hand, ownership o f livestock will increase the supply of manure available to the household, which may substitute for purchased inorganic fertilizer. Farmequipment may increase the productivity of labor incrop production, thus increasing the demand for labor, or may substitute for labor. Farm equipment and durable goods such as a bicycle or motorcycle may promote use o fbulky organic inputs by makingthem easier to transport andincorporateinto the soil, or may reduce use o f such inputs by increasingthe opportunity cost o f the farmer's labor. (iii) Human capital includes assets embodied in people's knowledge and abilities, such as education, experience (measured by primary livelihood strategy), sex, training, and the quantity o f labor endowment. These affect farmers' ability to make land management decisions. For example, due to imperfect labor markets, households that are well endowed with family labor are more likely to use labor intensive land management practices. Likewise, an experienced farmer knows well the biophysical and socio-economic environment to an extent that she makes informed decisions on land management. Holding all else constant, a better educated household head is likely to better collect and interpret extension messages, hence more likely to adopt improved landmanagement practices where these are being promoted by extension and suitable to the farmer's needs. On the other hand, education offers alternative livelihood strategies such as non-fann activities, which may increase labor opportunity costs and compete with agricultural production (Scherr and Hazell, 1994). Since education o f all household members may matter, and not only the education of the household head (Joliffe 1997), and since there may be differences in impacts o f female vs. male education on agricultural activities (Pender, et al. 2004), education is represented using the level education o f men and women in the household separately. (iv) Financial capital includes household liquid financial assets and access to credit. Access to financial capital is measured by whether farmers participate in rural credit and savings organizations. Limited access to credit has been cited by many studies as one o f the constraints to improved land management (Sharma and Buchemrieder, 2002; Fafchamps, 2000; Fafchamps and Minten, 1999). Lack o f access to financial capital may limit fanners' ability to purchase inputs such as fertilizer or to hire labor, and may limit their ability and incentive to invest inland improvements by causing households to have highdiscount rates (Pender 1996; Holden, et al. 1998; Pender and Kerr 1998). On the other hand, access to financial capital may enable households to invest more in non-fann activities and increase their opportunity cost o f labor, thus possibly reducing their interest in investing in agricultural production and land management activities (Pender and K e n 1998), especially iftheprofitabilityoftheseactivitiesislow. In general, household capital endowments have ambiguous impacts on land management, crop productivity and land degradation, depending on the nature o f market imperfections, as discussed inthe previous section. However, all endowments that require household investment are expected to contribute to higher household income (since this is part o f the reason why households invest in them), though larger household size may lead to lower income per capita if there are 9 diminishing retums to additional labor inthe household, or because larger households tend to have a higher share o f dependents. Land tenure relationships also can have important influence on land management decisions and agricultural productivity. Iflandtenure is insecure, this will tend to reduce farmers' incentive to invest inland conservation and improvement, since the returns to such investmentswill be at risk (Feder, et al. 1998; Place and Hazel1 1993; Besley 1995; Gavian and Fafchamps 1996). Tenure insecurity can also reduce farmers' ability to invest in land improvement and inputs, since it reduces the collateral value o f land and thus farmers' access to credit (Ibid.). The collateral value of land will also be reduced or even eliminated ifthere are restrictions on the transferability o f land (Pender and Kerr 1999). Transfer restrictions or imperfections in land markets can also inhibit investments in land improvement because farmers may be unable to recoup the value o f their investments by selling land assets, causing land investments to be irreversible investments, thus increasing farmers' option value o f waiting to invest inthe presence o f uncertainty (Fafchamps and Pender 1997; Pender and Kerr 1999). These arguments implythat landinvestment and adoption o f purchased inputs should generally be greater on freehold land that is fully titled, with secure and full rights to transfer and mortgage as well as use land, than on customary land that has more limited rights, or on leased or occupied land subject to greater insecurity and more limited rights. However, there are also theoretical and empirical counter-arguments, Often customary land i s quite secure in terms of use and bequest rights, and land titling efforts can actually increase rent seeking and hence tenure insecurity (Atwood 1990; Platteau 1996). Tenure insecurity may be associated with greater incentives to invest, if investment will help to increase tenure security (Besley 1995; Otsuka and Place 2001). Furthermore, land management may be influenced by regulations, community norms and responsibilities to manage the land sustainably as well as by farmers' formal rights, and these may be more influential in affecting management o f customary land than freehold or leasehold land. In Uganda, there are four major land tenure types: customary, mailo, freehold and leasehold. Each land tenure system i s associated with its own land rights and obligations and different degrees o fpermanence and security o f land rights (Republic o f Uganda, 1998). (a) Customary land tenure is the most common land tenure system in Uganda and is regulated by customary rules. Under customary tenure, an individual, family or traditional institution may occupy a specific area o f landas prescribed by the customary laws. Customary tenure often involves limitations on the individual's right to sell or mortgage land, though usufruct and bequest rights are usually fairly secure. Customary tenure may also carry informal obligations concerning land use and management that do not influence other tenure categories. Under the 1998 Land Act, customary landholders may apply for a certificate o f ownership from the District Land Board. Once such a certificate is issued, the land holder(s) may lease, mortgage, sell, sub-let, give or bequeath by will the land or part o f it (Ibid). However, implementation o f the Land Act i s still limited. (b) Freehold land tenure allows the landholder to own the land for an unlimitedtime. The landholder can use the land for any lawful purposes; may sell, rent, lease or use it as collateral to get a loan from a bank, may allow other people to use it or may give or 10 bequeath it by will, and has the first priority to buy land from persons who are occupying hidher land (tenants by occupancy) and are willing to sell their land (Ibid). This form of tenure provides owners the most complete rights, with the least obligations or restrictions on use. This has ambiguous impacts on landmanagement, depending on the nature o f obligations existing under other tenure systems. (c) Leasehold land tenure i s a form o f tenure created either by contract or by operations o f law. Under this system a person referred to as the tenant or lessee, occupies land through an agreement between h i d e r s e l f and the owner o f the landreferred to as the landlord or lessor. Under this system, the landlord allows the tenant to use the land for a specific period, usually five, forty-nine or ninety nine years without any disturbance by the owner as long as the lessee abides with the law. The tenants are usually but not necessarily required to pay rent or premiums or both or may be asked to render services (Ibid).The lesseemay change a lease ownership to freehold, can sell, sub-let, mortgage, give or bequeath by will the land for the period he or she is entitled to hold the land. The rights under leasehold are similar to those under freehold, except that the term i s limited. Where long-term leases are involved, the landmanagement of leasehold land i s therefore likely to be similar to management o f freehold land. (d) Mail0 land tenure is a system where the landholder owns the land forever inthe same way as a freehold owner. However, inmost cases, mailo land has longbeen occupiedby long-term occupants. The 1998 Land Act recognizes and protects the rights o f lawful and bonafide occupants4 o f that land as well as their improvements on that land. The landholder may lease, mortgage, pledge or sell, give away or bequeath by will his interest in the land or part of it. However, the Land Act prohibits landholders from evicting bonafide occupants from land. Thus bona fide occupants have a substantial degree of tenure security. Nevertheless, they are often restricted by owners from making landinvestments, since this reduces the rights o f the absentee owner, who owns the land but not the developments made on the land by bonafide occupants. Thus, occupants of mailo land may be inhibited from investing in land improvement, even though they may have secure use rights to the land. Access to agricultural technical assistance services can increase adoption o f inputs and land management practices by increasing farmers' awareness o f and ability to effectively use new agricultural inputs and practices. The impacts of extension will depend on the type o f enterprises and technologies that are promoted, however, as well as the suitability o f these to the farmers' conditions. Thus, extension may have mixed impacts on agricultural production and land The LandAct, 1998 recognizes three types o f occupants on registeredlandnamely; the lawful occupants, the bona fide occupants and the non bonafide (unlawful) occupants. The lawful occupant is a person who entered the land with consent o f the registered landholder or a personwho occupies land by virtue o fthe repealedbusuulu and envujo law o f 1928; or the Tooro or Ankole landlordand tenant law o f 1937. A bonafide occupant is a personwho before coming into force o fthe constitution had occupied or utilized or developedany landunchallengedbythe registeredowner or agent o fthe registered owner for twelve years or more. A bonafide occupant may also be a person settled on landby government or agent of the government, which may include a local authority. The unlawful occupant is the one who does not qualify as a lawful or bonafide occupant but holds landunder unlawful means. 11 management practices, depending on the approach and emphasis o f the program. Inthis study, households who are participating inthe traditional government agricultural extension programs are distinguished from those participating in the new extension approach, the National Agricultural Advisory Services The new extension approach is more demand-driveninnature than the traditional approach, emphasizes development of farmer organizations and promotion of new commercial agricultural enterprises that are expected to be more profitable for farmers than traditional production. The likely impacts on land management are not clear, since land management i s not a major emphasis o f the approach, although to the extent that more profitable cash crops are adopted, one could expect this to promote greater adoption o f purchased inputs such as seeds and fertilizer, and greater labor intensityincrop production. Inadditionto householdlevelcapital endowments, landtenure, andparticipationin technical assistanceprograms, there are other factors that affect landmanagement at village, regional, national or intemational level. Village or higher level factors that determine local comparative advantages and hence the profitability of labor use, landmanagementand input use include agro-ecological conditions, access to markets, infrastructure, localwages, andpopulation density.Rainfallregimes andother biophysical factors greatly influence farming systems and land management. Likewise, infrastructure development heavily influences farmer decisions on land management since it affects localprices, availability of inputs andmarket information, and other socio-economic aspects. Local wage levels reflect the scarcity of labor and can thus affect the labor intensity o f agricultural production, as well as affecting households' ability to finance purchase o f inputs. Controlling for wage levels, farm size and household size, population density reflects mainly scarcity of natural resources at the community level, since household level scarcity i s reflected by household endowments. This may influence land management on private land to the extent that there are interactions between use of common and private land. For example, greater scarcity of communal fuelwood or fodder supplies indenselypopulated communities may cause households to rely more on animal manure and crop residues for fuel and fodder, thus limiting the ability of farmers to apply such inputs to their private cropland. Below i s a description o fthese factors andhow they were measuredindetail. Agro-climatic zones: There are several classifications o f agroecological, agro-climatic and farming systems inUganda. The distinction among these classifications is fairly fuzzy. Kyamanywa (1987) and the Ministry o f Natural Resources (1994) divided Uganda into eleven ago-climatic zones and twenty ecological zones while Semana and Adipala (1993) identified four agro-ecological zones (AEZ). A study by Wortmann and Eledu(1999) divides Uganda into 33 agroecological zones that depict a detailed representation o f natural resource endowment and will therefore be used in this study. However, AEZ by Wortmann and Eledu fall into eleven major categories that are spatially representedinFigure 1. Below, are the six agro-climatic zones that were covered inthis study: (i) The Lake Victoria Crescent zone has a high level o f rainfall distributed throughout the year ina bimodal pattem ("bimodal highrainfall") and i s characterized by the dominant The old extension approach usedlocal government employed extensionworkers, who are still active inthe non- NAADS and to some extent inthe NAADS sub-counties. 12 banana-coffee farming system. The zone runs along the vicinity.of Lake Victoria from the east in Mbale district, through the central region to Rakai district in southwestern Ugandaalong the shores of LakeVictoria. Northwest farmland: This area is characterizedby unimodal low to medium rainfall and covers the west Nile districts o f Arua, Nebbi and Yumbe. Common crops grown in the zone are coarse grain (sorghum, millet, bulrush, etc), maize, tubers, andtobacco. North-moist farmland: This zone is also characterized by unimodal low to medium rainfall and covers most of the northern districts. The common crops grown are coarse grain, maize, tubers, cotton, anda variety o f legumes. Mount Elgon farmlands: This zone is on the slopes o f Mount Elgon in the east and is characterized by unimodalhighrainfall, highaltitude and hence cooler temperatures and relatively fertile volcanic soils. The only district covered by the survey in this zone is Kapchonva. The major crop in Kapchonva is maize. Farmers in this zone also plant bananasandraise livestock. Southwestern grass-farmland: This zone receives medium to low rainfall in a bimodal distribution. The only district covered by the survey in this zone i s Mbarara. The commoncrops inthe district are banana, coarse grains, maize andtubers. Many farmers inthe district alsokeepalargenumberoflivestock. Southwestern highlands (SWH) zone. This zone receives bimodal highrainfall and has high altitude, hence cooler climate, and relatively fertile volcanic soils. The common crops in the SWH are bananas, Irish potatoes and other tubers, sorghum, maize, and vegetables. Market and road access: The geographic coordinates o f the survey households were linked to geographic information on indicators o f market access and population density. Areas of relatively highmarket access were classified by Wood, et al. (1999) using the PotentialMarket Integration (PMI) index, an index o f travel time o f each location to the nearest five markets, weighted by the population size o f those markets (a higher value of PMIindicates better market access). The areas classified as havingrelatively highmarket access include most o f the Lake Victoria crescent region and areas close to main roads in the rest o f the country (Figure 2). Access to roads was classified based on information from the community survey on the distance o f the community to an all- weather road. Access to markets and roads are expected to favor adoption o f purchased inputs, by increasingtheir availability andreducing their costs relative to farm level commodityprices, andby favoring commercial production of higher value crops. Better access to markets androads are also expected to contribute to higher value o f crop production and higher incomes per capita, the latter both by increased value o f crop production as well as increased opportunities for other sources of income (e.g., non-farm activities, livestock production). The impacts on adoption o f labor or land intensive land management practices, however, i s ambiguous, since market and road access can increase the opportunity costs of labor and land, as well as increasing the marginal returns to labor andlandinputs. The impacts on landdegradation are also, therefore, theoretically ambiguous. 13 Population density and wage rates: The population density data were collected in the IFPRI- UBOS survey by measuringthe area of LC1andasking community leaders to report the number of people inthe LC1. As mentioned above, this variable reflects community level scarcity of natural resources, since household endowments are also being controlled. Greater scarcity o f resources may constrain households from using some organic land management practices, but may also promote greater investment in resource improvement at the household level. Local average agricultural wage rates inthe study communities were also included inthe analysis, as indicators of the scarcity of unskilled casual labor. Higher local wages are expected to contribute to lower labor intensity and less adoption of labor intensive land management practices, while they may promote greater use o f purchasedinputs by increasing households' access to cash. 14 n Figure 2. Classificationof Market Access in Uganda Source: Wood, et. al. (1999) 16 Data analysis Since there are considerable differences in how farmers manage land depending on the characteristics of specific plots, land management practices, crop productivity, soil nutrient flows andbalances at plot level are analyzed. Only householdincome is analyzed at household level since it is anaggregationo fall sources o fincome -farm andnon-farm. Descriptive analysis: Before turning to the determinants o f soil fertility management, the severity of soil nutrient depletion will be analyzed using descriptive data analysis. Even though knowingthe impact o f soil nutrient depletion on crop yield i s more important thanjust quantifying the depletion, there are no studies known to the authors that have measured agricultural productivity loss due to soil nutrient depletion inUganda. Therefore a simpler measure i s usedto estimate this impact. This measure is called the economic nutrient depletion ratio (ENDR) (der Pol, 1993). ENDR i s the share o f farm income derived from miningsoil nutrients.6 Soil nutrient miningis the practice o f growing crops with insufficient replacement o fnutrients taken up by crops. Mathematically, ENDR=NDMv x 100GM where: (NDMV) is nutrient deficit market value, which is the value o fnutrients minedper hectare ifsuchnutrients were to be replenishedby applying fertilizer purchased from the cheapest sources. GMis the gross margino fthe household from agricultural activities per hectare. ENDR measures the cost o f replenishing nutrient depleted relative to farm income, and not the benefit. Holding other factors constant, decreasing fertilizer prices will both increase returns to use of fertilizer and reduce ENDR. Econometric models: The econometric models estimated and the approach used to address the econometric problems arising inthe estimation i s presented inAppendix 13. 4. Discussion of Results DescriptiveResults Appendix 2 shows that only about 10% o f plots received plant or animal organic matter while around 20% were fallowed or had crop rotation. Use o f inorganic fertilizer is even lower as only about 9% o f the plots sampled received fertilizer, at an average rate 48 kg/acre on plots that received fertilizer. Inorganic fertilizer i s used mainly by large scale plantation farmers who account for 95% of fertilizer consumption in Uganda (NARO and FAO, 1999). The remaining 5% is accounted for by small scale farmers - mainly maize producers inKapchonva and tobacco farmers inthe west Nile. The majority o f smallholder fertilizer users inthe rest o fthe country use fertilizer Farm income includes income from crop, livestock and other agricultural activities. It excludes income from non-farm activities, transfers, etc. 17 on small plots planted with vegetables or other high value crops. Adoption o f soil and water conservation (SWC) measures i s also low, as only about 13% plots were affected by short-term SWC practices (including trash lines, deep tillage, zero tillage, and cultivation along contour lines). The results show the low level of use of organic landmanagement practices and even lower rate o f use o f inorganic fertilizer. The low adoption of improved soil fertility management technologies has important implications for soil nutrient depletion. Table 1 shows that the major sources o f nitrogen inflow are organic matter and BNF. Inorganic fertilizer contributes only 1% of N inflow. Plant organic matter is the major source of phosphorus while animal manure is the major source o fpotassium(Table 2 and 3). Table 1: Major sources of nitrogen inflows and channels of outflows at plot level Nutrient flow Nw NorthmoistMt. Elgon SW grass- Lake S W H All farmlands farmlands farmlands farmland Victoria zones crescent Total inflows (kidha) 13.79 18.79 25.58 25.38 19.53 12.13 18.05 % contributionto total inflow Inorganic fertilizer 5.OO 0.00 11.00 0.00 0.00 0.00 1.00 Plant organic matter 0.00 0.00 16.00 4.00 11.00 0.00 5.00 Animal manure & droppings 22.00 46.00 26.00 54.00 26.00 23.00 35.00 BNF 38.00 27.00 28.00 27.00 41.00 38.00 33.00 Atmospheric deposition 34.00 27.00 19.00 15.00 23.00 39.00 25.00 Total outflows (kg/ha) 55.00 75.89 116.75 132.56 114.38 137.00 104.20 % contributionto total outflow Crop harvest 33.00 21.oo 38.00 54.00 56.00 17.00 37.00 Animal grazing 26.00 41.00 24.00 22.00 4.00 1.00 15.00 Leaching & denitrification 21.00 29.00 13.OO 12.00 15.00 13.00 17.00 Soil erosion 20.00 8.00 25.OO 11.00 24.00 69.00 31.00 Table 2: Major sources of phosphorusinflows and channels of outflows at plot level Flow sources and channels Nw North' Mt.Elgon SWgrass-Lake S W H All farmlands moist farmlands farmland Victoria zones farmland crescent Total inflows (kgha) 1.30 1.74 4.09 4.00 3.37 1.51 2.46 % contribution to total inflow Inorganic fertilizer 10.00 0.00 25.0 0.00 0.00 0.00 3.00 Plant organic matter 0.00 0.00 28.00 12.00 37.00 1.00 17.00 Animal manure & droppings 30.00 52.00 28.00 73.00 42.00 47.00 50.0 Atmospheric deposition 60.00 48.00 19.00 16.00 22.00 52.00 31.00 Total outflows (kgha) 10.06 7.77 20.32 12.84 16.94 41.25 18.09 % contribution to total outflow Crop harvest 29.00 29.00 20.00 46.00 37.00 6.00 22.00 Animal grazing 17.00 42.00 19.00 24.00 3.00 0.00 9.00 Soil erosion 55.00 30.00 60.00 30.00 59.00 94.00 69.00 18 Table 3: Major sources of potassiuminflows and channelsof outflows at plot level Flow sources and channels NW Northmoisthlt. Elgon SW grass-Lake SWH All zones farmlands farmland farmlands farmland Victoria crescent Total inflows (kgha) 6.01 12.40 10.33 13.25 15.73 4.36 10.45 % contribution to total inflow Inorganic fertilizer 2.00 0.00 11.00 0.00 0.00 0.00 1.oo Plant organic matter 0.00 0.00 15.00 15.00 66.00 4.00 27.00 Animal manure & droppings 46.00 73.OO 44.00 66.00 16.00 25.00 44.00 Atmospheric deposition 52.00 27.00 30.00 19.00 18.00 72.00 29.00 Total outflows (kdha) 46.99 50.23 124.83 202.37 111.32 303.29 141.33 %contributionto total outflows Crop harvest 29.00 24.00 42.00 69.00 62.00 6.00 34.00 Animal grazing 30.00 65.OO 20.00 15.00 5.00 0.00 11.00 Leaching- 0.00 0.00 0.00 . 0.00 0.00 0.00 0.00 Soil erosion 41.00 11.oo 37.00 16.00 33.00 94.00 55.00 Crop harvesting i s the major outflow for N, contributing over one third o f total nutrient outflows. Soil erosion i s the most important channel o f outflows for both P and K, accounting for more than one half o f the total outflow. Soil erosion i s an especially important outflow for phosphorus as it contributes over two thirds o f its total outflow. This i s partly due to the fact that P does not leach significantly. The amount of P lost through crop harvest i s the lowest of the three macronutrients. These results underscore the low-extemal input agriculture practiced in Uganda and the consequent severe depletion o f soil nutrient stocks. In most plots surveyed, the total nutrient outflow exceeds total nutrient inflow. Only about 20% of plots had positive nitrogen or potassium balances, but about a quarter o f the plots had positive phosphorus balances (Table 4). The Lake Victoria Crescent region has the second largest rate o f nitrogen depletion after the southwestem grassland AEZ (Mbarara) and the second largest rate o f phosphorus depletion after the southwestem highlands (Kabale). The average amount o fnitrogendepleted inall regions during the 2002 cropping seasons was about 2% of total nitrogen stock in the top 20 cm. o f the soil (most critical zone for crops), which includes both the available and inert stocksa7 The corresponding averagerate of nutrient stock depletion for phosphorus andpotassium are 0.5% o f extractable P and 1%ofexchangeableK inthe top 20 cm. ofthe soil, respectively. A total nutrient stock is a sumof the inert nutrients that are not readily available and the soluble stock, which is readily available to plants inthe short term. The inert stock establishes a stable equilibrium with the soluble solution, whereby inert stocks dissolve and become available to plants over a long period of time, depending onthe parent material, weather condition and soil physical, biological and chemical characteristics. 19 Table4: Severityof soil nutrientdepletionanditseconomic magnitude NW North Mt. Elgon SW grass- Lake S W H All zones farmlandmoist farmland farmland Victoria farmland crescent Nitrogen Nutrient balances (kg/ha/year) -35.55 -53.11 -70.01 -99.22 -82.19 -73.18 70.60 % o fplots withpositive balances 21.16 19.17 22.58 14.73 14.75 28.40 20.14 Nstock (kg/ha) 1944.2 2897.0 6017.3 3842.0 3700.5 4746.1 3695.0 Nbalanceas %oftotalNstock 1.83 1.83 1.16 2.58 2.22 1.54 1.91 NDMV (US$)/farm' 66.17 139.06 106.50 190.41 145.16 75.65 124.80 ENDR~(%) 12.0 23.0 6.0 13.0 11.0 6.0 11.0 Phosphorus Nutrientbalances (kghdyear) -6.29 -4.97 -8.01 -7.33 -9.29 -18.55 -9.98 % ofplots with positivebalances 25.19 26.11 33.45 26.94 19.32 32.16 26.41 P stock (kg/ha) 1160.2 1412.1 3127.8 1655.2 1828.7 2759.8 1916.5 Nbalance as %oftotal Pstock 0.54 0.35 0.26 0.44 0.51 0.67 0.52 NDMV (US$)/farm' 13.21 14.69 13.75 15.88 18.53 21.62 19.91 ENDR' (%) 2.00 2.00 1.oo 1.oo 1.oo 2.00 2.00 Potassium Nutrientbalances(kg/ha/year) -31.97 -34.17 -81.25 -172.95 -78.75 -143.70 -94.85 % ofplots with positivebalances 23.11 30.53 14.42 15.50 14.10 30.70 22.99 K stock (kgha) 4207.5 3407.2 11992.6 10888.4 6560.1 18579.9 9618.9 Kbalance as %oftotal Nstock 0.76 1.oo 0.68 1.59 1.20 0.77 0.99 NDMV (US$)/farm' 30.71 46.17 63.79 171.30 71.79 76.56 86.54 ENDR~(%) 5.56 7.67 3.75 11.29 5.26 6.32 7.78 All Nutrients (N,P,K) Nutrientbalance (kgha) -73.82 -99.48 -159.27 -279.50 -178.10 -235.53 -178.80 Nutrientbalance as % of stock 1.01 1.29 0.75 1.71 1.47 0.90 1.17 % ofplots with positive balances 19.14 17.99 20.00 13.18 11.23 26.58 18.05 ENDR* (%) 19.94 33.21 10.82 24.90 17.25 14.34 20.80 1. Nutrient Deficit Market Value (NDMV) is the value o f nutrients mined per hectare if such nutrients were to be replenished by applying purchased fertilizer (der Pol, 1993). 2. Economic Nutrient Depletion Ratio (ENDR)is share (%) o f farmers' income derived from miningsoil nutrients (Ibid). Eventhough the depletion rates are 1.2% for all nutrients combined, this does not mean that the nutrient stocks would be depleted inless than 100 years. Firstly, the inert stocks are not readily available in a short term; hence their depletion rates are much slower. The amount depleted comes mainly from the soluble component o f the nutrient stock. Secondly, as crops deplete nutrients, their yields decline exponentially, decreasing the rate o f depletion since crop harvest i s the leading channel ofnutrient outflow. Evidence o f declining yields and soil fertility inUganda since the early 1990's (Deininger and Okidi 2001; Pender, et al. 2001) supports the hypothesis that soil fertility declines are causing yield declines. Thirdly, the regeneration o f soils from parent material is not included as a nutrient inflow. Finally, nutrient stocks below the top 20 cm. o f soil are not being 20 included, which can be available to deeper rooting crops and trees, or as a result o f fallowing or deep tillage. One measure o f the economic magnitude o f the loss o f soil nutrients is the economic nutrient depletion ratio (ENDR), which measures the share o f farm income that would be required to replenish the lost nutrients usingthe cheapest available fertilizers (van der Pol 1993). If farmers were to buy the cheapest source o f nutrients to replenish the nutrients depleted, the average cost o f fertilizer bought would be equivalent to one fifth o f the total household farm income in the eight districts studied.' Dueto the low farm income inthe northern moist farmland, farmers inthis AEZ would have to use more than a third o f their farm income to replenishminednutrients, as compared to only about 11% for the case o f the Mt. Elgon farmers who have greater income and practice better soil fertility management practices. The nutrient requiring the largest cost to replenish i s nitrogen, followed by potassium. These results show the heavy reliance of smallholder farmers on mining soil fertility. Using a fifth o f farm income to avoid nutrient depletion would be very difficult for most farmers, who depend on agriculture as their primary source o f income. This begs the question of what could be done to help farmers to practice sustainable landmanagement, which is the focus o f the next section. Household farm income includes only income from the farm enterprise, and excludes non-farm income, gifts, and other forms o f transfers. Appendix 2 shows that the average household income in2002/03 was Ush3.04 million, which i s about US$1788. 21 Econometric Results Determinants of land managementpractices Household ownership o f physical assets has mixed impacts on land management practices (Appendix 3). As expected, larger farms are more likely to fallow since they have enough land for crop productionwhile resting part o f their land. Larger farms are less likely to use short-term SWC measures such as trash lines, deep tillage, and zero tillage, and less likely to incorporate crop residues on a given plot. These results are consistent with Boserup's (1965) theory o f agricultural intensification and the findings o f Tiffen, et al. (1994), conceming the impacts o f population pressure on intensity o f land use and propensity to invest in SWC measures, but are contrary to a long-term study in Kabale district, which found that fallowing increased with population pressure (Lindblade, et al. 1996). Controlling for farm size and other factors, population density has no impact on fallowing or other landmanagement practices, however. Greater ownership of livestock i s associated with less likelihood o f using crop rotation and fallowing. This is perhaps because crop rotation and fallowing are less necessary for soil fertility management if farmers own more livestock, because o f the soil fertility benefits o f manure. Households who own more farm equipment are more likely to incorporate crop residues, probably because mechanical equipment such as plows makes this practice easier to accomplish. The human capital of the household has mixed impacts on land management. Secondary education o f males i s associated with lower probability o f using slash and bum for clearing land, possibly because secondary education increases households' awareness o f negative impacts o f slash and burn or reduces their need to clear new land for cultivation. Post-secondary education o f males i s associated with greater likelihood o f practicing crop rotation but lower likelihood o f using SWC practices. These results may be due to higher opportunity costs of labor in more educated households, reducing adoption of labor-intensive SW C practices, while possibly encouraging crop rotation as a less labor intensive means o f addressing concerns about soil fertility, pests and weeds. Other aspects o f human capital, including the gender o f the household head and the size o f the household, have statistically insignificant impacts on land management practices. The livelihood strategy o f the household, measured by the primary source o f income o f the household head, has limited impact on most land management practices. Non-farm activity as a primarysource of income increases the probability to fallow relative to households for whom crop production i s the primary activity. This suggests that non-farm activities enable and encourage less intensive crop production, by providing households with alternative sources o f income and increasing the opportunity cost o f family labor. Having non-farm activity as a primary source o f income reduces the probability to use slash and burn, possibly because such households have less need to clear new land for production. N o statistically significant differences in landmanagement practices i s found between households whose primary income source is livestock vs. crop production. Natural capital has significant impacts on several land management practices. Farmers are more likely to incorporate crop residues and practice short-term SWC technologies on steeper slopes. This is probably because the need for and benefits o f SWC practices are greater on steeper 22 slopes. Crop rotation is more likely to be used on deeper soils. This suggests that farmers take advantage o f deeper andmore fertile soils to practice better managementto maximize returns since the response to better land management practices on more fertile soils may be higher (Kaizzi, 2002). Surprisingly, fallowing i s more likely to be practiced on plots with higher soil stocks of nitrogen (N), though this result is only weakly statistically significant (10% level). This may reflect reverse causality (fallowing causes higher soil N stocks). Insignificant impacts of soil nutrientstocks are found on other landmanagementpractices. Prior investments on the plot also influence land management. The presence of SWC structures such as stone bunds, terraces, grass or vegetative strips, and irrigation structures increase the probability that the farmer applies crop residues to the plot, probably because such structures reduce losses and/or increasethe retumto applying such inputsby conserving soil moisture (Pender and Kerr 1998). Fallowing and crop rotation are more common on plots where agroforestry (non- crop) trees have been planted, perhaps because o f adoption of agroforestry trees in an improved fallow rotation system. Other land investments (fish ponds, fences, paddocks and pasture improvement)also increase the probability to incorporate crop residues and practice crop rotation. Some of these investments which are associatedwith livestock management (fences, paddocks and pasture improvement), the availability of which can facilitate incorporation o f crop residues using ox-plowing. These results are consistent with results o fNkonya, et al., (2004b), who also observed that prior landinvestments influence current landmanagementpractices. All of the landmanagementpractices considered (slash and burn, fallowing, crop rotation, incorporation of crop residues, and SWC practices) are less likely on plots where perennials dominate than where annual crops dominate. Clearly, these are practices associated with productiono f annual crops. Access to markets, as measuredby the potential market integration (PMI), and access to all- weather roads have limited impact on most land management practices. However, better access to markets is associated with higher probability to adopt SWC practices, while slash and bum practices are more likely farther from an all-weather road. These results are consistent with the findings of Tiffen, et al. (1994) that better market access can promote more sustainable land managementby increasing the retumto labor and other inputs investedinthe effort. Nevertheless, the impacts o f market and road access on land management practices in Uganda are generally mixed (Nkonya, et al. 2004b; Pender, et al. 2004). Access to agricultural technical assistance services (measured by the number o f contact hours o f the household with agricultural extension agents and participation of household in the NAADS program) has statistically ,insignificant impacts on the land management practices considered. These programs are apparently focusing more on other technologies such as use of inorganic fertilizer. Consistent with this, it i s found inresults discussedinthe next section that use o f inorganic fertilizer i s more likely where access to these technical assistanceprograms is greater. Access to rural finance organizations has statistically insignificant impacts on most land managementpractices, except a negative impact on crop rotation. The negative association o f credit with crop rotationmaybebecausecredit is used to facilitate non-farm activities, rather than efforts to increase soil fertility and crop production. Consistent with this, participants in rural finance 23 organizations use less fertilizer and obtain lower crop productivity, but higher per capita income (findings discussed below). These findings suggest that credit constraints are not a major impediment to adoption of improved land management practices, and that access to credit may promote less intensive land management by facilitating more remunerative non-farm activities. Thisresult is similar to findings ofNkonya, et al. (2004b) andPender, et al. (2004). There are significant differences in some land management practices across different land tenure types. Slash and bum and incorporation o f crop residues are more common on plots under customary tenure than freehold plots, while use of SWC practices is less common on mailo than freehold plots. Customary tenure is associatedwith cerealproduction (Nkonya, et al. 2004; Pender, et al. 2004), which is probably the reason for its association with slash and bum and incorporation o f crop residues. Mail0 tenure i s associated with perennial crop production, which, as already noted, is associatedwith less use of short-term SWC practices. Other factors that significantly influence land managementpractices include the size o f the plot, distance of the plot from the household residence, and the agro-ecological zone/farming system. The impacts of such factors inthis report will not be emphasized, as they are static factors and not directly related to the issues of poverty and access to markets and services, which are the main focus o f this report. Use of inputs Use of farm inputs, including labor, fertilizer, and purchased seeds, i s influenced by many of the same factors as land management practices. Larger farms are less likely to use purchased seeds and use less labor per acre (Appendix 4). These results are consistent with the Boserup theory o f intensification and the findings o f Nkonya, et al. (2004b) and Pender, et al. (2004), and with the findingreported below that larger farms obtain lower value ofcrop productionper acre. Ownership of physical capital significantly and positively influences use o f several inputs. Greater ownership of livestock is associated with greater use o f labor per acre in crop production, perhaps because o f greater ability o f wealthier households to hire labor. Ownership o f farm equipment is associated with greater use of labor per acre and greater likelihood o f using organic matter. Farm equipment is apparently complementary to labor use, and likely helps intransporting and usingorganic inputs. Human capital has mixed impacts on use o f inputs. Primary education o f both males and females i s associatedwith greater labor intensity. Perhaps, households having more members with primary education have more young members who contribute to labor intensity incrop production. Female secondary education i s associated with less likelihood of usingpurchased seeds but greater likelihood of using organic inputs, while female post-secondary education i s associated with lower likelihood of using organic inputs. The negative impact of post-secondary education on organic inputs use is as expected, and likely due to the higher labor opportunity cost of more educated women. It is not sure why secondaryeducation has mixed impacts on organic inputs andpurchased seeds. Male education at all levels is strongly associatedwith higherprobability o f usinginorganic fertilizer. This may be because more educated farmers are more aware o f the benefits o f using inorganic fertilizer, or becausethey are better able to afford to purchase fertilizer. 24 Male headed households are less likely to use inorganic fertilizer but more likely to use organic inputs than female headed households. These results may reflect labor constraints facing female headed households, limiting their ability to apply organic materials, and causing them to rely more on inorganic fertilizer instead. However, findings show no statistically significant difference between male and female headed households in terms o f labor intensity. Larger households use more labor per acre than smaller ones (weakly significant), probably due to their greater supply. Household livelihood strategies have limited impacts on input use in crop production. Householdswith non-farm activities as the primary income source use labor less intensively incrop production, consistent with the findings reported earlier that they are more likely to fallow. There are no differences found inother input use associated with livelihood strategies. Natural capital also influences input use incrop production. Labor is used more intensively on steeper slopes, probably because more effort is required to farm on slopes. Less labor is used and use o f organic fertilizer is less likely on deeper soils, probably because organic inputs and the associated labor are less needed and thus yield lower retum on deeper soils. However, labor i s used more intensively on plots that have more soil nutrients, suggesting that the retum to investing labor in land management and crop production is greater on more fertile soils, consistent with the findings o f Kaizzi (2002) in eastem Uganda. Use o f organic inputs is also more likely (weakly significant) on soils that have greater stocks o f K; the reason for this is not clear. The presence o f land investments also influences input use, but in mixed ways. The presence of SWC structures i s associated with greater labor intensity (weakly significant), while agroforestry trees are associated with less labor intensity. SWC structures may require labor to maintain, and may also increase the retum to labor in crop production by increasing soil moisture and responsiveness to inputs. The presence o f non-crop trees likely reduces labor requirements, by reducing the share o f the plot requiring labor inputs for crop production. The presence o f other land investments such as fish ponds, fences, paddocks, and improved pasture reduce the likelihood o f using purchased seeds, probably because these investments promote livestock or aquaculture rather than crop productionon the plot. Investment in perennials also influences input use. Labor is used less intensively on perennials than on annual plots, although use o f organic inputs i s more likely on perennial plots. These findings are consistent with findings o fNkonya, et al. (2004b). Use o f purchased seeds and organic inputs i s less likely on plots more distant from the residence, probably because o f the costs o f transporting such bulky inputs (especially organic inputs). This result is similar to findings o f Nkonya, et al. (2004b). By contrast, the amount o f labor used per acre is greater on more distant plots. This is probably because the time involved in walking to such plots is included inthe labor inputs for managing them. Access to markets and roads has a positive impact on labor intensity, likely because this increases the return to labor invested in crop production. Surprisingly, however, use o f purchased seeds i s more likely (weakly significant) further from an all-weather road. 25 Participation in traditional agricultural extension is associated with greater labor intensity, while both traditional extension and the new NAADS program are associated with greater likelihood o f using inorganic fertilizer. Surprisingly, however, participation in rural finance organizations has a negative association with labor intensity in crop production and use o f inorganic fertilizer. This finding suggests that farmers use credit to engage in non-farm activities, which are likely to have higher retums than agricultural production, and that access to credit is not a bindingconstraint to labor or inorganic fertilizer use. Consistent with this explanation, inorganic fertilizer use is not very profitable for farmers in the study districts (results discussed in next section). Land tenure has statistically insignificant impacts on labor intensity. However, inorganic fertilizer is less likely to be applied on plots under customary tenure than freehold plots, while organic inputs are more likely to be applied to plots under mailo tenure. The positive association o f freehold tenure with inorganic fertilizer use is consistent with the hypothesis that land titles facilitate purchased input use by increasing access to credit (Feder, et al. 1988; Place and Hazel1 1993; Besley 1995), although the importance o f this impact is questionable given other evidence already discussed that rural finance appears not to be a bindingconstraint to inorganic fertilizer use. The association o f mailo tenure with organic inputs may be due to the association o f mailo tenure with banana production, for which use o f mulch and other organic inputs is relatively common. This finding is consistent with findings o fNkonya, et al. (2004b) andPender, et al. (2004). Surprisingly, labor intensity i s lower in more densely populated communities and higher where wage rates are higher. The positive association with wage rates may be due to reverse causality; i.e., higher labor demand for crop production may lead to higher local wage rates. The negative association of labor intensity with population density is hard to explain, but this i s not the total effect o f population density on labor intensity, since population pressure likely affects other household variables that influence labor use, such as farm and household size. The total effect o f population density is thus not clear. Use o f fertilizer is more likely where wages are higher (significant in the reduced form model only), possibly because higher wages enable households to purchase fertilizer. There are also significant differences ininput use across the agro-ecological zones, as expected. Cropproductivity and incomeper capita Several inputs and land management practices have a positive impact on crop production in the structural OLS model, including labor, purchased seeds, organic fertilizer, and incorporation o f crop residues (Appendix 5). None of these inputs or practices has a statistically significant impact on productivity in the IV regression, although the magnitude o f the estimated coefficients was larger inthe IV model in all cases. This indicates that identification problems inthe IV model are limiting the ability o f that model to identify significant impacts o f these endogenous variables, despite the fact that the relevance tests show that the instrumental variables are highly relevant, The Hausman test fails to reject statistical exogeneity o f these inputs and practices, so the OLS model is the preferred model. Other inputs and land management practices had statistically insignificant impacts on crop productioninbothregressions. 26 The positive impact of labor intensity on crop production is consistent with the findings of Nkonya, et al. (2004b) and Pender, et al. (2004). The positive impact o f seeds i s also consistent with the findings o f Pender, et al. (2004). However, the positive impacts found for organic fertilizer and crop residues contrasts with the results o f Pender, et al. (2004) and Nkonya, et al. (2004b), who found insignificant impacts of organic fertilizer on crop production, perhaps because of differences in the sample frames or the way organic fertilizer was measured in these different studies. The insignificant impact o f inorganic fertilizer found inthis study also contrasts with the significant impact found by Pender, et al. (2004). However, the coefficient o f productionresponse to fertilizer in the OLS regression in Appendix 5 (0.027) is quite similar in magnitude to the magnitude of the coefficient of impact o f inorganic fertilizer estimated by Pender, et al. (2004) (0.036 in their OLS model). The smaller sample size in this study and limited number of households that use fertilizer limits the ability to identify the impact o f inorganic fertilizer use on crop productivity. Considering the magnitude of the coefficient for inorganic fertilizer use, use of this input appears not to be profitable where it i s being used. Households using fertilizer realize an average value o f production o f Ush 814,00O/acre and apply fertilizer at an average cost o f Ush43,457/acre. With an estimated elasticity of productionresponseto fertilizer of 0.027, a one percent increase in meanfertilizer use, worth Ush434/acre, would increase the predicted value of production by only Ush220/acre (Ush814,000 x 0.01 x 0.027), which translates to a marginalvalue/cost ratio (VCR) of fertilizer use o f only 0.51 (220/434). A minimumVCR of at least one i s needed for additional fertilizer use to be profitable, and it is estimated that a marginal value/cost ratio o f at least 2 is needed for significant adoption of fertilizer (CIMMYT, 1988). Thus, even if the true elasticity o f crop productionresponse to fertilizer were two to four times the estimate, fertilizer would only be marginally profitable for the sample households, and substantial increasesinfertilizer use would be unlikelywithout substantial reduction inthe price of fertilizer and/or increases incrop prices. The low profitability o f inorganic fertilizer explains its low adoption inUganda, and suggests that major improvements inthe market environment facing Ugandan farmers are a prerequisite for substantial adoption to occur. Similar findings were reported byPender, et al., (2004) andWoelcke (2002). Inadditionto the fact that Ugandais a landlocked country, there are manyother factors that contribute to the high cost of fertilizer in the country relative to her neighbors. For example, Omamo (2002) observed that Uganda fertilizer procurement and distributionis dominated byretail- level trade and high prices that discourage farmers to use fertilizer and low net margins that discourage traders to market fertilizer (Omamo, 2002). Faced with low smallholder demand for fertilizer, traders inUganda appear to beunwillingto invest inmeasuresthat might reduce fertilizer farm-gate prices. In early 2005, the retail price o f a ton o f Diammonium Phosphate (DAP) in Kampala was US$508 (APEP, 2005), while the same quantity costs US$265 in Nairobi and US$240 in Dar es Salaam (The Sunday News, 2005). Transportation contributes a large share o f the high fertilizer price inKampala. For example, transporting one ton o f fertilizer from Mombasa port to Kampala costs US$lOO (Sanchez, 2004). This does not include a number o f transit taxes, Inorganic fertilizer has a statistically significant positive impact on crop production ina medianregressionversion o f the model, suggesting that problems with outliers (together with the limited number o f cases o f fertilizer use) are limiting the ability to identify the effect of inorganic fertilizer. However, the results ofthis medianregressionare not reported because it does not account for the sample weights (the effects o f which were tested and found to be statistically significant at p = 0.01) andthus can causebiased coefficients. 27 warehouse costs, etc. All these factors contribute to the low profitability o f fertilizer inUgandathat i s observed inthis study. Controlling for use of inputs and landmanagementpractices, landquality, andother factors, larger farms have lower per acre value of crop produced, supporting the inverse farm size - land productivity relationship observed in many empirical studies in developing countries (e.g. Chayanov 1966; Heltberg 1998; Carter 1984, Sen 1975; Berryand Cline 1979; Barrett 1996; Bhalla 1998; Lamb 2003; Nkonya, et al. 2004; Pender, et al. 2004). This inverse relationship even when controlling for use o f labor, other inputs and land management practices, plot size and observable land quality indicators, implying not only that smaller farmers tend to farm more intensively, as already seen, but are more productive inthe use of their inputs.loThese results suggest that market imperfections (such as limitations inthe markets for some factors o fproduction or ininsurance and output markets) limit the productivity of larger farms (Carter 1984, Feder, 1985; Barrett 1996; Heltberg 1998), but unobserved differences in land quality operated by larger vs. smaller farms may also account for part of this (Bhalla 1988; Lamb 2003). For example, soils innorthern Uganda where f m s are larger tend to be o f sandier texture, and this aspect o f land quality has not been controlled for, although agro-ecological farming system zones and soil depth and nutrient stock, which are significantly correlated with soil texture, have been controlled for (Ssali 2002). Despite having lower land productivity, larger farms have higher per capita household income (Appendix 6), suggestingthat they have higher labor productivity (Pender 1998). Thus wealthier households have higher incomes, as expected. Livestock assets increase crop productivity and per capita income, as expected. Consistent with Nkonya, et al. (2004b), livestock ownership increases crop productivity, perhaps due to the synergies between the two enterprises. Farmers with livestock have a supply o f manure and insome areas use animal power for plowing and transportation. Livestock thus contribute to higher per capita income by contributing to crop income as well as to livestock income. Ownership o f farm equipment is surprisingly associated with lower crop productivity, controlling for land management practices and input use, but has a statistically insignificant effect inthe IVmodelandinthe reduced form modelthat excludesthese variables. Itis hardto see why ownership of farm equipment would reduce productivity. Perhaps ownership o f farm equipment is negatively correlated with unobserved aspects of land quality, such as soil texture (e.g., use of plows may be more common in lighter textured sandy soils, which are less productive than clay soils), or i s more associated with livestock than crop production. The main impacts of mechanization may be to enable farmers to farm on a larger area, rather than increasing their productivity on a given area. However, findings show statistically insignificant impacts o f farm equipment on household income per capita (Appendix 6). Thus, if farm equipment is enabling some farmers to farm on a larger area, this may mainly be offsetting the lower productivity per unit area that larger farmers attain. Post secondary education of females and secondary education o f males are associated with higher crop productivity as compared to those with no formal education. Thus, even though post- secondary education reduces labor intensity, as noted previously, it increases the productivity o f labor and other inputs in production. The net impact on value o f production per acre i s positive, 10Nkonya, et al. (2004b) noted a similar finding from their analysis of data from a different sample in Uganda. 28 despite reduced labor intensity (as indicated by the positive impact o f female post-secondary education in the reduced form model). Since male secondary education increases productivity o f inputs, and was not found to reduce labor intensity or use o f other inputs, it is not surprising that this also has a strong positive net impact on crop production (in the reduced form model). Male secondary education and post-secondary education also are associated with significantly higher income per capita, as expected, and consistent with other studies o f income determinants inUganda (Nkonya, et al. 2004b; Appleton 2001b; Deininger and Okidi 2001) andnumerous other developing countries.' ' Male headed households and larger households obtain higher crop productivity, possibly because o f labor and management constraints faced by female-headed households and households with a smaller family labor supply, inthe context o f imperfect markets for labor and management. However, family size decreases per capita household income (significant only in the IV model), suggesting that the agricultural intensification practiced by larger families and the resulting higher value o f crop per acre does not compensate for the effect o f dependency ratio, which tends to depress per capita income and which i s generally higher in larger families, and/or that there are diminishing (though positive) retums to increased labor supply.'2 Non-farm activity as the primary income source of the household head is positively associated with the value of crop production per acre and household income per capita: predicted crop productivity is 25% higher and per capita income i s 23% higher for households dependent on non-farm activity rather than crop production as the primary income source. The positive impact o f non-farm activity on the value o f crop production is consistent with findings o f Nkonya, et al. (2004b). The positive impact on crop productivity may be related to the fact that non-farm activities reduce the probability to practice slash and burn and increases the probability to fallow (Appendix 3), both o f which may improve soil fertility and increase productivity over time. Non- farm activity also reduces labor-intensity in crop production, which may lead to higher marginal labor productivity in crop production if there are diminishing marginal retums to labor in crop production. Notice that the impact o f non-farm activity is positive only in the structural model reported in Appendix 5 but not in the reduced form model, consistent with this interpretation (i.e., households pursing non-farm activities obtain higher crop productivity when controlling for labor input,butnot when excluding labor input from the regression, because they use less labor per acre). The positive impact o f non-farm activities on household income per capita is as expected, although Nkonya, et al. (2004b) did not find a statistically significant impact o f non-farm activities on household income. In contrast to the positive impacts of non-farm activities, households dependent upon livestock income as their primary source o f income obtain substantially lower crop productivity than primary crop producers. This result contrasts to the positive impact o f livestock ownership on crop productivity. Ranchers and pastoralists that depend primarily on livestock income are probably not much focused on crop production, and tend to live in areas that are less suitable for ''The impact o f post-secondary education was significant only inthe reduced form model excluding participation in extension and credit organizations, however. ''Thedependency ratio was excluded from our regression specification since it was highly positively correlated with family size. 29 crop production.l3 Thus, while crop-livestock producers tend to have higher crop productivity when they have more livestock assets, ranchers and pastoralists more focused on livestock production tend to have lower productivity, controlling for the amount o f livestock owned and other factors. However, there are no statistically significant differences betweenper capita incomes of households dependent on livestock vs. crop production. This result contrasts with findings of Nkonya, et al. (2004b), who found that households dependent on livestock income obtained higher incomes. Not surprisingly, natural capital also influences crop productivity. Deeper soils andNstock in the topsoil have large positive impacts on crop productivity, as expected. A 1% increase in topsoil depth is associated with 0.34% higher productivity, while a 1% increase inN stock in the topsoil is associated with 0.26% higher productivity. However, neither topsoil depth nor the N stock have a significant impact on per capita income; although the soil P stock has a positive association with income (the reasonfor this is not clear). Even though agroforestry trees and SWC structures have a potential of competing with crops for space, light, and moisture, these investments significantly increase crop productivity. Predicted productivity is 22% higher on plots with agroforestry trees and 38% higher on plots with SWC structures, These investments can help increase crop productivity by reducing soil erosion, fixing nitrogen if leguminous trees and shrubs are planted, improving moisture conservation and soil physical characteristics. By contrast, other land investments such as fish ponds, fences, paddocks and pasture improvement are associated with lower crop productivity, undoubtedly because these promote livestock or aquaculture production rather than crop production. SWC structures and other land investments are associated with significantly higher per capita incomes perhaps due to their positive impact on crop productivity. It i s possible that reverse causality contributes to this positive relationship @e., people with more income are more able to invest), although manyfactors that determinethe capacity to investwere controlled for. Investments inperennial crop productionalso increaseproductivity and income. The value of crop production per acre and income per capita i s significantly higher on plots and for households where perennial crop production dominates than where annual crop production dominates. Perennial crop production increases the predicted value of crop productionper acre by about 17% compared to annual crop production. These results are consistent with findings o f Nkonya, et al. (2004b). consistent with results of Nkonya, et al. (2004b) and Pender, et al. (2004). I Nevertheless, Access to markets and roads has statistically insignificant impacts on croP productivity, proximity to roads is associated with significantly higher per capita income, probably because it promotes off-farm activities. This result is contrary to that o fNkonya, et al., (2004b) who observed apuzzling negative association between access to an all-weather road and household income, but i s l3 Although many factors that influence suitability for crop productionhave been controlled for, omitted climatic and landquality factors may still be correlated withreliance o n livestock as a primary source of income. l4 These variables were included ina full version o f the structural models reported inAppendix 5 and found to be jointly statistically insignificant inboth models, as well as inthe reduced form model. These and other variables were dropped inthe reported models to improve efficiency o fthese models. Results o f other models are available upon request. 30 more consistent with results of Pender, et al. (2004), who found that better road access is associated with higherincomes inthe central regionofUganda (but insignificant association inother regions), and Fan, et al. (2004), who found that road investment (especially inrural feeder roads) contributes to higher income growth and reduced poverty in Uganda. Thus the impacts o f road access on income appear to be positive, though these impacts may be location-specific. Agricultural technical assistanceprograms appear to have favorable impacts on agricultural productivity. Participation in the regular traditional agricultural extension program and in the NAADS program is associatedwith significantly higher value o f crop production per acre. For example, based on the regression results in Appendix 5, the value of crop production per acre in 2002/2003 is predicted to be 15% higher for households that participated inthe NAADS program than those who didn't, while a 10%increaseinnumberof contact hours with traditional agricultural extension would lead to a predicted 1% increase in productivity (ie., elasticity = 0.10). The positive association of NAADS with value o f crop production per acre may have less to do with changing farming practices than with promoting production of higher value crops, since insignificant impacts of NAADS presencewas found on most landmanagement practices and input use (Tables 7 and 8). We do not find robust statistically significant impacts o f NAADS or other extension programs on income per capita, h~wever.'~ These results are consistent with the findings of Nkonya, et al. (2004) conceming the positive impacts of access to agricultural extension and training on the value of crop production, and Fan, et al. (2004) conceming the positive agricultural productivityimpacts of expenditures on agricultural research and extension inUganda.l6 It is possible that these positive associations are due in part to selection bias; i.e., these programs may be operating in areas where productivity was already higher prior to the NAADS program or program participants may be those who were more productive evenbefore the program. This concem has been addressed by including numerous explanatory factors influencing productivity potential in the regressions, but it i s still possible that some excluded factors that are associated with technical assistance program placement or participation are partly responsible for these positive associations.l7 To address the endogeneity o f program participation and possible selection bias, the model was estimated usinginstrumental variables (IV) regressions. The impact of participation inregular extension or the NAADS program on crop productivity is not significant inthe IVmodel, though similar inmagnitude for bothtypes ofextension. Thisresult maybedueto difficulties of identifying such impacts in the IV models, rather than due to actual lack o f impact. Since the Hausman test fails to reject the OLS model, that i s the preferredmodel because it is more efficient (and less prone to bias causedbyweak instruments (Bound, et al. 1995). A largepositive andstatistically significant impact o fNAADS participation onincome per capita inthe IV model is found (reported inAppendix 5), but thls impact i s not robust inthe OLS model, which is the preferred model given the Hausman test results supporting that model. Giventhe weakness o f the instrumental variables usedinpredicting participation inNAADS (p-value =0.353 inthe relevance test for this variable), this could result inmore biased results inthe IVmodelthan inthe OLSmodel(Bound, et al. 1995). However, Pender, et al. (2004) found mostly statistically insignificant impacts o f agricultural extension on crop productivity and income. Shortcomings inthat study, as discussed previously, may have limited its ability to discern such effects. Participants to regional seminars that were carried out to disseminate findings o fthis study also expressed concem o f potential NAADSprogrambias. However, NAADS program site selection criteria were based on compliance with local government development program, which is not supposed to be influencedby income levels, and to reflect variety withrespect to nature oflocal agricultural economy and agro-ecological zones. 31 Also, the model was estimated using the presence o f NAADS in a sub-county rather than household level participation as the explanatory variable, and found similar positive impacts o f the presence o f N A A D S on productivity.'* This finding reduces the concern that household level selection bias i s responsible for the positive association o f productivity with NAADS participation. However, there still could be bias caused by the initial placement o f NAADS in more productive sub-counties. The possibility o f bias in selection o f NAADS sub-counties was investigated using data from the 1999/2000 U N H S and survey data from this study. Appendix 7 and 8 show the differences inmean value o f crop production per acre and per capita income in 1999/2000 between NAADS and non-NAADS sub-counties in the districts where the study was cond~cted.'~The results in these tables show that there was no bias towards selecting sub-counties where productivity or income was already higher. In only one district (Kabale) was there a statistically significant difference inpre-NAADS productivity betweenN A A D S and non-NAADS sub-counties and inthat case pre-NAADS productivity was higher inthe non-NAADS sub-counties. Inall other cases, average pre-NAADS productivity was quite similar in the N A A D S vs. non-NAADS sub- counties, and for all six sub-counties, the average difference inpre-NAADS productivity was less than 2% (slightly lower in the NAADS sub-counties). In no district was there a statistically significant difference in pre-NAADS income per capita between N A A D S and non-NAADS sub- counties, and the average differences are quantitatively small (less than 0.5% difference in all six districts). These results strengthen the confidence that NAADS is indeed having significant positive impacts on crop productivity. It i s still theoretically possible that some other factors besides the introduction o f N A A D S or the factors that are controlled for inthe regressions have changed since 1999/2000 in N A A D S and non-NAADS sub-counties, and are responsible for the higher current productivity in the NAADS sub-counties and among N A A D S participants. But it is difficult to imagine what those factors are, given that so many factors affecting productivity have been controlled for, or why such factors would have affected N A A D S vs. non-NAADS sub-counties differentially, in favor o f NAADS sub-counties. These results therefore provide support to the emphasis in the PMA on increasing the availability o f agricultural technical assistance in Uganda through expansion o f NAADS. Nevertheless, more focused research on the impacts o f NAADS to better understandwhether and how the NAADS program is having these favorable impacts in the trail-blazing districts and whether such favorable impacts are being scaled-out in these and other districts would be very usehl. Participation in rural finance organizations i s associated with significantly higher crop productivity and per capita incomes. These results are robust inthe IV as well as OLS versions o f the models, where the instrumental variables used to predict participation in rural finance organizations are strong predictors. These results indicate that selection bias or reverse causality are not the likely explanation for these findings. These results are consistent with findings o f Results of these regressions usingpresence o fNAADS ina sub-county, rather than household level participation were included inan earlier version o f this paper, and are available uponrequest. l9NAADS hadnot yet begun to operate inthe other two districts covered by this study (Masaka and Kapchorwa) inthe year covered by the IFPRI-UBOS survey (2002/03). 32 Pender, et al. (2004), who also found a positive impact o f access to rural credit on the value o f crop production, but not with those of Nkonya, et al. (2004b), who found insignificant impacts o f credit access. It is not clear how such organizations contribute to crop productivity, since they are associated with less intensive use of labor and less use of inorganic fertilizer. Perhaps the availability of rural finance is more important in helping farmers in marketing their crops and shiftingto highervalue crops, which canresult inhighervalue ofproductioneven ifthe quantityof production i s not affected. For example, households with access to credit may have no liquidity constraints that force many farmers to sell their produce immediately after harvest when agricultural prices are normally at their lowest levels. Rural finance also can help households to pursueoff-farm activities, which maybepart of the reasonfor its positive contributionto income. Land tenure has statistically insignificant impacts on crop productivity and income per capita.20. Consistent with Nkonya, et al. (2004b) and Pender, et al. (2004), these results suggest that lack of land titles and other differences in land tenure are not major constraints to crop productivity and income in Uganda. Similar findings of limited impacts o f land titles in areas of secure customary tenure have also been observed elsewhere inAfrica (e.g., Place and Hazel1 1993; Platteau 1996). Not surprisingly, agroecological zones influence both crop productivity and income, as expected. Productivity i s highest in the Mt. Elgon highlands and the southwest grasslands, and lowest in the northem zones. Income per capita is highest in the southwest grasslands. These results are fairly consistent with the findings o f Nkonya, et al. (2004b) and Pender, et al. (2004) (although the classification of zones inthis study i s somewhat different). Soil erosion Predicted soil erosion i s not significantly affected by the size of the farm or the household's physical assets (Appendix 9). Female primary and post-secondary education are associated with more erosion, though the reasons are not clear. In the case o f primary education, this may be related to the association o f primary education with labor intensive crop production. Larger households have significantly hi er erosion, probably because of more labor intensive crop production by larger households. P This contradicts the optimistic "more-people, less-erosion" hypothesis (Tiffen, et al. 1994) at the household level, and is consistent with findings o fNkonya, et al. (2004b). Not surprisingly, soil erosion is greater on steeper slopes, and i s reduced by investments in agroforestry or other land investments. Access to markets and roads has insignificant impacts on erosion, as does access to agricultural technical assistance and credit. These findings are consistent with the limited impacts o f these factors on land management practices, discussed earlier. Erosion is lower for households dependent on livestock activities as their primarysource of income than for households dependent upon crop income, probablybecause such households use the land less intensively, and with greater permanent soil cover on pastures 2oThe coefficients o fthe land tenure variables were jointly statistically insignificant ina fuller version o f the OLS and IV models for crop productionpresentedinAppendix 5, and were dropped fromthe regressions. They were also jointly insignificant inthe reduced form model. Results available uponrequest. 21Eventhough labor availability may theoretically positively influence adoption oflabor intensive soil erosion control practices, Appendix 3 shows that greater family labor endowment is not associated with adoption o f SWC and other organic land management practices. Furthermore, labor availability may leadto adoption o f erosive practices such as frequent weeding and cultivation. 33 than annually cropped fields. Erosion i s greater on mailo tenure than freehold, though the reasons are not clear. This result contradicts a finding of Nkonya, et al. (2004b), who found that mailo tenure was associated with lower erosion. Erosion differs across agro-ecological zones, being the worst inthe southwest highlands and least inthe northernzones. Soil nutrient balances The determinants o f soil nutrient balances are shown in Appendix 10. Most o f the factors investigated have statistically insignificant impacts on soil nutrient balances. Livestock ownership is, surprisingly, associated with more rapid depletion o f N. This i s likely due to feeding crop residue to livestock after harvest, which i s a common practice inareas with large cattle population. The resulting nutrient outflows through crop harvests andgrazing outweigh their positive impact on nutrient inflows o f organic matter. Human capital endowments have mixed impacts on nutrient balances. Female primary education is associated with more negative N balances, and all levels o f female education are associated with greater depletion o f total N, P, and K. This likely relates to the association o f female education with erosion noted earlier. In the case o f households with higher female education, they are less likely to use organic inputs but obtain higher productivity, which also causes nutrient depletion. By contrast, male education is not significantly associated with nutrient balances, except a weakly statistically significant negative association o f male post-secondary education with P balances. This may be related to the fact that male post-secondary education is negatively associated with use o f SWC practices, as noted earlier. The gender o f the household headand size ofthe householdhave insignificant impacts on soil nutrientbalances. Soil nutrient balances are more negative on steeper slopes as a result o f greater erosion. They are more favorable on plots with SWC or agroforestry investments. By contrast, soil nutrient balances are much more negative on perennial than annual plots, especially for K. This is due to highrates o f soil nutrient depletion inbanana production (Kaizzi and Kato, 2004; Wortmann and Kaizzi 1998). Access to markets has insignificant impacts on soil nutrient balances, while better road access is associated with more favorable nutrient balances. The beneficial impacts o f road access may be in part because this reduces use o f slash and bum (Appendix 3), which depletes soil fertility. Participation in extension and credit has insignificant impacts on soil nutrient balances, even though these have significant impacts on use o f inorganic fertilizer, as noted previously. The use o f inorganic fertilizer i s too uncommon for this to have much effect on average nutrient balances, while extension and credit have limited impacts on organic landmanagement practices. Customary land has more favorable K balances than landunder freehold tenure, while mailo land has more negative balances o f Nand total NPK than freehold land. The association o f mailo landwith banana production may be part o f the reason for greater nutrient depletion on mailo land. It is not sure why K balances are more favorable on customary than freehold land, though this may be due to less banana production on customary than freehold land. Inany case, these results do not 34 support the common concern that land degradation may be greater on customary land due to inadequate incentives o f farmers to conserve such land. There are also differences across agro-ecological zones in soil nutrient depletion, with the greatest depletion rates inthe southwest zones, due to banana production inthese zones and higher erosionrates inthe southwest highlands. 35 5. Conclusionsand PolicyImplications Land degradation in the form o f soil erosion and soil nutrient depletion is an important problem in Uganda. The study shows that farmers in the eight districts studied (representing six major ago-ecologicavfarming system zones) deplete an average o f 179 kg/ha o f nitrogen, phosphorus and potassium, which is about 1.2% o f the nutrient stock stored in the topsoil (0 - 20cm depth). The value of replacing the depleted nutrients using the minimum price o f inorganic fertilizer is equivalent to about one fifth o f the household income obtained from agricultural production. This underscores the reliance o f smallholder farmers on soil nutrient mining for their livelihoods and the high costs that would be required to solve this problem. The findings o f this study also underscore the great concem that soil nutrient depletion poses since it contributes to declining agricultural production in the near term as well as the longer term. For example, a 1% decrease in the nitrogen stock in the topsoil leads to a predicted 0.26% reduction in crop productivity. This loss inagricultural productivity may contribute to food insecurity. Furthermore, soil nutrientdepletion may contribute to deforestation and loss o f biodiversity since farmers maybe forced to abandon nutrient-depleted soils and cultivate more marginal areas such as hillsides and rainforests. These results highlight the challenges that Uganda faces as it accelerates the implementation o f the Plan for Modemization o f Agriculture (PMA) and rolling out the new extension program, the National Agricultural Advisory Services (NAADS), to more districts. To forestall potential medium and long-term impact of land degradation, policy makers need to design strategies to reduce soil nutrient depletion and other forms o f land degradation. Such strategies include, but are not limited to, reducing the cost o f inorganic fertilizer and developing and promoting organic soil fertility technologies that are cost effective and relevant to local farming systems. Such strategies could contribute to increasing agricultural productivity and farm income as well as reducing land degradation. The qualitative results o f the econometric analysis are summarized in Appendix 11. The results provide evidence o f linkages between poverty and land management. Natural capital inthe form of some prior land investments contributes to better current land management practices, higher productivity and income, reduce erosion and, in the case o f SWC structures, improve soil nutrient balances. These findings imply that SWC investments can lead to win-win-win outcomes since they increase income andcrop productivity and conserve natural resources. As expected, land management practices, namely use o f purchased seed, labor, value o f organic fertilizer, anduse o f crop residues increase the value o f crop production per acre. However, inorganic fertilizer does not have a statistically significant impact on crop production, and its estimated profitability i s low. The estimated marginal value cost ratio o f fertilizer is much less than 1, suggesting that adoption o f fertilizer is likely to remain low unless its price is reduced substantially or crop prices improve substantially. Hence efforts to improve the market environment through investments in infrastructure and market institutions are necessary for substantial adoption o f fertilizer to occur inthe regions studied. Not surprisingly, the quality of the land also influences land management and outcomes. For example, average plot slope increases the likelihood to incorporate crop residues, use short- term soil and conservation (SWC) practices but it leads to greater erosion and consequently lower 36 nutrientbalances. Plots with deeper soils are more likely to be used for crop rotation and likely to give higherproductivity. The observation was an inverse farm size - crop productivity relationship, due to lower farming intensity by larger farms. Although smaller farms obtain higher value o f production per acre, this does not fully compensate for the fact that they have less land, and they earn lower per capita incomes as a result. These findings are consistent with those o f Nkonya, et al. (2004b) and Pender, et al. (2004) for Uganda, and with numerous studies from other developing countries. Despite these differences related to farm size, no significant differences in soil erosion or soil nutrient depletion were found due to farm size. Thus, improving access of small farmers to land, for example by improving the functioning o f land markets, can increase aggregate agricultural production and small farmers' incomes in Uganda, with no apparent tradeoff in terms o f land degradation. Non-land assets, including livestock and value o f equipment, have mixed impacts on land management and outcomes. Value of equipment increases labor intensity and but decreases crop productivity. Livestock ownership decreases the probability to fallow and the level of nitrogen balances but increases crop productivity andper capita income. These results suggest that livestock poor farmers are likely to remain inpoverty with low productivity. Human capital has mixed impacts on landmanagement, productivity and land degradation. Female primary education i s associatedwith more erosion and soil nutrient depletion, while female post-secondary education is also associatedwith more soil erosion as well as less labor intensityin crop production, but also with higher crop productivity. Male education i s associated with greater use of inorganic fertilizer, higher crop productivity, and, inthe case of secondary education, higher income per capita. In general, female education has less positive impacts on land management, productivity and sustainability than male education. This may be due to a greater tendency o f educatedfemales to focus on other livelihoodactivities. These results imply that simply investing in education will not solve the problem o f land degradation in Uganda, even though education i s critical to the long-term success of poverty reduction efforts. These findings support the objective within the Plan for Modemization o f Agriculture to introduce agricultural and natural resource management education into school curricula in order to prepare students to become better farmers who manage natural resources sustainably. Larger families use more erosive practices but realize higher value of crop production per acre but have lower per capita income, suggestingthat the retums from agricultural intensification do not compensate for a higher dependency ratio in larger families, which tends to depress per capita income. These results imply that population pressure has negative impacts on per capita income and land degradation at the household level, though it contributes to more intensive and productive use of the land (similar to the effects o f farm size, and also consistent with Boserup's (1965) theory). These results demonstrate the need to promote reproductive health as one way o f reducingpoverty and landdegradation. 37 Access to financial capital, in the form o f household participation in programs and organizations providing financial services, decreases the probability to practice crop rotation or apply inorganic fertilizer but i s associated with lower crop productivity and per capita income, suggesting that households borrow mainly to finance non-agricultural activities that appear to have greater returns than agriculture. Access to markets and roads significantly affects some land management practices and outcomes. In areas closer to major markets, fanners are more likely to use SWC practices. However, this does not contribute to higher per capita income or crop productivity or better nutrient balances. Farmers closer to all-weather roads are less likely to use destructive slash and burn practices, obtain higher per capita income, andhave less soil nutrient depletion. These findings are consistent with the favorable impacts o fmarket and road access found insome other studies inEast Africa (e.g., Tiffen, et al. 1994; Pender, et al. 2001; Fan, et al. 2004), though findings o f such favorable impacts are not universal (e.g., Nkonya, et al. 2004b). These results support the Ugandan govemment's efforts to build rural roads as investments that can reduce poverty, as well as potentially helpingto reduce landdegradation. However, the impacts ofroads andpotential market integration on most landmanagement practices are not clear andrequire further investigation. Access to agricultural technical assistance services, measured by contact with government extension agents and participation in the NAADS program, has a positive association with crop productivity, as expected. Investigationswere carried out to see whether the positive associations o f participation inNAADS program with higher production may be due to selection bias (ie., initial operation o f NAADS in higher productivity sub-counties), and the results ruled out this explanation. The findings thus provide support for the NAADS approach, suggesting that N A A D S is already having substantial positive impacts due to the introduction of profitable strategic enterprises. However, further focused research is needed to better understand whether and how the NAADS approach is leading to greater crop productivity, and whether such benefits are being realized inother districts and sub-counties as the program expands. The positive impacts o f agricultural extension on productivity that are found are consistent with the findings o f Nkonya, et a1 (2004b), and suggest that remote areas with poor access to technical assistance (Jagger and Pender 2003) are likely to continue to face low productivity and poverty. This suggests the need to give incentives for technical assistance programs to operate in remote areas. The agricultural technical assistance programs analyzed in this research have generally limitedimpacts on organic landmanagement practices, which are important given the highcost and low profitability of inorganic fertilizer. This suggests the urgent need for N A A D S to give greater attention to promoting organic land soil fertility practices in order to address the potential soil fertility depletion resulting from promotion o f adoption o fmore profitable farmingenterprises, Changes inhousehold livelihood strategies, whether promoted byN A A D S or resulting from other factors such as education, population growth or market and road development, can have important implications for land management, productivity, incomes and land degradation. Households pursuing non-farm activities are more able to fallow their land and less likely to use slash and burn practice, and obtain higher value o f production per acre and per capita income. 38 These results imply that non-farm activities can be complementary to crop production, by enabling households to fallow and by reducing households' exposure to agricultural price and production risks. Hence promotion o f non-farm activities has potential o f achieving win-win-win outcomes, increasingproductivity, reducing poverty and conserving natural resources. Efforts to increase rural households' formal and vocational education, rural electrification programs, road development and development o f rural microfinance institutions can help increase opportunities to participate innon- farm activities. Such efforts are likely to be especially important for poor farmers andwomen, who often lack access to off-farm opportunities (Barrett, et al. 2001; Gladwin 1991). Householdspursuing livestock production as their primary livelihood strategy also have less erosionbut their productivity incrop production i s lower. Thus, promotion o f livestock production can help to improve the sustainability o f natural resource management, though it may involve some tradeoff interms o f reduced crop production (though recall that controlling for livelihood strategies andother factors, greater livestock ownership is associated withhigher crop productivity). Among crop producers, perennial crop producers use less slash and burn, crop rotation, short-term SWC or fallow but obtain higher value o f crop production and income than annual crop producers. However, they deplete soil nutrients more rapidly (especially nitrogen and potassium), despite the common application o f mulch and other organic materials to perennial crops. Thus, perennial crop production involves tradeoffs among the objectives o f increasing productivity, reducing poverty and ensuring sustainable use o f natural resources, at least given the land management practices currently used in Uganda. Promoting measures to restore soil nutrients in perennial (especially banana) production should be a high priority for agricultural technical assistanceprograms. The land tenure system also is associated with some differences in land management practices and land degradation. For example, use o f slash and burn, crop residues and labor are greater on customary land than freehold land, leading to more favorable potassium balances. Despite these differences, no differences are found in crop productivity or income per capita associated with differences in land tenure systems. There i s still need to facilitate access to credit incustomary tenure areas, since owners of landunder customary tenure are unable to pledge their land as collateral inthe formal credit service, and this research has shown that such services could help to reduce poverty. The findings suggest that some modernization strategies can achieve win-win-win outcomes, simultaneously increasing productivity, reducing poverty, and reducing land degradation. Examples of such strategies include promoting investments in SWC and road development. Some strategies appear able to contribute to some positive outcomes without significant tradeoffs for others, such as promotion o f non-farm activities, agricultural extension programs andrural finance. Other strategies are likely to involve tradeoffs among different objectives. For example, investing in livestock appears to improve crop productivity and household income, but also i s associated with more rapid soil nutrient depletion. Expansion o f banana production i s likely to cause more soil nutrient depletion as well as higher income and productivity, unless greater efforts to restore soil nutrients are made. Female education may contributes to improved health, nutrition 39 or other development indicators not analyzed in this research, but also appears to contribute to some indicators of land degradation. The presence o f such tradeoffs is not an argument to avoid such strategies; but rather i s an argument to recognize and find ways to ameliorate such negative impacts where they may occur. For example, incorporating teaching o f principles of sustainable agriculture and natural resource management into educational curricula, as well as inthe technical assistance approach o fNAADS andother organizations, is one important way o f seeking to address such tradeoffs. Overall, the results provide support for the hypothesis that promotion of poverty reduction and agricultural modernization through technical assistance programs and investments in infrastructure and education can improve agricultural productivity and help reduce poverty. However, they also show that some o f these investments do not necessarily reduce land degradation, and may contribute to worsening land degradation inthe near term. Thus, investing in poverty reduction and agricultural modernization is not sufficient to address the problem o f land degradation inUganda, andmust be complemented by greater efforts to address this problem. The results of this study provide important information on the marginal returns, interms o f income and crop productivity, to such factors as use o f soil andwater conservationpractices, use o f fertilizers, farm size, access to credit, market and road access, and agricultural extension. However, using these results to make specific policy choices will require additional information on the marginal costs o f different policy interventions. Beyond these policy implications, it i s recommendedthat the government of Uganda, led by UBOS, NARO and Makerere University, continue to collect systematic data on natural resource management and degradation linked to the socioeconomic surveys o f UBOS, building upon the surveys conducted for this project. Future crop production surveys led by UBOS would be more usekl ifthey collected information on landmanagement, input use and crop production at the plot level, and also collect soil samples from the same households surveyed previously every five to ten years, so that changes in soil conditions and their linkages to productivity and poverty assesses, This project has established a baseline o f information that can be used inconjunction with future surveys to be able to analyze the dynamic linkages between poverty and natural resource degradation, and the medium and longer term effects o f the PEAP, PMA and other policies and programs inaddressing these problems. 40 References Agrawal, A. and C.C. Gibson. 1999. "Enchantment and disenchantment: The role o f community in natural resource conservation." WprldDevelopment, 27(4): 629-49. APEP (Agricultural Productivity EnhancementProgram). 2004. Agri-input markets - Uganda Newsletter (1)1. Appleton S. 2001a. Poverty reduction during growth: the case of Uganda. Mimeo, Center for African Studies, Oxford University, U.K. Appleton, S. 2001b. What canwe expect from Universal Primary Education? In: R.Reinikkaand P. Collier (eds.), Uganda's recovery: the role of farms, firms, and government. The World Bank, Washington, D.C. Agricultural Policy Secretariat (APSEC) o fUganda. 2001. Characterization o fpolicies and institutions affecting land management. APSEC, Kampala, Uganda. Mimeo. Atwood, D.A. 1990. LandRegistration inAfrica: The Impact on Agricultural Production, World Development 18(5), 659-671. Bardhan, P., 1993. Analytics ofthe institutions ofinformalcooperationinruraldevelopment. World Development, 21(4): 633-639. Barrett, C.B. 1996.Onpricerisk and the inverse farm size-productivity relationship. Journal of Development Economics, 51:193-215. Barrett, C.B., F.Place, A. Aboud. 2002. The challengesof stimulating adoption o f improved natural resourcemanagementpractices inAfrican agriculture. In:Barrett, C.B., F.Place, A.A. Aboud(eds). NaturalResourcesManagement in African Agriculture. ICRAF and CABI, Nairobi Kenya. Barrett, C.B., T. Reardon, P. Webb. 2001. Non-income diversification andhousehold livelihood strategiesinruralAfrica: Concepts, dynamics andpolicy implications. Food Policy 26(4):315-331. Baum, C. F., M.E.Schaffer, and S. Stillman. 2002. Instrumental variables and GMM: Estimation and testing. Unpublishedworking paper no. 545, Boston College Department of Economics. http://fmwww.bc.edu/ec-p/WP545.pdf Bhalla, S.S. 1988. Does landquality matter? Theory andmeasurement.Journal of Development Economics, 29:45-62. Berry, R.A.andW.R. Cline. 1979. Agrarian Structure and Productivity in Developing Countries. Baltimore: Johns Hopkins. 41 Besley, T. 1995. Property rights andinvestmentincentives: Theory and evidence from Ghana. Journal of Political Economy 103(5): 903-937. Bekunda, M. 1999. "Farmers responses to soil fertility decline inbanana-basedcropping systems o fUganda." Managing Africans Soils 4(February, 1999): 1-20. Boserup, E. 1965. Theconditions of agricultural growth. New York: Aldine Publishing Company. Bound, J., D.A. Jaeger, and R.M. Baker. 1995. Problems with instrumental variables estimation when the correlation between the instruments and the exogenous explanatory variable is weak. Journal of theAmerican Statistical Association 90,443-450. Camey, D. 1998. Implementing the sustainablerural livelihoods approach. In:Carney, D.(Ed.), SustainableRuralLivelihoods: What Contribution CanWe Make? Departmentfor International Development, London. Carter, M.R. 1984. Identificationof the inverse relationship between farm size andproductivity: An empirical analysisofpeasantagricultural production. Oxford Economic Papers 36(1): 131-145. Chayanov, A., 1966. The theory of peasant economy. Homewood1llinois:Richard Irwin. CIMMYT, 1988.From agronomic data tofarmer recommendation: An economics training manual. Completely revised edition. Mexico DF. Cleaver, K.M.and G.A. Schreiber. 1994. Reversing the Spiral: ThePopulation, Agriculture and Environment Nexus in Sub-Saharan Africa. The World Bank, Washington, D.C. Davidson, R. and J.G. MacKinnon. 2004. Econometric Theory and Methods. Oxford: Oxford UniversityPress. Deininger, K.andJ. Okidi. 2001. Rural households: Incomes, productivity andnonfarm enterprises. In Uganda 's Recovery: TheRole of Farms, Firms, and Government. Washington, D.C., The World Bank. de Janvry, A., M.Fafchamps, and E. Sadoulet. 1991. Peasant household behavior with missing markets: some paradoxes explained. Economic Journal 101. der Pol, V. F. 1993.Analysis and evaluation options for sustainable agriculture with special referenceto southem Mali. In:Van Reuler H.,Prins, W.H. (Eds), Therole ofplant nutrients for sustainablefood cropproduction in Sub-Saharan Africa. VKP Leidschendam Fafchamps, M., 2000. "Ethnicity and Credit in African Manufacturing." Journal of Development Economics, 61:205-235. 42 Fafchamps, M., and B. Minten, 1999. "Relationships and Traders in Madagascar." Journal of Development Studies, 35(6): 1-35. Fafchamps, M.,andPender, J.L. 1997. Precautionary Saving, Credit Constraints, and Irreversible Investment: Theory andEvidence from SemiaridIndia, Journal of Business and Economic Statistics, 15(2), 180-194. Fan, S., N.Rao. 2003. Public spending indeveloping countries: Trends, determination andimpact. International Foodpolicy ResearchInstitute IFPRI) Environment andProduction Technology Division (EPTD) Discussion paper # 99. Washington DC. Feder, G. 1985. The relation betweenfarm size and farmproductivity. Journal of Development Economics 18: 297-313. Feder, G., T. Onchan, Y. Chalamwong, andC. Hongladaron. 1988. Landpolicies andfarm productivity in Thailand. Baltimore: The Johns HopkinsUniversityPress. Gavian, S. and Fafchamps, M. 1996. LandTenure andAllocative Efficiency inNiger,American Journal of Agricultural Economics, 78(May), 460-471. Gladwin, C. H. 1991. Structural adjustment andAfrican women farmers. Gainesville: Universityo f Florida Press. Grepperud, S. 1997."Poverty, landdegradationand climatic uncertainty," Oxford economicpapers 49(4): 586-608. Griffin, R.andJ. Stoll. 1984."Evolutionary processesinsoil conservation policy," Land economics 60:30-39. Hammer, J. 1986."Subsistence first: farm allocation decisions inSenegal," Journal of development economics 23:355-369. Heltberg, R. 1998. Ruralmarket imperfections and the farm size-productivity relationship: Evidence from Pakistan. WorldDevelopment 26(10): 1807-1826. Holden, S., B. Shiferaw., M.Wik. 1998. Poverty, credit constraints andtimepreference o f relevance for environmental policy Environment and Development Economics, 3:105-130. Jagger, P. andJ. Pender. 2003. Impacts o fprograms and organizations onthe adoption of sustainable landmanagement technologies inUganda. Environment andProduction Technology Division Discussion PaperNo. 101,Washington, DC: InternationalFood PolicyResearchInstitute. Kaizzi, C. K., Ssali, H.andKato, E. 2004. Estimated nutrientbalances for selectedagroecological zones o fUganda. National Agricultural ResearchOrganization ("0) Mimeo, Kampala Uganda. 43 Kaizzi, C.K. 2002. The potential benefit of green manures andinorganic fertilizers incereal productionon contrasting soils ineasternUganda. PhD Dissertation submittedto the Universityo fBonn and GermanCenter for Development Research(ZEF):64-69. Kyamanywa, C. (1987). Case Studies inEnvironmental Managemento fUrbanAreas. EcosystemsManagement in Developing Countries Vo1.W: 243. Lamb, R.L. 2003. Inverseproductivity: Landquality, labor markets, andmeasurementerror. Journal of Development Economics, 71(1):71-95. Leonard, H.J. 1989. Environment and the Poor: Development Strategiesfor a Common Agenda. Transaction Books, New Brunswick, N.J. Lindblade, K.,J. Tumuhairwe and G.Carswell. 1996.Morepeople, more fallow: environmentally favorable land-usechanges insouthwesternUganda. Report for the Rockefeller Foundation and CARE International. Maddala, G.S. 1983. Limited-Dependent and Qualitative Variables in Econometrics. Econometric Society Monographs No. 3. Cambridge: Cambridge UniversityPress. Magunda, M.K. and M.M. Tenywa. 1999. Soil and water conservation. Mimeo, Uganda National Agricultural ResearchOrganization (NARO), Uganda. MinistryofNaturalResources. 1994.State of theEnvironment Reportfor Uganda, NationalEnvironment InformationCentre. Mukherjee, C., H.White, andM.Wuyts. 1998.Econometric and data analysisfor developing countries. London Routledge. NEAP (National Environmental Action Plan. 1992. Landtenure andlandmanagementinUganda. Report o f task force on landmanagement. Ministry o fNaturalResources, Kampala Uganda. Nkonya, E., J. Pender, E. Kato, J. Muwonge. 2004a. Community natural resource management and its linkage to poverty inUganda. InternationalFoodPolicyResearchInstitute, Washington D.C. Mimeo. Nkonya, E., J. Pender, P. Jagger, D. Sserunkuuma, C. Kaizzi and H. Ssali. 2004b. Strategies for Sustainable Land Management and Poverty Reduction in Uganda. Research Report No. 133, International FoodPolicy ResearchInstitute, Washington, D.C. Otsuka, K.and F.Place. 2001. Issues andtheoretical framework. InK.Otsuka andF.Place (eds.), Land Tenureand Natural ResourceManagement: A Comparative Study of Agrarian Communities in Asia and Africa. Baltimore: Johns Hopkins. Omamo, W. 2002. Fertilizer trade andpricinginUganda. Paperpresentedat the Association for StrengtheningAgricultural ResearchinEastern andCentral Africa (ASARECA) workshop 44 on the assessment of the fertilizer sub-sector ineast Africa, Hilton HotelNairobi, July 15- 17,2002 Pender, J.L. 1996.Discount Rates andCredit Constraints: Theory andEvidence from RuralIndia, Journal of Development Economics, 50,257-296. Pender, J. 1998. Populationgrowth, agricultural intensification, induced innovationandnatural resource sustainability: an application ofneoclassicalgrowth theory. Agricultural Economics 19,99-112. Pender, J., P. Jagger, E.Nkonya andD. Sserunkuuma. 2001. Development Pathways andLand Management inUganda: Causes andImplications. Environment andProduction Technology DivisionDiscussionPaperNo. 85. Washington, DC: InternationalFoodPolicy ResearchInstitute. Pender, J., S. Ssewanyana, K.Edward, E.Nkonya. 2004. Linkages betweenpoverty andland managementinrural Uganda: evidence from the UgandaNational Household Survey, 1999/2000. Environment andProductionTechnology Division Discussion PaperNo. 122, International FoodPolicy ResearchInstitute, Washington, D.C. Pender, J. andKen, J., 1998. Determinants of farmers' indigenous soil andwater conservation investments inIndia's semi-arid tropics. Agricultural Economics 19, 113-125. Pender, J. and J. Ken. 1999. The effects of land sales restrictions: Evidence from south India. Agricultural Economics 21,279-294. Place, F. andP. Hazell. 1993. Productivityeffects o f indigenous landtenure systems insub- Saharan Africa. American Journal of Agricultural Economics 75(1): 10-19. Platteau, J.P. 1996. The evolutionary theory o f land rights as applied to sub-Saharan Africa: A critical assessment. Development and Change 27:29-86. Rausser, G. 1980. "Economics o f soil conservation from the fanner's perspective: Discussion," American Journal of Agricultural Economics, 62:1093-1094. Reardon, T. and S. Vosti. 1995. Links between ruralpoverty andthe environment indeveloping countries: Asset categoriesand investmentpoverty. WorldDevelopment 23(9), 1495-1506. Renard, K.G., G.R. Foster, G.A. Weesies, andJ.P. Porter. 1991.RUSLE: RevisedUniversal Soil Loss Equation. Journal of Soil and Water Conservation 46( 1): 30-33. RepublicofUganda. 1998.LandAct, 1998.Government Printer, Kampala, Uganda. Sanchez, P.A., K.D.Sheperd, M.J. Soule, F.M.Place, R.J. Buresh, A-M.N. Izac, A.V. Mokwunye, F.R.Kwesiga, C.G. Ndiritu, andP.L. Woomer. 1997. Soil fertilityreplenishment inAfrica: An investment innaturalresourcecapital p 1-46. In:R.J. BureshR.J., A. Sanchez, andF. 45 Calhoum (eds), Replenishing soilfertility in Africa. American Society of agronomy and Soil Science Society of America: 1-46. Sanchez, P. 2004. "Innovative investmentsinhealthy soils andbetter landmanagement." Paper presented at a Highlevel seminar onInnovative approaches to meet the hunger Millennium Development Goals (MDS) in Africa. July 5,2004, Addis Ababa Ethiopia. Scherr, S. andP. Hazell. 1994. Sustainableagricultural development strategies infragile lands. Environment and Production Technology Division DiscussionPaperNo. 1, Washington, DC: Intemational FoodPolicyResearchInstitute. Semana, A. R. and Adipala, E. 1993. Towards Sustaining Crop Production inUganda. African Crop Science ConferenceProceedings 1: 19-22. Sharma, M.,G. Buchemrieder. 2002. Impact ofmicrofinance on food security andpoverty alleviation. In:Zeller, M., R.Meyer (eds). The Triangleof microJinance, Johns Hopkins University Press, Baltimore: 221-240. Singh, I.,L.Squire, andJ. Strauss. 1986. Agricultural householdmodels: Extensions, applications andpolicy. Baltimore: Johns Hopkins. Smaling, E.M.A., Stoorvogel, J.J.and P.N. Windmeijer. 1993. Calculating soil nutrientbalances in Africa at different scales. 11.District scale. Fertility Research. 35,237-250. Ssali, H.2002. Soil organic matter inUganda andits relationship to major farming systems. Resourcepaper submitted to the Intemational FoodPolicy ResearchInstitute, Washington D.C. Stokey, N.LandR.E. Lucas. 1989.RecursiveMethods in Economic Dynamics.Cambridge,Massachusetts:Harvard UniversityPress. Stoorvogel, J. J., and E.M.A. Smaling. 1990. Assessment of soil nutrient depletion in sub- Saharan Africa: 1983-2000. Report 28. Wageningen, The Netherlands: Winand Staring Centre for IntegratedLand, Soil andWater Research. SundayNews, February 13,2005. Tanzaniaweekly Newspaper, Dar es SalaamTanzania. Swinton, S.M.,G. Escobar and T. Reardon. 2003. Poverty and the environment inLatinAmerica: Concepts, Evidence andPolicy Implications. WorldDevelopment 20(10): 1-8. Tiffen, M., M.Mortimore, andF. Gichuki. 1994. Morepeople - less erosion: Environmental recovery in Kenya. London, UK: Wiley and Sons. UBOS (UgandaBureau of Statistics) 2003. UgandaNationalHousehold survey2002/2003. Report on the socio-economic survey, UBOS EntebbeUganda. 46 WCED. 1987."World commission on environment anddevelopment, our common future," Oxford University Press, Oxford andNew York. Wortmann, C. S. and C. A. Aledu 1999. Uganda's Agroecological Zones: A guide for Planners and Policy Makers. Kampala, Uganda: Centro Intemacional de Agricultura Tropical (CIAT). Wortmann, C.S. and C.K. Kaizzi. 1998. Nutrient balancesand expectedeffects o f altemative practices infarming systemso fUganda. Agriculture, Ecosystems and Environment 71(1-3): 115-130. Woelcke, J., Berger, T., Park, S. 2002. Landmanagement and technology adoption inUganda: An integratedbio-economic modeling approach, In:Nkonya, E., Sserunkuuma, D.,Pender, J. (eds). Policiesfor improved land management in Uganda: Second national workshop. EPTD workshop summarypaperNo. 12:131-136.Wood, S., K.Sebastian, F.Nachtergaele, D.Nielsen, andA. Dai. 1999. Spatial aspects ofthe designandtargetingofagricultural development strategies. Environment andProduction Technology DivisionDiscussion Paper 44. Washington, DC: Intemational FoodPolicy ResearchInstitute. 47 Appendix 1:SelectedDistricts,Communitiesandhouseholds (bimodal low rainfall) Soroti 10 70 79.0 High Low potential Northeast (Unimodal medium) Total 123 851 60.4 48 Appendix 2 :Descriptive statistics of plot and household levelvariables Variable Observations Mean Std. Dev. Min Max Use slash & burn?Yes=l, 0-0 3738 0.198 0.399 0 1 Practicefallow? Yes=l no=O 3738 0.206 0.404 0 1 Practicecrop rotation? yes=l no=O 3738 0.220 0.414 0 1 Top slope 3363 0.140 0.347 0 1 Bottom 3363 0.578 0.494 0 1 Flat 3363 0.110 0.313 0 1 Average slope ("A) 2750 8.024 9.363 0 60 Topsoil depth(cm) 2504 27.660 11.389 4 80 Practiceagroforestry?` Yes=l no=O 3625 0.399 0.490 0 1 1 ~~~~ 3625 0.209 0.407 0 1 Have other NRMin~estment?~ Yes =1 no =0 3625 0.053 0.223 0 1 Farmsize (acres) 851 4.316 5.087 0.123 51.819 Secondary 851 0.148 0.315 0 1 Post-secondary 851 0.070 0.229 0 1 Sex of householdhead (male =1, female=O) 851 0.817 0.387 0 1 49 Non-farm activity 851 0.306 0.461 0 1 Livestock 851 0.022 0.146 0 1 population density 851 9.679 3.025 1.264 402.333 Community wage rate (Ush) 851 1279.683 1.881 475 10000 Type o fcrop produced (cf annual crop) Perennial 3570 0.231 0.422 0 1 Pasture 3570 0.031 0.172 0 1 Northwestern farmlands 851 0.133 0.340 0 1 Northernmoist farmland 851 0.203 0.402 0 1 Mt.Elgonfarmland 851 0.041 0.198 0 1 Southwestern grass-farmland 851 0.137 0.343 0 1 Southwestern highlands 851 0.241 0.428 0 1 Includes: Live barriers, planting trees inplot and on bunds. Includes: stone bunds, fanya juu & fanya chini (bench terraces), drainage trenches, irrigation structures, and grass or other vegetative strips. Includes fish ponds, fences, paddocks, and pasture management A standard animal with live weight of250 kgis calledTLU(Defoer, et al, 2000). Average TLUfor common livestock inUganda area: C o ~ 0 . 9oxen = 1.5, sheep or goat =0.20, and calf =0.25. , 50 Appendix 3: Determinantsof landmanagementpractices(Probit models) Variable Slash & bum fallow Crop Crop Short- rotation residue term SWC Natural capital Ln(Average slope (%)) 0.074 0.000 0.060 0.181** 0.262*** Ln(Top soil depth (cm)) 0.048 0.140 0.218** 0.133 -0.175 Nutrient stock (kg/ha) Ln(nitrogen ) 0.005 0.140* -0.033 -0.138 0.024 Ln(Phosphorus) 0.083 -0.076 0.039 -0.021 0.019 Ln(Potassium) -0.082 -0.030 0.016 0.046 0.022 Investmentonplot dummies (yes= 1, no=O) Agroforestry -0.087 0.182** 0.138* 0.137 0.052 SWC structures 0.077 0.104 0.055 0.444 *** 0.198 Perennialcrop (cf annual crop) -0.265*** -0.353*** -0.240** -0.199* -0.252** Other NRMinvestment 0.186 0.108 0.268* 0.475*** 0.062 Ln(p1ot area inacres) 0.120*** -0.015 -0.001 0.067 0.124** Ln(farm areainacres) -0.062 0.168*** 0.059 -0.132* -0.264*** Physical capital Ln(TLU) 0.010 -0.127*** -0.112* -0.088 0.146* Ln(va1ue o f equipment in000' Ush) 0.023 -0.009 0.015 0.055* -0.008 Human capital Proportion o f female householdmembers with ....(cf no formal education) Primary education -0.038 -0.043 -0.097 0.169 -0.008 Secondary education -0.265 -0.074 -0.211 0.034 0.133 Post secondary -0.315 -0.413 0.118 0.162 -0.105 Proportion o fmale household members with .... (cfno formal education) Primary education 0.009 -0.065 -0.038 -0.027 -0.145 Secondary education -0.372** -0.159 0.204 0.054 -0.329 Post secondary 0.003 0.118 0.385 ** 0.213 -0.732** Male household head 0.107 -0.122 0.147 0.282 0.137 Ln(househo1d size) -0.082 -0.058 0.009 -0.142 -0.007 Proportion o f land owned by women 0.144 -0.054 0.281 0.380 0.134 Primary source o f income (cf crop production) Nonfarm -0.237** 0.248*** 0.004 -0.119 -0.135 Livestock production -0.183 -0.482 -0.231 0.075 -0.032 Village leveifactors Ln(distance from plot to residence inkm) 0.119 0.099 -0.289*** 0.054 0.232*** P M I 0.000 -0.000 -0.000 0.001 0.002** Ln(distance to all weather road inkm) 0.129** -0.060 -0.049 0.020 -0.007 Household has access to credit 0.011 -0.060 -0.249*** -0.172 -0.090 Ln(# o f contact hours with extension 0.074 -0.036 -0.007 0.081 0.124 workeriyear) Householdparticipates inNAADS 0.244 -0.119 -0.190 0.176 -0.29 1 program Landtenure system(cf freehold and leasehold) Customary 0.329** -0.146 -0.143 0.537*** -0.244 Mailo -0.014 -0.320 -0.096 0.193 -0.506** Ln(popu1ation density per km2) 0.006 -0.002 -0.016 0.015 0.024 Ln(vil1age wage rate inUsWday) 0.147 0.003 0.166* 0.151 0.123 Agroecological zones (cf Lake Victoria crescent) Northwest moist zone (West Nile) -0.109 0.421 * -0.087 0.107 -0.085 Northern moist zone -0.501** 0.807*** 0.522*** 0.254 -0.461 M t Elgonzone -0.570* -0.023 0.404 1.286*** 0.236 Southwestern grassland -0.499** 0.042 -0.035 0.471* -0.632** Southwestern highlands -0.554** 0.641*** 1.080*** -0.104 -2.701*** Constant -1.932 -1.617 -3.187*** -3.163** -2.122 51 Appendix 4: Determinantsof inputuse (purchasedseed, inorganicfertilizer andorganic residues applied) (Probit models) Variable Purchased seed Inorganic Organic residues fertilizer Natural capital Ln(Average slope (%)) -0.027 -0.121 0.025 Ln(Top soil depth (cm)) -0.062 0.209 -0.218**- Nutrient stock -(kgiha) Ln(nitrogen ) -0.074 -0.041 0.048 Ln(Phosphorus) -0.020 0.195 0.078 Ln(Potassium) -0.058 -0.069 0.106*+ Investment on plots dummies (yes=l, no = 0) Practice agroforestry -0.039 -0.094 -0.109 Have SWC structure? 0.102 0.228 -0.060 Perennial as dominant crop on plot? (cf annual crop) 0.008 -0.070 0.354***+++ Have other NRMinvestment? -0.285*- -0.075 -0.301 Ln(D1ot area inacres as measured by GPS) 0.093**++ -0.057 0.03 1 Log(farm area in acres) -0.185***--- 0.178 -0.013 Physicalcapital ln(Trooica1 livestock unit) -0.011 0.039 0.060 Ln(vaiue o f equipment in'Ush `000) -0.024 -0.053 0.044*+ Human capital Share o f female household members with ....(cf no formal education) Primary education -0.066 -0.151 0.099 Secondary education -0.362**-- -0.468 0.345*++ Post-secondary education -0.038 -0.940 -1.094*- Share o f male household members with ....(cf no formal education) Primary education 0.169*+ 1.154***+++ -0.147 Secondary education 0.035 1.128***+++ -0.222 Post-secondary education 0.250 1.576***+++ -0.155 Sex o f household head. Male = 1, N o = female -0.034 -0.430- 0.283* Ln(Househo1d size) 0.023 0.264 0.164 Share o f farm owned by women -0.227 -0.595- 0.278 Primary source o f income o f household head (cf crop production) Non-farm -0.087 0.055 -0.160 Livestock 0.107 -0.391 Village levelfactors Ln(Distance from plot to residence inkm) -0.168**-- -0.067 -0.277**-- Potential market integration (PMI) -0.000 0.004 0.001 Ln(Distance from plot to all-weather road+l) 0.100*+ 0.187 -0.097 Ln(Number o f extension visits+l) 0.020 0.321** 0.062 Householdparticipate inNAADS activities? Yes=l, no=O 0.127 0.608* -0.025 Householdhas access to credit? Yes=l no =O -0.157* -0.875*** 0.059 Land tenure o f plot (cf freehold and leasehold) Customary 0.067 -0.751*- 0.028 Mailo 0.005 Dropped 0.557**++ Ln(popu1ation density per km') 0.003 -0.092 -0.034 Ln(vil1age wage rate per day inUsh) 0.034 0.228++ -0.098 Agroecological zone (cf (Lake Victoria crescent) Northwest moist zone -0.557***-- **++t -0.065 Northern moist zone -1.032***--- 3.083*0.691+ -1.162***--- MtElgonzone -0.664**-- 3.518***+++ 0.946***+++ Southwestem grassland -0.329*-- 0.401*+ Southwestem highlands -0.041 0.488 -0.354 Constant 1.581 -7.973**--- -2.362 Legend: * p<.l; ** p<.05; *** p<.o1 +I-,++/--, +++I---means the associated coefficient is significant at p<.1; p<.05; and p<.O1 inthe reducedmodel equation that excludedpotentially endogenous variables. 52 Appendix 4 (cont'd): Determinantsof inputuse Variable Ln(pre-harvest labor) Natural capital OLS IV Ln(average slope %) 0.003 0.011 Ln(topsoi1 depth (cm)) -0.036 0.013 Ln(nitrogen stock kg/ha) 0.140*++ 0.143* Ln(P stock kgha) 0.030 -0.000 Ln(Kstock kgiha) 0.001 0.001 Land investment on plot dummies (yes=l no=O) Practice agroforestry? -0.055 -0.034 Have SWC structure? 0.060 0.078 Perennial crop as dominant crop grown on plot? (cf annual crop) -0.034 -0.052 Have other NRMinvestment? -0.039 0,008 Log(farm area in acres) -0,260***--- -0.251*** Physicalcapital ln(Tropica1 livestock unit) 0.000 0.034 Ln(va1ue of equipment in Ush `000) 0.005*** 0.008 Human capital Share o f female household memberswith ...... (cf no formal education) Primary education 0.230***+++ 0.229** Secondary education -0.022 -0.008 P-secondaryeducation -0.511**-- -0.454' Share of male householdmembers with ...... (cf no formal education) Primary education 0.121 0.111 Secondary education -0.020 -0.003 Post-secondaryeducation -0.045 -0.025 Sex o fhousehold head. Male = 1, No= female -0.035 -0.061 Ln(Househo1d size) 0.100 0.094 Share o f farm owned by women 0.010 0.001 Primary source of income o fhousehold head (cf crop production) Non-farm -0.156*- -0.164* Livestock -0.109 0.022 Village level factors Ln(Distance from plot to residence in km) 0.016 0.019 Potential market integration 0.000 0.000 Ln(Distance from plot to all-weather road+l) -0.024 0.012 Ln(Number ofextension visits+l) 0.084 -0.141 Household has access to credit? (yes=l no=O) -0.178* -0.216 Does household participate in NAADS activities?Yes=l, no=O 0.019 0.613 Land tenure of plot (cf freehold and leasehold) Customary 0.315***++ 0.266** Mailo land 0.291*+ 0.335* Ln(popu1ation density per km2) -0.078*- -0.075** Ln(vil1age wage rate per day in Ush) 0.005 0.002 Agroecological zone (cf Lake Victoria crescent) Northwestmoist zone -0.144 -0.138 Northem moist zone -0.077 -0.085 M t Elgon zone 0.104 0.084 Southwestem grassland 0.390**+ 0.336* Southwestem highlands 0.404**+ 0.310 Constant 4.052***+++ 4.138*** Wu-Hausman test of exogeneity of participation variables (P>?) 1.000 Relevance tests of excluded variables (P>?); Participation in ... Extension 0.000 NAADS 0.000 Credit 0.387 Hansen J test overidentification restrictions (P>?) 0.067 Legend: * p<.l; ** p<.05; *** p<.o1 +/-,++/--,+++/---meansthe associatedcoefficient is significant atpC.1; p<.05; and p<.Ol inthereduced modelequationthat excludedpotentially endogenousvariables. 53 Appendix 5: Factorsaffectingplotproductivity Variable Ln(crop productivity) OLS IV' Natural capital Ldvalue of seedmrchased inUsh+1) 0.016*** 0.080 Lntvalue ofinorganicfertilizer purchasedinUsh+1) 0.027 0.060 Ln(value of organic fertilizer appliedinUsh+l) 0.024** 0.043 Ln(pre-harvest labor usedonplot+l) 0.086*** 0.131 Were the crop residuesincorporatedinto plot?Yes=l no=O 0.241*** 1.267 Ln(average slope %) 0.022 Ln(topsoi1depth(cm)) 0.338 **++ *0.004 0.223 ** Ln(nitrogenstock kgha) 0.256***+++ 0.257*** Ln(P stockkgha) 0.018 0.001 Ln(Kstock kgha) 0.001 0.070 Landinvestmentsonplot dummies(yes=l no=O) Practiceagroforestry 0.195***+ 0.214* Have SWC structure? 0.321***++ 0.277* Perennialas dominantcrop grownon plot?(cf annualcrop) Yes=l, no=O 0.156***++ 0.123 Have other NRMinvestment? -0.301*** -0.179 Ln(plot areainacres as measuredby GPS) -0,101 ***--- -0.083 Lo&farm areainacres) -0.549***_-- -0.499*** Physical capital ln(Tropica1livestockunit) 0.118*** 0.170** Ln(va1ue ofequipment inUsh'000) -0.033*** -0.046 Humancapital Share of female householdmemberswith ....(cf no formal education) Primaryeducation 0.030 0.036 Secondaryeducation -0.146 0.921** -0.044 Post-secondaryeducation 0.803***+ Share of malehouseholdmemberswith . . ..(cfno formal education) Primaryeducation 0.727***+++ 0.147** -0.031 Secondaryeducation 0.640*** Post-secondaryeducation 0.105 0.099 Sex ofhouseholdhead. Male= 1, No = female 0.522***+++ 0.244** 0.272 Ln(Househo1dsize) 0.495*** Shareof farm owned by women 0.038 0.242 Primarysource ofincomeof householdhead(cf cropproduction) Non-farm 0.223*** 0.198 Livestock -0.529***-- -0.950*** Village level factors Ln(Number of extensionvisits+l) 0.099***+ 0.105 Doesthe householdparticipateinNAADS activities?Yes=l, no=O 0.152*** ' 0.406*** 0.187 Householdhas accessto credit? (yes=l no=O) 0.392***+ Agroecologicalzone (cf LakeVictoria crescent) Northwestmoistzone -0.629***--- -0.649***-- -0.617** Northemmoistzone -0.364 MtElgonzone 0.384** 0.134 Southwestemgrassland 0.485***++ 0.613*** Constant 0.753 -0.078 Wu-Hausmantest of exogeneity of landmanagementpractices andparticipationvariables (P>?) 0.948 Relevancetests of excludedvariables (P>& Value of seed 0.000 Value o f inorganicfertilizer 0.0072 Value of organic fertilizer 0.000 Labor 0.000 Crop residue 0.000 Hansen J test overidentificationrestrictions (P'?) 0.596 The identification o f the landmanagement and input use variables was a problemwhen the participation variables (participation inNAADS programand traditional extension services and access to credit) were included as endogenous variables. The exogeneity o f the participation variables was tested by runninga model that excluded the land management and input use variables -thus assuming the participation variables were the only endogenous variables. The Wu-Hausman test of the model failed to reject the exogeneity o fthe participation variables at p = 1.OOO. The 54 relevance test o f the excluded variables also showed a P>2=O.OOO for NAADS, Extensionand Credit endogenous variables. The corresponding Hansen J test of overidentification P>2 = 0.552. Hence to improve identification ofthe land management practices and input use variables, the participationvariables were treated as exogenous variables in the IV model reported inthis table. Legend: * p<.l; ** p<.o5; *** p<.O1 +I-,++I--,+++I---means the associatedcoefficient is significant at p<.1; p<.O5; and p<.Ol inthe reduced model equation that excluded potentially endogenousvariables 55 Appendix 6: Factors affectingper capitahouseholdincome Ln(percapitaincome) Variable OLS I V Natural capital Ln(average slope %) -0.004 -0.003 Ln(topsoi1depth (cm)) 0.005 0.005 Ln(nitrogen stock kgha) 0.060 0.074 Ln(P stock kg/ha) 0.265***+++ 0.250*** Ln(Kstockkg/ha) -0.083 -0.101 Land investmentonplot dummies (yes=l no=O) Practiceagroforestry? 0.180 0.172 Have SWC structure? 0.474***+++ 0.508*** Perennialas dominant croponplot?(cf annualcrop) Yes=l, no=O 0.421***+++ 0.408*** Haveother NRMinvestment? 0.953**++ 1.102** Log(farm area inacres) 0.388***+++ 0.418*** Physicalcapital In(Tropica1livestock unit) 0.222***+++ 0.209* ** Ln(value of equipment inUsh`000) 0.025 0.030 Human capital Share of femalehouseholdmemberswith ..(cf no formal education) .. Primaryeducation 0.052 0.096 Secondaryeducation -0.244 -0.210 Post-secondaryeducation 0.378 0.198 Share ofmale householdmembers with ....(c fno formal education) Primaryeducation 0.168 0.099 Secondaryeducation 0.404**++ 0.342* Post-secondaryeducation 0.369+ 0.164 Sex ofhouseholdhead. Male = 1, No = female 0.228+ 0.218 Ln(Househo1dsize) -0.206 -0.301** Share offarm owned by women 0.100 0.206 Non-farm as primary sourceof income for householdhead?Yes=l, no=O 0.207*++ 0.162 Livestock as primary sourceof incomefor householdhead?Yes=l, no=O -0.516 -0.766** Village levelfactors Ln(distancefrom plot to residence+1 inkm) 0.155*++ -0.173** Potentialmarket integration 0.001 0.000 Ln(distancefrom plot to all-weather roadinkm+1) -0.155**- 0.147 Householdmemberbelongsto savings& credit association 0.449*** 0.874* Doesthe householdparticipateinNAADS activities?Yes=l, no=O 0.035 1.376** Ln(Number of extension visits+l) 0.004 -0.048 Share of landunder .. tenure of (cf freeholdandleasehold) ... Mailo -0.053 -0.023 Customary 0.204 0.182 Ln(populationdensityper kmz) 0.013 0.023 Ln(vil1age wage rateper day inUsh) 0.078 0.056 Northwestmoistzone -0.354 -0.538* Northernmoistzone 0.241 0.069 Mt Elgonzone 0.066 -0.124 Southwesterngrassland 0.651***+++ 0.410 Southwesternhighlands 0.229 -0.004 Constant 2.685 **++ 3.215** Wu-Hausmantest of exogeneity of participationvariables (P>?) 1.ooo Relevancetests of excludedvariables (P>?): Participationin ..... Extension 0.353 NAADS 0.000 Credit 0.000 HansenJtest overidentificationrestrictions (P>?) 0.235 Legend: * pc.1; ** pC.05; *** pc.01 +I-,++I--,+++I---meansthe associatedcoefficientis significant atpc.1; pC.05; and pc.01 inthe reducedmodel equationthat excludedpotentiallyendogenous variables 56 Appendix 7: Pre-NAADSvalue of crop productionper acre inNAADSvs non-NAADSsub- counties of sample districts MeanValue of Crop Production per Acre, 199912000 Statistical First Year in (`000 USh.lacre) (no. of observations) significance District NAADS Non-NAADS NAADS (plevel) sub-counties sub-counties mean Standard mean Standard error error h a 200112002 246.2 18.1 253.5 (65) 20.7 0.8417 (234) Kabale 200112002 470.5 60.7 303.3 23.3 0.0105** (137) (n=146) Soroti 2001/2002 124.8 26.4 126.4 (65) 12.9 0.9405 (143) Iganga 200212003 272.8 11.3 318.0 (65) 47.4 0.1598 (303) Lira 200212003 99.3 6.5 78.5 (61) 10.5 0.1420 (229) Mbarara 200212003 298.4 12.9 342.2 (31) 50.5 0.2912 Source: Data from 199912000UNHS Appendix 8: Pre-NAADSincomeper capitainNAADSvs. non-NAADSsub-countiesof sample districts MeanIncome per Capita, 199912000 Statistical First Year in (`000 USh.) (no. of observations) significance District NAADS Non-NAADS NAADS (p level) sub-counties sub-counties mean Standard mean Standard error error Arua 200112002 238.5 12.5 213.5 (65) 23.3 0.3524 (233) Kabale 200112002 258.0 14.5 264.8 17.4 0.7686 (137) (146) Soroti 200112002 205.8 39.6 226.5 (65) 26.6 0.7002 (143) Iganga 200212003 232.9 16.2 254.8 (65) 20.9 0.4654 (303) Lira 200212003 143.5 8.4 131.7 (61) 28.3 0.5972 (229) Mbarara 200212003 328.8 17.7 320.7 (31) 32.6 0.8835 Source: Data from 199912000UNHS 57 Appendix 9: Determinants of soil erosion [ln(Soil erosion)] Variable OLS IV Natural capital Ln(average slope %) 0.760* *+++ * 0.726** * Ln(topsoi1 depth (cm)) -0.080- -0.099 Land investment on plot dummies (yes=l no=O) Practice agroforestry? -0.227**-- -0.177* Have SWC structure? 0.070 0.098 Perennial as dominant crop on plot? (cf annual crop) yes=l no=O 0.011 0.052 Have other NRMinvestment? -0.359** -0.372* Ln(plot area inacres as measuredby GPS) -0.031 0.001 Log(farm area in acres) -0.055- -0.040 Physical capital In(Tropica1 livestock unit) -0.087 -0.21I** Ln(va1ue o f equipment in Ush `000) 0.012 -0.008 Human capital Share o f female household members with .....(cfno formal education) Primary education 0.195** 0.105 Secondary education -0.184 -0.317 Post-secondary education 0.348*+ -0.006 Share o f male household members with ..... (cf no formal education) Primary education -0.045 -0.116 Secondary education 0.194 0.175 Post-secondary education -0.038 -0.093 Sex o f household head. Male = 1, No = female 0.052 0.370* Ln(Househo1d size) 0.295 **+++ 0.269* Non-farm as primary source of income for household head? Yes=l. no=O -0,007 -0.009 Livestock as primar; source o f income for household head? Yes=]; no=O -0.697 *** -1.445*** Village levelfactors Ln(Distance from plot to residence inkm) 0.028 0.074 Potentialmarket integration 0.000 0.000 Ln(Distance from plot to all-weather road+]) 0.041 0.058 Ln(Number o f extension visits+l) 0.001 0.782*** Does the household participateinNAADS activities? Yes=l, no=O 0.166 -0.282 Household has access to credit? Yes=l no=O -0.016 0.043 Land tenure o f plot (cf freehold and leasehold) Customary 0.161 0.276* Mailo 0.394**+ 0.239 Share o f farmowned by women 0.137 0.521** Ln(popu1ation density per km') -0.030-- -0.002 Ln(vil1age wage rate per day in Ush) -0.245***--- -0.276 *** Agroecological zone (cf Lake Victoria crescent) Northwest moist zone -1.054*** Northem moist zone -1.279*** MtElgonzone -0.022 Southwestern grassland -0.753 *** Southwestern highlands 1.036*** Constant 3.93 1***+++ 3.848*** Wu-Hausman test o f exogeneitv o f DarticiDation variables ( P > h 0.763 _ _ I _ \ I., Relevance o f excluded variables (P;?) Participation in: Extension 0.000 NAADS 0.000 Credit 0.000 . ... Hansen J test o f over identifyingrestrictions (Pz?) 0.377 +/-;++/--,+++/---means Legend: * p<.1; ** p<.05; *** p<.O1 the associated coefficient is significant at p<.1; p<.O5; and p<.Ol inthe reducedmodel equation that excluded potentially endogenous variables 58 Appendix 10: Determinantsofnutrientbalances Nitrogen (N) Phosphorus(P) Variable OLS I V OLS I V Naturalcapital Ln(average slope %) -21.657***--- -21.541*** -5.064* **--- -6.430 Ln(topsoi1depth (cm)) 6.082 5.474 0.321 -2.560 Land investmenton plot dummies. Yes=l no=O Practiceagroforestry? 9.758 8.732 2.416++ -0.714 Have SWC structure? 26.928**++ 22.581* 5.069***+++ -5.059*** Perennialas dominant crop onplot?(cf annual) -29.334***--- -29.739*** -2.712- -2.848 Haveother NRMinvestment? 3.461 6.932 -3.052 -0.262 Ln(p1ot area inacres) 4.169+ 3.993 1.155*+ 2.037 Log(farm area inacres) -2.613 -2.762 -0.499 3.814* Physicalcapital ln(Tropica1livestockunit) -12.191**--- -6.130 -0.455 -3.470 Ln(va1ueof equipment in Ush `000) -1.090 -0.179 0.291 1.047 Human capital Share of female householdmemberswith .....(cf no formal education) Primary education -17.569**- -14.809 -2.455 -0.595 Secondaryeducation -15.613 -1 1.389 3.306 1.065 Post-secondaryeducation -2.957 15.927 -4.065 0.541 Primary education 10.930 14.545 1.557 -2.247 Secondaryeducation 12.046 15.403 0.268 4.960 Post-secondaryeducation -35.376- -26.943 -7.774*- 1.252 Sex of householdhead. Male = 1, No = female -0.789 -8.034 -1.890 2.45 1 Ln(Househo1d size) 8.671 9.697 2.844 1.842 Shareof farm ownedby women -3.485 -14.508 -2.158 -4.365 Major source of incomeof householdhead(cf cropproduction) Non-farm 4.005 6.356 3.414 4.247* Livestock -12.006 15.386 11.646 17.334 Village levelfactors LdDistance fromdot to residenceinkm+l) 4.738++ 3.220 1.407+ -4.897 Potentialmarket integration 0.029 0.014 -0.013 -2.940 Ln(Distancefrom plot to all-weather road+l) -8.665 -8.296 -4.024***-- 3.984 Ln(Number of extensionvisits+1) 0.817 -26.776 -1.426 0.730 HouseholdparticipatesinNAADS activities?Yes=l, 8.626 12.064 3.030 -0.017 no=O Householdhas access to credit 4.213 -6.177 1.826 -4.097* ** Landtenure ofplot (cf freeholdand leasehold) Customary 13.559 10.475 0.374 -0.131 Mailo land -32.360* -22.5 19 -0.562 1.898 Ln(popu1ationdensityper km2) -5.438 -6.639* -0.872 -1.123 Ln(vil1age wage rateper day inUsh) 11.029 11.395 1.480 1.392 Agroecologicalzone (cf Lake Victoria crescent) Northwestmoistzone 16.006+ 23.812 -0.398 1.012 Northem moistzone 5.203 9.336 1.391 2.855 MtElgonzone 17.815 25.987 4.670+ 8.168 Southwesterngrassland -10.801 -5.111 6.364 Southwesternhighlands 6.823 9.723 -9.2713.394+ ***--- -7.092 Constant -127.400 -116.484 -9.682 -6.338 Wu HausmanTest of exogeneity of participation variables (P>?) 1.000 1.000 Relevanceof excluded variables(P>?) Extension 0.000 0.000 NAADS 0.780 0.780 Credit 0.000 0.000 HansenJtest of over identifying restrictions(P>?) 0.708 0.259 +/-T Legend: * p<.l; ** p<.05; *** p<.o1 ++/--,+++/---meansthe associatedcoefficientis significantat p<.1; p<.05; and p<.Ol inthe reducedmodel equation that excludedpotentiallyendogenous variables 59 Appendix 10: Determinantsof nutrientbalances (cont'd) Potassium (K) N P K Variable OLS IV OLS IV Natural capital Ln(average slope %) -24.822**--- -81.727** -57.470***--- -111.710** Ln(topsoi1depth (cm)) 5.341 32.778 22.167 43.447 Land investment on plot dummies (yes=l no=O) Practice agroforestry? 23.324 140.264* 39.669*+ 132.028 SWC structure? 7.703 -22.719* 29.617 -56.014*** Perennial as dominant crop on plot? (cf -49.909**-- -46.808** -89.473***--- -87.451*** annual crop) Yes=l, no=O NRMinvestment? 7.205 15.222 28.907 32.278 Ln(p1ot area inacres +1) 2.367 22.624 9.915+ 40.870* Log(farm area inacres) 8.016 11.340 0.523 29.388 Physicalcapital Ln(Tropica1 livestock unit) -7.037 24.136 -20.679-- 52.029 Ln(va1ue o f equipment inUsh `000) 3.499 2.641 -0.486 10.626 Human capital Share o f female household members with ... (cf no formal education) Primary education -5.207 8.992 -28.749 1.340 Secondary education -26.403 -1.061 -26.458 -9.583 Post-secondary education 16.468 4.106 3.840 1.340 Share o f male household members with . .(cfno formal education) Primary education -4.652 11.169 4.718 -10.038 Secondary education -3.412 -31.331 16.497 -31.931 Post-secondary education -33.209 26.848 -67.749 26.234 Sex o fhouseholdhead. Male = 1, N o = 11.360 2.567 17.810 13.313 female Ln(Househo1d size) 13.834 -16.320 28.303 1.124 Share o f farm owned by women 9.305 -18.864 -2.022 -42.155 Primary source o f income o fhouseholdhead (cf crop production) Non-farm 25.328 22.449 26.015 23.557 Livestock 143.301 213.152 36.236 125397 Village level factors Ln(Distance from plot to residence inkm) 7.808 -40.475 11.475 -71.358 Potentialmarket integration -0.026 -17.525 0.013 -19.217 Ln(Distance from plot to all-weather -34.231***-- -1.277 -41.986**- 13.620 road+l) Ln(Numberofextension visits+l) 0.457 13.904 -1.564 18.028 Does the householdparticipate inNAADS 16.472 -0.054 28.152 -0.013 activities? Yes=l, no=O Household has access to credit 3.558 -31.724*** 13.768 -38.866** Landtenure ofplot (cf freehold and leasehold) Customary 43.307**+ 32.559 50.608 35.662 Mailo -28.277 -20.292 -87.790* -74.356 Ln(popu1ation density per km2) -8.572 -10.684 -15.925 18.889* Ln(vil1age wage rate per day inUsh) 6.319 13.556 17.783 25.926 Agroecological zones (cf Lake Victoria crescent) Northwest moist zone -6.158 11.983 14.694 45.168 Northem moist zone -14.710 -16.377 -7.969 -3.752 MtElgonzone -20.962 -41.043 -15.476 -34.609 Southwestem grassland -108.721***--- -137.370*** 119.893***-- -.148.522*** Southwestem highlands -79.381*-- -112.145** -89.355 -122.200** Constant -93.835 -121.973 -242.256 -267.752 Wu-Hausman test of exogeneity o fparticipation variables -- 0.976 60 (P>?> Extension 0.000 Extension 0.000 NAADS 0.780 NAADS 0.780 Credit 0.000 Credit 0.000 Hansen J test of over identifyingrestrictions(P>>l) 0.583 0.580 Legend:* p<.l; ** p<.05; *** p<.o1 +/-, ++/--,+++/---means the associated coefficient is significant at p<.l; p<.05; and p<.O1in the reducedmodel equation that excluded potentially endogenous variables 61 4 I I T ++ Appendix 12. TheoreticalDynamicHousehold In this Appendix, a theoretical household model of livelihood strategies and land management is developed. The model incorporates household investment decisions-with investments broadly defined to include investments in physical, human, natural, and financial capital-as well as annual decisions regarding crop choice, labor allocation and adoption o f land management practices. Consider a householdthat seeks to maximize its lifetime welfare: where ct is the value o f consumption inyear t, ut() is the single period consumptionutilitg3andthe expectation (Eo) is taken with respect to uncertain factors influencing future income at the beginning o f year t=O. Consumptioninyear t i s givenby: where Lt is gross crop income, 11~ gross livestock income, Iwnet wage income, and I,,, is is is income from nonfarm activities inyear t.24INVwt is a vector o f investments (or disinvestments) in assets during year tyincluding investments inphysical capital (PC,) (livestock, equipment), human capital (HC,) (education, experience, training), "natural capital" (NC,) (assets embodied in natural resources, including land quantity and quality, land improving investments), and financial capital (FCt) (access to liquid financial assets). pw i s the price o f marketed assets, or in the case o f non- marketed assets (e.g., experience), pw i s interpreted as the cost o f acquiring an additional unit o f these assets. Household gross crop income is givenby: where y( ) represents the value o f production per acre farmed, Ati s the area farmed (part o f NCtl, Ct represents the vector of area shares of different annual crops grown by the household2', pctis the vector o f farm level prices o f the different crops, Lcti s the amount o f labor per acre applied, LMtis 22This appendix is adapted from the theoretical model developedinNkonya, et al. (2004). 23This is a generalization of the commonly used discounted utility formulationut(ct) =@u(c,) (e.g., see Stokey and Lucas (1989)). 24The value ofhiredlabor usedincrop andlivestock production is subtracted from net wage income. Costs o fother purchased inputs used inagricultural production canbe treated inexactly the same way. For simplicity o f exposition, labor was treated as the only variable input inagricultural production(it is by far the most important for small farmers inUganda). 25Perennialcrops available for harvest inthe current year are the result o f investment inprior years, and are taken as part o f the land investments on the plot (included inNC,). a vector o f land management practices and input use (use of fallow, crop residues, fertilizer, etc.) used, Tt represents tenure characteristics of the land, ASt represents household access to information and services (e.g., agricultural extension), BPt are other biophysical factors affecting the quantity o f crop production (e.g., rainfall, temperature, etc.), and other variables (NCt, PCt, HCt, FCt) are as defined previously. The physical, human and financial capital o f the household are included as possible determinants of crop production because these assets may affect agricultural productivity ifthere are imperfect factor markets(de Janvry, et al. 1991). Biophysical conditions are modeledina givenyear as dependentupon observable agroecological conditions (a subcomponent ofXvt)and random factors (ubt): where Xvt is a vector including observable agroecological characteristics, market access and populationdensity o fthe village, anduCtrepresents unobservedrandom factors affectingprices. Substituting equations 4) into 3), the value of crop productionfunction i s redefined: In a similar way, livestock income is determinedby labor allocated to livestock activities (Lit),ownership ofland, livestock, andother physicalassets, thehumanandfinancial capital ofthe household, access to information and services, biophysical conditions, access to markets and infrastructure, andpopulation density: Net wage income is givenby: where Lotand Litare the amounts of labor hiredout and inby the household, respectively, and wOt and wit are the wage rates paid for hiredlabor. It i s assumedthat wages may be affected by village level-factors such as agroecological conditions, market access and population density (X,,) that influence the local supply and demand for labor, and by household-level human capital, and other random factors (uwot,Uwit). 65 Nonfarm income i s determined by the labor allocated to nonfarm activities, the physical, human and social capital of the household, access to information and services, the local demand for nonfarm activities as determined by Xvt,and random factors: Labor demand by the householdmust be no greater than labor supply: where Lfti s the supply of household family labor. Most forms o f capital must be nonnegative: io) PC, 2 0, HC, 2 0, NC, 2 0 Financial capital may be negative, however, if borrowing occurs. It is assumed that the household's access to credit is determined by its stocks o f non-financial capital (which determine the household's collateral, potential for profitable investments and transaction costs o f monitoring and enforcing credit contracts): 11) FC,,, 2 -B(PCt ,HC,,NC,) where B is the maximum credit obtainable. Financial assets (or liabilities) grow at the household- specific rate o f interest r, which may be affected by the same factors affecting prices and wages, as well as factors affecting the borrowing constraint: 12) FC,,, =(1 r(Xvt,PC,,HC,,NC,))FC, INV,, + + where INVFct i s investment (or disinvestment) infinancial capital inyear t (a subvector o f INVwtin equation 2)). Physical capital also may grow or depreciate over time, inaddition to changes instocks resulting from investments: 13) PC,,, =(1+g)Pc, +INVpct where g i s a vector of asset-specific growth (or depreciation if negative) rates and INVpctis investment inphysical capital inyear t. Natural capital may depreciate (degrade) over time as a result o f unsustainable resource management practices, as well as being improved by investment. For example, if soil depth is thought as one component of natural capital, this may be depleted by soil erosion as well as restoredby investments insoil conservation: 66 where NCptis taken here to represent soil depth on plot p, E the rate o f erosion (net of the rate of soil formation), uetare random factors affecting erosion, and INVNct is investment inincreasing soil depth inyear t. A similar relation for change insoil nutrient stocks also holds. It is assumed that humancapital does not depreciate or grow without investment. Since these are also non-marketed assets, they are subject to irreversibility constraints: 15) HC,,, 2 HC, Maximizationo f 1) subject to the constraints defined by equations 2), 3), and 5) - 15) defines the household optimizationproblem. Ifthe optimized value o f 1) ("value function") is defined as Vo andnotice that this i s determined by the value ofthe state variables at the beginning of period 0 (PCO, HCo, NCo, FCo), andbythe other exogenous variables inthis system that are determined at the beginning o fperiod 0 (XVo,ASo, Lm), then: V o ( P c o , H c o , ~ ~ o , ~ ~ o y ~ ~=o y ~ ~ o , ~ v o y ~ f , ~ 16) T max E o [ ~ u , ( c , )subject to equations 2), 3), 5)-15) ] t=O Defining Wt = (PCt, HCt, NCt, FCt) and defining VI as the value function for the same problem as in l), butbeginning inyear t=l, the Bellman equation canbe writtenbydetermining the solution inthe first period: where Lois a vector of all labor allocation decisions, LMois a vector o f land management choices on all plots inyear 0, and INVwo i s the vector of investments indifferent forms o f capital inyear 0. Solution of the maximization inequation 17) implicitly defines the optimal choices o f LO,LMo, and INvwo: 67 The optimal solutions for crop choice, labor allocation and land management determine the optimized value of production, land degradation, and household income. Substituting equations 18) and 19) into equation 5), the optimalvalue of crop production26is obtained: Equation 21) forms the basis for empirical estimation of the determinants o f value o f crop production. It will be estimated in structural form, including the impacts o f the endogenous variables (bo,LM,o). The modelwill also be estimatedinreduced form: The reduced form income function is derived by substitutingthe crop value o f production function from equation 22) into crop income equation 3), the labor allocation functions inequation 18) into the other income equations 6)-8), andthen summingup total household income27: Equations 18) -23) are the basis o fthe empiricalwork. 26 The terms relatedto randomvariations inprices (h0) inbiophysical factors ( u ~have been combined into a and ) single random variable reflecting random fluctuations invalue o f crop production (ug) inequation 22). 27 Inthe last part of equation 23),UIOcombines the effects o f the different randomfactors included inthe middle expression (ug, UIO,uw0o,uwio, uno). 68 Appendix 13. EconometricModelsandApproach It is assumed that the value of crop production per acre by household h on plot p (Yhp) is determined by labor use per acre on the plot (Lhp), land management practices on the plot (including use o f inputs), the natural capital (size and quality) of the plot(Nchp), the tenure o f the plot (Thp), the household's endowments of physical capital (PCh), human capital (HCh), and financial capital (FCh), the household's access to agricultural technical assistance (Ash), village and higher level factors influencing comparative advantage (agro-climatic potential, access to markets androads, population density, andwage level) (Xh), and other random factors such as weather ina given year and location (eyvhp). Some o f these factors may have only indirect impacts on crop production, by influencing use o f labor and land management practices (e.g., population density and the wage level). However, these are includedinthe fill specification of the structural model, and then use hypothesis testing to eliminate such factors that have statistically insignificant impacts. The structural modelof crop production is thus: Itis also estimatedthat the following generalreduced formmodel for eachset ofthe dependent variables**: Where 0 Ih i sincome per capita o f household h; 0 Ehpis estimatederosion onplot p ofhouseholdh, usingthe RUSLE; 28See Nkonya, et al. (2004b) for a derivation ofthis empirical model. 69 0 N U t b a l h ,i s a vector o f soil nutrient balances o fmacronutrients, namely nitrogen (N), phosphorus (P), potassium(K) and total nutrient balance (NPK) from householdhat plot p; 0 eYrvh,, elmvhp,eIvh, eInvhp, eOutvhp, andeNBvhp are unobserved random factors affecting the dependent variables invillage v for householdhat plot p. It is likely that the error terms across equation (1)-(7) are not independently distributed hence the need to estimate the models using a system o f equations. Estimating them as single equations reduces the efficiency o f estimation because correlation in error terms across equations cannot be accounted for and cross equation restrictions cannot be imposed. However, estimation o f a system o f equations using such methods as three-stage least squares is not possible because some o f the dependent variables are limited dependent variables, hence their determinants cannot be consistently estimated using standard linear models (Maddala 1983). Fortunately, the inability to estimate a system of equation to account for cross equation relationships does not cause the estimated coefficients to be inconsistent or biased. Hence, each equation i s estimated independently usingeconometric models suitable to the nature o f each dependent variable. Equation (4) i s estimated using a probit model since the dependent variables are dichotomous (e.g. whether or not farmer used inorganic fertilizer, organic fertilizer, purchased seeds, crop rotation, slash-and-burn land preparation method, short-term soil and water conservation practices such as trash-lines, deep tillage, zero tillage, fallow, incorporation o f crop residues). All other equations are estimated using ordinary least squares (OLS), correcting for sample weights and plot clustering (possible non-independence o f error terms across plots within a household) at household level. Equation (1) includes endogenous choices that could cause endogeneity bias. The endogenous choices are land management practices (including inputs) and pre-harvest labor input. The participation variables, namely participation in agricultural extension or rural finance organizations could also lead to endogeneity bias. To address this problem, IV estimation i s used for equations (1) - (3) and (5) - (7); i.e., those equations whose dependent variable i s a continuous variable. IV estimation results in consistent estimates o f the model coefficients, provided that a unique solution to the estimation problem exists and the instrumental variables are uncorrelated with the error term inthe model (Davidson andMacKinnon2004). However, infinite samples, IV estimates are generally biased, and can be more biased than OLS estimates if the instrumental variables used are weak predictors o f the endogenous explanatory variables (Ibid., pp. 324-329; Bound, et al. 1995). Furthermore, identification o f the coefficients o f a linear IV model i s impossible unless restrictions are imposed on the model, such as excluding some o f the instrumental variables from the regression. Inlinear IV estimation, it is necessary to have as many restrictions as endogenous explanatory variables to be able to identify the model, and additional restrictions ("overidentifying restrictions") can help to increase the efficiency o f the model, provided that these exclusion restrictions are valid and that the excluded instrumental variables are significant predictors o f the endogenous explanatory variables. 70 Inthe IV regressions, several community levelvariables are usedas instrumental variables that are excluded from the regressionmodel, including whether or not a community had enacted a bylaw related to natural resource management and the degree o f cropland degradation in a community (which are indicators of awareness of the need for improved land management in a community), the number of program and organizations of different types present in a community (indicators of access to extension and credit), and ethnicity (a proxy for social factors that may influence participationinprograms, livelihood and landmanagement decisions). It i s hypothesized that such variables are significant predictors o f the endogenous variables (i-e., they are "relevant"), but that they do not add additional explanatory power to the regression after controlling for the participation variables and other variables (i.e., the overidentifying restrictions are valid). In estimating equation (l), those explanatory variables that werejointly statistically insignificant also, inthe less restrictedversion ofthe model(including factors such as landtenure, access to markets and roads, population density and wage levels) were excluded from the regression and used as instrumental variables. These are factors that were found to influence crop production only indirectly, via their impacts on labor use andlandmanagementdecisions. In all cases, the relevance of the excluded instrumental variables is statistically tested by computing their joint statistical significance in predicting the endogenous explanatory variables (Bound, et al. 1995). The overidentifying restrictions were using Hansen's J statistic (Davidson and MacKinnon 2004, pp. 366-368), which is consistent under heteroskedasticity (Baum, et al. 2002). Also tested is the consistency of OLS relative to IV using a Durban-Wu-Hausman test (Davidson and MacKinnon 2004, pp. 338-340). Since OLS estimation i s more efficient than IV estimation if the OLS model i s consistent, the OLS model is preferred are reportedif the Hausman test fails to reject the consistency of OLS. Regardless o f the results o f these tests, the OLS and IV results are reported, since IV estimation may be biased infinite samples, as noted above.29 Other estimation and data issues considered included samplingweights, heteroskedasticity, multicollinearity, and outliers. The distribution o f each variable was examined and an appropriate monotonic transformation towards normality was determined using the ladder o f power test, because this improves the model specification (i.e., reduces problems o f nonlinearity, outliers and heteroskedasticity) (Mukherjee, et al. 1998; Stata 2003). The following variables were found to be severely skewed and were transformed towards normality using natural 10garithm:~' stock o f soil nutrients, tropical livestock units (TLU),31 distance from plot to residence and roads, value of equipment, crop productivity, household income, pre-harvest labor intensity, soil erosion, populationdensity, and village wage rate. Regression statistics (coefficients and standard errors) were also adjusted for sample weights, stratification and cluster sampling. Multicollinearity was tested using pair-wise correlations and variance inflation factors (VIF). Pair wise correlation showed very strong correlation o f some variables. For example, ethnicity showed a very strong correlationo f over 0.7 significant at p=O.OO 1 with agroecological zones. Therefore, ethnic group variables were dropped 29As observed earlier, we also report medianregressionresults for comparisonpurposes. 30To preserve observations with zero, the log-transformation was done as follows ln(x+l), where x is the variable being transformed. Hence zero of the untransformed variable will correspond to zero o fthe transformed variable. 31A standard animal with live weight of 250 kg is called TLU(Defoer, et al., 2000). Average TLUfor each livestock category is: Cow = 0.9, oxen = 1.5, sheep or goat =0.20, and calf = 0.25. 71 from the original specification. The overidentification tests (none of which were significant) verified that this and other exclusion restrictions in the IV models are valid. In the final specifications, multicollinearity was not a major concern (maximum VIF = 7) (Mukherjee, et al. 1998). 72