FARM EFFICIENCY, VALUE CHAIN PARTICIPATION, AND FOOD ACCESSIBILITY FOR WHEAT GROWERS OF TANZANIA By WILLIAM BARNOS WARSANGA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMEN T OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2016
Â© 2016 William Barnos Warsanga
To my parents and my family
4 ACKNOWLEDGMENTS This dissertation was made possible with valuable contributions by many individuals and institutions . I would like to a cknowled ge the support, supervision, and encouragement provided by my chairman Dr.Evans, my co chair Dr. Gao, my other committee members Dr. Useche and Dr. Slutsky, and my T anzanian advisor Dr. Msuya , for their guidance, dedication, recommendations , and constructive critic i sm during the preparation of this dissertation. My gratitude thanks go to USAID through IAGRi for their generous financial support throughout my graduate studi es and to the Tanzanian government through MoCU for granting me a study leave abroad. Heartfelt thanks to all government officials including the village leaders and extension officers for their assistance in loc ating the study areas, providing the sa mpling frame, and for briefing the crop understudy. Also much thanks to my research assistants for data collection and data entry. Special thanks to my wife and my two daughters for their emotional support, patience, and understanding during these four yea rs away from home for my graduate studies. I would also like to recognize the invaluable contribut ion from my parents and my brothers and sisters for my past and present accomplishments . You taught and push ed me harder to reach this far, no t enough words can express how thankful I am to you , dad and m o m . Lastly, m uch thanks to classmate s for sharing with one another the struggles and stress es unti l we are here today celebrating our fruitful and success ful accomplis h ment s .
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ ........ 10 ABSTRACT ................................ ................................ ................................ ................... 13 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 15 2 TECHNICAL, ALLOCATIVE, AND ECONOMIC EFFICIENCY OF WHEAT FARMS IN TANZANIA: PROPENSITY SCORE MATCHING AND STOCHASTIC FRONTIER ANALYSIS ................................ ................................ ... 18 Introductory Remarks ................................ ................................ .............................. 18 Concept and Approach of Efficiency in Agriculture ................................ ................. 22 Theoretic al and Analytical Framework ................................ ................................ .... 24 Technical Efficiency ................................ ................................ .......................... 28 Allocative and Economic Efficiency ................................ ................................ .. 31 Testing the Hypothesis ................................ ................................ ..................... 36 Determining Factors Influencing Efficiency ................................ ....................... 36 Value Chain Participation E ffect on Efficiencies; Bias Selection Consideration ................................ ................................ ................................ 38 Empirical Model ................................ ................................ ................................ ...... 41 Technical Efficiency ................................ ................................ .......................... 41 Allocative Efficiency ................................ ................................ .......................... 42 Data ................................ ................................ ................................ ........................ 44 Results and Discussion ................................ ................................ ........................... 45 Household Characteristics and Farm Features ................................ ................ 45 Household and Farm Characteristics by Wards ................................ ............... 48 OLS a nd SFA Estimates for Translog Production Function .............................. 50 OLS and SFA Estimates for Translog Cost Function ................................ ....... 51 Efficiency Scores ................................ ................................ .............................. 52 Distribution of Efficiency Scores (Pooled Sample) ................................ ........... 53 TE distribution ................................ ................................ ............................ 53 AE distribution ................................ ................................ ............................ 53 EE distribution ................................ ................................ ............................ 53 Efficiency Distribution by Wards ................................ ................................ ....... 54 TE distribution by wards ................................ ................................ ............. 54 AE distribution by wards ................................ ................................ ............ 54 EE distribution by wards ................................ ................................ ............ 54 Factors Influencing Efficiency ................................ ................................ ........... 55
6 TE factors ................................ ................................ ................................ ... 55 AE factors ................................ ................................ ................................ .. 57 EE factors ................................ ................................ ................................ .. 59 Impact of Value Chain Participation on Efficiency ................................ ............ 60 Vertical coordination impac t ................................ ................................ ....... 61 TE effect vertically ................................ ................................ ...................... 61 AE effect vertically ................................ ................................ ..................... 62 EE effect v ertically ................................ ................................ ..................... 62 Horizontal coordination impact ................................ ................................ ... 62 TE effect horizontally ................................ ................................ ................. 62 AE effect horizontally ................................ ................................ ................. 63 EE effect horizontally ................................ ................................ ................. 63 Concluding Remarks ................................ ................................ ............................... 64 3 ASSESSMENT OF WELFARE IMPACTS OF TANZANIA WHEAT FARMERS PARTICIPATION IN THE VALUE CHAIN ................................ ............................... 86 Overview ................................ ................................ ................................ ................. 86 Int roductory Remarks ................................ ................................ .............................. 86 Literature Review ................................ ................................ ................................ .... 91 Concept of Value Chain and Value Chain Development ................................ .. 91 Theoretical Framework ................................ ................................ ..................... 94 Propensity Score Matching ................................ ................................ ............... 95 Matching algorithms ................................ ................................ ................... 98 Overlap and common support ................................ ................................ .. 100 Testing the matching quality ................................ ................................ .... 101 Sensitivity analysi s ................................ ................................ ................... 102 Studies with applied PSM approach ................................ ........................ 104 Method and Data ................................ ................................ ................................ .. 106 Method ................................ ................................ ................................ ........... 106 Data ................................ ................................ ................................ ................ 112 Results and Discussions ................................ ................................ ....................... 113 Value Chain Stru cture ................................ ................................ .................... 113 Value Chain Coordination ................................ ................................ ............... 115 Vertical coordination ................................ ................................ ................ 115 Horizontal coordination ................................ ................................ ............ 116 Mean Characteristics of Participants and Nonparticipants of Value Chain ..... 116 Characteristics of vert ical coordination participants and nonparticipants . 116 Characteristics of horizontal coordination participants and nonparticipants ................................ ................................ ..................... 117 ............................. 118 Vertical coordination factors ................................ ................................ ..... 118 Horizontal coordination f actors ................................ ................................ . 119 Covariate Balancing ................................ ................................ ....................... 120 Vertical coordination balancing property ................................ .................. 121 Horizontal coordination balancing property ................................ .............. 121
7 Profit ................................ ................................ ................................ ............ 122 Vertical coordination effect ................................ ................................ ....... 122 Horizontal coordination effect ................................ ................................ ... 123 Sensitivity Analysis ................................ ................................ ......................... 123 Vertical coordination sensitivity analysis ................................ .................. 124 Horizontal coordination sensitivity analysis ................................ .............. 125 Concluding Remarks ................................ ................................ ............................. 125 4 AN ECONOMIC ASSESSMENT OF THE CONTRIBUTION OF WHEAT INCOME TO HOUSEHOLD FOOD ACCESSIBILITY ................................ ........... 14 6 Overview ................................ ................................ ................................ ............... 146 The Issue ................................ ................................ ................................ .............. 146 Literature Review ................................ ................................ ................................ .. 149 Concept of Food Security ................................ ................................ ............... 149 Poverty and Food Accessibility ................................ ................................ ....... 151 Theoretical Framework ................................ ................................ ................... 155 Analytical Framework ................................ ................................ ..................... 158 Empirical Model ................................ ................................ ................................ .... 161 Data ................................ ................................ ................................ ...................... 165 Results and Discussion ................................ ................................ ......................... 167 Land Allocation ................................ ................................ ............................... 167 Price Variation by Crops ................................ ................................ ................. 168 Share of Wheat Income on Actual Food Expenditure ................................ ..... 169 Household Food Poverty Line ................................ ................................ ........ 170 Mean Characteristics of Fo od Accessible and Inaccessible Household Based on Wheat Income ................................ ................................ ............. 171 Factors Influencing Food Accessibility through Wheat Income ...................... 173 Food Accessibility based on Total Crop Income (Restricted Sample Analysis) ................................ ................................ ................................ ...... 174 Food Accessibility based on Total Household Income (Unrestricted Sample Analysis) ................................ ................................ ................................ ...... 175 Concluding Remarks ................................ ................................ ............................. 176 5 CONCLUSION ................................ ................................ ................................ ...... 190 LIST OF REFERENCES ................................ ................................ ............................. 197 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 207
8 LIST OF TABLES Table page 2 1 Household and farm characteristics ................................ ................................ ... 66 2 2 Household and farm characteristics by wards ................................ .................... 67 2 3 Estimates for production function (OLS) and stochastic frontier (MLE) .............. 68 2 4 OLS and MLE estimates of translog cost function ................................ .............. 70 2 5 TE, AE, and EE for wheat production (pooled sample) ................................ ...... 71 2 6 TE distribution by wards ................................ ................................ ..................... 71 2 7 AE distribution by wards ................................ ................................ ..................... 71 2 8 EE distribution by wards ................................ ................................ ..................... 71 2 9 Factors influencing TE (Tobit model) ................................ ................................ .. 72 2 10 Factors influencing AE (Tobit model) ................................ ................................ .. 73 2 11 Factors influencing EE (Tobit model) ................................ ................................ .. 74 2 12 TE effect due to participation in vertical coordination ................................ ......... 74 2 13 AE effect due to participation in vertical coordination ................................ ......... 75 2 14 EE effect due to participation in vertical coordination ................................ ......... 75 2 15 TE effect due to participation in horizontal coordination ................................ ..... 75 2 16 AE effect due to participation in horizontal coordination ................................ ..... 75 2 17 EE effect due to participation in horizontal coordination ................................ ..... 75 3 1 ................................ ........................... 129 3 2 horizontal coordination by wards ................................ ....................... 129 3 3 Characteristics for vertical coordination, participants and nonparticipants ....... 130 3 4 Characteristics for horizontal coordination, participants and nonparticipants ... 131 3 5 Logit estimates of propensity score model for contracted farmers .................... 132 3 6 ................ 133
9 3 7 Covariate balancing for contract and noncontract farmers (caliper 0.07) ......... 134 3 8 Covariate balancing for members and nonmembers farmers (caliper 0.022) ... 136 3 9 Vertical coordination effect (ATT) ................................ ................................ ..... 137 3 10 Horizontal coordination effect (ATT) ................................ ................................ . 137 3 11 Vertical coordination Rosenbaum sensitivity test ................................ .............. 138 3 12 Horizontal coordination Rosenbaum sensitivity test ................................ ......... 139 4 1 Crop land allocation ................................ ................................ .......................... 179 4 2 Crop price variation ................................ ................................ .......................... 179 4 3 Staple food expenditures on various food items per week ............................... 180 4 4 Food (in)accessibility by wards through wheat income ................................ ..... 181 4 5 Mean characteristics of food accessible and inaccessible farmers ................... 182 4 6 Probit model estimation for food a ccessibility by wheat income using unrestricted sample ................................ ................................ .......................... 183 4 7 Food (in)accessibility by wards through total crop income for restricted sample ................................ ................................ ................................ .............. 183 4 8 Probit model estimation for food accessibility through total crop income for restricted sample ................................ ................................ .............................. 184 4 9 Food (in)accessibility by wards through total household income ( un restricted sample) ................................ ................................ ................................ ............. 184 4 10 Probit model estimation for food accessibility with total household income for unrestricted sample ................................ ................................ .......................... 185
10 LIST OF FIGURES Figure pag e 2 1 Per capita consumption gap between wheat and staple food maize and rice in SSA ................................ ................................ ................................ ................ 76 2 2 Production consumption tre nd of wheat in Tanzania ................................ .......... 76 2 3 The distribution of TE scores of the pooled sample in the study area ................ 77 2 4 The distribution of A E scores of the pooled sample in the study area ................ 77 2 5 The distribution of EE scores of the pooled sample in the study area ................ 78 2 6 Technical efficiency distribution by wards ................................ ........................... 78 2 7 Allocative efficiency distribution by wards ................................ ........................... 79 2 8 Economic efficiency dist ribution by wards ................................ .......................... 79 2 9 Vertical coordination histograms before and after matching by Nearest Neighbor algorithm of ratio 1:1 ................................ ................................ ........... 80 2 10 Vertical coordination histograms before and after matching by caliper radius of 0.005 ................................ ................................ ................................ .............. 80 2 11 Vertical coordination histograms before and after matching by caliper radius of 0.01 ................................ ................................ ................................ ................ 81 2 12 Vertical coordination histograms before and after matching by caliper radius of 0.02 ................................ ................................ ................................ ................ 81 2 13 Vertical coordination histograms before and after matching by caliper radius of 0.03 ................................ ................................ ................................ ................ 82 2 14 Vertical coordination histograms before and after matching by caliper radius of 0.1 ................................ ................................ ................................ .................. 82 2 15 Vertical coordination histograms before and after matching by caliper radius of 0.13 ................................ ................................ ................................ ................ 83 2 16 Horizontal coordination histograms before and after matching by Nearest Nei ghbor algorithm of 1:1 ratio ................................ ................................ ........... 83 2 17 Horizontal coordination histograms before and after matching by caliper radius of 0.03 ................................ ................................ ................................ ...... 84
11 2 18 Horizontal coordination histograms before and after matching by caliper radius of 0.04 ................................ ................................ ................................ ...... 84 2 19 Horizontal coordination histograms before and after matching by caliper radius of 0.05 ................................ ................................ ................................ ...... 85 3 1 Wheat grain value chain map ................................ ................................ ........... 140 3 2 Distribution of propensity scores before and after matching the contract and noncontract fa rmers by caliper radius of 0.01 ................................ ................... 141 3 3 Distribution of propensity scores before and after matching the contract and noncontract farmers by caliper radius of 0.03 ................................ ................... 141 3 4 Distribution of propensity scores before and after matching the contract and noncontract farmers by caliper radius of 0.05 ................................ ................... 142 3 5 Distribution of pr opensity scores before and after matching the contract and noncontract farmers by caliper radius of 0.07 ................................ ................... 142 3 6 Distribution of propensity score before and after matching the contract and non contract farmers by nearest neighbor algorithm 1:1 but not good match .... 143 3 7 Distribution of propensity scores before and after matching the members and nonmembers by caliper radius of 0.005 ................................ ............................ 143 3 8 Distribution of propensity scores before and after matching the members and nonmembers by caliper radius of 0.02 ................................ .............................. 144 3 9 Distribution of propensity scores before and after matching the members and nonmembers by caliper radius of 0.022 ................................ ............................ 144 3 10 Distribution of propensity score before and after matching the membe rs and nonmembers by nearest neighbor algorithm of 1:1 ratio but not a good match 145 4 1 The average land allocation across crops and their respective average output per acre for unrestricted sample (N=310) ................................ ......................... 186 4 2 The average land allocation across crops and their respective average output per acre for restricted sample (N=298) ................................ ............................. 186 4 3 Crop price variations in the study area for unrestricted sample ........................ 187 4 4 unrestricted sample ................................ ................................ .......................... 187 4 5 Food (in)accessibility by wards through wheat income for unrestricted sample 188 4 6 Food (in)accessibility by wards through w heat income for restricted sample ... 188
12 4 7 Food (in)accessibility by wards through total crop income of restricted sample 189 4 8 Food (in)accessibility by wards through total household income of unrestricted sample ................................ ................................ .......................... 189
13 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy FARM EFFICIENCY, VALUE CHAIN PARTICIPATION, AND FOOD ACCESSIBILITY FOR WHEAT GROWERS OF TANZANIA By William Barnos Warsanga August 2016 Chair: Edw ard Evans Cochair: Z h i fe ng Gao Major: Food and Re source Economics This dissertation is motivated by the seeming ly unresponsiveness of domestic wheat production to the ever increasing demand for wheat and wheat products in Tanzania. Currently, there exists a gap in wheat demand and supply of at least 820 thousand metric tons that cost the country about 30% of its total food import bill . Th is study hypothesize s that the low farm efficiency, nonparticipation of farmers in the w heat value chain, and low returns for wheat are among the main fact ors leading to the lack luster response to this market opportunity . The first essay , therefore, examines the technical, allocative, and economic efficiencies (TE, AE, and EE) of wheat farms in Tanzania . It empl oys the stochastic frontier analysis to determine the effici en cy levels and the propensity score matching technique to measure the effect of participation in the value chain on efficie ncy . Results show that the average TE, AE, and EE are 79%, 80%, and 64% , respectively . Also , the findings reveal that farmers partici pation in the value chain (vertically) could improve by 6.8%, 5.7%, and 8.7% , respectively. On the other hand,
14 participation in horizontal coordination could improve TE, AE, and EE by 6.3%, 9.5%, and 11.6%, respectively. The second essay wheat value chain. It was found that few farmers were vertically (~17%) and horizontally ( ~39%) coordinated. The PSM procedure reveals that the net income effect of participation i n the val ue chain vertically is significant at 1% while horizontally is significant at 5% level. Generally , the results a re insensitive to hidden bias es that could have be en caused by unobservable s . The third essay assesses the economic contribution of wheat income on food accessibility. The recursive agricultural household model is applied to formulate the food accessibility equation. This study finds that 71 % of wheat farm households can not cross the national food poverty line based solely on income generated from wheat . This finding implies that wheat has low net returns that could conceivably provide a disincentive of reinvesting in the best technology to boost prod u ction .
15 CHAPTER 1 INTRODUCTION Agriculture, which includes crops, livestock, hunting, forestry an d fisheries, is a vital economic sector in Tanzania, accounting for roughly US $ 7.35 billion (all currency dollars), or 26% US$ 28.25 billion Gross Domestic Product (GDP), an d employing more than 80% of the labor force (NBS, 2011). Of this amount, crop production c ontributes the most (17.8%) , followed by livestock (3.8%), hunting and hectares ( ha ) , of which, land area is 88 million ha. Of the 88 million ha of land area, 44 million ha are arable (i.e. , land) , b ut only 9.5 million ha (~10%) are currently under cultivation (Promar Consulting, 2012) . Appro ximately 84% of the cultivated land is held by small scale farmers, with a parcel of land ranging from about 0.2 to 2.2 ha. Although more than half of the total harvested land area is allocated to cereals /grains , and despite its favorable agroecological co nditions, Tanzania remains a net importer of cereals , with wheat grain being the majority. Over the last two decades, the demand for wheat in Tanzania has risen considerably. Several factors have contributed to the sharp rise in demand , including the rapid ly increasing population, increased urbanization, rising incomes, and a noticeable change in food preferences from customary cereal s such as maize and rice toward wheat and wheat products. Despite this rise , the country produces only 10% of the total wheat consumed per year ; the rest is imported . While the gap signifies a market opportunity for wheat farmers to generate more income and an opportunity for the country to save valuable foreign exchange, production has yet to respond to this seeming opportunity . This study, therefore, investigates some of the reason s for this
16 dis connect ion through three essays examining wheat farm efficiency, value chain participation, and food accessibility among wheat growers. The first essay in C hapter 2 examines the technic al, allocative, and economic efficiencies (TE, AE, and EE) of wheat farms in Tanzania. It employs the stochastic frontier analysis to determine the efficiency levels and propensity score matching to measure the effect of participation in the value chain on efficiency levels. The importance of this essay is to explain why there are discrepancies in the levels of output among wheat farmers while the y homogenously use more or less the same farm production technology. I t also examines to what extent participati on in the value chain could improve the level of effi ciencies among wheat farms to increase overall wheat production level . The second essay in C hapter 3 participation in the wheat value chain. A possible explanation for the lackluster response of domestic farmers to what appears to be a clear market opportunity might be their failure to participate formally in the value chain and a breakdown in communication s (information flow). The rationale of this essay is to demo nstrate the potential benefits regarding increased income that farmers are likely to obtain through formal involvement in the value chains. This essay employs the propensity score matching (PSM) technique to measure the impact. The third essay in C hapter 4 assesses the economic contribution of wheat income on food accessibility. It disti l s from the two essays above, whereby the underlying premise is that wheat growers could improve their level of food security as a result of improvement in wheat productiv ity and value chain participation from the
17 resultant increase in on farm income. This study hypothesizes that the net wheat returns generated are likely to be below the respective hous ehold food poverty threshold , leading to a disincentive to expand product ion . The recursive agricultural household model is employed to establish the probability of food accessibility by wheat growers through wheat income .
18 CHAPTER 2 TECHNICAL, ALLOCATIVE, AND ECONOMIC EFFICIENCY OF WHEAT FARMS IN TANZANIA: PROPENSITY SCORE MATCHING AND STOCHASTIC FRONTIER ANALYSIS Introduc tory Remarks Demand for wheat in Tanzania has risen considerably within the last two decades. Several factors contribute to the rise in demand , includ ing the rapidly increasing population, increased urbanization, rising incomes, and a noticeable change in food preferences from the customary cereal s such as maize and rice toward wheat and wheat products. D espite the relative abundance of suitable land fo r the production of this crop in Tanzania and the obvious market opportunity, domestic production ha s lagged considerably behind demand , leading to a heavy re liance on imports to satisfy domestic demand. Apart from the exposure to international price fluct uation s , this untenable situation continues to drain the economy of valuable foreign exchange which could otherwise be deployed to other areas t o promote economic development. In order to meet the emerging demand and food preferences, the marketing literat ure suggest s coordination between the business actors (Boselie et al., 2003 ; Neven & Reardon, 2004) . I ncreased coordination (vertical and/or horizontal) is likely to impact a farm however , there is little informatio n on the subject within the literature. While a few studies have focused on the coordination effect on productivity (Hernandez et al., 2007 ; Minten et al., 2007 ; Neven et al., 2009) , none, to the best of our knowledge, has considered the impact of smallholders participation in value chains on levels of farm efficiency. Efficiency in utilizing and allocating factors of production by smallholders can be a major path way for poverty reduction and food security . T hus , a critical analysis of th e potential sources of efficiency and their policy
19 variables are essential for the development of food and agricultural policies for a country such as Tanzania. There are several ways to which participation in the value chain (PVC) c ould impact farmer s te chnical, allocative , and overall economic efficiencies. For one , PVC , especially when the postharvest value chain actors have strict requirements regarding quantity and quality of a product. For example , the minimum volume of output required may demand that farmer s modif y their production practice s t o expand farm operation s and employ improved technologies such as fertilizers and irrigation technology . Likewise , the improvement of quality may requir e tha t farmers use hybrid seed s , pes ticides , herbicides, and other modern inputs. PVC also brings market assurance s that may encourage farmers to invest in innovation technolog ies such as grading, semi processing, and packaging to meet the quality demanded. PVC may , therefore, at least, influence techni cal and allocative efficiency since it facilitates access to factors of production and their respective prices, and more importantly , by gaining access to production and market related information for the crop of interest. A review of the literature suggests that the analysis of the efficiency of wheat production has not been conducted for Tanzania. Most of the whe at studies in Tanzania (SAGCOT, 2013; Wolter , 2008; Kilima, 2006; Minot, 2010) cover production and m arketing in general without critically analyzing the input output relationship. U nderstanding the efficiency of farmers in utilizing their resources and the factors influencing those efficiencies is critical in developing countries such as Tanzania, as eff iciency gain s by farmers can contribute to economic gain s . Thus , efficien cies
20 measure the magnitude of gains achieved by the factors of production applied in a given technology set (Zhu, 2000; Tauer, 2001; Rahman, 2003; Armagan, 2008). The current study fo cuses on the wheat crop for several reasons. The growth in per capita consumption of wheat has outstripped that of maize and rice which are staple foods in Tanzania. Between 1990 and 2010, per capita consumption of maize increased by 0.3 kg per year, wheat consumption increased by an average of 0.35 kg per y ear, and rice consumption grew at 0.32 kg per year (Mason et al., 2012) . Figure 2 1 depicts the trends in per capita consumption of maize, wheat and rice for Sub Saharan Africa (SSA) over the period 1998 t o 2008 (the latest year for which data are available). These trends typify the situation in Tanzania and indicate that from 1998 to 2008 the gap between the per capita consumption of maize and wheat narrow ed due mainly to a rise in per capita consumption o f wheat. Figure 2 2 also shows the wheat consumption and production trends from 1964/65 to 2016/17. The general visualization of figure 2 2 justifies the widening gap between production and consumption of wheat in Tanzania, a sharp increase of the gap is observed from 1997 to 2016. While the gap signifies the increasing importance of wheat as part of the diet of Tanzanian households, domestic production has not responded to this seemingly good market opportunity. Of the 44 million ha of arable land in Tan zania ( a fair amount of which is suitable for wheat producti on ) , only 80,000 ha (0.18% ) is under wheat production. Moreover, the average yield is estimated to be around 1.18MT/ha (Promar Consulting, 2012) which is relatively low , especially when compared to its neighbor country , Kenya whose average yield is 2.2 MT/ha (MAFAP, 2013).
21 Although Tanzania has the potential to increase its production of wheat, improving the lev el of production faces several challenges, notably expensive inputs (chemicals, seeds, and fertilizers); insufficient farm machineries; high fuel prices; unstable producer prices; and the subdivision of large scale farms into smaller units (Mburu et al., 2014) . In addition , it is worth noting that while small scale farmers comprise the majority of the producers, they differ significantly from large scale farmers in their use of inputs, agronomic practices, and productivity . Seventy percent of small scale farmers use hand tools to plow their fields, and 10% use a cow . Only 20% have access to more modern technology such as tractor s and machine operated threshers. Fertilizer usage is also r eported to be very low, only 15% o f producers apply fer tilizer in their farms (Temu, 2006) Bringing about a noticeable increase in wheat production will likely require a combination of activities , including the expansion of production plot s , increase d access to modern technologies , and efforts to ensure that current factors of production are being utilized in an efficient manner. Because there are scarce resources for wheat cultivation , determining the extent to which such resources are being utilized in an efficient manner as well a s determining the factors responsible for any noticeable inefficiency will be crucial in taking advantage of wheat marketing opportunities. The results of such an analysis will reveal areas of inefficiencies as well as pr ovide critical information to form policies that will lead to improvement s in an enabling environment. This study uses the stochastic frontier and propensity score matching (PSM) approach es to estimate the impact of participation in the value chain on the technical, allocative, and economic efficiencies ( TE, AE , and EE ) of wheat farms in Tanzania. The
22 PSM is used to account for selection biases on the impact of value chain partic ipation. The significance of this study is in line with the policies espoused by the Government of Tanzania relate d to the achievement of broad food self sufficiency, rural employment creation, and poverty reduction through agricultural productivity. Concept and Approach of Efficiency in Agriculture Within the economic literature, the origin of discussion and measurem ent of efficiency and productivity date back to the 1950s , with much recognition given to the early work of Farrell in 1957. Firm (farm) performance is evaluated based on the concept of economic efficiency which is composed of technical and allocative e ffi ciencies . Farell (1957) defines efficiency as the ability of the firm to produce a particular level o f output at the lowest cost. This definition implies that producers will be considered efficient if they can produce their output s on a frontier under a gi ven set of inputs and also at a minimum co st given the prices of inputs. Technical efficiency (TE) defines the ability of the firm ( farm ) to produce output at the maximum level given the same level of input requirements. A llocative efficiency (AE) is the e xtent to which farmers equate the marginal value product of a factor of production to its price. When both technical and allocative efficiency are achieved the firm/producer is said to have achieved economic efficiency. Therefore, economic efficiency (E E) is attained when the producer correctly combines a set of input requirements to produce the maximum level of output . This implies that the technical and allocation aspects of efficie ncy will lead to maximum profit . The importance of efficiency measureme nt in this study is threefold. First , a comparison is made of the efficiency level of farms with similar technolog ies operating under the same growing condition s in the study area. Second, further analysis is
23 undertaken to explain the causes of variations in efficiencies among farms . Third, an analysis is conducted to disclose the policy implication s for improving the efficiency level of wheat farms. The main assumption in production theory is that producers make a rational decision by maximizing profit s us ing the best combination of factors of production at the lowest cost possible. However, the imperfection of markets, climatic factors , and demographic factors are likely to cause a violation of assumption, thus the importance of examining the efficiencies and identifying any causes of not producing at the frontier level. Firms ( in this case , farmer s) , could be operating below their production frontier due to imperfect knowledge of best farm management practices or due to their poor participation or nonparti cipation in input and output linkages (value chains). This study incorporates the technical, allocative , and economic efficiencies in production and cost functions to address the wheat production challenges faced by farmers in Tanzania. In order to estimat e the actual contribution of value chain participation on efficiency , we use the PSM approach together with stochastic frontier production and cost functions to reduce sample selection bias. The stochastic frontier is common ly used in the literature o f effi ciencies for analyzing farm production data. The Battese and Corra (1977) study has been cited as among the first to anal yze farm production data using the stochastic frontier model. The foremost importance of the stochastic production frontier , as well as the cost frontier, is the estimation of the efficiency level of the indi vidual production unit and the associated sources of inefficiencies. The efficiency estimation is associated with the comparison of outputs ( i.e. , observed and potential output s) . The potential output is the maximum possible output that cou ld be attained using the existing technology set, thus the need to use
24 stochastic frontier model. Further studies with the grounded concept of efficiency and production frontier include Fare and Knox Lovell ( 197 8) , Kalirajan and Shand, ( 199 9) , and Battese and Coelli ( 1988) . Theoretical and Analytical Framework In studies dealing with efficiency, the production frontier approach estimates the production function , or its dual cost function, or a combination of the two, and the profit function (Greene, 2005) . These estimations signify (i) the maximum yield to b e achieved give n a set of input requirements; (ii) the minimum cost of production given the prices of the input requirements ; and (iii) the maximum profit possible given the inputs requirements, the outputs produced, and the prices of input and output . In general , the purpose of developing the production, cost , or profit models is to measure efficiency by analyzing residuals from production or cost functions. This st udy adopts production and i ts dual cost functions to analyze the TE , AE , and EE of wheat producers in Tanzania. In the deterministic frontier analysis, it is assumed that each producer faces the same production technology represented by the conversion of a vector of inputs into a single output . Greene (1993) argued that given the observed production and the potential or ideal production, it is possible to estimate the efficiency of inputs with the production frontier. The stochastic frontier function can inc orporate a composite error term and a one sided error term. The composite error term captures a random effect that arise s outside the c ontrol of a firm (farmer) while the one sided error term represent s an inefficiency component of the production unit (Jondrow et al., 1982) Generally, the multiplicative production fr ontier is modeled as (2 1)
25 where is the firm output, is the firm inputs used for production, is unknown vector of parameters to be estimated, is the exponent term, is the stochastic error and represents inefficiency term. The terms and are independently and identically distributed with variance and respectively. The stochastic effect is outside the farmers control due to factors such as rainfall, temperature, natural disaster, and others. These factors are measurement errors and are unobservable statistical noises. Following Pitt and Lee (1981 ) and Jondrow et al. (1982 ) is assumed to be normally distributed while u is half normal distributed and reflects technical inefficiency. E quation 2 1 above can b e transformed into logarithm as (2 2) The transformed equation 2 2 above can now be used to estimate by applying maxim um likelihood (ML) method as follows: At maximum Applying the first order expansion (2 3) (2 4) Since the likelihood function of the stochastic frontier is n on linear in its parameters, give a closed form solution. (2 5)
26 The process involves obtaining the initial using for example , the OLS estimation technique , and repeat ing the formula above by replacing with newly calculated value . T he iteration continues until the newl y generated value is indifferent from . In order to use maximum likelihood estimation in SFA we need a distribution density function. Total error However , the distributi on of are assumed to follow normal and half normal respectively. Accordingly, the density functions for are (2 6) (2 7) Total error term distribution becomes the convolution of the distribution of and which is given by (2 8) Integrating the above equation results in the following: (2 9) where and is the distribution (density) function of the standard normal distribution with mean 0 and variance 1. The log form of this density function is given by (2 10) For the N firms, the log like lihood function becomes (2 11) (2 12)
27 where , the are estimated by the stochastic frontier model using the maximum likelihood method. The R pa ckage (SFA) from Benchmarking i s used to conduct the analysis (Bogetoft & Otto, 2015) . The percentage of total variances due to inefficiency is obtained from the formula this is the percentage of inefficiency variation to total variation . We can further obtain both the variance for inefficiency and the variance for random errors from the total varianc e and lambda equations simultaneously as (2 13) (2 14) From equations 2 13 and 2 14 , (2 15) (2 16) In early studies such as of Battese and Corra (1977) in their estimation of production function model for the pastoralism , they found that farm effect variance ( ) was highly significant proportion of the total variability ( ) of the value of sheep production in Eastern Australia. The parameter esti mate exceed ed 0.95 in all cases. They concluded that the stochastic frontier production functions and deterministic frontiers were statistically significant ly different from one another. However, the technical, allocative, and economic efficienc ies were not e stimated in their study . Sekhon et al. ( 2010) found a gamma value of 0.52 in their study of regional wise analysis of crop technical efficiency in Punjab. They concluded that the difference between the observed and frontier output is explained ( 52% ) by factors that can be
28 controlled by farmers. Therefore , the gamma value explains variation of yield due to inefficiency factors that are identified in the system (Essilfie et al., 2 011 ; Asra vor et al., 2016 ; Amor & Muller, 2010) Technical Efficiency After completing the theoretical procedures to estimate the SFA parameters and to decompose the error term into noise and inefficiency, the firm specific level of efficiency can be estimated. The firm specific efficiency by Shephard lem m a is obtained from the ratio of the observed output to the maximum output or its inverse (Fa rrell approach) that can be produced with the observed input quantities. Accordingly, the technical efficiency is given by (2 17) Since is not observed in the SFA analysis, the approach outlined by Bogetoft and Otto (2010) is used to obtained this value. Specifically, Bogetoft and Otto show that can be manually calculated with some estimated variables from the SFA results as follows: (2 18) where the variables in the equation are as defined earlier. Measuring technical efficiency ( TE ) is well documented in the literature . Coelli and Battese ( 1996) are among the first to apply the stochastic frontier approach to estimate and determine inefficiency factors simultaneously; unlike previous studies that first estimated the efficiency score and then regressed the scores on factors sti pulated to influence TE . C hang and Wen (2011) investigated how farmers with and without off farm work differ ed in yield level, efficiencies, and in production risk. Th ey applied the stochastic production frontier models to estimate simultaneously the TE an d production
29 risk between the groups . They found that farmer s with off farm work are more technical ly efficient than the other group. L yubow and Jensen (1998) analyze d TE using trans logarithmic stocha stic frontier production for grain p roduction from1989 to1991 in Ukraine. Based on a survey of 80 farmers they found that factors such as the number of farm labor/ha, proportion of a positive i nfluence on TE . Onyen weaku and Okoye (2005) estimate d the TE of cocoyam and its determinants by using a translog stochastic frontier productio n function in Nigeria. They found that the TE range d from 69% to 98% , with a mean TE of 93%. The factors analyzed in the efficiency scores that were found to be significant a re the size of the membership . However, they d id not find any significant relation ship between efficiency and household size, extension, credit acquisition, and a ge of the farmer. Onyenweaku and Ohajianya (2005) measure d the TE level and its determinants for rice production in Nigeria by employing the stochastic frontier production model. Their finding shows that TE var ied from a minimum of 17% to a maximum of 93% , with a mean TE of 65%. C redit accessibility, size association membership, farming experience, improved rice seed, production systems, and extension contacts significantly impact ed TE. However, farmer s age, tenur e status, and off farm employment had no significant impact on TE in their study. Seidu et al. (2006) examined TE for small scale rice farms in Ghana during 2002 to 2003 cropping years using the production frontier and maximum likelihood approach. The ir results indicate d that the smallholder rice farmers we re technically inefficient with
30 an average score of 34% which is far below the maximum possible attained. They also found a significant difference between irrigation and non irrigation farms in effic iency level. When they introduce d gender , the y found a significant difference efficiency level between male owned and female owned farms. C redit availability, size of the family, and non farm employment were found to impact TE significantly in their study . The review of TE studies in Tanzania are limited . The few of them that can be accessed include Shapiro and Muller (1977) who measure d TE by emplo ying a deterministic Cobb Douglas production frontier derived from linear programming. The main objective of their study was to analyze the roles of information and modernization in the production process of 76 cotton farms in the Geita D istrict of Tanzani a using data collected in 1970/71. However, only 40 cotton farms were used for their analysis because of data problems. Using correlation analysis, Shapiro and Muller (1977) found that TE had a high ly positive association with both general modernization an d information. The se measures of modernization include the knowledge concerning input and output prices, and local agricultural officials, as well as an overall index of moderni zation. The highest correlation found was between the overall index of moderniz ation and knowledge of local agricultural officials. The average level of TE obtained from a Cobb Douglas production frontier was 66%. The authors conclude d that traditional agriculture suffers significantly from inefficiencies. Msuya and Ashimogo (2007) applied the stochastic frontier model to estimate the TE of outgrowers and non outgrowers of sugarcane in the Mvomero D istrict (Mtibwa) of Tanzania. A p pl ying the Cobb Douglas model , TE levels we re 76.43% and 80.65% , respectively, implying that the non outgrowers were more efficient than the ir
31 counterpart s . Factors found to significantly influence TE a re age, education, and experience. However , the ir stu dy did not consider selection bias (discussed in a later section) between the two groups in which this study address es for wheat grower participants and nonparticipants of the value chain. Ilembo and Kuzilwa (2014) applied stochastic frontier model with Cobb Douglas production function to measure the TE of tobacco pr oduction farms in Tanzania. Their study found an average TE of 64%, with the m ost significant factors contributed to this level being farm size, input credit use, off farm income, and education. They recommended linking farmers to the inputs credit system to increase TE because tobacco production is capital intensive and unaffordabl e for most small scale farmers. Rajendran et al. (2015) use d the Cobb Douglas stochastic production frontier to measure the TE of traditional vegetables in five regions of Tanzania. The average mean TE was 67% in which their further analysis suggested th e strengthening of farmers associations to increase the level of TE. Others , like Oleke and Isinika (2011) on the TE of egg production , utilize d the stochastic production frontier specified in the Limdep software. Mlote et al. (2013) examine d the TE of small scale beef cattle fattening by applying the stochastic production frontier in the lake zone of T anzania. Sarris et al. (2006) estimate d technical and allocative efficiency for farmers of the Kilimanjaro and Ruvuma regions. Allocative and Economic Efficiency As mentioned e arlier, allocative efficiency (AE) represents the least cost combination of inputs to obtain the maximum level of revenue. When output and input price are known, cost function which is the dual to production function shown in Equation 1 can be used to measure AE and TE . The cost function is expressed as a
32 function of input prices and output . The extra advantage of the dual cost function is that it can provide answers that are also provided by the production function by , then the cost minimizing point of a given good with a given price is unique. The cost functions show the minimum cost of producing the output when the input prices and the technology set T are given as follows: (2 19) The above formula explains that the observed cost is higher than or e qual to optimized cost. That is, (2 20) Therefore , the cost efficiency which is the ratio of the optimized cost t o the observed cost is given by, (2 21) Thus, (2 22) As in the production function, we can express c ost efficiency by using which become s , (2 23) w here By introducing the multiplicative expression of error term , the cost efficiency can now be expressed as : (2 24)
33 (2 25) Thus, (2 26) Given that Equation 2 26 is similar to Equation 2 1, the estimation procedures fo llowing MLE are the same. In the case of economic efficiency (EE); if allocative efficiency (AE) is defined by the least cost combination that produces cost efficiency (CE), then EE is just the product of AE and technical efficiency (TE) which implies the firm is both technically and allocative efficient (Bravo Ureta and Rieger, 1991) . S tudies addressing AE and its determinants are relative ly scanty compared to TE studies , particularly in Tanzania. According ly, Henderson and Kingwell (2002) argue that when AE is improved, its contribution is greater on profitability than TE because AE farms are most likely to be technically efficient , and not vice versa. This is because any rational producer would think of maximizing profit s through which TE can be realized . TE and AE analyzed by Henderson an d Kingwell (2002) on a broad range of farmers in Western Australia over three years use Data Envelopment Analysis (DEA) and Stochastic Frontier Analysis (SFA) . The ir finding reveal ed that the yearly inefficiencies were decreasing over time. Their empirical TE . Ogundari and Ojo (2006) use d production and cost frontier functions to estimate the TE, AE, and EE of cassava farms in Nigeria. With efficiency levels of 0.903, 0.89 and 0.807 for TE, EE, and AE , respectively , they conclude d that cassava farms are efficient in al locating resources given the scope of production. Sauer and Mendoza (2007) investigated the well
34 (19 64) that small scale farmers can be allocative ly efficient in utilizing their scarce resources d espite their poor condition, especially farmers of developing countries in was assumed that the be explained by scale effects (i.e., relative ly allo cative ly efficien t but inefficien t at a specified scale of operation ) . The authors employed the Generalized Leontief (GL) profit function to discuss the theoretical consistency of the concept. The empirical findings suggest a revised hypothesis that consid er ed the scale of operation. Thus, the small allocative ly efficien t ly inefficien t with respect to the scale of operation . M any other studies also show ed evidence of a llocating agricultural practices resources (Hopper, 1965; Chennareddy, 1967; Sahota, 1968; Saini, 1968; Srivastava & Nagadevara, 1972). Okoye et al. (2006) estimate d the AE level of 120 cocoyam farmers by applying a translog production and cost functions in the stochas tic frontier in Nigeria. The average level of allocative efficiency found was 65% and was significantly influenced by age, education level , size of the farm, fertilizer application, and farming experience. F actors found to have an insignificant impact on t he AE level we re extension visit s , membership in cooperative s , and size of the family. Asogwa et al. (2011) use d the stochastic frontier models to analyze the TE and AE of Nigerian rural farmers. The ir study show ed that TE and AE in farm production among the farmers could be increased by 68 % and 65 % , respectively, through better use of available resources. They conclude d that AE could be achieved through improved farmer specific factors which include farmer education and support to young
35 male farmers into farm productions. Osborne and Trueblood (2006) examine d the EE, AE, and TE of crop production for Russian corporate farms from 1993 to 1998. The ir findings reveal ed that that, over the period, EE has been declining owing to the declines in TE and AE . The authors indicate d that there was the possibility of maintaining output levels while reduci ng factors of production by 29 % to 31% (TE aspect). On the other hand , their finding show ed that there was the possibility of reducing the cost of factors of production by 30% ( AE aspect). They, therefore, conclude d that EE could increase by lowering the use of all factors of production such as fertilizer and fuel. Nwaru et al. (2011) also confirm ed that household size, access to credit, membership in cooperatives , and farming experience are the variables positively and significantly related to allocative efficiency. The coeffi cients for farm size and gender we re negative , indicating that female farmers we re more allocative ly efficient than their male counterparts. The analy si s was conducted within the framework of the stochastic frontier function of primary data from a random sample of 120 farmers. Accordingly, Obare et al. (2010) use d the dual stochastic efficiency decomposition technique and a two limit Tobit model to study the allocative efficiency of the Irish potato in Kenya. The ir paper establishe d that Irish potato production is characterized by decreasin g returns to scale , with a mean allocative efficiency of 0.57%. It was further established that farming experience, access to extension services and credit, together with membership in a d allocati ve efficiency. Bifarin et al. (2010) utilize d a stochastic frontier to determine the TE, AE, and EE of plantain in Nigeria. Their findings reveal a gap variation from 20% to 87% for TE scores, with a mean of 61%. For AE , they found the indices var ied from 14 % to 83% ,
36 with a mean of 57% , and EE has a minimum of 3% and a maximum of 67% , with a mean of 35%. Chiona (2011) found that Zambian smallholder maize farmers had very low levels of TE and AE, with a mean TE of 15% and a mean AE of 12% . The author found s ize of the farm, hybrid variety, extension service accessibility, household size, and level of education to significantly influence EE. Testing the Hypothesis It is important to determine whether or not the variation in efficiency is significant. This is accomplished by testing the hypothesis below. Accepting the null hypothesis implies that the variation is not significant and suggests that such variations in the firms output or its dual cost are mainly caused by noises . In contrast, rejecting the null h ypothesis implies that inefficiency contributes to the variations. (2 27) (2 28) Since testing is equivalent to testing and is obtained directly from t he SFA estimates, thus the t test can be used. Alternatively, the likelihood ratio test can be used where the OLS production/ cost function estimates can be comp ared to SFA estimates to test whether there is significant variation between the two approaches . In that case , the log likelihood values would be used and the result would be compared to the critical value from chi Square distribution with 1 degree of freedom. Determining Factors Influencing Efficiency Bravo Ureta and Rieger (1991) note d that within the literature several researchers have conducted studies relating to efficiency to various social economic
37 variables by focusing on two approaches. One approach uses correlation coefficients to conduct simple nonparametric analyzes . The second approach , which is usually referred to as a tw o step procedure, involves first estimating the farm efficiency level and then regressing the efficiency scores on possible socioeconomic attributes. This study utilizes the lat t er procedure where the censored Tobit regression model is commonly used and is a generalization of the standard Tobit model. In order to use the censored Tobit model , the dependent variable can either be left censored, right censored, or both left censored and right censored, where the lower and upper limit of the observed dependent variable can take any value. Since our efficiency scores are bounded between 0 and 1 , the approach proposed by Henningsen (2010) is used and is described as follows: (2 29) (2 30) where is the efficiency scores, are the factors believed to influence efficiency levels, is the minimum efficiency level , and is the maximum efficiency level. Censored regression models are usually estimated by maximum likelihood (ML). Assuming that the disturbance term follows a normal distribution with mean 0 and variance , the log likelihood function is (2 31)
38 w here (.) and (.) denote the probabil ity density function and the cumulative distribution function, respectively, of the standard normal distribution, and and are indicator functions with (2 32) (2 33) The log likelihood function of the censored regression model (2 31 ) can be maximized on the parameter vector ( , linear optimization algorithms. The the R software i s used to estimate the model (Henningsen, 2015) . Value Chain Participation Effect on Efficiencies ; Bias Selection Consideration The stochastic frontier approach marks the differences in TE, AE, and EE for each firm (farmer) considered in the analysis, so do the group s of par ticipants and nonparticipants in the value chain. However, differences in these two groups are not simple to attribute because of selection bias (self selection). That is, the factors that e xplain value chain p articipation could also have an impact on efficiency . The selectivity bias problem has been addressed by several authors ( e.g. , Sipilainen & Lansink, 2005 ; SolÃs et al., 2007 ) by employing the Heckman two step model. However, the Heckman procedure is useful in linear but not nonlinear analysis such as st ochastic frontier analysis. P ropensity score matching wou ld be appropriate and has been ap p lied by other authors in the stochastic frontier analysis ( e.g ., Mayen et al., 2010) . The literature has considered various matching procedures for covariates but the most common o ne s used are Nearest N eighbor matching , Caliper radius matching,
39 Kernel matching, Mahalanobis matching, and Stratifi ed radius matching (Caliendo & Kopein ig 2008). This study employs nearest neighbor and caliper radius matching. N earest neighbor matching pairs the participants and nonparticipants of the value chain who are closest in terms of probability of participati on as identical partners. Caliper radius matching, conversely, uses a radius distance in which all the control unit s falling within the radius are matched to the treated unit in order to construct a counterfactual outcome. That is , the outcome for nonp articipants is matched with the outcome for participants, and the difference is the effect of the program for each participants. The average difference for all matched sample is t he estimate of ATT, or the net impact of the program. A drawback of PSM is th at it can only control for observable characteristics, wh ich means there could still be a selection bias due to unobservable factors (Smith & Todd 2005). This study assumes that the distribution of unobservable is the same for participants and nonparticipa nts. Accord ingly, Rosenbaum (2002) proposed the standard bounding test to evaluate how strong the unobservable could influence the selection bias to nullify the implication of matching process. A more extensive discussion of the PSM approach is given in th e next essay where we measure the wheat income effect from value chain participation. A search of the literature uncover s a study conducted by Seyoum et al. (1998) who used the Cobb Douglas stochastic model to estimate TE and productivity by compari ng The farmers with technology were found to be more efficient than their counterpart s with a mean TE of 94% and 79%, respectively. Although the general idea of th at paper is
40 similar to our study , a major diffe rence is that their approach does not consider select ion bias n or the ignorability condition accounted for in this study. Bravo Ureta et al. (2012) used the stochastic fronti er framework and PSM approach for e valuating TE between the beneficiaries and non beneficiaries of the MERANA program in Honduras. They control led for both the observable and unobservable selection bias using PSM and the selection correction model proposed by Greene (2010) . They found that the m ean TE was higher for the treated group than for the control group and that the hypothesis of the presence of selectivity bias could not be rejected . They further found that the treated group perform ed well in both TE and frontier output. However, the ir st udy applie d only the nearest neighbor matching without replacement and is silent on the extent to which ATT would improve, opting only to give the ranges. Abate et al. (2014) applied PSM to compare the TE of cooperative farmers and independent farmers in Ethiopia using the kernel regr ession method. Their study found that th e hidden selectivity bias was insensitive to the analysis and that farmers participation in agricultural cooperative improved their effic iencies due to support offered by the cooperatives. This study departs from that by Mayen et al. (2010) in that this study uses the matching procedures after estimation , rather than before, to avoid losing observations that are useful in the frontiers anal ysis because the procedure normally drops unmatched observations. The PSM matches the covariates of the farm and farmer characteristics rather than simply comparing the outcome variables (TE, AE, and EE) between the participants and nonparticipants of the value chain. The PSM matches only
41 the outcomes between the participants (treated) and nonparticipants (control) of the value chain that are similar in terms of observable features. This closes the bias gap to some extent that would exist when the two clust ers are systematically different (Dehejia & Wahba, 2002) . PSM involves three procedure s. I n the first procedure, the propensity scores (predicted probabilities, ) are generated from a logit or probit model that gives the probability of a farmer participat ing in the value chain. The conditional observed vector of participation overlaps with explanatory variables which are farm and farmer characteristics. In t he second procedure , the control group (nonparticipants) is constructed by matching their propensity scores generated from logit or probit models. The treated group (participants) with no good matches and the control group (nonparticipants) not used in the matching are dropped from the analysis. In t he third procedure , the average treatment effect on the treated (ATT) is calculated for the outcomes (TE, AE, and EE). The ATT is the outcome difference between the two groups (treated and untreated) where their covariate s are matched and balanced by the propensity score. Empirical Model Technical E fficiency This study employ s one of the flexible functional for m s of to estimate equation 2 1. This functional form was chosen because it allows variations in elasticity unlike Cobb Douglas which assumes that the elasticity of production is constant. Thus, equation 2 1 is specified as (2 34)
42 where Qu antity of wheat produced in kg Size of wheat plot in acres Amount of fertilizer applied in Kg/a cre Amount of chemicals (herbicides insecticides and pesticides) applied in Lt/acre Labor used in man days Amount of local /hybrid seed planted in Kg/acre , are unk nown parameters to be estimated = noise or disturbance follows normal distribution N (0, ) = technical inefficiency which is positive and half normal distributed N (0, ) Allocative E fficiency The stochastic cost frontier for this study i s also specified as a translog cost function with a two part error structure, v + u , where v and u are described as above. approximation (C hristensen et al., 1973) and is expressed as a total cost of wheat production in terms of input pri ces and output level as follows: (2 35) where is the in dex of different inputs considered and , is total cost, is output and the are the prices of the factor inputs. For a cost function to be well behaved it must be homogeneous of degree one in prices, implying that, for a fixed level of output, total cost must increase proportionally when all prices increase proportionally. Explicitly, the conditions above is full filled by checking on ; : Homogeneity : Symmetry and Homogeneity , : Monotonicity
43 : Hessian matrix , negative semidefinite. The above restrictions ar e accomplished by normalizing the cost f unction by one of the input prices. While any price can be used for normalization and the analysis would result in the same parameter values, in this study , we use the labor price. The translog cost function also requires that costs b e monotonically increa sing and concave in input prices. Specifically , for monotonicity , the requirement is that and , for concavity , the matrix has to be negative semidefinite. The distance from the cost frontier is measured by (Coelli 1996). Cost efficiency for the farm is then computed as . With this fo rmulation, cost efficiency is greater than or equal to unity. Its inverse ( ), therefore, is the percentage reduction in cost necessary to bring total cost to the frontier. The inefficiency is modeled in terms of farm sp ecific and household feature s ( ) as follows: (2 36) where Contract with buyers dummy Age of the household head in years Education level of the household he ad in years of schooling Experience in wheat farming in years Household composition proportional number of people share and live in one roof Family members with age below 18 years old in number Family members with age betwee n 18 and 50 years old in number Family members with age above 50 years old in number Rental land acquired dummy Extension visits in number of times per year Village and technical meetings attended in number of times per year Mbulumbulu ward dummy Rhotia ward dummy Monduli juu ward dummy
44 Transport ownership (car, motorcycle, and oxcart) dummy Farm equi pment ownership (tractor, plow , knapsprayer, wheelbarrow) dummy Lives tock keeping dummy Hybrid seed planted dummy Off farm income dummy Parameters to be estimated Disturbance term Data Data were collected through a field survey co nducted by trained enumerators using a pre tested questionnaire d esigned to study wheat farmers in n orthern Tanzania where 90% of the domestic wheat supp l y is grown (FAO, 2013). Two regions Arusha and Kilimanjaro were chosen because they are relatively homogenous in agricultural land use, production practices, and ecol ogical condition. Two districts from Arusha (Karatu and Monduli) and one from Kilimanjaro (Hai) were selected for the survey based on their level of wheat production. The corresponding wards located in the high lands were selected because wheat is grown in highland areas. T he Mbulumbulu and Rhotia wards were selected in the Karatu D istrict , the Monduli juu ward was selected in the Monduli D istrict , and the Ngarenairobi ward was selected in the Hai District to form more hom ogenous strata by location to represent the variability in wheat growing conditions by the wards . The farmers from villages in each of the major homogeneous wards were selected from the list pro vided by village officials. A combination of random and snowba ll sampling techniques were used to select farmers from the sampling frame provided by village officials. The sampling frame consisted of a list of farmers who grew wheat in the 2014/2015 season. A structured questionnaire was used to obtain information re lated to production, costs , and marketing practices. Backgroun d information solicited includes household size, age, gender, education , and occupation of the respondents, contracts, membership in an organization, and challenges farmers
45 face when producing and marketing wheat. In addition , formal discussions with key informants such as government officials and traders were conducted where we first described the importance of the survey so they could give us their opinion about the wheat crop. The information ob tained supplemented the information that was collected with the structured questionnaire. A total of 350 farmers were sampled despite several farm er s switched from wheat to barley production or significantly reduced the ir wheat production. Barley competes directly with wheat since both crops are grown under the same condition using the same inputs . The differences include seed s and buyers . Barley has the advantage in that it sells for a slightly higher price than wheat on a unit basis and receives full supp ort from private brewery companies. Such support includes the provision of inputs, assistance with harvesting , and tran sportation of the crop . Despite the dif ficulty encountered, a total of 310 out of 350 farmers completed the questionnaires and we re used in the analysis. T he focus of this study is small scale farmers who are the majority of farmers in Tanzania, with land size ranging from 0.2 to 2 ha ( the equivalent of 0.5 to 5 acres ) . Results and Discussion Household C haracteristics and Farm F eatu res Table 2 1 shows the households and farm characteristics for the pooled sample of wheat farmers. The information obtained shows that the majority of farm operators are relatively older with a low level of formal education. Most of the younger generation has migrated to urban areas in search of work. The average age of the household head is 44 years old with just a primary level of education (7 years of schooling) achieved.
46 The level of education has implication s f or farm production and understand how to better market their agricultural products . On average, the surveyed wheat farmers have 14 years of experienc e in producing and marketing wheat grain. While i t can be argued tha t more experience should have a positiv e impact on efficiency , this is not always true. On the one hand , highly experienced farmers know the technicality of how the production is conducted because they have been involved in the same activity for many years. However, on the other hand, they may be reluctant to accept new technolog ies which could suggest a negative impact on efficiency , thus resulting in ambiguity to its expected sign when evaluating it s relationship with efficiency levels. The average household composition is 7 family members who eat and share from the same cooking pot every day and live together under one roof. This may imply that a good number of family labor ers are available for is expected that larger family size will have a positive implica tion for efficiency level because it is easier for the household to supply the labor required for the production of wheat. In a household with 7 family member s on average , 3 members are younger than 18 years old , 3 members are between 18 years old and 50 y ears old , and 1 member is older than 50 years old . Table 2 1 also shows that the majority of the farmers surveyed (62%) rented land for wheat cultivation on a seasonal basis. This implies that the size of owned lands might be in sufficient for growing wheat , thus the need to rent ad ditional land to increase production. However , the average land area under wheat cultivation is approximately 5 acres , which is too small to realize a profitable wheat output. The small acreage
47 allocated for wheat production could be attributed to the high cost of owning/renting land and/or the relatively low incomes of the farmers , thus making it challen ging to acquire additional land . The information in Table 2 1 indicates that on average farmers receive at least one visit by ext ension officer s and participate in at least one meeting on an annual basis. The se small number s may suggest that there is an insufficient number of extension officers or that the farmers might not be making a special effort to participate in meetings organ ized by village officials. Irrespective of the reason, the farmer s may be missing crucial information related to farm and market operations , which ultimately could improve their levels of efficiency. As sh own in Table 2 1 only 7% of the farmers own a mean s of transport ation, namely an oxcart , a motorcycle , o r a car. The majority have to pay for service s to transport their crops to the ir house or to the market after harvest. This could be an area where substantially increase s the cost of their operation s an d thus an area where cost savings may be realized . Only 9% of farmers own tech nical farm equipment such as tractor s , knapsack sprayer s , plo w s , and wheelbarrow s . T ractor s are necessary because wheat soil s need to be properly tilled , thus requir ing multiple (repetitive) cultivations before planting. Because of the relatively high cost to hire a tractor service, invariably the soils are not properly cultivated . Likewise, the majority of farmers does not have spray equipment and must also hir e this service whic h substantially increases their production costs. Added to this is the fact that in some cases hired service s are not readily available when needed thus adversely impacting overall levels of efficiency.
48 Only 10.3% of farmers purchase and use hybrid wheat s eed s; the rest (~90%) use recycled hybrid or local seeds . The farmers surveyed report that local seed is cheaper and readily available while the hybrids seeds are more expensive and inaccessible. Farmers also report ed traveling long distance s to access the agrochemical selling points to purchase hybrid seeds. Thus, they become discouraged and use stored wheat grain from the previous harvest. In situations where they did not store sufficient grain , they purchase grain from neighboring farmers at the prevaili ng grain market price. This practice could adversely impact productivity since there is no guarantee that the variety of the seed they are buying from their neighbors is of an acceptable standard. Once the grains have been compromised with less than desira ble ones , this could have a long lasting negative impact on the quality and standard of grains harvested in subsequent seasons. Livestock production is common in the study area. The results of the survey indicate that 94% of wheat farmers keep at least 1 c ow, goat, or sheep. In addition , only 10% of farmers have off farm activities that generate household income. The r e are very few farmers with nonfarm wage s or self employment activities. O ff farm income is important in supplementing the total household inc ome which can also be allocated to farm activities such as purchases of factors of production and marketed food items. Wheat output ranges from 33.3kgs/acre to 1600kgs/acre with an average of 705Kgs/acre. This suggests that there is considerable opportunit y for increasing wheat output by boosting the production of farmers who reported lower productivity per acre. H ousehold and Farm C haracteristics by W ards Table 2 2 shows that on average, farmers from the Ngaren airobi and Monduli juu wards are much younger than those from the Mbulumbulu and Rhotia wards . This is n ot
49 surprising because the former are closer to town s , and most of the young generation tends to migrate to town. The number of farmer s in the Mbulumbulu, Monduli juu , Ngaren airobi , and Rhotia wards who had contract s with wheat buyers were 20 (17%), 18 (16%), 4 (18%) , and 9 (16%) , respectively. The information in Table 2 2 also indicate s that the percent ages of farmers that belong to associations or farm groups in each of the aforementioned wards a re 38%, 44%, 41% and 28%, respectively. The majority of farmers in Mbulumbulu, Ngarenairobi , and Rhotia have at least accomplished the primary level of education (more than 7 years in school) while the majority in Monduli juu have not (less than 7 years of s chool) . Rhotia has the most experienced farmers (17 years of farming experience), followed by Mondoli juu and Ngaren airobi ( 14 years ), and Mbulumbulu (12 years). The household composition variable shows that Rhotia ha s the most (7) household members , while the other wards average 6 family members. Likewise, the Rhotia w ard ha s the most (4) household members younger than 18 years old , while the other wards have 3 members per household in this category . For the 18 years old to 50 years old category, all ward s report an average of 3 househo ld members. Similarly, each ward has at least 1 family member who is older than 50 years old . With respect to land tenure, the Ngarenairobi ward ha s the least number of farmers with land ownership; the majority of the farmers (72%) in this ward cultivate crops on leased land. This is because most of the farmers are not originally from there, with most of the land area being owned by the government before be ing privatized. The privatized land is now leased out to prospective wh eat producers. On the other hand,
50 only 7%, 8%, and 19% of farmers in Mbulumbulu, Monduli juu , and Rhotia wards , respectively , rent/lease l and for cultivation seasonally. Information regarding the number of visits by extension office rs shows that farmers in the Mbulumbulu and Rhotia wards receive on average less than one visit per annum , while those in the Monduli juu and Ngarenairobi wards receive at least one visit per annum . For all the wards , the data indicate that the average number of meetings attende d by farmers is less than one, implying that they might not be interested in participating in such meeting s . Transport ation ownership shows that 8% of farmers from Mbulumbulu and Monduli juu own an oxcart, a motorcycle , or a car , while 9% and 5% of farmers from Ngarenairobi and Rhotia , respectively , own some means of transport ation . In the case of farm equipment (namely, tractor, sprayer, wheelbarrow , plo w ), farmers in Monduli juu (12%) have the highest rate of ownership , followed by Mbulumbulu (8%), Rhotia (7%), and Ngarenairobi (4%). Regarding livestock ownership , almost every household own s at least 1 cow, goat , or sheep. The data reveal that the majority of the farmers do not use hybrid seed to grow wheat, only 11% from Mbulumbulu and Monduli juu and 9% from Ngarenairobi and Rhotia plant hybrid seed s . The majority of farmers do not participate in off farmactivities , and only 10%, 11%, 5%, and 9% of farmers from Mbulumbulu, Monduli juu , Ngarenairobi , and Rhotia , respectively , have off farm income. OLS and SFA E stimate s for Translog Production F unction The maximum likelihood (ML) estimate of the translog production function is presented in T able 2 3 below. The standard ordinary least square (OLS) is also presented for comparison purpose. The results show th at 6 of the coefficients are significant a nd have the expected signs. L and size has the expected sign and impacts the output level in k gs/acre positively at the 1% significance level. I nsecticide s seem to
51 decrease the level of output (n ot as expected) when more is applied , and it is significant at the 5% level and 1% level for the OLS and MLE models , respectively. However , t his could be explained by inappropriate use of insecticides , or bad timing of spraying , or improper application ratio (Chen et al., 2013) . The second derivatives for all inputs are negative , implying the translog concavity property holds for existing farm technology. The is 95% , which implies that the wheat output as a dependent variable is well explained by the inputs utilized. The value is 189.1 and is significant at the 1% level which means that the general model is well specified. Lambda , which measures the inefficiency variation in relation to idiosyncratic variation computed as the ratio of standard errors of to has a value of 4.29 , which is significant at the 1% level. G amma which is the ratio of efficiency variati on to the total variation of parameters is 0.95. This implies that 95% of the total output variation is due to production inefficiency while 5% is due to variation from unobserved and measurement errors . OLS and SFA E stimat e s for Translog Cost F unction The translog cost frontier which is dual to the translog production frontier is shown in Table 2 4. It is econometrically estimated t o provide the basis for computing both the allocative efficiency (AE) and the economic effici ency (EE). As mentioned earlier, the total cost and input prices were normalized by labor price per acre. The inputs p rices included in the model were seed s (Ps), fertilizer (Pf), herbicides (Ph), insecticides (Pi), and pesticides (Pp) prices.
52 Efficiency S cores The technical, allocative and economic efficiencies are reported in T able 2 5 below. As shown in T able 2 5 , the average technical efficiency (TE) of the sampled wheat farms is 79% , with a minimum of 37% and a maximum of 97%. This implies that wheat f armers are not operating on the production frontier and can increase their efficiency level by as much as 21%. If the average producer is to achieve TE of the most efficiency producer , then he/she would realize an output gain of about 19%; in the case of t he most inefficient farmer (TE=37%) , the o utput gain would be as much as 62%. With respect to all ocative efficiency (AE), the average AE was determined to be 80% , with a minimum of 24% and a maximum of 98%. This implies that on average allocative inefficie ncy accounts for a 20% loss in wheat income o r the potential to reduce cost (cost savings) by as much as 20%. It , therefore, implies a lack of cost minimizing behavior on the part of the farmers. If the average producer is to achieve an AE similar to most AE producer, he/she would realize a savings of about 18% of the total cost; while the most inefficient AE producer would reduce costs by as much as 76% of the total cost. Of the three measures of efficiency, the farm household economic efficiency (EE) has the lowest score , averaging 64%, with a min imum score of 9% , and a maximum score of 93%. This implies that under the existing production technol ogy, the average wheat producer can realize substantive increases in wheat output by becoming more technically efficient as well as cost savings by optimizing the use of inputs given the prevailing prices by 36%.
53 The average economically efficient producer could increase output and reduce cost s by 31% if he/she were to achieve the EE level of the most economical ly efficient producer. On the other hand, the least economically efficient producer could increase out put and reduce cost by 90% if he/she is to achieve the EE level of the most economically efficient producer. Distribution of Efficiency Scores (Pooled S am ple) TE distribution F igure 2 3 depicts the distribution of TE scores for the pooled sample under the study. The histogram shows a negative skewness w hich is a good indication that the majority of the sampled farmers are closer to being technically efficie nt. The average technical efficiency is ~ 80% as shown in T able 2 5, and 43% o f the total sampled farmers scoring below the average. Although the majority of the growers (57%) are above the average, there are still substantial amounts (43%) that need to in crease their effort to operate above average, given the objective of increasing domestic wheat production. AE distribution Figure 2 4 shows the histogram of AE scores and has a negatively skewed shape. The mean AE for the pooled sample is 80%, with 37% of the sampled farmers scoring below the average and 63% scoring above the average. EE distribution The minimum economic efficiency (EE) score as shown in Table 2 5 is 64% , with 42% of the sampled wheat farmers operat ing below the average EE score. F igure 2 5 depicts the negative skewness , impl ying that the median is on the right of mean score.
54 Efficiency Distribution by W ards TE distribution by wards The Monduli juu and Ngarenairobi wards have farmers with the least TE score s of 37% and 38% , respectively , whi le the minimum score s for the Mbulumbulu and Rhotia wards are 44% and 43% , respectively (Table 2 6). On the other hand, the max imum TE score of 97% is determined for the wards of Mbulumbulu, Monduli juu , and Rhotia while , in Ngarenairobi , the maximum TE sc ore is 92%. The average TE for the study area does not seem to vary much , as both Mbulumbulu and Ngarenairobi score 78% each , while Monduli juu and Rhotia each score 79%. Figure 2 6 shows the visual of histogram comparison of TE scores across wards . AE dis tribution by wards The Monduli juu farmer s have the lowest AE score at 23% compared to the minimum scores from Mbulumbulu, Ngarenairobi , and Rhotia farmers at 45%, 44% and 46% , respectively (Table 2 7). The maximum AE scores obtained in the study area are 98%, 96%, 94% , and 95% recorded for Mbulumbulu, Monduli juu , Ngarenairobi , and Rhotia , respectively . Overall , Monduli juu has the highest (82%) mean TE score, followed by Rhotia (80%), Ngarenairobi (79%) , and Monduli juu (78%). Figure 2 7 display s a visual comparison of the AE distribution by wards. EE distribution by wards The minimum EE for the wards is 20%, 9%, 17%, and 23% for Mbulumbulu, Monduli juu , Ngarenairobi , and Rhotia , respectively (Table 2 8). The highest maximum EE score from these 4 wards is 93% for both Mbulumbulu and Monduli juu , while the least maximum EE score is 85% for Ngarenairobi and 92% f or Rhotia . O n average, Mbulumbulu and Rhotia ha ve identical EE scores of 65% while Monduli juu and
55 Ngarenairobi each score 64%. Figure 2 8 shows the visual distribution of EE by the wards . Factors Influencing E fficiency In contrast with input and output factors used in estimating efficiency scores, the household idiosyncratic factors were analyzed to explore sources of inefficiencies for pooled sample data of the study area. Each ward specific location was also considered to capture implicitly their heterogeneity in soil fertility, distance to urban areas , and infrastructure accessibility effect on TE, AE, and EE. TE factors Of particular interest to this study are the coefficients of the variables associated membership since they are the 9 shows that with the ignorability condition (wi thout considering sample selection bias) ; the contract variable is positively and strongly significant at the 1% level on influencing TE. With contracts, farmers become more committed to farm work in order to and eventually bec groups/associations significantly (5% level ) influence TE. This implies that farmers who belong to groups/associations prove to become more efficient because of their collective action in transforming f actors of production to wheat output and their information sharing about good farm practices during their group meeting s . Further, the information contained in T able 2 9 reveals that age has a negative and significant (at the 5% level) effect on TE . This s uggests that older farmers are less efficient than younger farmers. This result is plausible given that older farmers tend to be more traditional and conservative , especially when it comes to adopting modern
56 farming technology. The coefficient of education is positive and statistically significant at the 10% level as expected and is in accordance with the findings of several other studies. This result indicates that the higher the level of education, the higher the TE. A possible explanation for this is tha t farmers with more education are likely to have better access to information and hence execute better farm management practices which ultimately improve their levels of efficiency. The coefficient of the household composition is positive though insignifi cant at the 10% level, impl ying that large family size provides enough labor for farm work. However , the composition of household members seems to ma tter with all age categories negatively influencing TE. A possib le explanation for this finding is that the greater the number of household members younger than 18 years old and the older than 50 years old of age categories , the less efficient the household become s because the time needed for adult s of age between 18 years old and 50 years old category to atten d to the needs of the other two age cate gori es , thus reducing the number of hours of labor allocated to farming activit ies . As to be expected, extension services (visits) influence farmers TE positively and significantly at the 5% level. Farm equipment is essential for wheat production in the study area , with ownership influenc ing TE positively at the 10% significant level . Ownership of modern farm equipment ensures carrying out various operations in a timely manner compared to delays which might be encoun ter ed when such services must be hired . Hybrid seeds are known to yield high er production output in various crops. Not surprisingly , the results show that the use of hybrid seeds ha s a positive and significant (5% level) impact on TE.
57 O ff farm income infl uences TE positively and significantly at the 1% level. This could be because off farm income facilitates purchases of factors of production at a certain point in time. Further, the result s reveal that there is no significant variation between wards , thus signifying the homogeneity of climatic conditions such as the soil and weather characteristics of the sampled wards. AE factors As in TE above, of particular interest to this study are the coefficients of the variables associated with contract participatio memberships since they are the indicators for 10 shows that contract s positively and strongly (significant at 5%) influence AE. With contracts, farmers receive factors of product ion that best lower the cost per unit of This implies that farmers through their associations could take advantage of buying factors of production in bulk , thus reduc i ng the individual cost per unit. Age has a negative coefficient in AE at the 5% significan ce level (Table 2 10). Th is could be because less modern and might not be optimizing the best inputs on the farm. Y ounger fa rmers are more eager to learn new skills and are more receptive of new technologies that help increase production and minimize cost s . Our findings are consistent with those of other studies that found the managerial capability to allocate factors of produc tion in cost minimizing behavior decreases with age (e.g. , Coelli et al., 2002). Land rent positively influences farmers AE at the 10% significan ce level. The finding suggests that farmers work hard on rented land to obtain sufficient returns to cover the rental cost as well as generate profits. The motivation for renting land is ,
58 therefore, to maximize profit / minimize cost. T hus , farmers select best farm management practices that minimize their production costs to realize the highest profit possible at pre vailing market prices. A p ossible explanation is that the on farm training conducted by extension officers enables farmers to acquire better information on good farm pract ices that lead them to select the least cost co mbination of farm inputs for wheat production. Farmers who live in the Mbulumbulu ward seem to be better at minimizing costs than their counterparts in the other wards . Th is result is somewhat surprising becau se Mbulumbulu is the most disadvantage d of the 4 wards in terms of distance to the city, road infrastructure, and market accessibility . It is the only ward where its farmers significantly operat e with a least cost combination at the 10% significan ce level. Th is finding seems to contradict the conventional wisdom that farmers with poor access to markets have less incentive in maximizing profit (minimizing cost) compared to those with better access to market s and those located near cities. It could also be ex plained that the farther away from the city, the lesser the advantages to market access, thus more attention is paid to cost reduction. Farm equipment is key to production . F or this study , farm equipment includes tractor s , sprayer s , plo ws, and wheelbarrow s . The results reveal that farmers who own somewhat modern farm implements are allocative ly efficient at the 5% significan ce level. O wnership of farm equipment/i mplements could save on rental expenses and the cost of hiring operators , thus leading farmers t o operate at low er cost s .
59 Hybrid seed influence s AE positively and is significant at the 10% level. Despite its higher relative price than local seed, its potential output realized spread s the cost per unit of output . The cost of purchasing hybrid seed is spread over the higher level of wheat production. Thus, per unit cost of hybrid seed becomes lower than that of local seed because the production level of local seed is lower. Off farm income influences AE positively and is strongly significant at the 1% level. A similar result is found in Chavas ( 2005 ) . O ff farm employment reduces labor ( household ) availability and enables the household to hire labor er s with the increased off farm income. The income further stimulates the use of industrial inputs on the f arms and the use of the best technology that link s farmers to least cost farm practices. EE factors As in the cases of TE and AE discussed earlier, o f particular interest to this study are the coefficients of the variables associated with contract particip ation and farmers association membership value chain. Table 2 11 shows that with ignorability condition (without con sidering sample selection bias), the contract variable is positively and str ongly significant at 1% on influencing EE. With contracts, farmers become more committed to farm work in order to a best factor of production that lowers the cost per unit of output produced, eventually they become efficient in wheat This implies that farmers who belo ng to groups/associations become more efficient because of their collective action in transforming factors of producti on purchased at lower bulk prices .
60 Age has a negative influence on EE at the 5% significan ce level (Table 2 11). The implication is that older farmers are traditional practitioners of farming and do not realize the potential output that could be produced with the least cost combination of modern factors of production. Education is significant at the 10% level, impl ying that knowledge obtained from formal education helps farmers acquire better farm management practices at the least cost combination of input requirements. Farmers receiving more extension visits have shown to be more efficient in terms of EE at the 1% significan ce level. Th is finding suggests that the training and information provided by extension officers are crucial not only for TE , but also for AE , which helps farmers to attain EE at the 1% level. Farm equipment ownership has a positive and significant influence on EE at the 5% level. Farm equipment o wnership saves farmers from having to hiring farm equipment and helps them attain best farmi ng management practices in an appropriate time needed for season al crops. By doing so , the production increase and the cost per unit go down. Thus , EE is attained. The use of hybrid seed s influence s EE at the 5% significan ce level. Since EE is about least cost and maximum output, then hybrid seed , ceteris paribus , work s for wheat farmers who plant hybrid seed s . Off farm income supports farming activities in hiring labor, and purchasing farm equipment and industrial inputs , t hus it positive ly influences EE a t the 1% significan ce level. Impact of Value Chain P articipation on E fficiency As mentioned earlier, a novel contribution of this study is the linking of efficiency measure s to value chain participation. Specifically, we examine the impact of value chain p articipation on TE, AE, and EE. Th is study uses propensity score matching (PSM) to ascertain the sample selection bias of observables. Nearest neighbor and caliper radius matching algorithm s were used to match the characteristics of both
61 participants and n onparticipants . In general, our results , as discuss ed in detail below, support the notion that levels of efficiency rise when farmers participate in vertical and horizontal coordination. We first examine the impact of vertical coordination on TE, AE and EE , followed by an examination of the impact of horizontal coordination on the same aforementioned efficiency measures. Vertical coordination impact TE effect vertically The nearest neighbor (NN) of 1:1 implies that all the vertical coordination participants were matched by one nonparticipant from the control group that has the closest characteristics. The matching was done with out replacement. However , the balancing property did not provide a good match as can be seen from the histogram in Figure 2 9. The hi stogram in Figure 2 9 shows the distribution of the sample before and after matching, where after matching the par ticipant distribution (matched treated) is not similar to the distribution of matched nonp articipants (matched control). Therefore , the TE eff ect of 6.2% under NN is still biased. Matched participants and nonparticipants distributions under a caliper radius of 0.005 were similar (Figure 2 10 ), implying good matches and that t he impact of participation can be evaluated . However, of 51 vertical c oordination participants , only 20 were used , and others were dropped in a given caliper radius (Table 2 12). Thus, we need to increase the caliper radius to accommodate more observations as long as the matches allow for more observations. Accordingly , we in creased the caliper radius to 0.13 and obtained suitable matches as confirmed by th e histogram (Figure 2 15 ) wh ich looked the same for matched treated and matched controlled observations . More importantly , the radius accommodates 49 of 51 participants; onl y 2 treated were
62 dropped . An attempt to further increase the radius to accommodate at least 1 more particip ants did not prove worthwhile because the distributions between matched participants and nonparticipants became dissimilar. The result s show that pa rticipation in vertical coordination as measured by farmers being in contract with their buyers , increases technical efficiency by 6.8% more for participants than for n onparticipants. This finding is significant at the 5% level (Table 2 12). AE effect vert ically As in the TE procedure, the vertical coordination participation under the caliper radius of 0.13 improves AE by 5.7% more than the counterpart at 10% significant level (Table 2 13). EE effect vertically As in both the TE and AE cases, vertical coo r dination participation under a caliper radius of 0.13 improves EE by 8.7% more for participants than for nonparticipants at the 5% significance level (Table 2 14). Horizontal coordination impact TE effect horizontally As in the previous case with the verti cal coordination , an attempt was made to use the nearest neighbor (NN) technique for horizontal coordination matching. The results are shown in T able 2 15 for farmers who particip ated in horizontal coordination as measured by being a member of ssociation. The results indicate an improvement in TE of 7.7% for those who participated compared with non participants. However, the results are bias ed because the histogram distributions for matched
63 treated and control groups are dissimilar (Figure 2 16 ) . T hus , the need for a better matching procedure for the sample is important . Accordingly , we employed caliper radius matching for 122 observations . The first caliper radius considered was 0.03, which had 109 good matches, with 13 observations being dropped . When w e increased the radius to 0.05 , 114 treated observations found good matches with only 8 observations being dropped (Table 2 15). Further increase of the radius did not give good matches; the matched distributions for treated and control as shown by the histograms were dissimilar like in the nearest neighbor matching procedure in F igure 2 16 . Therefore, with a caliper radius of 0.05, the horizontal coordination participants were found to improve their TE by 6.3% more than nonparticipants at the 1% sig nifican ce level. Figure 2 19 corresponds to a caliper radius of 0.05 that gives a visual look of matched participants and nonparticipants by histogram s . AE effect horizontally As with TE, horizontal coordination as measured by being a member of as sociation s improve s AE by 9.5% more for participants than nonparticipants at the 1% significan ce level. Table 2 16 shows that there is no significant difference in the AE effect across the matching methods , which impl ies that there are good observations for the matches. EE effect horizontally T able 2 17 shows that with a caliper radius of 0.05 , the participants improve their EE by 11.6% more than nonparticipants at the 1% significance level.
64 Concluding Remarks The general objective of t he chapter was to expl ain why wheat production is not responding to the ever increasing demand for wheat products in Tanzania. One factor considered to expla in the situation was the efficiency levels of wheat production units. TE, AE, and EE were first estimated over the pooled sample without the consideration of selection bias over participants and nonparticipants of the value chain. The translog functional form for production and cost functions were used in the stochastic frontier analysis. Land, fertilizer, and insecticides w ere found to significantly influence the level of wheat output at 1%, 10%, and 10% , respectively. However , insecticides application had a negative impact by lowering the level of wheat output . P reviou s research concluded that either the insects are resista nt to the type of chemical used or farmers are not using the recommended amount due to the cost (Chen et al., 2013) . On the other hand , th e trans log cost function revealed that prices of seed, fertilizer, and insecticides monotonically increase with production cost, except for fertilizer price which violates the monotonicity assumption at the 5% significance level. T otal production cost was also fo und to significantly increase with the quantity of wheat output at the 5% level. The mean TE, AE, a nd EE found in the study area wer e 79%, 80%, and 64% , r espectively. This result implied that benefits could be accrued by increasing TE and AE by 21% and 20% , respectively using the same factors of production at the existing market prices. The overall efficiency ( EE ) could be increased by 36% when TE and AE are attained . The factors that seemed to influence the efficiency levels (TE, AE , and EE) the most were contract , association membership, education, extension visits, farm
65 equipment (tractor, plow , knapsack sprayer, and wheel barrow ), hybrid seed, and off farm income. On the other hand, measuring the net effect of value chain pa rticipation on TE, AE, and EE using the PSM technique reduces biases associated with observed vari ables. U sing caliper radii , vertical coordination participants improve d their TE, AE , and EE score s by 6.8%, 5.7%, and 8.7% respectively more than nonparticipants , while horizontal coordin ation participants improve d their TE, AE, and EE scores by 6.3%, 9.5%, and 11.6% , respectively , more tha n nonparticipants . The study recommends t he expansion of wheat plots, application s of fertilizer, and u se of appropriate or recommended insecticides to increase wheat production in the study area. In order t o improve farm unit efficiencies , farmers need to be empowered with farm equipment, hybrid seeds, formal education for the young er generation, and on farm training by extension officers. Farmers are al so encouraged to participate in off farm activities when the opportunity arises as the findings revealed that farm efficiencies improve with the off farm income. Moreover , farmers are advised to participate in the value chain through contract and associati on m embership because these also improve farm efficiency . .
66 Table 2 1 . Household and farm characteristics Household characteristics min max mean Age (years) 20 85 43.532 Education (in years) 1 15 7.142 Experience (years) 1 46 13.826 Household composition (number) 2 17 6.523 <18 years old 0 11 3.139 18<50 years old 0 9 2.974 >50 years old 0 2 0.439 Land leased (dummy) 0 1 0.619 Extension visits (frequency) 0 10 0.936 Meetings attended (frequency) 0 5 0.461 Transport o wnership (dummy) Oxcart, Motorcycle, Car 0 1 0.07 Farm equipment (dummy) Tract or, Sprayer, Wheelbarrow, Plow 0 1 0.09 Livestock ownership (dummy) Cow, Goat, Sheep 0 1 0.936 Hybrid seed (dummy) 0 1 0.103 Off farm income (dummy) 0 1 0.1 Farm featur es Land(acre) 0.5 50 5.12 Output (Kgs/acre) 33.3 1600 704.788 Seed (Kgs/acre) 40 150 79.917 Fertilizer (Kgs/acre) 0 225 23.904 Herbicides (Lts/acre) 0 3 0.698 Insecticides (Lts/acre) 0 2.5 0.569 Pesticides (lts/acre) 0 2.5 0.383 * Pool n=310
67 Table 2 2. Household and farm characteristics by wards Mean values Household Mbulumbulu (n=116) Monduli juu (n=114) Ngare Nairobi (n=22) Rhotia (n=58) Age 46 40 39 47 Contract a 20 18 4 9 Members a 44 50 9 16 Education (in years) 7.1 6.8 7.7 7.4 E xperience (years) 12 14 14 17 Household composition (number) 6 6.8 6 7.1 <18 years old 2.8 3.3 2.5 3.6 18<50 years old 2.7 3.1 3.3 3.1 >50 years old 0.5 0.4 0.23 0.47 Land leased (dummy) 6.7 8 71.8 15.9 Extension visits (frequency) 0.56 1.26 1.05 0.6 6 Meetings attended (frequency) 0.57 0.43 0.45 0.31 Transport ownership Oxcart, Motorcycle, Car 0.08 0.08 0.09 0.05 Farm equipment (dummy) Tract or, Sprayer, Wheelbarrow, Plow 0.08 0.12 0.04 0.07 Livestock ownership (dummy) Cow, Goat, Sheep 0.92 0.96 0. 95 0.90 Hybrid seed (dummy) 0.11 0.11 0.09 0.09 Off farm income (dummy) 0.10 0.11 0.05 0.09 Farm features Land (acre) 4.2 6.0 6.0 4.9 Output (Kgs/acre) 691.1 716.1 726.3 701.8 Seed (Kgs/acre) 78.5 83.3 79.6 76.3 Fertilizer (Kgs/acre) 27.8 18.6 2 9.5 24.5 Herbicides (Lts/acre) 0.7 0.8 0.7 0.6 Insecticides (Lts/acre) 0.5 0.6 0.6 0.6 Pesticides (Lts/acre) 0.3 0.4 0.4 0.3 a = frequency
68 Table 2 3. Estimates for production function (OLS) and stochastic frontier (MLE) OLS MLE Output (kgs) Estimat e Std. Error t value Parameter Std.err t value (Intercept) 1.926 5.737 0.336 2.676 3.982 0.672 Land (acres) 2.373*** 0.620 3.827 1.964*** 0.499 3.936 Seed (kgs/acre) 0.970 2.291 0.423 0.785 1.590 0.494 Fertilizer (kgs/acre) 0.328* 0.188 1.742 0.205 0.1 71 1.203 Herbicides (lts/acre) 2.895 2.083 1.390 0.837 1.412 0.593 Insecticides (lts/acre) 3.444* 1.793 1.921 2.938* 1.535 1.914 Pesticides (lts/acre) 2.714 1.977 1.373 1.517 1.367 1.110 Labor (man days) 0.245 1.225 0.200 0.547 0.997 0.549 Land* land 0.283*** 0.058 4.875 0.267*** 0.055 4.854 Land*Seed 0.124 0.127 0.975 0.077 0.106 0.729 Land*Fertilizer 0.008 0.014 0.545 0.011 0.013 0.851 Land*Herbicides 0.204 0.139 1.467 0.192 0.130 1.473 Land*Insecticides 0.005 0.146 0.034 0.139 0.1 34 1.037 Land*Pesticides 0.201 0.139 1.450 0.035 0.125 0.279 Land*Labor 0.009 0.095 0.090 0.033 0.085 0.392 Seed*Seed 0.219 0.483 0.453 0.095 0.336 0.282 Seed*Fertilizer 0.029 0.038 0.758 0.009 0.034 0.262 Seed*Herbicides 0.504 0.423 1.1 92 0.154 0.290 0.531 Seed*Insecticides 0.800** 0.370 2.164 0.652** 0.319 2.042 Seed*Pesticides 0.123 0.417 0.294 0.094 0.306 0.308 Seed*Labor 0.126 0.239 0.528 0.118 0.204 0.579 Fertilizer*Fertilizer 0.050 0.034 1.489 0.041 0.032 1.261 Fer tilizer*Herbicides 0.009 0.044 0.218 0.006 0.041 0.150 Fertilizer*Insecticides 0.034 0.041 0.812 0.026 0.039 0.667 Fertilizer*Pesticides 0.004 0.044 0.080 0.053 0.044 1.215 Fertilizer*Labor 0.009 0.029 0.299 0.016 0.026 0.595 Herbicides*Herbici des 0.710 0.538 1.319 0.833* 0.452 1.843 Herbicides*Insecticides 0.292 0.363 0.804 0.277 0.310 0.893 Herbicides*Pesticides 0.161 0.374 0.430 0.450 0.351 1.282 Herbicide*Labor 0.210 0.351 0.598 0.037 0.338 0.110 Insecticides*Insecticides 0.68 6 0.465 1.477 0.363 0.460 0.790 Insecticides*pesticides 0.607* 0.338 1.795 0.575* 0.297 1.937 Insecticides*Labor 0.005 0.271 0.019 0.065 0.253 0.258 Pesticide*Pesticides 1.424** 0.565 2.519 1.694*** 0.479 3.534 Pesticides*Labor 0.602** 0.28 6 2.103 0.380 0.248 1.531
69 Table 2 3. Continued Estimate Std. Error t value Parameter Std.err t value Labor*Labor 0.009 0.159 0.055 0.037 0.144 0.254 R 2 0.950 Adjusted R 2 0.943 F 35,274 147.9*** 2 0.050 0.114 Lik. 42.770 57.698 4.003*** 0.881 4.542 2 v 0.007 2 u 0.108 0.95 ***,** and * represent significant level at 1%, 5% and 10% respectively
70 Table 2 4. OLS and MLE estimates of translog cost function OLS MLE Cost Estimate Std. Error t value Parameter Std.err t value (Intercept) 3.056 2.903 1.052 0.454 1.962 0.231 Ps/Pl 4.073*** 1.136 3.585 2.857*** 0.809 3.533 Pf/Pl 2.499** 1.154 2.166 2.534*** 0.902 2.81 0 Ph/Pl 0.945 0.678 1.393 0.697 0.738 0.944 Pi/Pl 0.602 0.694 0.868 1.436** 0.654 2.197 Pp/Pl 0.187 0.676 0.276 0.849 0.643 1.322 Q 0.551* 0.319 1.727 0.622** 0.278 2.233 Ps/Pl*Ps/pl 0.532* 0.275 1.930 0.223 0.217 1.027 Ps/Pl*Pf/Pl 0.042 0.212 0.197 0.086 0.211 0.407 Ps/Pl*Ph/Pl 0.233* 0.138 1.684 0.252* 0.134 1.878 Ps/Pl*Pi/Pl 0.066 0.143 0.460 0.126 0.117 1.074 Ps/Pl*Pp/Pl 0.013 0.154 0.082 0.106 0.148 0.719 Ps/Pl*Q 0.128** 0.058 2.214 0.194*** 0.057 3.375 Pf/Pl*Pf/Pl 0.288 0.2 47 1.166 0.201 0.216 0.929 Pf/Pl*Ph/Pl 0.014 0.163 0.088 0.103 0.173 0.596 Pf/Pl*Pi/Pl 0.214 0.156 1.371 0.422*** 0.155 2.727 Pf/Pl*Pp/Pl 0.081 0.189 0.428 0.286 0.201 1.428 Pf/Pl*Q 0.156** 0.062 2.499 0.181*** 0.054 3.389 Ph/Pl*Ph/Pl 0.016 0.117 0.136 0.107 0.112 0.955 Ph/Pl*Pi/Pl 0.068 0.089 0.767 0.023 0.084 0.272 Ph/Pl*Pp/Pl 0.041 0.082 0.496 0.019 0.082 0.226 Ph/Pl*Q 0.028 0.038 0.744 0.039 0.040 0.968 Pi/Pl*Pi/Pl 0.045 0.104 0.437 0.017 0.099 0.171 Pi/Pl*Pp/Pl 0.008 0.077 0.101 0.020 0.078 0.258 Pi/Pl*Q 0.007 0.040 0.167 0.013 0.042 0.305 Pp/Pl*Pp/Pl 0.056 0.109 0.513 0.075 0.112 0.670 Pp/Pl*Q 0.004 0.038 0.112 0.017 0.043 0.394 Q*Q 0.052*** 0.020 2.623 0.022 0.024 0.911 R 2 0.934 Adjusted R 2 0.928 F 27,282 147.7*** 2 0.046 0.103 Lik. 51.261 70.480 3.681*** 0.825 4.464 2 v 0.007 2 u 0.096 0.93 ***,** and * represent the significance level at 1%, 5% and 10% respectivel y
71 Table 2 5 . TE, AE, and EE for wheat production (pooled sample) Min Max Mean Std. TE 0.37 0.97 0.79 0.133 AE 0.24 0.98 0.80 0.125 EE 0.09 0.93 0.64 0.178 Table 2 6. TE distribution by wards Wards Min M ax M ean Std. Mbulumbulu 0.44 0.97 0.78 0.127 Monduli juu 0.37 0.97 0.79 0.139 Ngarenairobi 0.38 0.92 0.78 0.127 Rhotia 0.43 0.97 0.79 0.140 Table 2 7. AE distribution by wards Wards Min M ax M ean Std. Mbulumbulu 0.45 0.98 0.82 0.109 Monduli juu 0.23 0.96 0.78 0.142 Ngarenairobi 0.44 0.94 0.79 0.116 Rhotia 0.46 0.95 0.80 0.121 Table 2 8. EE distribution by wards Wards Min M ax M ean Std. Mbulumbulu 0.20 0.93 0.65 0.169 Mondulijuu 0.09 0.93 0.64 0.191 Ngarenairobi 0.17 0.85 0.64 0.164 Rhotia 0.23 0.92 0.65 0.181
72 Table 2 9. Factors influe ncing TE (Tobit model) TE Estimate Std. Error t value (Intercept) 0.712*** 0.053 13.555 C ontract 0.057*** 0.020 2.850 M embership 0.042*** 0.015 2.825 Age 0.002** 0.001 2.166 E ducation 0.004* 0.002 1.942 Experience 0.001 0.001 1.551 H ouse h old compo sition 0.025 0.019 1.301 Age18 below 0.026 0.019 1.356 Age18 to 50 0.029 0.019 1.480 A ge 50 up 0.016 0.021 0.744 L and leased 0.012 0.015 0.776 E xt ension visit 0.007** 0.003 2.045 M eeting 0.000 0.007 0.009 Mbulumbulu 0.012 0.029 0.419 Rhotia 0 .025 0.031 0.795 Monduli juu 0.011 0.029 0.378 Transport ownership 0.003 0.027 0.128 Farm equipment 0.044* 0.025 1.785 Livestock 0.046 0.028 1.618 Hybrid seed 0.048** 0.023 2.082 Off farm income 0.051*** 0.020 2.562 Logsigma 2.119*** 0.040 52.751
73 Table 2 10. Factors influencing AE (Tobit model) AE Estimate Std err t value (Intercept) 0.730*** 0.047 15.411 C ontract 0.037** 0.018 2.093 M embership 0.074*** 0.013 5.558 Age 0.001** 0.001 1.985 E ducation 0.003 0.002 1.357 Experience 0.001 0.0 01 1.353 H ousehold composition 0.011 0.017 0.639 Age18 below 0.009 0.017 0.517 Age18 to 50 0.016 0.018 0.888 A ge 50 up 0.008 0.019 0.434 L and leased 0.024* 0.014 1.786 E xt ension visit 0.009*** 0.003 2.878 M eeting 0.004 0.006 0.705 Mbulumbulu 0.047* 0.026 1.774 Rhotia 0.040 0.028 1.409 Monduli juu 0.006 0.026 0.228 Transport ownership 0.010 0.024 0.426 Farm equipment 0.047** 0.022 2.096 Livestock 0.006 0.026 0.224 Hybrid seed 0.037* 0.021 1.776 Off farm income 0.051*** 0.018 2.848 Log sigma 2.223*** 0.040 55.343
74 Table 2 11. Factors influencing EE (Tobit model) EE Estimate Std. Error t value (Intercept) 0.521*** 0.068 7.695 C ontract 0.069*** 0.026 2.685 M embership 0.086*** 0.019 4.519 Age 0.002** 0.001 2.127 E ducation 0.005* 0.003 1.757 Experience 0.002 0.001 1.564 H ousehold composition 0.031 0.024 1.271 Age18 below 0.031 0.025 1.253 A ge18 to 50 0.038 0.025 1.511 A ge 50 up 0.012 0.028 0.438 L and leased 0.029 0.020 1.462 E xt ension visit 0.012*** 0.004 2.875 M eeti ng 0.003 0.009 0.356 Mbulumbulu 0.043 0.038 1.146 Rhotia 0.049 0.040 1.218 Monduli juu 0.017 0.038 0.439 Transport ownership 0.015 0.035 0.420 Farm equipment 0.077** 0.032 2.408 Livestock 0.035 0.037 0.966 Hybrid seed 0.073** 0.030 2.476 Off farm income 0.080*** 0.026 3.108 Logsigma 1.864*** 0.040 46.417 Table 2 12. TE effect due to participation in vertical coordination TE effect SE t value p value treated sample matched Nearest Neighbor (1:1) 0.062 0.031 2.011 0.044 51 Caliper (0.005) 0. 030 0.014 2.141 0.032 20 Caliper (0.01) 0.045 0.019 2.432 0.015 27 Caliper (0.02) 0.067 0.023 2.899 0.003 35 Caliper (0.03) 0.072 0.023 3.013 0.003 37 Caliper (0.10) 0.074 0.030 2.495 0.013 47 Caliper (0.13) 0.068 0.031 2.216 0.028 49
75 Table 2 13. AE effect due to participation in vertical coordination AE effect SE t value p value treated sample matched Nearest Neighbor (1:1) 0.051 0.031 1.628 0.103 51 Caliper (0.005) 0.048 0.014 3.45 0.001 20 Caliper (0.01) 0.054 0.017 3.263 0.001 27 Caliper ( 0.02) 0.059 0.022 2.619 0.009 35 Caliper (0.03) 0.060 0.023 2.635 0.008 37 Caliper (0.10) 0.062 0.030 2.050 0.040 47 Caliper (0.13) 0.057 0.031 1.848 0.065 49 Table 2 14. EE effect due to participation in vertical coordination TE effect SE t value p value treated sample matched Nearest Neighbor (1:1) 0.077 0.043 1.779 0.075 51 Caliper (0.005) 0.061 0.021 2.940 0.003 20 Caliper (0.01) 0.075 0.026 2.901 0.004 27 Caliper (0.02) 0.091 0.032 2.892 0.004 35 Caliper (0.03) 0.097 0.032 2.984 0.003 37 Ca liper (0.10) 0.095 0.0414 2.298 0.022 47 Caliper (0.13) 0.087 0.043 2.031 0.042 49 Table 2 15. TE effect due to participation in horizontal coordination TE effect SE t value p value treated sample matched Nearest (1:1) 0.077 0.021 3.785 0.000 122 Cal iper (0.03) 0.061 0.017 3.544 0.000 109 Caliper (0.04) 0.064 0.018 3.585 0.000 113 Caliper (0.05) 0.063 0.018 3.534 0.000 114 Table 2 16. AE effect due to participation in horizontal coordination AE effect SE t value p value treated sample matched Ne arest (1:1) 0.101 0.018 5.688 0.000 122 Caliper (0.03) 0.096 0.016 5.927 0.000 109 Caliper (0.04) 0.096 0.0167 5.769 0.000 113 Caliper (0.05) 0.095 0.017 5.722 0.000 114 Table 2 17. E E effect due to participation in horizontal coordination EE effect SE t value p value treated sample matched Nearest (1:1) 0.132 0.026 5.080 0.000 122 Caliper (0.03) 0.115 0.023 5.076 0.000 109 Caliper (0.04) 0.117 0.023 5.019 0.000 113 Caliper (0.05) 0.116 0.023 4.959 0.000 114
76 Figure 2 1. Per ca pita consumption g ap be tween wheat and staple food maize and rice in SSA Source: Mason et al., (2012) Figure 2 2. Production consumption trend of wheat in Tanzania 0 200 400 600 800 1000 1200 1964/1965 1966/1967 1968/1969 1970/1971 1972/1973 1974/1975 1976/1977 1978/1979 1980/1981 1982/1983 1984/1985 1986/1987 1988/1989 1990/1991 1992/1993 1994/1995 1996/1997 1998/1999 2000/2001 2002/2003 2004/2005 2006/2007 2008/2009 2010/2011 2012/2013 2014/2015 2016/2017 Total ConsumptionÂ (1000 MT) Total Production (1000 MT)
77 Figure 2 3 . The distribution of TE scores of the pooled sample in the study area Figure 2 4 . The distri bution of AE scores of the pooled sample in the study area
78 Figure 2 5 . The distribution of EE scores of the pooled sample in the study area Figure 2 6 . Technical e fficiency distribution by wards 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Mbulumbulu Monduli juu Ngarenairobi Rhotia TE min TE max TE mean
79 Figure 2 7 . Allocative efficiency distribution by ward s Figure 2 8 . Economic efficiency distribution by wards 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Mbulumbulu Monduli juu Ngarenairobi Rhotia AE min AE max AE mean 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Mbulumbulu Monduli juu Ngarenairobi Rhotia EE min EE max EE mean
80 Figure 2 9 . Vertical coordination histograms before and after matching by Nearest Neighbor algorithm of ratio 1:1 Figure 2 10 . Vertical coordination histograms before and after matching by cal iper radius of 0.005
81 Figure 2 11 . Vertical coordination histograms before and after matching by caliper radius of 0.01 Figure 2 12 . Vertical coordination histograms before and after matching by caliper radius of 0.02
82 Figure 2 13 . Vertical coord ination histograms before and after matching by caliper radius of 0.03 Figure 2 14 . Vertical coordination histograms before and after matching by caliper radius of 0.1
83 Figure 2 15 . Vertical coordination histograms before and after matching by calip er radius of 0.13 Figure 2 16 . Horizontal coordination histograms before and after matching by Nearest Neighbor algorithm of 1:1 ratio
84 Figure 2 17 . Horizontal coordination histograms before and after matching by caliper radius of 0.03 Figure 2 18 . Horizontal coordination histograms before and after matching by caliper radius of 0.04
85 Figure 2 19 . Horizontal coordination histograms before and after matching by caliper radius of 0.05
86 CHAPTER 3 ASSESSMENT OF WELFARE IMPACTS OF TANZANIA WHEAT FA RMERS PARTICIPATION IN THE VALUE CHAIN Overview In the previous chapter , participation in the valu e chain improves the farm level efficiency of wheat production. This chapter provides further evidence of the link between particip ation in the value chain and welfare changes for wheat farmers in Tanzania. Specifically , it is hypothesized that, ceteris paribus, participation in terms of profit per unit of wheat sold. Introduct ory Remarks The e ver increasing gap between domestic wheat production and wheat demand suggest s the potential for wheat producers in Tanzania to meet this demand by increasing their production. Although economic theory holds that an increase in demand in a relatively compe titive industry should provide the price incentive to attract new entrants , thus increasing the supply of the commodity and returning prices to its long run equilibrium level, this general ly has not been the case observed in Africa and particularly in Tanz , undoubtedly , foreign suppliers have responded to the market signals, the majority of domestic wheat producers have shown either an unwillingness to do so or an inability to respond to such market signals. P ossible explanation s for the lackluster response of domestic farmers to pursue an obvious market opportunity might be their failure to participate formally in the value chain and a breakdown in communication s (information flow) that could enable these farmers to access high v alue wheat markets. Most small scale farmers in developing countries such as Tanzania sell their crops at the farm gate to intermediaries (brokers),
87 often at a low price (Fafchamps & Hill, 2005) . They produce an d sell to SPOT markets without directing their production to the market requirements, thus creating the failure for the country to satisfy the bulk of its demand without imports. The importing wheat price is generally low (e.g. , in 2015 , ranged from US$ 19 5 US$ 236/MT) compared to the domestic price . H owever, from 2005 to 2010, wheat farmers and traders in Tanzania received higher prices due to the effects of trade policies such as the East African Community Common External Tariff (35% ad valorem) and expens ive import procedures at the Dar es Salaam port in Tanzania . Although producers rece ived significant incentives after the tariffs and high import costs, wheat production increased but did not keep pace with consumption. The share of wheat production to con sumption was even lower , with wheat impor ts continue to account for 30% of the total food import bill (URT, 2013 ; USDA, 2016 ) Among many factors, nonparticipation in the value chain and broken communications information flow hinder farmers from taking adv antage of market opportunities. Lack of strong linkages between farmers themselves and with postharvest actors in the wheat value chain marginalizes welfare gain because prices received from intermediaries are much lower and rarely cover the cost of production. The linkage between farmers themselves is associated with their participation in farmers groups / associations or more formally in the cooperatives . In the value chain s in agricultural groups /associations refer t o their participa tion in horizontal coordination . Although these names have slight differences,
88 involvement in associations has shown progressive outcome s through their col lective actions as is vastly documented in the literature. Acting collectively, farmers might reduce their transaction costs for acc essing inputs and transporting output s , easing access to market information and extension services, and improving their barg aining power with postharvest actors (Abebaw & Haile, 2013 ; Roy & Thorat, 2008 ; Holloway et al., 2000 ; Wollni & Zeller, 2007 ; Thorp et al., 2005) . Despite such apparent advantages to farmers from group /association participation, the findings in the literature are not clear differ according to their local location and a gro economic condition (Hoken & Su, 2015) . The linkage between farmers and postharvest actors is associated with their involvement in contract farming. In the value chain studies, this type of linka ge is referred to as vertical coordination. Several benefits could arise as a result of such coordination. For example , vertical coordination through contractual arrangements between actors could enable farmers to gain valuable market intelligence. Farmers may also mitigate production and marketing risks by having a guaranteed market for their output, secure immediate market outlet s , and the ability to easily monitor quality and safety ( Reardon et al., 2004 ; Gulati et al., 2007 ; Miyata et al., 2009 ; Abebe et al., 2013 ; Boger, 2001 ; Barrett et al., 2012) . Moreover, some chain actors make it easier for inputs, modern technology, and credit (Bolwig et al., 2009 ; Dries et al., 2009 ; (Maertens & Swinnen, 2009) ; (Rao & Qa im, 2011) ; (Barrett et al., 2012) . (Masakure & Henson, 2005) listed the benefits farmers are likely to enjoy from participating in contracts as
89 those associated with reducing market uncertainty, enhancing knowledge acquisition , and increasing fa Despite the economic importance of co operatives and contracts to actors, most small scale farmers in Tanzania are not part of the formal value chain, opting rather to do their production and marketing activities by themselves. As a conseque nce , they operate within a framework/system that is characterized by weak or poor coordination with little or no legal enforcement between the main po stharvest actors. Moreover, few farmers belong to associations/coope ratives which could help them gain acce ss to guaranteed markets and increased power of negotiation for their produce. The lack of access to collective bargaining power by small scale farmers represents a monetary loss to them because individually they cannot influence the market. Within the lit erature, several studies have shown small scale farmers to be uncompetitive in high value markets w hile other studies have shown farmers are productive when are involved in associations and are being provided with institutional support (Narrod et al., 2009) . A review of the literature indicates that there has not been any formal comprehensive assessment of the value chain participation of wheat farmers i n Tanzania. Those studies close t to this one did not address the issue of horizontal and example SAGCOT (2009) only covered the mapping aspects of the value c hain where mapping means tracing the flow of inputs, good s, and services from production point to the ultimate consumer . A USAID (2010) report on staple food value chain analysis focused mainly on the pro duction and consumption trends, constraints , and opportunities.
90 In addition , only a few studies have addressed the impact of contracts and cooperatives separately after controlling for endogeneity. In this study , the possible endogeneity problem could be f The problem of not controlling endogeneity cause d by unobservables could result in bias ed estimates, when , for instance, the unobservables affect the participation decision of the individual in the value chain (Rosenbaum & Rubin, 1983 ; Heckman et al. , 1997 ; Caliendo & Kopeinig , 2008 ; Smith & Todd , 2005 ; Dehejia & Wahba , 1999) . These unobservable factors as pointed out by (Barrett et al., 2012) may include individual r isk aversion behavior, social capital, personal technical ability, the interest of crop grown, or trust /distrust of association s and contracts. P ropensity score matching (PSM) solves for observed factor bias , thus produc ing less biased results. However , ad ditional analysis is required to account for the likely impacts of the unobservables. This study, therefore, departs from others in that it employs the PSM technique in assessing the impact of and performs a sensitiv ity analysis for unobservables. The spec ific purpose of this study is fourfold . First, we provide a detailed description of the wheat value chain from production to consumption. Second, we examine the mean statistical differences be tween the participant and nonparticipants of the value chain. Third, we decision in value chain through vertical and horizontal c oordination. Fourth, we asse ss overall farmers in terms of net returns fr om their participation in two coordination systems (horizontal and vertical coordination). The rationale is to demonstrate the potential benefits in terms of increased wheat profit that farmers are
91 likely to obtain through formal involvem ent in value chains. Increased profits for farmers will boost household welfare and eventually enhance their food accessibility. Literature Review Concept of V alue Chain and Value Chain D evelopment The v alue chain concept carries various definitions based on the question the researcher wants to address. Donovan et al. (2015) grouped the definitions into three main areas based on the ir critical review of various value chain guide books. Accor dingly, the authors note that value chain can be defined as a set of activities, or a set of actors, or as a strategic network to engage in exten sive collaborations. This study adopts t he first definition whereby value chain is defined as the full range of activities required to bring a product or service from conception, through the different phases of production, delivery to final customers, and final disposal after use (Kaplinsky & Morris 2001) . It, therefore, describes how value is created from the conception of a product to i ts final consumption, including the different stages of input supply, design, production, distribution, retailing and support services. It can also be viewed as an organized system of transforming the products in various forms from production to consumptio n. The value chain, therefore, embodies two main notions , notably value and chain. Value is associated with value added and chain suggests a somewhat linear movement of the product from one actor to another (chain of events ). As the product moves along the chain , it increases its value through transformation/processing, relocation, and distribution , thus the concept of value chain. In agriculture, food safety and food functionality also add value to the products through product differentiation. The incremen tal value of the resultant products can be identified by their price differences .
92 Value Chain Development (VCD) is geared toward analyzing the value chain and addressing key weakn esses in a manner that contribute s to the development or improvement in the v alue chain . Therefore, VCD is a positive or desirable change in a value chain to extend or improve productive operations and generate soci o economic benefits toward poverty reduction, income and employment generation, economic growth, environmental performa nce, gender equity , and other development goals (UNIDO, 2011) . The value chain concept in agriculture involves linkages of actors and their agri food products toward adding value for their consumers. According to this view, the features of value chain development include mapping ( a schematic representation of the entire value chain), coordination, governance, upgrading , meeting consumer demand , and being competitive ness . P roducts gain value as they move along the value chain to various actors , say from input suppliers (such as seed providers) to farmers, then to intermediaries such as processors, wholesalers, retailers, and ultimate ly to consumers. Therefore, there must be linkages between actors to facilitate the movements of these products. These links need to be effective so that the benefits of the value chain are distributed among the chain actors. The value chain is not sustainable if only one actor receives all the benefits . Typically, farmers receive the lowest share of the consumer dollar unless special arrangements are put in place to ensure that they benefit from some of the downstream activities. The low returns to farmers in the value chain may be attributed to several factors , including the relatively small quantities traded by individual farmers, lack of organization, l ack of access to market information by farmers, or the risk of product
93 damage that is passed on to buyers. Other factors include high product transport costs to urban market and weak linkages with actors further up the chain. In order to deal with these im perfections , value chain actors need to be organized and have external support to participate effectively in the high value markets (Markelova, 2009). E xternal support as highlighted in the literature include s better rural infrastructure , education al insti tutions , and research and extension services (Hazell et al., 2007). Many studies have shown the impact of value chain participation in various farm Bi rthal et al.(2005) on vertical coordination in high value commodities found that contract s reduce transaction costs and improved market efficiency to benefit smallh olders. C oordinated f armers were paid better prices and enjoyed the benefit of assured pro curement of their products . Valkila et al. (2010) employed the value chain approach to assess whether the Fair Trade system empowers traders. They found that the retailers of the c onsuming countr ies appropriate d the largest share of the price paid by the ultimate consumer despite the premium prices received by farmers that are set by Fair Trade . Warsanga, ( 2014) employed the marketing margins to assess price variations among actors of banana value chai n in Tanzania. He found that farmers received lowest price share and prices. The study concluded on the establishment of strong coordination of banana actors to have mu tual benefit among actors. Swinnen & Maertens (2006) in their paper on globalization, privatization , and vertica l coordination in the value chains of developing and transition al countries argue d that private vertical coordination (after collapsed state controlled coordination) is growing rapidly despite production constraints
94 caused by factor market imperfections. T hey further showed that vertical coordination impact s both the quality, equity, and efficiency of the agri food system , and welfare. Olukunle (2013) found that improvements in the rural infrastructure strengthened the value chain for cassava in Nigeria by increas ing income and employment for smallholders in that country . Alemu et al. (2016) analyzed the coordination of honey producers in Ethiopia. They found that participation through contract ha d a significant effect on productivity and income but not through cooperatives. They argue d that the mismanagement of cooperatives could be the reason for of the in significant effect on performance and well being. It is , therefore, welfare, which has not been conducted on the wheat industry in Tanzania. Theor etical F ramework A frequent problem encountered by social scientists in the qu est to determine the impact of a decision taken by a particular group (treated) vs . a group that has not made that decision (untreated) . This problem formally described in the literature as self selection. Briefly , self selection implies that individuals in a particular group decide for themselves to join a particular group based on a set of unobservable characteristics that influence their decision and therefore are not randoml y assigned to any given group. Apart from statistical estimation issues, this pre sents a challenge to the researcher who in most cases only observes the outcome of the choice made by the individual but not the effect had that same (or identical similar) individual been randomly assigned to the alternate group. For instance, in our stud y, we only observe the outcome of participation in the value chain, not the outcome of nonparticipation in the value chain. C omparison s
95 are easier if the data are randomized experiment s rather than nonrandomized data (observations ) as is the case in the so cial sciences such as economics. In light of the above, several approaches have been advanced in the literature aim ed at circumventing this issue. In general , the aim is to evaluate the treatment effect by select ing a control unit (untreated group) that is identical in terms of characteristics to the self selected unit (treated group) in order for the researcher to make inferences about the choice that was made . Moreover, it is important to account for observable and unobservable factors when evaluating the treatment effect in order to make good inferences about the choice. Failure to do so could result in selection bias (endogeneity) problems, thus leading to faulty inferences being drawn regarding the choice, program, or the particular treatment. The stati stical technique aimed at matching the control (untreated) group and the treated group so as to reduce selection bias due to observables was advanced in the early1980s (Heckman et al., 1997) . One difficulty with earlier approaches was the difficult y in match ing each covariate when there we re many observations. This led Rosenbaum and Rubin (1983) to propose using the approach of function of covariates. The approach of using the function of covariates is formally known as pr opensity score matching (PSM).For our study case, we use PSM to estim ate the probability (propensity score) of participating in the value chain. Propensity Score M atching PSM is a statistical technique in which treatment individuals (beneficiaries of the program) are matched with one or more of the controlled individuals b ased on scores obtained from the function. For PSM to yield unbiased inferences, conditional independence must exist between the participants and nonparticipants of the program.
96 It is a well recognized approach/technique in the literature that can be appli ed to various fields/disciplines such as statistics ( Rosenbaum, 2002 ; Rubin & Waterman, 2006) , sociology (Morgan & Harding, 2006) , econ omics (Imbens, 2004 ; Dehejia & Wahba, 2002 ; Dehejia & Wahba, 1999 ; Abadie & Imbens, 2006) , and political science (Ho et al. 2007 ; Hansen & Bowers, 2008 ; Herron & Wand, 2007 ; Bowers & Hansen, 2005) . S tatistical inference about the treatment effect on the individual outcome involves having prior information before participation in the program. The missing information (unobserved outcome) is referred to as a counterfactual outcome. In such circumstances, estimating individual treatment effect is impossible, so we need to consider either obtaining an average treatment effect (ATE) or the average treatment effect on the treated (ATT) where consideration is given to the entire sample population . ATE and ATT are the main two parameters that are estimated in the lite rature where ATE is the difference of the expected outcomes of participants and nonparticipants in the program. For policy implication, ATE is not very useful because it does not exclude anyone (Heckman et al., 1997) . For instance, if a given program is for non educated individuals, there i s no interest in the participation of educated individuals . D etermining ATE will not bring sense as it does not exclude those who were educated before the program . Therefore, the preferred technique will be to determine ATT, which focuses on the intended g roup that needed the program. words, the ATT parameter is the actual gain from part icipation in the program and can be compared with its cost to determine whether the program should proceed, assuming
97 it has a positive impact (Heckman et al., 1997) . Accordingly , this study focuses on ATT by fixing counterfactual information from a group of untreated using the covariate mat ching procedure proposed by Rosenbaum & Rubin (1983) . The major assumption of the treatment effect for evaluation studies is that the treatment sat isfies the exogeneity condition , referred to in the literature by several names. For example, Rosenbaum & Rubin (1983) called it unconfoundedness; Heckman and Robb (1985) used the term, selection on observables ; and Lechner (1999) refer red to it as condi tional independence. The Unconfoundedness assumption state s that (3 1) w her e denotes independence, i f treatment was received , and if no treatment was received. The expression above implies that with a given set of covariates but not affected by treatment, the potential outcome s , would be independent of treatment assignment . Further, it implies that all covariates that might affect the treatment and the outcome simultaneously must be observed to redu ce any biasness that could alter i nference. Another assumption is Overlap which states that: (3 2) The condition above implies that the participants and nonparticipants with the same values both have a positive probability of being treated (Heckman et al., 1997) . A ssumptions 1 and 2 are referred as strong ignorability by Rosenb aum and Rubin (1983) where ATE and ATT can be defined for all values of . When estimating ATT using propensity score, we must choose the model and variables to include in the model. Because the procedure involves participation and
98 nonparticipation , any discrete choice model can be used. For binary treatment where there are two choices , either the logit or probit model can be used; the use of either yield the same results ( Caliendo & Kopeinig, 2008) . However, for multiple treatment effects, the choice of the model is important . Multinomial logit has a strong assumption compared to multinomial probit; thus, the latter is ideal when we have more than two choices (G r eene, 2003). Since our study has a binary dependent variable, the logit model i s then the first step chosen to obtain the propensity scores. Matching algorithms After the propensity scores have been obtained , the second necessary step is to choose matching algorithms. There are various matching algorithm techniques used in the literature, but the most common ones are Nearest Neighbor Matching (NN), Caliper and radius, stratification and interval, and Kernel and Local linear ( Caliendo & Kopeinig, 2008) means an individual from the control group can be used more than once to match with individuals from the treated group. In contrast, means the individuals from the control group are paired only once. However invol ve a tradeoff between bias and efficiency (variance). If we allow with replacement the average quality of the matching will increase and the bias will between the control an d treated groups. The choice will also depend on the nature and availability of the observations for the treated group than for the control group, and to opposite situation. A nother common approach in situations where there are more observations in the control group than in the treated group is to use more than one
99 observation from the former to match with that of the latter. However, quality (suitability) of those to be matche d needs to be carefully checked as discussed in a later section. Of the various matching techniques, NN is the one most frequently used, often in combination with others. NN matches individuals with the closest propensity score from the control group to those in the treated group. Caliper and radius matching resolve the problem of NN when the closest neighbor is far away. A tolerance level of how close the neighbor should be imposed by using the caliper (distance or radius) technique . Caliper is one of the algorithms that impose s a common support condition, whereby observations that are out of radius are dropped (Dehejia & Wahba, 2002) . One advantage of the caliper technique is that it uses all the individuals with in the caliper range. When there are suitable matches within the range , extra individuals can be used ; otherwise fewer individuals are used. Thus, caliper shares the attractive feature that avoids the risk of unsuitable matches ( Caliendo & Kopeinig, 2008) . Another type of matching algorithm is stratification and internal matching which involves partitioning the common support of the propensity score into intervals (strata) and calculating the impact within each interval by taking the mean difference in outcomes between the treated and control observations. Th is algorithm is sometimes referred to as interval matching, blocking , and sub classification (Rosenbaum & Rubin, 1984) . The literature suggests the use of five strata (subsampl es) to remove more than 95% of the bias ; however , checking the balance score (discussed later) is mandatory to determine how many strata should be considered (Cochran, 1968 ; Imbens, 2004 ; Aakvik, 2001) . In contrast to the matching techniques discussed above, kernel and linear are nonparametric matching algorithms that u se the average of all the
100 observations to construct counterfactual outcomes. Because more observations are included , these non parametric techniques usua l ly result in lower variance ; however , it is possible to get unsuitable matches when all observations ar e used . Additional detailed information about matching algorithms can be found in the literature ( Caliendo & Kopeinig, 2008 ; Becker & Ichino, 2002 ; Dehejia & Wahba, 2002 ; Rosenbaum & Rubin, 1983 ; Heckman et al., 1997) . Once a decision has been made regarding the choice of matching algorithm s and their combination s , the next stage of the process is to check for overlap and region of common support between pa rticipants and nonparticipants. Overlap and common support Overlap and common support is the third step that ensure s only comparable observations are used in the matching algorithm before proceeding with the analysis (Dehejia & Wahba, 1999) . Several techniques to accomplish this can be found in the literature, including visual distribution of propensity score before and after matching, minima and maxima comparison, and trimming (Smith & Todd, 2005) . In general, all the aforementioned techniques for determining the overlap and common support region involve identifying and keeping those individuals who fall within the region of a suitable match and discarding those outside of the region. In other words, once the propensity been determined , individuals below the minimum or above the maximum of the control (untreated) group are discarded. A s imilar proc ess i s followed in the trimming technique as described by Smith and Todd (2005). Any individual with propensity score of exactly 0 and very small at some percent are dropped (see also Heckman et al., 1997, 1998). Since the process of finding the ar ea of common support only utilizes the propensity score , the
101 next step requires revisiting the covariates and assessing the quality of matching. The process of doing so is di scussed in the next subsection. Testing the matching quality The fourth step is th e assessment of the matching quality , which involves checking all the covariates to determine if the balancing property is achieved from relevant variables of both the control and treated group. Specifically , the intent is to determine w hether or not there any systematic differences between the groups remain after the matching is completed i.e. after conditioning on the propensity score s . The matching quality, therefore, checks if (3 3) w here are the covariates which are indepe ndent to treatment ( ) after conditioning to thei r probability of participation (i.e. after conditioning on: ) (3 4) If it is found that such differences still exist , for example , if there is still a dependency on covariates, then it can be concluded that either the model is misspecified or lack s good matches between the groups (Blundell et al. 2005) . In other words, there should be no more new sig nificant information about the treatment decision. In applying the test, various methods have been suggested , including standardized bias, t test, joint significance and pseudo , and the stratification test (see Lechner, 1999; Caliendo et al. , 2007 ; Sianesi, 2004) . This study uses the t test method for part icipants and nonparticipants for the reasons described below . The t t est is used to test the means of covariates before and after matching. It is expected that the difference will be significant before matching but not after matching
102 because the covariates should be balanced in both groups after matching. It is the preferred test because it gives statistical ly significant results , even though it does not tell to what extent bias es have been reduced after matching, just an indication of balanced covariates. After the matching quality has been checked and tested, the impact of participation is measured using the matched sample. The parameter value to be obtained is the ATT ( the average treatment effect on the treated ) . However, the PSM only reduces the observa ble bias , so the re is the need to conduct a sensitivity test for en dogeneity or unobservable bias. Sensitivity analysis The fifth (last) step for this analysis is to check the sensitivity of confounders on our results. The treatment effect estimation is ba sed on major two assumptions ; one of which is unconfoundedness , or the selection of observable s . As mentioned earlier, the unconfoundedness assumption is a strong assumption that can lead to bias estimates if indeed there are confounders, which affect both the participation and the outcome simultaneously (Rosenbaum, 2002). This is because the estimators from matching will not be robust to the hidden bias. Because it is not possible to estimate the magnitude of selection bias with nonexperimental data, the p roblem can be addressed by sensitivity analysis. The sensitivity analysis in PSM was first proposed by Rosenbaum and Rubin ( 1983) . Recent studies ( DiPrete & Gangl, 2004 ; Caliendo et al., 2007) have used sensitivity analysis in combination with PSM. A sensitivity analysis gives answers on whet her the inference about the outcome can be altered by unobservables or confounders . Stated slightly different ly , it tells how strong ly the unmeasured variables could alter the inference made from the analyzed model. In order to conduct such an analysis, we will need to redefine our model by including the unobserved coefficient.
103 That is, the participation probability , say , is determined by observables and unobservables such that (3 5) w here show s the extent to which could affect the participation decision of individual . The n if there is no hidden bias and is only determined by . If , this implies there is an unobserved effect on partic ipation and that two observations , say , and with the same (identical value) differ in their probability of receiving treatment. The sensitivity of the results from hidden bias can be checked by varying the value of . By varying the values of , the bounds for significance and the confidence interval can be generated (Rosenbaum , 2002; Aakvik 2001; Becker & Caliendo, 2007) parameter , , which determines the degree of departure from treatment or participation. Thus, two individuals , and wit h the same covariates differ in their odds of participating in the program by at most a factor , . In non observational studies (i.e. , experimental studies) , the randomization ensures that the value is always 1 to control for bias , but in observatio nal studies , if the gamma value is 2 means that two individuals with identical covariate value s differ in their odds by twice as much. In other words, one individual may be twice as likely as the other to participate because the two differ in unobservables (L. Keele, 2010) . In the odds criteria, values of are normally generated and tested in the model to see whether the findings will change . Odd ratio in sensitivity analysis is frequently used to answer the question of to what extent the unobserved factors could alte r the inference. In other words it answers how great the differences in would need to be, to
104 change our estimated resu lts. If is the probability of participation for individual , then the odds that individual participates in the program is . The odd ratio is bounded by gamma ( ) such that (3 6) The expression implies that there would be a hidden bias if two individuals with the same covariate values have a different probability of participating in the program. That is, we would have hidden bias if = but for individuals and . The basic process for a sensitivity analysis has two steps. F irst , is the selection of values for . In social science, often the values range from 1 to 2 and are adjusted for the results. Second , the values can either be used on p values or on the effect (outcome) to see how the values change as gamma increases. For binary outcomes, the sensitivity on the Wilcoxon sign rank tes t and the Hodges Lehmann point estimate for the signed rank test. This study uses the latter because the outcome is a continuous variable. Studies with applied PSM a pproach Wainaina et al. (2012) examined participation in contract s on broiler production in Kenya. T hey found that factors such as risk averse farmers and farm and nonfarm income have a significant influence on participation in contract s . On the other hand, factors such as male farmers, extension services, education level, and distance to the main road have a negative influence on participat ion in contract s . They also found that by using NN, kernel, and radius matching algorithms, farmers who participate in contract s may increase their income per bird by
105 27% more than nonparticipants. Their sensitivity analysis using rbounds revealed an absence of a close gamma value for unobservable s that could have nu llified their inference result. Saigenji and Zeller (2009) found that factors such as household s located closest to the s , and the number of household members in the C ommunist P arty significantly influence participation in contract farming. E ducation of household head seems not to be an influential factor reg arding contract participation. They further applied PSM on measuring the impact of contract participation on income and found a significance effect of 8000 VND in daily per capita. However, they did not report either the overlap condition or the sensitivit y of unobservables which could affect their findings. Using PSM, Martey et al. (2015) examined the impact of participation in the agricultural value chain project on efficiency and income in Ghana. They found that by using a probit model, participation in the value chain is significantly determined by years of farming experience , education al level, credit use, accessibility to and training received from extension services, and access to electricity. The balancing property was satisfied after using nearest neighbor, kernel, radius, and local linear matching procedures . They further found that with the existence of exogeneity and non randomization assignment of the project in a sample of 200 farmers, the project had a positive significant impact on TE by 28% more but was not significant on income effect. Using sensitivity analysis for all the aforementioned matching algorithms , they also found that their outcome results were insensitive to hidden bias es that could undermine
106 the impact of the project. However, their sensitivity analysis relied only on significance level, un like in this study which uses both significance level and outcomes . Alemu and Adesina, (2015) in usin g the bivariate probit model found that the factors of land size and livestock ownership were less likely influence rs for membership in cooperative s than for contract s. On the positive side, they found that dairy farmers located closest to rural developmen t office had a higher percentage of memberships in cooperatives and that, membership s in cooperative s increase d income from milk sales by 60%. Using kernel and nearest neighbor matching , cooperative members increased their income by 48% and their per capit a income by 49%. The sensitivity analysis revealed a reliable estimate even after the inclusion of a neutral confounder. They, therefore, concluded that cooperatives offer ed better opportunities to dai ry farmers than contracts in milk production. Method an d Data Method P articipation of farmers in the value chain is associated with linkages among themselves (horizontal coordination) and between postharvest actors (vertical coordination). Horizontal coordination implies participat ion in cooperative s that deal with agricultural commodities. Vertical coordination implies participation in contract s with buyers of agricultural commodities. One approach that determine s the extent to which farmers benefit from value chain participation , is through the implementation of an experimental approach (treatment and control framework) in order to identify a causal relationship between participation and an outcome or set of outcomes . However, it is well documented in the literature that the estimation of effect for non experim ental (social) studies rules out respondents (sample selection bias) with no treatment and,
107 therefore, leads to biases . As mentioned earlier, PSM is a popular method employed to address sample selection bias by matching the outcome s of treated (participant s) and untreated (nonparticipants) group s, assuming that the outcome is explained by identified observed variables . T reatment effect correction model was first proposed by Rosenbaum and Rubin (1983) to reduce observable bias when estimating the effect of t reatments. This study employ ed PSM to evaluate the impact of participation in contracts ( ) and associations ( In order to accomplish the analysis, five steps were followed as documented by Heckman et al. (1997) and Dehejia a nd Wahba (1999). First, the propensity scores were estimated using logistic probability regression. Second, the algorithms for matching were selected. Third, common support conditions were checked for each variable that influence d both vertical and horizon tal coordination participation. Fourth, the ATT was estimated, and fifth a sensitivity analysis was conducted to check for any confounder effect. The treatment groups for this study are the participants in contract s and associations , while the control gro ups are the nonparticipants in contracts and associations . The out come ( ) is the wheat net profit per kg. The impact of participation in contract s and association s on household wheat profit ( ) is estimated by taking the average difference for across both the treatment and control groups after controlling for differences in participation due to observable variables ( ) . Next, the five steps mentioned earlier are followed. F irst , the logit mo del i s used to estimate the probability ation , assuming that the error term is logistic distributed (Saigenji & Zeller , 2009; Ravallion , 2001; Baker , 2000). The logit model for and are specified as (3 7)
108 (3 8) where represent s contract s (dummy), and represents membership in an association (dummy). takes on a value of 1 if the farmer had a contract during the wheat sales, and zero otherwise . L ikewise , takes on a v alue of 1 if the farmer belong ed to a wheat association , and a value of zero if not. Following the probability estimation, the nearest neighbor (NN) and calip er algorithms with varied radii are used , respectively, to match the control and treatment groups based on propensity scores. The matched sample is used to determine the average treatment ef fect on treated (ATT) group for net profit . Explicitly, the treatment effect for individual is written as (3 9) where is the outcome of an individual with treatment , and is the outcome of the same individual without treatme nt. However, because is not ob served the counterfactual profit is used. In this case, the expected treatment effect of participation or average treatment effect on treated ( ) is the diff erence between the actual profit and the profit if they d id not participate in the contract ( ) or association ( ). (3 10) (3 11) where and are the observed net profit , and and are the counterfactual net profit s for contract participation and membership in a , respectively . Counterfactual Observe d Counterfactual Observe d
109 A proxy is needed for the counterfactual. A ready candidate for the proxy is to use an outcome observed from the untreated group (or a subset of the group). If researchers simply compare the average difference in the outcome of the treated vs the proxy as counterfactual, the estimated ATT is (3 12) (3 13) where and are the net profit proxies for contract and membership participation, respectively , as obtai ned from the matched control group. The difference between true ATT and estim ated ATT is the estimation bias due to some farmers being selected (or self selected) for the treated group and others for the untreated group such that proxies have to be used fo r counterfactual outcomes. This bias is referred to as (3 14) (3 15) where and are biases given by unobserved pre existing differences between the groups. Thus, the true parameter of ATT is only identi fied if the counterfactual net profit is similar to proxy net profit without considering unobserved biases. That is (3 16) (3 17) Proxy Proxy Counterfactual Proxy Proxy Counterfactual
110 After the ATT is obtained , we can now further check for unobserved effect. Let the probability of participation in value chain for individual be and for matched individual be . Assuming each individual is exactly matched by individual , their treatments odds are given by (3 18) (3 19) The odds ratio for the paired matched individuals is given by: (3 20) w he re is the treatment odd ratio in s the probability ratio of participants to the matched nonparticipants of the value chain. From (3 21) A ssume th e function has a logistic distribution. The n odds ratio equation becomes (3 22) Since the values of after matching, then (3 23) The individuals still differ in their odds of participation by a factor and their unobserved covariate . If there are no diffe rences in their unobserved variables or if unobserved variables have no influence on the probability of participating , the odds ratio is 1 , implying the absence of hidden or unobserved
111 selection bias. Following Aakvik (2001) and Ros enbaum (2002) , the bounds for the odds ratio in equation 3 23 above is given by (3 24) w here is the probability of individual to participate in value chain and the individual did participate, while is the probability of individual to participate in the value chain but did not participate despite the similarity in the covariate value with individual . Similarly, shows the difference in the odds of treatment and u nobservable covariates between two individuals of the same covariate values. The paired individuals have equal chance s of participating in value chain only if , otherwise , impl ying that individual who appear to be similar in terms cou ld differ in their odds of participation by as much as a factor of 2. This implies that for every pair, t he individual with the higher profit is twice as likely to have participate d in the value chain. Or the individual with the lesser profit is twice as l ikely to have not participate d in the value chain. Therefore is the measure of the degree of departure from the participation that is free of hidden bias. The package rbound in the r program is used such that is the log odds of differential assignment to treatment due to unobserved factors ( Keele, 2015) . The profit outcome is a continuous variable ; thus the sensi tivity test for p value is conducted using the Wilcoxon signed rank p va lue test. For the profit effect , the Hodges Lehmann point estimate test is used . The null hy pothesis is implying that there is no unobserved bias that would badly affect our inference ( ATT ) .
112 Data Data were collected through field survey in the northern part of Tanzania where 90% of total cultivated wheat is produced (FAO, 2013). Two regions , nam ely Arusha and Kilimanjaro , which are relatively homogenous in agricultural land use, production practices, and ecological condition were chosen . Two districts from Arusha (Karatu and Monduli) and one from Kilimanjaro (Hai) were selected based on thei r level of wheat production. The corresponding wards located in high lands were then selected since wheat is grown in highland areas. T he Mbulumbulu and Rhotia wards were selected in the Karatu District , the Mondulijuu ward was selected in the Monduli Dist rict , and the Ngarenairobi ward was selected in the Hai District to form more homogenous strata by location to represent the variability in wheat growing conditions by the wards . The with a list of farmers who grew wheat in the 2014/2015 season were obtained from village officials. The combination of random and snowball sampling techniques were used select the respondents . A pre tested questionnaire was administered by trained enumerators to obtain information r elated to production, costs, and marketing practices. Background information on solicited heads of households include d household size, age, gender, education, and occupation of the respondents, contracts, membership in an organization, and challenges that farmers face regarding production and marketing of wheat. In add ition , formal discussions with key informants such as government officials and traders were conducted when we first solicited their opinion s about the wheat crop. Th is information supplemented the data collected from the structured questionnaire for farmers . A total of 350 farmers were sampled despite several farm er s switch ing from wheat production to barley production or significantly reduc ing the ir wheat production.
113 Barley competes directly w ith wheat since both crops are grown under the same condition s using the same inputs. The differences include seeds and buyers. Barley has the advantage in that it sells for a slightly higher price than wheat and receives full support from private brewery companies. Such support includes the provision of inputs, assistance with harvesting and transport ing the crop . Despite the difficulty encountered, a total of 310 out of 35 0 farmers completed the questionnaires and participated in the analysis. Incomplete questionnaires were discarded . T he focus of this study is small scale farmers who are the majority of farmers in Tanzania, with land size ranging from 0.2 to 2 ha ( the equivalent of 0.5 to 5 acres ) . Results and D iscussions This section reports and discusses t he findings of our investigation. It commences with a presentation of the structure of the wheat industry in northern Tanzania within the framework of value chain analysis. This is followed by an examination of the main results from the propensity scoring matching method and associated analyses aimed at assessing the impact of participating in value chain on f Value C hain S tructure The wheat value chain in the study area consists of four main chains ; the wheat input chain , the wheat grain c hain, the wheat fl our chain, and the wheat product chain. The wheat input chain consists of private and public (research institutes) manufacturers, the wholesalers and the retailers of seeds, fertilizers ( common NPK, DAP, and urea) , and chemicals applied t o the wheat farm. The wheat grain chain consists of producers, brokers, wholesalers, and retailers. The wheat flour chain consists of processors (millers), wholesalers , and retailers. The product chain consist s of bakeries,
114 wholesalers, retailers , and cons umers. For the purpose of this study, we focus mainly are at the source of wheat production for which we claimed to be very low despite its market opportunity. Although landholdings in the study ar ea range from 0.5 acres to more than 50 acres , the bulk of farmers are small scale farmers with land averaging about 5 acres. Figure 3 1 depicts the wheat value chain. As can be seen in Figure 3 1, the farmers sell wheat grain to local retailers, brokers, and wholesalers. The wheat brokers in the study area are the major/dominant players in the wheat grain value chain because they ae involve d in organizing most of the transactions between the traders and the farmers. They participate in the harvesting and t ransport ing of crops to the urban The farmers work on their own until harvest time when brokers and traders visit at the farm gate to bid for price s before the harvest . Sometimes br okers and traders agree to harvest the crops themselves in exchange for 1 bag (100kg) of wheat to every acre harvested as the cost of harvesting. The brokers and traders harvest and transport the wheat to the processors for further value adding activities . Once the wheat is sold , the brokers and traders then deduct their costs and give the cash balance to the farmers. One disadvantage that the farmers face in selling their wheat to the brokers and traders at the farm gate is that they value one bag of wheat at 100 kgs, while in reality , th e bags can hold from 110 kg s to 130 kg s , depending on how much ex tra (overflow) the bags can extend. In the Swahili language, the overflow bags are called lumbesa (extended bags).
115 Value Chain Coordination Coordination along t he chain is achieved by means of contractual arrangements and groups/associations between and within actors . Two types of coordination are identified ; the first one is vertical coordination as revealed by farmers having verbal contracts with brokers and tr aders and the second one is horizontal coordination as verified by memberships in farmer s groups / associations. Vertical coordination Table 3 1 shows that only 16.5% of all the farmers surveyed had a contract with their traders while 83.5% did not. This im plies that the majority of small scale farmers are not coordinated vertically and sell their wheat grain to spot markets or at the farm gate . For the Mbulumbululu, Rhotia, Monduli juu , and Ngarenairobi wards , 17.3%, 15.5%, 15.8% , and 18.2% , respectively , o f those surveyed had contract s . The small percentage of farmers with contract s implies weak vertical coordination between farmers and other actors of the wheat grain value chain. Moreov er, in the majority of cases, growers who reported having contract s , ha d only verbal agreements based on prices and quantities of wheat t o be purchased for that season. Because most contracts are verbal, it is easy for either party to renege on their commitment , especially in situations where business is less than promising. In other words, oral contract s are not strong enough to make any one liable if the agreement is not honored . Since the relation ship between farmers and traders starts at harvest time, there are no set predetermined prices or forward contract s . Neither is th ere any type of arrangement that would compel buyer s to provide agricultural inputs nor farmers to supply wheat. In addition , verbal contracts do not force the wheat buyer to provide
116 technical or extension support which leaves farmers in a dilemma during h igh peak/good harvest s . This is in sharp contrast to the situation that exists for barley, whereby the companies purchasing the barley also finance the inputs and provide extension and transportation services. Barley contracts are a type of centralized agr eement where the buyers take care of almost everything while the farmer only takes care of the operations costs ( land preparation, planting, spraying, and fertilization ) . As a consequence, m any farmers have ex changed/ reduced wheat cultivation for barley. H orizontal coordination Table 3 2 shows that 39.4% of the total producers survey ed belonged to wheat associations while 60.6% did not. The fact that the vast majority were not part of /association indicates weak horizontal coordination. Table 3 2 also shows that 42.2%, 24.1%, 43.9%, and 40.9% of the producers surveyed in the wards of Mbulumbulu, Rhotia, Monduli juu , and Ngareneirobi , respectively , were part of a group or belonged to an association. However, the associations were not specific fo r wheat since they also deal with other crops. Most of these asso ciations are not formed officially; oftentimes it is just a group of farmers working together on various eco nomic and community activities. Mean C haracteristics of Participants and Nonpartici pants of V alue C hain Charac t eristi c s of v ertical coordination participants and nonparticipants The participants and nonparticipants of vertical coordination differ significantly in farm and farmer characteristics. Nonparticipants are more experienced in fa rming, have a larger number of household members, and a larger number of people aged between 18 years old and 50 years old in the household. Participants of vertical coordination , on th e other hand, are characterized b y larger number of farmers leasing the land (1%
117 significan ce level), higher frequency of extension visitations (1% significan ce level), attending more village meetings (5% significan ce level), higher off farm income (5% significan ce level), larger proportion of land used for wheat production (5% significan ce level), more output per acre (1% significan ce level), and lesser amount of seed planted per acre than the ir counterpart s (5% significan ce level). Further, t hey have a higher TE and AE compared to nonparticipants at the 1% significan ce leve l (Table 3 3). However , for the rest of the characteristics, the differences between participant and nonparticipant farmers were not significant. This implies that, although not perfect, relatively suitable matches would be available for vertical coordinat ion participants and nonparticipants to analyze the impact of vertical coordination welfare as measured by wheat profit per kgs . Characteristics of h orizontal coordination participants and nonparticipants Table 3 4 shows the chara cteristics of participants and n onparticipants of farmer groups/ association s. Many variables are insignificant which implies that the mean differences between participants and non participants in farmer groups / association s are not associated with farm and farmer characteristics. Table 3 4 demonstrates that participants differ positively and significantly from non participants in the level of education (10% significance level), frequency of meetings attend ance (5% significance level), output per acre (1% sig nifican ce level), TE (1% significan ce level), and AE (1% significan ce level). Conversely, nonparticipants differ positively and significantly from participants on better farm equipment owne rship and number of herbicide application s at the 10% and 5% signif icance level , respectively. Since most of the variables are either not significant or have a weak significance level, there is a good
118 match from nonparticipants to analyze the impact of horizontal coordination participation welfare as measured by wheat profit per kgs . Vertical coordination factors As mentioned earlier, the logit model was used to determine the factors that participation in contract s and to generate the fitted values (estimated coefficients) that were used to create the propensi ty scores. Table 3 5 shows that age of the household head , land leased, frequency of extension visits, frequency of meetings attend ance , off farm income, land size allocated fo r wheat , and TE are significant factors that determine farmers participation in vertical coordination. Unlike younger farmers, older farmers might have longtime partners who buy their wheat thus, more likely to participate in vertical coordination . Also, f armers who receive more extension visitations are more likely to participate in vertical coordination. A possible explanation could be the fact that the extension officer s provid e both technical and marketing information , including identifying traders will ing to negotiate contract s . Also, farmers who attend village meetings are more likely to participate in contract s . Village meetings allow farmers to gain access to various production and marketing information and meet traders interest ed in establishing cont ractual arrangements. Farmers with off farm activities have a greater opportunity of meeting wheat traders either at the off farm work place or in the market and agree to abide for wheat business . The larger the land size allocated for wheat production , th e higher the probability of participating in vertical coordination. T his is probably because traders are likely to procure more wheat from farmers with larger acreage, thus mak ing it easier to enter into business agreement s . Not surprisingly , farmers with higher TE have a greater chance of participating in
119 vertical coordination because their effort in production has a connect ion to ready markets and the availability of assured buyers. The marginal effect for age indicates that all other factors being held c onstant, aged farmers are more likely (by 0.3%) to participate in the contract than younger farmers. The marginal effect of the extension visits indicates that the probability that visited farmers participating in the contract is 1.1% more than non visited farmers. Attendance in several village meetings increases probability of participating in a contractual arrangement by 3.0% . The marginal effect for off farm income shows that farmers with higher off farm income have 6.3% more chances of particip ating in vertical coordination than the counterpart. Farm ers with larger land size are more likely to participate in vertical coordination with a probability of 0.8%. Further, farme rs with higher TE scores have 48 % more probability to participate in vertic al coordination than those with less TE scores. However, in general, the low probability participation in contracts suggests week vertical coordination in the value chain. It is possible that if there had been better institutional arrangements, more farmers would have participated in the contractual arrangements . Horizontal coordination factors Horizontal coordination is associated collective action s in agricultural activities. T able 3 6 shows the farm and f armer characteristics influencing participation in horizontal coordination. Few factors significantly influence ch as meetings attendance (5%) , farm equi pment (5%), and output (5%) . The factors of farmers located in Rhotia and the ownership of somewhat high tech fa rm equipment negatively influence participation in horizontal coordination. The Rhotia ward is close to the town center in
120 the Karatu D istrict where most of the agricultural dealer shops and traders are found . Un like in the other wards , farmers in Rhotia have direct access to the markets and most of the factors of production , thus reducing the need for groups/associations. Farmers with high tech farm equipment are somewhat capable of standing alone and are less li kely to participate in groups/associations . On the other hand , village meetings attendance and level of output positively and significantly influence participation in horizontal coordi nation by 5% each. Meetings attendance influence s participation in groups/associations because of the camaraderie and the sharing of information and expenses , which encour ages others to join/form groups. The marginal effect of meeting s attendance indicates a 7% increase in the probability of farmers who attend village meetings to participate in horizontal coordination. However, f armers liv ing in the Rhotia ward and farmers wit h high tech farm equipment are 20 % and 2 1.1 % , respectively, less likely to participate in horizontal coordination (groups/association). Covariat e B alancing Overlapping of propensity score s between participants and nonparticipants is one of two basic assumptions in PSM. PSM estimates that lie between 0 and 1 are used to determine the common support region and to check whether this assumption has be en met . Results for both vertical and horizontal coordination are provided in Tables 3 7 and 3 8, respectively , to show the covariate balances of observables. Visual proofs of histogram distribution (Figures 3 2 to 3 6 and 3 7 to 3 10) are also provided to show the balances of the matched treated and control sample s . As mentioned elsewhere, checking the overlap assumption before further analysis is necessary in order to ensure that reliable estimates are presented .
1 21 Vertical coordination balancing property T able 3 7 shows that the matching for the control and treated group s is properly overlapped for the selected variables . That is, there is no significant difference between the means of the control and treated group s after matching. As indicated in T able 3 7 , all of group s are insignificant even for farming experience, household composition, number of family members age s 18 years old to 50 years old , land leased, extension visits, fre quency of meetings attendance by farmers, off farm income, land size for wheat production , quantity of output, and TE which seemed to var y significantly before matching . Figure s 3 2 to 3 6 show the visual look of the distribution before and after matching for various caliper levels and nearest neighbor matching. distributions before and after matching for treated (contract) and control (noncontract) group s reveal that after matching, the shapes for the treated and control group s are similar and that there are no significant differences between the two groups, thus suggesting that we can further use the matched sample group to examine the effect of farmers participation in vertical coordination on wheat profit per kg. Horizontal coordination balancing property Table 3 8 shows that the matching for the control and the treated group s were properly overlapped . That is, there is no significant difference between the means of the control and treated group s after matching. This is revealed by the f the treated and control group s differ insignificantly even for variables such as education levels, frequency of meetings attendance by farmer s , farmer s locat ed in the Rhotia ward, farm equipment ownership, outp ut, TE, and AE , which varied significantly before matching. Figure s 3 7 to 3 10 show the visual
122 distribution before and after matching for the control and treated groups. The histograms for the nearest neighbor and caliper algorithms indicate that the after matching distribution s are similar between the control and treated group s . This implies that there are no significant differences between the two groups. Thus, the matched sample can be used for analyzing the effect of horizontal coordination participatio n on wheat profit per kg. Profit Vertical coordination effect The vertical coordination effect was measured using both the nearest neighbor and caliper radius matching algorithms. As seen in F igure 3 6 , the nearest neighbor visual distribution did not result in the best matches, so its profit effect of Tsh 230 /kg in T able 3 9 will still be bias ed . C aliper radius matching is a flexible form of an algorithm that checks matching at various radii . A caliper radius of 0.01 showed a profit effect of Tsh 136 /kg and was significant at the 1% level. Despite the visual diagram (Figure 3 2) showing similar distribution between the matched treated and control groups , few treated farmers were used for the analysis. We further checked for possible increase s in the sample s by increa sing the caliper radius. A caliper radius of 0.07 gave us the maximum number of treated farmers that matche d with the control group and the visual look after matching show ed a simi larity between the matched treated and matched control groups . A caliper radius of 0.07 showed a profit effect of Tsh 126 /kg and was significant at the 1% level for a sample of 43 treated farmers included in the analysis. This finding implie s that vertical coordination participants earn a higher net profit than nonparticipants by o ver 126 Tsh/kg. However , the unobserved confounder effect was not considered in the outcome because we assume d that matches were done for all
123 relevant farm and farmer characteristics . Thus, a sensitivity analysis need s to be conducted to see how the unobserved confounding factors might alter our inference. Horizontal coordination effect Our analysis also reveals a positive and significant impact of participation in the value chain through membership in agric ultural associati ons/ group s . The nearest neighbor matching as in vertical coordination did not show suitable matches for horizontal coordination (Figure 3 10). Caliper matching radii of 0.0 05 , 0.02, and 0.02 2 were used to c heck the visual balances of the matched sample . The caliper radius of 0.022 accom modated the maximum sample of 100 treated farmers out of 122 participants. The average net profit effect on treated (ATT) was found to be Tsh 46/ kg more for participants than for nonparticipants and is si g nificant at the 5 % level . We further check ed the robust ness of our result f r om unobserved exposure s (confounder s ). Sensitivity A nalysis The results in the previous section relied only on the assumption of unconfoundedness, or con ditional independ ence. That is to say, no systematic differences in the distribution of the covariates between the two groups caused by observable or unobservable (hidden bias) factors. Hidden bias can arise, for instance, by the non inclusion of variables that may affect both the valu e chain participation and the profit , simultaneously. When they exist , unobservable heterogeneity of our estimates arise s and our matching estimators would not be r obust (Becker and Caliendo 2007; Keele 2010; Rosenbaum 2002; R osenbaum & Rubin 1983). Thus, this study perform ed a sensitivity analysis to check the extent to which the inferences made from vertical and horizontal coordination participation could be altered by unobservables .
124 Vertical coordination sensitivity analysis Table 3 11 shows t he Rosenbaum bounds sensitiv ity analysis from signed rank test and Hodges Lehmann point estimate test. The Hodges Lehmann test gives the median range of ATT for every value of gamma while the Wilcoxon provid es their correspond ing ranges of significance levels for each ATT generated from gamma values . The parameter values generated from gamma for unobserved covariates explain how hidden biases of various magnitudes could alter the profit effect of value chain participation. The values of in Table 3 11 give the range of the profit effect along with the corresponding range of significance levels. When implies , and this means that the unobserved covariates have no effect on profit inference and that no hidden biases infl uence the results. This explains why we have single values at both bounds of Tsh131.92 /kg and why the result is significant at the 1% level. If the values increase up to 4, the participation effect would range from Tsh 38.7/kg to 222.6/kg and the result would still be significant at the 10% level; this is considered the upper bound significance threshold for this study. In other words, two farmers may differ in their odds of participation by a factor of 4 because they differ in terms of unobserved covariates. If the values increase beyond 4, the impact of participation would still be positive but not significant; this means that the inference made from participation would be sensitive to hidden bias. In other words, our significant result would become q uestiona ble due to unobserved covariates despite the similarities in the observed covariates values. Thus, our net profit effect from vertical coordination participation is insensitive to the hidden biases unless the value of is beyond 4.That is, our
125 signi ficant result would become questionable due to unobservables if despite the observed covariates being similar in values. Horizontal coordination sensitivity analysis The same sensitivity tests of Wilcoxon and Hodges Lehmann were conducted for ho rizontal coordination participation. Table 3 12 shows that when , the unobserved covariates are not relevant to participation yielding to the single value unconfoundedness profit estimate of Tsh 48.568/kg and its corresponding significance level o f 0.0039. The value of which indicate s the magnitude of unobserved parameter values shows that for a slight increase of 0.3 , from 1 to 1.3 , the profit effect ran ge could be as low as Tsh 25.768/kg or as high as Tsh 70.868/kg and the corresponding s ignificance level could be as low as 0 (1%) or as high as of 0.07 (10%) depending on the unobserved value of the covariates. The worst scenario is when the value of is 1.8 and beyond, where the participation effect is both insignificant and negativ e , impl ying a loss at the lower bound and a gain at the higher bound depending on the value of the unobserved covariates. That is, a small increase in the odds of horizontal coordination participation would make our null hypothesis of no effect from the un observed covariates rejected. I t would require a gamma value of 1.4 or more to alter the significance effect of horizontal coordination participation due to unobservable covariates. Concluding R emarks The objective of th is study was to investigate why dom estic wheat production in Tanzania has been slow to respond to the market opportunity that presents itself due to the widening demand and supply gap that exist s at the national level . For over two decades, this gap has being satisfied by rising levels of i mports. This essay postulate d
126 that the slow response could be due to the failure of wheat growers in Tanzania to formally participate in the value chain . Nonparticipation breaks the information flow about this market opportunity and restricts the potential contribution of this crop to the welfare of farmers. In explor ing that broad objective, th is essay specifically describe d the wheat grain flows from production point to ultimate consumption, analyze d the coordination of wheat actors with the main focus on farmers, explore d factors influences participation in the value chain, and examine d the value chain. The wheat value chain in the study area consists of four main chains : the wheat input chain , the wheat grain c hain, the wheat flour chain , and the wheat products chain. For the purpose of this study, we focused mainly on the because farmers are the primary source of wheat production in the chain . Farmers sell wheat grain to local retaile rs, brokers, and wholesalers at the farm gate. The wheat brokers in the study area are the major/dominant players in the wheat grain value chain because they ae involve d in organizing most of the transactions between traders and the farmers. Th is study fou nd that only a few farmers are vertically (only ~17%) and horizontally (only ~39%) coor dinated as indicated by participation in contract s and groups/associations, respectively. At the vertical coordination level, farmers with contract s had characteristics that were significantly different from those without contract s in terms of wheat land size, technical efficiency, allocative efficien cy , output/acre, frequency of extension visits, frequency of village meeting s attendance, and off farm income. At the horiz ontal coordination level, similar findings were obtained
127 in that the farmers who were members of groups/ associations differ significantly with nonmembers in terms of level of education, frequency of meetings attend ance , output per acre, technical efficienc y, and allocative efficiency. The propensity scoring technique was used to explore the causal relationship between participation and nonparticipation in the wheat value chain and the impact on the welfare of wheat farmers as reflected in wheat profits per kg . A logistic model was use d to explore the factors influencing farmers participation in the value chain and estimate th e propensity scores that were la ter used to match the covariates for participants and nonparticipants. The results indicate that parti cipation in vertical coordination was influenced by the age of the farmer, land leased, frequency of extension service visits , frequency of meetings attend ance by farmers, off farm income, land size allocated for wheat , and technical efficiency. On the oth er hand, participation in horizontal coordination was influenced by the frequency of meetings attend ance , ownership , and technical efficiency . F armers who are close to town center s and those who have better factors of prod uction and high technical efficiency scorers were more likely to work independently than joint groups/ associations. The fitted values from the logit model generated propensity scores that were used to match the participants and nonparticipants of the value chain. The overlapping and unconfoundedness assumptions were fulfilled by applying the nearest neighbor and caliper radius matching algorithms. The vertical coordination participation impact on ed that participants received Tsh 126/ kg of wheat more than did nonpartici pants and the difference was significant at the 1% level. On the other hand,
128 horizontal coordination participants received Tsh 56/ kg of w heat more than nonparticipants and that th e difference was significant at the 5% le vel . The sensitivity chain is generally insensitive to unobserved covariates . H owever , we cannot ignore the fact that horizontal coordination is somewhat more sensitive to hidden bias than is vertical coordination. As the findings suggest, policy makers and other beneficiaries need to consider upfront investment to wheat farmers in order to facilitate production through binding contracts. We have seen that even with wea k oral contracts, farmer s manage to secure more profit than those without contract s . In order to get most farmers to participate in contracts , more emphasis is needed on offering farmers unlimited extension services, better agricultural related meetings, g reater land size for wheat production, more off farm work opportunities, and higher level s of technical efficiency. More emphasis should also be placed on improving the efficiency of h orizontal coordination as it has a positive and significant impact on fa associations that specifically deal with wheat production.
129 Table 3 Ward noncontract contract Total Mbulumbulu 96 (82.7) 20 (17.3) 116 Rhotia 49 (84.5) 9 (15.5) 58 Monduli juu 96 (84.2) 18 (15.8) 114 Ngarenairobi 18 (81.8) 4 (18.2) 22 Total 259 (83.5) 51 (16.5) 310 ~ Values in bracket are percentages Table 3 Ward Nonmembers members Total Mbulumbulu 67 (57.8) 49 (42.2) 116 Rhotia 44 (75.9) 14 (24.1) 58 Monduli juu 64 (56.1) 50 (43.9) 114 Ngarenairobi 13 (59.1) 9 (40.9) 22 Total 188 (60.6) 122 (39.4) 310 ~ Values in brackets are percentages
130 Table 3 3. Characteristics for vertical coordination , participants and non participants Variable Nonparticipants (n=259) Participants (n=51) t value p value A ge 43.54 43.49 0.03 0.9792 E ducation 7.06 7.51 0.83 0.4083 E xperience 14.26 11.63 1.81 0.0739* H ousehold compo sition 6.69 5.69 2.92 0.0044*** A ge below18 3.20 2.82 1.3 5 0.1798 Age 18 to 50 3.06 2.53 2.74 0.0076*** Age 50 up 0.46 0.35 0.97 0.3376 Land leased 0.58 0.80 3.45 0.0009*** Ext ension visit 0.90 2.18 2.10 0.0080*** Meeting 0.38 0.86 2.10 0.0401** Transport ownership 0.07 0.12 1.08 0.0000*** Farm equipm ent 0.09 0.10 0.20 0.8402 Livestock 0.93 0.94 0.19 0.8531 Hybrid seed 0.10 0.10 0.13 0.8934 Off farm income 0.14 0.29 2.34 0.0224** Land size used 4.60 7.75 2.36 0.0220** Output per acre 686.84 795.91 270 0.0088*** Seed per acre 80.73 75.81 2.28 0.0250** Fertilizer Lts/acre 24.63 20.21 1.14 0.2593 Herbicides Lts/acre 0.69 0.73 0.57 0.5688 Insecticides Lts/acre 0.55 0.64 1.32 0.1921 Pesticides Lts/acre 0.37 0.42 0.75 0.4573 TE scores 0.78 0.85 4.77 0.0000*** AE scores 0.79 0.85 4.09 0.0 001*** *significant at 10% level, ** at 5% level, *** at 1% level
131 Table 3 4. Characteristics for horizontal coordination , participants and nonparticipants Non participants (n=188) Participants (n=122) t value p value A ge 42.80 44.66 1.31 0.1928 E duc ation 6.89 7.54 1.67 0.0955* E xperience 13.29 14.66 1.20 0.2303 H ousehold compo sition 6.51 6.55 0.14 0.8854 Age below 18 3.16 3.11 0.24 0.8130 Age18 to 50 2.97 2.98 0.09 0.9302 Age 50 up 0.39 0.52 1.51 0.1329 Land leased 0.62 0.62 0.10 0.9167 Ext ension visit 1.07 1.18 0.42 0.6777 Meeting 0.32 0.67 2.56 0.0110** Transport ownership 0.05 0.11 1.64 0.1021 Farm equipment 0.11 0.06 1.74 0.0833* Livestock 0.94 0.93 0.52 0.6031 Hybrid seed 0.10 0.11 0.53 0.5989 Off farm income 0.15 0.17 0.4 1 0.6801 Land size used 5.23 4.96 0.39 0.6974 Output/acre 666.21 764.25 3.33 0.0010*** Seed/acre 80.21 79.47 0.38 0.7068 Fertilizer/acre 22.45 26.15 1.16 0.2453 Herbicides/acre 0.73 0.64 2.06 0.0402** Insecticides/acre 0.60 0.53 1.53 0.1269 Pestic ides/acre 0.39 0.38 0.25 0.8004 TE scores 0.77 0.81 2.73 0.0067*** AE scores 0.77 0.84 5.72 0.0000*** *significant at 10% level, ** at 5% level
132 Table 3 5. Logit estimates of propensity score model for contracted farmers MLE Marginal Effect Contract (dummy) Estimate Std. Error Pr(>|z|) dC/dx Std. Err. P>|z| (Intercept) 7.976 *** 2.073 0.000 Age 0.040 * 0.022 0.069 0.003 * 0.002 0.066 Education 0.023 0.056 0.673 0.002 0.005 0.673 Experience 0.035 0.025 0.154 0.003 0.002 0.155 H ousehold co mp osition. 0.311 0.441 0.482 0.026 0.037 0.479 Age below 18 0.145 0.454 0.749 0.012 0.039 0.748 Age18 to 50 0.056 0.452 0.902 0.005 0.039 0.902 Age 50 up 0.211 0.523 0.686 0.018 0.045 0.686 Land leased 1.234 *** 0.452 0.006 0.096 *** 0.033 0.003 E xt ension visit 0.126 * 0.068 0.063 0.011 * 0.006 0.071 Meeting 0.355 ** 0.141 0.012 0.030 ** 0.012 0.012 Mbulumbulu 0.182 0.731 0.803 0.016 0.065 0.807 Rhotia 0.120 0.786 0.879 0.011 0.071 0.883 Monduli juu 0.377 0.735 0.608 0.034 0.069 0.625 Transport 0. 360 0.602 0.550 0.035 0.065 0.595 Equipment 0.554 0.624 0.375 0.039 0.037 0.282 Livestock 0.317 0.755 0.674 0.024 0.051 0.638 Hybrid seed 0.411 0.611 0.502 0.031 0.040 0.442 Off farm income 0.912 ** 0.458 0.047 0.100 0.063 0.111 Land size 0.094 *** 0.030 0.002 0.008 *** 0.003 0.003 Output 0.000 0.001 0.830 0.000 0.000 0.830 TE scores 5.619 *** 2.038 0.006 0.479 *** 0.165 0.004 *significant at 10% level, ** at 5% level, *** at 1%level
133 Table 3 6. Logit estimates of propensity score model for farmers membership MLE Marginal Effect Membership (dummy) Estimate Std. Error Pr(>|z|) dA/dx Std. Err. P>|z| (Intercept) 3.332*** 1.286 0.010 Age 0.007 0.014 0.601 0.002 0.003 0.601 Education 0.062 0.040 0.118 0.015 0.009 0.118 Experience 0.002 0.01 6 0.890 0.001 0.004 0.890 H ousehold compo sition 0.564 0.468 0.228 0.133 0.111 0.229 Age below 18 0.613 0.476 0.198 0.144 0.112 0.199 Age18 to 50 0.603 0.476 0.205 0.142 0.113 0.206 Age 50 up 0.791 0.505 0.117 0.187 0.119 0.118 Land leased 0.033 0.27 5 0.906 0.008 0.065 0.906 Ext ension visit 0.055 0.062 0.377 0.013 0.015 0.377 Meeting 0.294** 0.125 0.018 0.069** 0.030 0.019 Mbulumbulu 0.016 0.528 0.976 0.004 0.124 0.976 Rhotia 0.938 * 0.579 0. 056 0.200* 0.108 0.064 Monduli juu 0.260 0.530 0.6 24 0.062 0.126 0.626 Transport 0.671 0.487 0.168 0.165 0.121 0.173 Equipment 1.043** 0.496 0.035 0.211*** 0.081 0.009 Livestock 0.322 0.522 0.538 0.078 0.129 0.545 Hybrid seed 0.190 0.426 0.655 0.044 0.096 0.648 Off farm income 0.114 0.366 0.756 0.027 0.088 0.758 Land size 0.021 0.023 0.348 0.005 0.005 0.348 Output 0.002** 0.001 0.014 0.000** 0.000 0.014 TE scores 1.255 1.225 0.305 0.296 0.288 0.305 * S ignificant at 10% level, ** at 5% level, *** at 1%level
134 Table 3 7. Covariate balancing f or contract and noncontract farmers (caliper 0. 07 ) Sample Mean treatment (n=51) Mean control (n=259) std mean diff p valu e Age Before 43.49 43.54 0.401 0.979 After 43.44 43.35 0.702 0.974 Education Before 7.51 7.07 12.640 0.408 After 7.70 7.06 18. 515 0.382 Experience Before 11.63 14.26 28.256 0.074 * After 11.56 12.32 8.057 0.714 Household comp . Before 5.69 6.69 46.572 0.004 *** After 5.67 5.86 8.587 0.716 Age below 18 Before 2.82 3.20 20.867 0.180 After 2.81 2.93 6.313 0.762 Age18 to 50 Before 2.53 3.06 44.148 0.008 *** After 2.51 2.66 11.647 0.565 Age 50 up Before 0.35 0.46 14.927 0.338 After 0.37 0.28 12.304 0.531 Land leased Before 0.80 0.58 55.093 0.001 *** After 0.79 0.83 8.475 0.609 Ext ension visit Before 2.18 0.90 39 .882 0.008 *** After 1.65 1.97 13.308 0.618 Meeting Before 0.86 0.38 30.507 0.040 ** After 0.84 1.22 24.391 0.298 Mbulumbulu Before 0.39 0.37 4.360 0.776 After 0.44 0.45 2.314 0.916 Rhotia Before 0.18 0.19 3.303 0.830 After 0.16 0.18 4.151 0. 837 Monduli juu Before 0.35 0.37 3.671 0.811 After 0.35 0.32 5.626 0.777 Transport Before 0.12 0.07 15.984 0.284 After 0.09 0.13 11.869 0.585 Equipment Before 0.10 0.09 3.075 0.840 After 0.12 0.12 0.000 1.000 Livestock Before 0.94 0.93 2.867 0. 853 After 0.93 0.94 4.511 0.815 Hybrid seed Before 0.10 0.10 2.067 0.893 After 0.09 0.08 3.956 0.856 Off farm income Before 0.29 0.14 34.548 0.022 ** After 0.28 0.29 1.708 0.925 Land size Before 7.75 4.60 33.983 0.022 *** After 6.00 5.19 12.702 0.469
135 Table 3 7. Continued Sample Mean treatment (n=51) Mean control (n=259) std mean diff p valu e Output Before 795.91 686.84 40.779 0.009 *** After 744.38 798.30 23.257 0.272 TE scores Before 0.85 0.78 78.885 0.000 *** After 0.84 0.82 17.957 0.432
136 Table 3 8. Covariate balancing for members and n onmembers farmers (caliper 0.022 ) Sample Mean treatment Mean control Std mean difference p value Age Before 44.66 42.80 15.898 0.193 After 43.69 42.94 6.543 0.674 Education Before 7.54 6.88 1 8.864 0.096 * After 7.22 7.61 11.913 0.387 Experience Before 14.66 13.29 15.036 0.230 After 13.50 14.51 11.895 0.405 H ousehold comp . Before 6.55 6.51 1.630 0.885 After 6.40 6.37 1.321 0.917 Age below 18 Before 3.11 3.16 2.609 0.813 After 3.02 3.00 1.273 0.920 Age18 to 50 Before 2.98 2.97 1.013 0.930 After 2.94 3.08 9.204 0.527 Age 50 up Before 0.52 0.39 17.029 0.133 After 0.45 0.31 18.951 0.137 Land leased Before 0.62 0.62 1.219 0.917 After 0.64 0.68 7.946 0.560 Ext ension visit Befo re 1.18 1.07 4.739 0.678 After 1.07 0.86 9.588 0.441 Meeting Before 0.67 0.32 27.255 0.011 ** After 0.44 0.50 5.620 0.718 Mbulumbulu Before 0.40 0.36 9.194 0.425 After 0.42 0.44 3.360 0.820 Rhotia Before 0.11 0.23 37.273 0.005 *** After 0.14 0. 13 3.823 0.764 Monduli juu Before 0.41 0.34 14.056 0.221 After 0.38 0.34 8.883 0.537 Transport Before 0.11 0.05 17.225 0.102 After 0.08 0.06 7.335 0.594 Equipment Before 0.06 0.11 23.263 0.083 * After 0.07 0.07 1.950 0.892 Livestock Before 0.93 0 .94 5.814 0.603 After 0.92 0.91 4.890 0.749 Hybrid seed Before 0.11 0.10 5.940 0.599 After 0.10 0.12 6.633 0.670 Off farm income Before 0.17 0.15 4.716 0.680 After 0.17 0.12 14.569 0.285
137 Table 3 8. Continued Sample Mean treatment Mean con trol Std mean difference p value Land size Before 4.96 5.23 4.834 0.697 After 5.07 5.27 3.423 0.822 Output Before 764.24 666.21 36.845 0.001 ** * After 727.94 722.31 2.393 0.849 TE scores Before 0.81 0.77 34.501 0.007 ** * After 0.80 0.80 0.873 0.94 7 Table 3 9. Vertical coordination effect (ATT) Method Profit effect SE (AI) t value p value Treated observation Neighbor (1:1) 129.53 23.75 5.454 0.000 51 Caliper 0.01 136.42 21.338 6.394 0.000 30 Caliper 0.03 129.00 21.056 6.127 0.000 38 Caliper 0 . 05 127.64 21.106 6.047 0.000 40 Caliper 0. 07 126.00 21.954 5.739 0.000 43 Table 3 10. Horizontal coordination effect (ATT) Method Profit effect SE t value p value Treated observation Neighbor (1:1) 41.096 25.527 1.610 0.107 122 Caliper 0.0 05 39.268 11.29 3.478 0.001 40 Caliper 0.02 53.011 22.211 2.387 0.017 79 Caliper 0.02 2 46.023 21.282 2.163 0.031 100
138 Table 3 11. Vertical coordina tion Rosenbaum sensitivity test Hodges Lehmann point estimate (ATT) value) Gamma ( ) Lower bound Upper bound Lower bound Upper bound 1 131.92 131.92 0 0 1.1 124.92 138.52 0 0 1.2 115.82 145.72 0 0 1.3 111.22 151.82 0 0 1.4 105.42 156.62 0 0.0001 1.5 101.02 161.22 0 0.0001 1.6 96.417 166.82 0 0.0003 1.7 94.117 171.32 0 0.0005 1.8 90.017 173.72 0 0.0008 1.9 86.917 178.82 0 0.0013 2 83.717 182.02 0 0.002 2.1 80.517 184.62 0 0.0029 2.2 77.617 186.92 0 0.0041 2.3 75.617 189.42 0 0.0056 2.4 73.117 192.22 0 0.0074 2.5 70.517 194.42 0 0.0096 2.6 67.417 196.32 0 0.0123 2.7 6 5.417 198.02 0 0.0154 2.8 63.617 200.22 0 0.0189 2.9 61.517 203.32 0 0.023 3 58.917 205.72 0 0.0275 3.1 56.317 206.52 0 0.0325 3.2 53.517 207.92 0 0.038 3.3 51.117 209.12 0 0.044 3.4 48.017 211.32 0 0.0505 3.5 46.317 213.92 0 0.0575 3.6 44.017 216 .12 0 0.065 3.7 42.317 217.32 0 0.0729 3.8 40.717 218.32 0 0.0813 3.9 40.317 220.52 0 0.0901 4 38.717 222.62 0 0.0992 4.1 37.217 224.32 0 0.1088
139 Table 3 12 . Horizontal coordina tion Rosenbaum sensitivity test Gamma ( ) Lower bound Upper b ound Lower bound Upper bound 1 48.568 48.568 0.0039 0.0039 1.1 40.068 57.168 0.0009 0.0132 1.2 32.768 64.468 0.0002 0.0343 1.3 25.768 70.868 0 0.0724 1.4 20.268 76.968 0 0.1305 1.5 14.568 82.968 0 0.2076 1.6 9.2679 87.368 0 0.2995 1.7 3.8679 92.36 8 0 0.3992 1.8 0.03213 97.968 0 0.4998 1.9 4.8321 103.57 0 0.5951 2 9.3321 108.47 0 0.6807
140 Figure 3 1. Wheat grain value chain map
141 Figure 3 2. Distribution of propensity scores before and after matching the contract and noncontract farm ers by c aliper radius of 0.01 Figure 3 3. Distribution of propensity scores before and after matching the contract and noncontract farme rs by caliper radius of 0.03
142 Figure 3 4 . Distribution of propensity scores before and after matching the contract and noncon tract farme rs by caliper radius of 0.05 Figure 3 5 . Distribution of propensity scores before and after matching the contract and noncontract farme rs by caliper radius of 0.07
143 Figure 3 6. Distribution of propensity score before and after matching the co ntract and noncontract farmers by nearest neighbor algorithm 1:1 but not good match Figure 3 7. Distribution of propensity scores before and after matching the members and nonmembers by caliper radius of 0.0 05
144 Figure 3 8 . Distribution of propensity sc ores before and after matching the members and nonm embers by caliper radius of 0.02 Figure 3 9 . Distribution of propensity scores before and after matching the members and nonm embers by caliper radius of 0.022
145 Figure 3 10. Distribution of propensity s core before and after matching the members and nonmembers by nearest neighbor algorithm of 1:1 ratio but not a good match
146 CHAPTER 4 AN ECONOMIC ASSESSMENT OF THE CONTRIBUTION OF WHEAT INCOME TO HOUSEHOLD FOOD ACCESSIBILITY Overview This chapter exa mines the contribution that income obtained from the production of wheat improves food security among selected group of farmers. The underlying premise is that wheat growers can improve their income levels as a result of improvement in wheat productivity a nd value chain participation , leading to an improvement in their level of food security. The Issue Food security and poverty among smallholder farmers remain a significant challenge in a large part of the world. Improving levels of food security naturally requires a reduction in the level of poverty. Smallholders , in general, face many agricultural constraints in the production and marketing of their crops due to a combination of economic and non economic factors that affect their levels of food security an d poverty status. Economic factors such as low output prices constrain profit margins per unit and , in some cases , could result in a loss of any profits , thus ultimately reducing purchasing power and food accessibility/security. Likewise, non economic fact ors such as climate (drought and flood) can lower production level with more or less the same consequences as economic factors . Lobell et al. (2011 ); Lobell and Fiel d (2007) ; and Schlenker and Roberts (2009) make the observation that the increasing temperatures and declining precipitation associat ed with climate change adversely affect production worldwide , causing a widening of the yield gap for primary crops such as corn, wheat, and rice.
147 While some factors are beyond the control of farmers, limited access to factors of production such as improve d technologies, extension services , and market outlets are more tractable and require the necessary policy support. Brown and Funk (2008) point out that low level of food production can result in higher levels of poverty and food insecurity. They noted tha t for a household to be truly considered food secured, it must have access to adequate food at all times, and should be risk free from losing access to food as a consequence of sudden economic, climatic, or political shocks (i.e. , stability as defined late r). Therefore, analyzing various sources of household income and their relative contributions to household food accessibility is important for policy makers to formulate and implement an effective and sound food security policy that takes into consideratio n the most likely sources of income for agricultural households. The ability of a household to generate adequate income assumes a greater level of importance when consideration is given to the fact that many underdeveloped and developing countries suffer from hunger due to inadequate income to purchase their minimum food requirements rather than food unavailability . While food availability involves production, processing, marketing, and trading , food accessibility involves ll as physical access to the source. E conomic means is determined by such things as employment status , income level , and food prices, which are critical factors in buying adequate quantities of food for the household . The principle should be to set food pr ices to ensure economic incentive is mutually shared by both consumers and producers . Notwithstanding the fact that farmers are simultaneously consumers and producers of food , setting prices too low can result in low profit margins which often discourage f armers from reinvesting in improved
148 technologies or expand ing their farming operations to take advantage of market opportunities. For example, in Tanzania, the demand for wheat is 905,000MT, but the country only produces a paltry 83,000MT (less than 10% se lf sufficiency) despite the existence of suitable agronomical conditions in seve ral areas of the country (Promar Consulting, 2012) .A s we have demonstrated elsewhere, in participation in the value chain could lead to improvements in farm productivity and thus farm income. Higher farm income would increase the probability of farmers accessing adequate food for a healthy life style . It is within th is context that this study seeks to examine the contribution that wheat production is making to the welfare of a selected group of farmers, given the apparent market opportunity and the increasing attention being paid to this crop . Specifically, the research focuses on the extent to which income obtained from wheat production by a selected group of smallholders in Tanzania contributes to improved levels of food security, as reflected in an improvement in their purchasing power/food accessibility. To conduct this inqu iry, the following steps are taken: (1) examine the sources of household income of wheat growers, (2) analyze the portion of wheat income spent on staple foods, (3) examine whether the wheat income crosses the food poverty threshold (food access), and (4) identify some of the major factors preventing farmers from achieving a greater degree of food accessibility as measured by household wheat income , total crop income, and total household income . The rationale of this investigation is to provide information that will allow decision makers to implement measures that address some of the major bottlenecks impeding wheat production in
149 Tanzania . Such an outcome is likely to increase the income of wheat producers , thus increasing their purchasing power and ability to improve their household food security. Literature Review Concept of Food Security The literature reveals that the concept of food security has taken on several meanings over time . Reutlinger and Knapp (1980) defined food security as access to a minimum level of food consumption . Siamwalla and ValdÃ©s (1980) defined it as the ability to meet a national target level of food . Barraclough and Utting (1987) defined it as the adequate supply and distribution of food to meet nutritional needs. Kracht (1 981) defined it as enough and healthy lifetime food for growth.The World B ank (1986) defined food security as enough food to supply the energy needed for households to live healthy, active , and productive lifestyles . USAID (1992) defined the concept as the situation where all the people have physical and economic access to sufficient, safe, and nutritious food to meet dietary needs and food preference for an active and healthy life. The definition of food security (FAO,1996) adopted in 1996 by the World Foo d Summit specifie d f ood security exists when all people, at all times, have physical, social and economic access to sufficient, safe and nutritious food that meets their The World Fo od Summit definition , which is now widely accepted , embrace s four specific aims to ensure food availability, accessibility, utilization, and stability . Food availability mainly depends on domestic and international supply, storage, and trade and should be available at all levels ( i.e. , national, regional, district , and village /local levels ) (Riely et al., 1995; WFP, 2010). Food accessibility depends on income, prices , and employment to enable the household to purchase enough food for consumption ,
150 and on the distance to local food markets (WFP, 2010; Riely et al., 1999). Food consumption/utilization , which refers to the individual ability to process the energy and nutrients from foods consumed , it depend s on factors such as quality, sanitation, and calorie intake . Food stability combines continuous physical and economic accessibility for proper consumption of food (Maxwell & Frankenberger, 1992; WFP, 2010). The stability dimension also aims to monitor the robustness of the food security situation to cyclical , predictable variations connected with season al weather patterns. Food security can be measured in several ways depending on the level of aggregation in the economy (IFPRI, 1994). Food availability may not assure food accessibility due to poor disseminati on or inadequate income to purchase food ( FAO , 1983) . At the household food security level , food is acquired from production, food gifts , or income to fulfill the nutritional needs of all members of the household . For the net buyer s of food in urban are as, income is a large determinant of the degree of food security. URT (2003) offers a somewhat similar definition of food security at the household level whereby food security is defined as the ability of the household to acquire food either through produ ction, purchases, transfers , or exchange that are adequate in quantity and quality to fulfill the nutritional needs of all members of the household. In this context , the URT definition is concerned with intra household microeconomics , which describes the u se of food in the household and the influencing factors such as culture, beliefs, practices , and food preparation. Thus, for the household to achieve food security , it must have either the means to produce (land, production tools, and inputs) or the abilit y to purchase (job and
151 income) the food that the households need to ensure that the nutritional needs of the household members are met . This essay concentrate s on the ability to purchase foods with the income derived from wheat production , total c rop income, and total household income . The objective is to examine whether the income from wheat is enough to cover the based on a comparison to the national food poverty threshold of Tanzan ia . Poverty and Food A ccessibility The poverty measures require one to specify the poverty line that demarcates the poor and non poor individuals. The literature identifies three types of poverty lines : absolute, relative, and subjective (Hagenaars & van Praag, 1985 ; Zheng, 2001) . The absolute poverty line refers to the minimum amount of resources needed at a point in time ; it is updated when the price changes (adjust only for inflation). The relative poverty line specifies a point of distribution of income (income inequality) or expenditures ; it is update d over time when the standard of living changes. The subjective method derives the poverty line based on public opinion o f the minimum income or expenditure levels to survive . Compared with the first two approaches, the subjective method is rel atively less popular and is rarely used. In general, the literature identif ies two types of poverty threshold s: food and nonfood poverty line s . The food poverty line measures the level of expenditure necessary to achieve a defined daily caloric intake (Greer & Thorbecke, 1986 ; Pradhan et al., 2001) . C alorie intake alone does not define the food poverty line because intake can be achieved through infinit e basket bundles. Calories are just a proxy for an ov erall nutritional adequacy. The poor are those individuals whose expenditure (income) falls below a
152 defined poverty line of a region or a country. Food p overty is, therefore, consumption below the target level defined by the poverty line (threshold). The a mount of poverty in a country is determined relatively to a poverty threshold by the cultural and ethical condition s of that nation (Greer & Thorbecke, 1986) . Thus, the poverty line is the threshold at which households below that threshold a re considered poor, and households above that threshold are considered non poor. While v arious ways have been proposed to set poverty lines in the literature , their distinctions are based on relative or absolute poverty lines (Madden, 2000) . The re lative poverty line is set relative to distributions of income or consumption in a country or region of a reference and is used mainly in high income countries (Hagenaar s & van Praag, 1985) . As the name implies, the relative poverty line is not fixed in the domain of poverty comparisons; it is based on a referenced average level income or consumption in a country/region. It changes with the average earnings of the wage and salary earners; it may show a reduction in poverty even when people are starving . Among other things, the relative poverty line is best employed when the intent is to identify and target a section of the population who are relatively poor compared to o thers in the particular domain being considered . In this regard , a portion of society will always be considered poor . Absolute poverty lines are set as an absolute level below which consumption is considered to be too low to meet the minimum level of welf are and are typically used in low income and middle income countries. In these countries where some people may be unable to meet the minimum standards, an absolute poverty line is preferred to identify those who need intervention. For example . Greer and Thorbecke (1986)
153 establish ed the poverty threshold based on the nutritional standard where the minimum consumption level has to meet the expected minimum nutritional requirement (Food Energy I ntake Method). However , they did not allow for regional or intertemporal price variation s becau s e they did not include a basket of goods (Pr adhan et al., 2001) . Further, the absolute poverty line represents a fixed standard of living over the domain of poverty comparisons (Foster, 1989) . This implies a level of income that is just enough to maintain the basic minimum standard of living and is the approach used by many developing countries to ensure that nobody in the society is below that minimum (Ravallion et al., 1991) . It should be noted that the Untied States of America is one of the few developed countries that has a fixed absolute poverty based on inflation. An advantage of the absolute poverty line is that it allows for comparisons over the years. In this regard , it is useful for assess ing the impact of anti poverty programs over time or for judg ing the impact of a particular project on poverty. Regarding the relative and absolute poverty lines, thi s study focuses on the latter because it is more appropriate for low developing countries w h ere a large part of the income is used for accessing food, as is the case of the rural farmers in Tanzania. Several methods have been advanced for developing an abs olute poverty line. These methods include the food energy intake (FEI), the cost of basic needs (CBN), the consumption insufficiency method (CI) and the budget standard method (BS) (FAO, 2005). Each of these methods is briefly discussed below. The FEI met hodology is strictly food based and defines the minimum foo d intake cannot afford the cost of FEI are poor (Ravallion & Bidani, 1994) . Since the
154 methodology o nly takes into account expenditure s on food , it is more suited for developing countries where a substantial portion of the income is spent on food. FEI is the method most often used for drawing the base food poverty line . CBN includes both food and non foo d items to generate the poverty line . N on food items may include housing, utilities ( electricity and water ), and clothing. CI includes food, nonfood, essential, and non essential items to develop a poverty line. BS extends the list to include social life ( e.g. , recreational activities) to generate a poverty line . In all the methods above , the idea is to identify a basket of goods that would allow for an adequate standard of living and to convert such a basket of good s to a monetary value to establish a pove rty line. Since our concentration is only on the food poverty line, FEI is appropriate for this study. Two methods which a re frequently used to price the food basket , are the least cost method and the expenditure method (Chen & Ravallion, 2007) . The approach of the least cost method requires identifying one or more food baskets that give similar energy intake s . The cost of each basket is then calculated and the one that gives the minimum cost is chosen to set the value of the food poverty line. In contrast, t he expenditure method compares the minimum calori c intake to that of the actual average caloric intake by a select group of households. If the minimum is bel ow the actual, then the minimum caloric intake is reevaluated and priced, and the new value is chosen as the food poverty line. If the minimum is above the actual by household, then the cost of the minimum caloric intake is chosen as the food poverty line. Rather than calculating the food poverty line , this study adopt s the one developed by the World Bank (2015) as the threshold for food accessibility. The food
155 poverty line estimated by the World Bank is based on the cost of the basic need s method. Accordin g to the household budget survey (HBS) of 2011/2012 , the food poverty line was estimated to be T s h 26,085.5 based on 2,200 calories per adult per day. The World Bank further estimated the basic need s poverty line to be Tsh 36,482 per adult per month which add s the allowance for basic nonfood necessi ties to the food poverty line. Other challenges may arise when measuring food security is the food supply shortage at the household level (Shiferaw et al., 2003) . The presence of food shortage s at the household level , when there is plenty of food at the national level, implies that the problem is food accessibility. Sen (1981) is credited for bringing about the paradigm shift from the global and national level s to the household level for food accessibilit y. Over time, the agricultural household model was developed in which agricultural households are simultaneously producers and consumers . The agricultural household model provides a comprehensive framework integrating both demand and supply sides of the ec onomies of rural households. The theoretical and analytical framework for t his study , therefore, follows the agricultural household model to address the food accessibility of wheat producers in Tanzania. Theoretical F ramework The origin of the household mod el can be traced back from Becker (1965) , Chayanov (1966), and Nakajima (1969). They were among the first who belie ved that the behavior of farmers c ould best be understood using the household firm model. The basis of their argument was that the household firm framework allows for potential interactions between external labor markets (nonfarm labor markets), the farm ope ration, and household consumption . The treatment by Singh et al. (1986) provides a
156 detailed explanation o f was developed . In brief , agricultural households are constrained by two rational ities that need to be made at the same time . As producers, households need to make decisions on the allocation of labor and other inputs to produce crops. As consumers, households need to decide on the allocation of income generated from the farm (as profit) and from labor (as a wage) to the consumption of food and services. Underpi n ning the theory of the agricultural household model are issues of se parability (recursivity) and nonseparability (simultaneity) . T he challenge faced by the researcher is in deciding whether to model the production, consumption , and labor of the agricultural household separabl y (recursive) or jointly (nonseparable) . If we a ssume separabil it y / recursively, it implies that the agricultural household maximizes utility first by maximizing profit through production, and then through consumption of produced and marketed goods. The application of separabi li ty / recursivity assumes that farmers have complete knowledge about the prices of factors of production and outputs. Th is assumption implies that farmers can equate the marginal product of labor to wage as a condition to attain maximum profit (Singh et al., 1986). In this model , househ olds are price takers of all commodities , including the price they receive for their labor. Consumption of a product is determined independent ly of production of the same product ; hence the assumption of separate production and consumption decision s . Househo ld labor supply can also be determined independently of consumption as long as the difference between household labor supply and labor demand can be hired at the market wage. Consumption is constrained by income which , in turn , is determined by
157 profit. Acc ording to Singh et al. ( 1986) , since income and utility are positively related it makes sense first to maximize profit and then decide on con sumption and leisure. Straus (1986) and Sicular (1986) point ed out that the assumptions of the utility and production functions ensure that the second order conditions for maximization are met and , therefore , the two decision problems can be solved separab ily / recursively despite their simultaneity in time. Barnum and Squire (1979) show ed that household characteristics c ould be introduced into the model as linear functions. Straus (1986) reinforce d th is point , noting that since the household characteristics can be treated as fixed variables, introducing them as linear functions will not change the analysis. Based on the arguments of Straus , the recursive model can be used to show that farm technolo gy, quantities of fixed inputs , and prices of variable inputs and outputs affect consumption decisions . H owever , the reverse is not true as a result of the recursiveness of the model. The assumptions f or joint production and consumption decision s are diffe rent from the separable household model. The separable model is widely used by researcher s and is easier compared to nonseparab ility in econometric settings (Tayl or & Adelman, 2003) . In this context , Singh et al. (1986) suggest that the recursive assumptions should be applied unless there is strong evidence to assume otherwise. It is worth noting that if the recursivity is mistakenly assumed , the parameter to be estimated will be inconsistent, and the elasticities will be erroneous. However, separation fails when the utility is misspecified, the disutility is attached to working off farm, or the family and hired labor is considered to be imperfect substitutes ( Benjamin, 1992 ; Lopez, 1986 ). In addition to the above, separation cannot be assumed in a
158 situation where the physical strength of household labor is hampered such that dietary intake will affect production since the marginal product of labor will be directly related to consumption. The early applications of the basic agricultural model to developing countries include Lau et al. (1978) , Kuroda and Yotopoulos (1978), Adulavidhaya et al. (1979) , Barnum and Squire (1979) , Strauss (1984) , Benjamin (1992) , Jacoby (1993) , and Skoufias (1994) . The agricultural household m odel applies to all but agribusiness operated commercial farms, which consume only a small share of their own output and supply few of their own inputs (Taylor & Adelman, 2003) . Analytical F ramework Yotopoulos et al. (1976) , Lau et al. (1978) , Kuroda and Yotopoulos (1978), and Barnum and Squire (1979) are among the first studies to apply agricultur al household models with the econometrics approach. These studies assume d that farm household production and consumption decision are separable . They showed that the profit effect is an important part of the household production and consumption decision to maximizing utility . Based on the economic efficiency of farm household s in Gambia , Chavas et al. (2005) theoretically deco mposed the agricultural household model so as to capture the importance of off farm activities. B uild ing on the above studies , we apply the separable (recursive) agricultural household model to derive an equation for food accessibility to rural wheat growe rs of Tanzania. As stated earlier, the objective of this essay is to examine the extent to which wheat income in Tanzania to identify factors that hinder their attainment of food accessibility throug h their main income generating activity. In order to conduct such an analysis, the agricultural household model framework is applied following the stud ies by Chavas et al. (2005) and
159 Singh et al. (1986) . The consumption decision of the agricultural households is constrained by on farm incom e, nonfarm income , and labor allocation. Thus, the utility function representing the preference for all household members i s specified as (4 1) Subject to; w here is the amount of time allocated by households to own farm activities, represents the amou nt of time spent by the household on off farm activities to generate income , and is the time allocated for leisure activities and social obligations. Also, the households use the hired labor , input to produce a vector of farm output . The is the fraction of consumed by household. and are the prices of consumed goods purchased at market and produced goods evaluated at market prices, respectively. The solution to the first order conditions yields standard demand and supply that are functions of market price and income. Barnum and Squire (1979) , Straus (1986b) , and Sicular (1986) show that the first order derivatives can be arranged into a system of linear equati ons in matrix form and the solution for commodity demands, the marginal utility of full income, output supplies, variable input demands and associated multipliers can be solved. Thus, from the above utility function, the optimal values for which de r , and w) can be obtained. From equation 1 above , let be the quantity of own produced goods consumed by household members. Without losing generality and by
160 decomposing and normalizing equation 1 above, the marginal utilities derived from consuming own produced and purchased goods become (4 2) (4 3) (4 4) (4 5) (4 6) w here is the total household income earned from farm and off farm activities. However, t he on farm income is obtained by maximizing profit subject to inputs x and the total labor supplied to the farm (i.e. , family and hired labor). Thus in terms of the production function, the household profit maximization problem is to choose (4 7) w here the optimal supply values to be obtained from equation 7 are (4 8) (4 9) (4 10) Substituting back into equation ( 4 7) we obtain (4 11) Now the full income for the household ca n be rewritten as (4 12) From equation ( 4 12) above it can be depicted that, production decisions made by the household on utilizing inputs , and affect consumption decisions of goods
161 and through farm profits ( ). Thus, the agricultural households would have little income to purchase enough food if less of output is produced. L evel of output produced is influenced by farm specific features (e.g. Land size) and household characteristics (e.g. education level). The spillover effect goes into total household income and would lower the consumption level leading to food inaccessibili ty as measured in terms of income. In order to determine the contribution that the wheat crop enterprise makes to the household ability to access food, both the income associated with wheat activity as well as the overall household income need to be assess ed. The extent to which total household income surpasses the poverty line indicates the level of food security by the household . On the other hand, the income derived from wheat operation, especially in cases where wheat farming is the main income generati ng activity, will determine the extent to which the farmer /household can depend on this crop to cross over the poverty line. Empirical M odel The dependent variable for the model is food accessibility. First, agricultural household food accessibility was c lassified depending on whether the wheat income was sufficient to meet the basic food needs of household members, ( i.e., whether the income generated from wheat was below or above the food poverty line ) . Second, the food accessibility equation was determin ed based on income obtained from all farming activity plus income derived from other non farming activity, in order to assess the overall food security status of the household. The difference between the total income and poverty line value was then determi ned. The household is , therefore, food
162 insecure/ inaccessible if the difference is negative ; otherwise , the household is considered food secure/ accessible. The reason for considering only wheat income in the first case is that this crop has the largest share o f household income ( a major source of income ) among the selected group of growers . Thus, we need ed to establish the contribution that wheat has made toward household food security. Moreover, as pointed earlier, this crop is one that could contribute substa ntially to improving levels of food security given the existence of vast areas that are suitable for its production, the number of small scale growers who are currently engage d in its production, and most importantly the large wheat deficit and ever increa sing demand for wheat products due to population gro wth and changes in diet and food preferences. This approach would offer an explanation as to why farmers are not responding to what appears to be an obvious wheat market opportunity. Our working hypothesis is that given the importance of the crop among many smallholders in the study area, if the income generated from wheat is insufficient to satisfy the basic food needs, then it will not provide enough incentive for motivating farmers to produce more despit e the high demand, prompting the need for policy intervention to rectify the situation Given the demand function for food, the food accessibility was measured first by the difference between the wheat household income ( ) and income required to purchase minimum food requi rements for a healthy life style . (4 13) The second analysis was done using the total crop income as depicted in equation 4 1 4 below.
163 (4 1 4 ) Then the third analysis was conducted using the total household income to understand the overall food security status of wheat farmers as depicted in equation 4 15 below. (4 1 5 ) w here is the wheat net r evenue for household obtained by taking the product of selling price and quantity produced an d l es s the total cost ; is the food poverty line (minimum income needed for the household to acquire the minimum household food requirement) for household ; and deno tes food accessibility for household when and inaccessibility when . presents the tot income from household obtained by summing up the income from common crops (wheat, barley, beans, and maize) gr own in the stu dy area ; and is the total household income for household obtained by taking the summation of incomes from wheat, barley, maize, beans, livestock, and off farm activities . The food poverty line ( ) is the minimum food expenditure necessary to achieve minimum food requirements as measured by caloric intakes per person per day (Greer, & Thorbecke, 1986 ; Pradhan et al., 2001) . The minimum caloric intakes differ from country to country. In Tanzania , the minimum caloric intake is 2,200 kilocalor ies (Kcal) per adult pe r day at a cost of Ts h 26,085 per adult per month (World Bank, 2015) . Thus, a food accessibility model can further be defined with the dependent variable being a discrete variable as (4 1 6 ) w here is the food accessibility indicator that household would access minimum food requirement.
164 The analytical framework as discussed above implies that the income derived from agricultural production could be altered by farm specific fea tures and household characteristics. Equation (12) shows that production decisions made by the household when utilizing inputs x, F , and H affect consumption decisions of goods Z and g through farm profits ( ). Farm profit is a function of the quantity of output produced and its respective price. Since the quantity of output is influenced largely by farm features and household characteristics , it stands to reason that the net wheat income is also influenced by those features through the farm output and thu s the probability of food accessibility is altered. On the basis of the above, the probability of food a ccessibility can be modelled as (4 17 ) where Contract with buyers dummy tion dummy Age of the household head in years Education level of the household head in years of schooling Experience in wheat farming in years Household composition in num ber of family members share a pot and lives under one roof Family members with age below 18 years old in number Family members with age between 18 and 50 years old in number Family members with age above 50 years old in number Rental land acquired dummy Extensio n visits in number of times per year Village and technical meetings attended in number of times per year Mbulumbulu ward dummy Rhotia ward dummy Monduli juu ward dummy Transport ownership (car, motorcycle, and o xcart) dummy Farm equi pment ownership (tractor, plow , knapsprayer, wheelbarrow) dummy Livestock keeping dummy Hybrid seed planted dummy Off farm income dummy Land size allocated for wheat production in acres
165 TE AE Parameters to be estimated Disturbance term Various studies have applied somewhat similar to above equation to address the food security issues . For example, Omonona and Agoi (2007) used food security index as a dependent variable by taking the ratio of per capita food index to two third s of the mean per capita food index to identify factors influencing food security in Nigeria. They classified the households into food secure and food insecure if the ratio was above 1 and less than 1 , respectively. Babatun de et al. (2007) applied the logit model to determine factors that influence food security in urban Nigeria. The dependent variable was the ratio of daily actual calories intake to the recommended calori c intake per day. Departing from above studies , this study examine s the contribution that wheat income has on household food accessibility for rural wheat growers . Further, determine s the share of wheat income on total household food expenditure. Data Data were collected through a field survey conducted by trained enumerators using a pre tested questionnaire designed to study wheat farmers in n orthern Tanzania where 90% of the domestic wheat supply is grown (FAO, 2013). Two regions , Arusha and Kilimanjaro , were chosen because they are relatively homogenous i n agricultural land use, production practices, and ecological condition. Two districts from Arusha (Karatu and Monduli) and one from Kilimanjaro (Hai) were selected based on their level of wheat production. The corresponding wards were then selected becaus e wheat is grown in highland areas. T he Mbulumbulu and Rhotia wards were selected in the Karatu District , the Monduli juu ward was selected in the Monduli District , and the
166 Ngarenairobi ward was selected in the Hai District to form a more homogenous strata by location to represent the variability in wheat growing conditions by the wards . The farmers from villages in each of the major homogeneous ward were selected from the list provided by village officials. A combination of random and snowball sampling tec hniques were used to select farmers from the sampling frame provided by village officials. The sampling frame consisted of a list of farmers who grew wheat in the 2014/2015 season. A structured questionnaire was used to obtain information on the type of fo od consumed within the last seven days (week) from the reference period of the interview, the quantity of food consumed, the amount purchased and their respective prices, the amount of own production consumed, and the amount of food received as gifts. Back ground information solicited included household size, age, gender, education, and occupation of the respondents, contracts, and membership in an organization /association . A total of 350 farmers were sampled despite several farm er s switch ing from wheat to barley prod uction or significantly reduced their wheat production . Barley competes directly with wheat since both crops are grown under the same condition using the same inputs except seed and has different buyers. Barley has a slight advantage in that it sells for a slightly higher price than wheat on a unit basis and receives full support from private brewery companies. Such support includes the provision of inputs, assistance with harvesting and transportation of the crop to their companies . Despite the difficult y encountered, a total of 310 out of 350 farmers completed the questionnaires and participated in the analysis. T he focus of this study
167 was small scale farmers who are the majority of the farmers in Tanzania, with land size ranging fro m 0.2 to 2 ha ( the equivalent of 0.5 to 5 acres ) . Results and Discussion Land A llocation Wheat farmers in the study area also grow other crops , including barley, maize, beans, and potatoes. Barley is grown mainly as a cash crop for brewery companies. Maize and beans are commonly grown for food consumption , and the surplus is sold at mar ket. Wheat and barley compete for land resources as they are grown under similar farm ing conditions. Table 4 1 shows the land allocation for the pooled (N=310) and restricted (N= 298) sampled farmers. The restricted sample considers only those farmers who have allocated a larger share of their total farm land to wheat production than to other crops. Using the restricted model helps to analyze the food access status of farmers who d epend on wheat as the main source of income. T he average land allocated to wheat and barley is ~ 5 acres separately (wheat at 5.1 acres and barley at 4.8) for the pooled sampled farmers . Maize and beans are grown practicing intercropping system s . Since mai ze and beans are intercropped , their average was taken as a unit land (i.e., the average of 3.6 and 3.2 is 3.4) . Thus , the total farmland for the unrestricted sample is 5.1+4.8+3.4 = 13.3. Overall , barley and wheat accounted for the bulk of l and use with s hares of 38% and 36 %, respectively. However , it appears that barley growth has been at the expense of wheat production because most wheat farmers reported switching out their wheat land for barley because of the slightly higher barley prices and the fact t hat barley buyers (brewery companies) offer farmers extension services and inputs on credits that is paid after the harvest.
168 Table 4 1 further show s that with respect to the restricted sample, for a representative farm of 11.0 acres the crops distribution would be as follows: wheat ( 5.2 acres ), barley (1.6 acres), maize (2.2 acres), and beans (2.0 acres) on average . Thus for the restricted sample used in our analysis , the share of wheat land is the highest ( 47 %) followed by maize and beans ( 20 %), and barl ey ( 15 ). Figure s 4 1 and 4 2 further show the land and output variations between crops for pooled and restricted sample respectively. Price V ariation by C rops Price analysis for various crops was conducted . Table 4 2 shows price variability of wheat, barle y, maize and beans, the most common crops that are grown in the study area. The result s indicate that b eans ha ve the highest farm gate price followed by barley , wheat, and lastly maize (Table 4 2; Figure 4 3) . If higher price implies the highest net price , then farmers would allocate more land to the crop with highest price incentives, ceteris p a ribus. Howev er, given that most farmers tend to be risk averse , i t is understandably that they would adopt a crop diversified strategy to mitigate the impact of any si ngle crop failure . As shown in our previous analysis , wheat is the main crop grown in the area but other food and cash grains are grown to forestall the possibility that wheat might fail , and other grains might be available for selling probably at a high er pr ice. The possible failure of the main crop (wheat) and possibility of higher prices from affect the income and food accessibility status of the household . In general we observed that w heat production has not b e en increasing significantly , possibly due to the fact that price receive d by the farmers is relative ly low compared to barley that is grown on the same type of land. This situation provides a disincentive to potential and actual farmers
169 causing them to diversify/change the land use patterns to other crops that provide relative higher prices . The relatively low pr ices received for wheat makes it difficult for the income generated from wheat production to purchase the minimum food requirem ent of the households leading to food inaccessibility . This analysis is linked to food accessibility in order to explain how low the wheat income is by comparing to the purchase of the minimum food requirements that is taken as food poverty threshold. Shar e of Wheat I ncome on A ctual Food E xpenditure The consumption of fo od and beverages information was analyzed. The common types (sources) of food eaten by the sampled farmers were grouped into cereals and cereal products, roots and tubers, sugar and sweets, pulses, nuts and seeds, vegetables, fruits, meat and meat products, milk and milk products, oil s and fats, and spices. Beverages such as tea, coffee, soda, juice, beer, and local brews were also considered . C ereals that are consumed the most in the study a rea include rice, maize and maize flour, millet and sorghum, and wheat and wheat products. Maize and wheat products are generally sourced from own farm production while others are purchased from the markets. Roots and tubers , including bananas, fresh cassa va, potatoes (sweet and Irish), and bananas/plantains are consumed less frequently. Most of these food crops are bought in the market , but few like bananas and cassava are grown for home consumption and sometimes the surplus is marketed . Sugar and sweets s uch as honey, jams, and jellies are consumed but are purchased from markets. Pulses such as peas and beans are grown , with the vast majority being sold at market . Nuts and seed s consumed include peanuts and coconuts , with all purchased from market s . Vegeta bles such as spinach, carrots, tomatoes, onions, and carrots are consumed by the households and are mainly purchased from markets . Fruits , including ripe bananas,
170 oranges, lemon, and mangoes , are part of the household diet, with the majority being purchase d from markets. Meats from goats, sheep, cows , and chicken are consumed by household s and mainly come from their own production , except for beef which is mainly purchased from butchers. Milk, cheese, yogurt from cow s, and eggs from chicken are part of the household dietary. Cooking oil, spices, and salt are purchased from market s for food preparation. Beverages such as soda, juice, beers , and local brew s are occasionally consumed and are purchased from markets . The share of wheat income spent on food consum ption was calculated based on the most frequently consumed food (staples) in the households. These food include rice (kgs), maize flour (kgs), bread (pieces), bans (pieces), sugar (kgs), beans (kgs), cooking oil ( lts ), tomatoes (kgs),onion (kgs), vegetable s (pieces), banana (pieces), orange (pieces), goat meat (kgs), beef (kgs), mutton (kgs), chicken (pieces), milk ( lts ), yogurt ( lts ), salt (kgs), and tea away f rom home (cups). Table 4 3 shows that , of the total household food expenditure, wheat income cove rs only 38.8 % , which implies that the net income derived from wheat is not enough to cover household food expenditures. Thus, farmers tend to diversify their labor to other sources by growing other crops at the expense of wheat. These other sources of inc o me include off farm employment and sale of other crops , including barl ey, maize, and beans. Figure 4 4 further depicts the food expenditure pattern per week. Household F ood P overty L ine The food poverty line was established base d on two categories within t he household : adults and children. For adult s , t he poverty line was obtained by multiplying the number of adults (ages 18 years old and older) in the household by the value of the minimum caloric intake (2,200 calories) per day per adult (evaluated at Tsh 26,085.0
171 per month). While for the children (ages younger than 18 years old), half the value of the minimum caloric intake for adults (1,100 calories) was multiplied by the number of children in the household (evaluated at Tsh 13,042.5 per month) . The valu e for the food poverty line is higher for household s with many family members. Table 4 4 shows that only 29% of the pooled sam ple of wheat households can cover the minimum food expenditure of their household from the income derived from wheat p roduction. T he vast majority (71%) are unable to do so from this source. Several explanations for this situation includ e the very low price s received by farmers; high production costs; and low quantity of wheat sold at prevailing market prices. Table 4 4 and Figure 4 5 sh ow a breakdown by wards of the per c entage of wheat farmers sampled who were able to satisfy their minimum food requirement s from revenue generated by wheat production. Of the four wards considered, Monduli juu had the highest percentage (75 %) of wheat producers who were unable to meet their household minimum food expenditure, followed by Rhotia (72%) , Mbulumbulu (69 %), and Ngarenairobi (64 %) for the pooled sampled farmers . For the restricted sample, there is a slight positive change of food accessibil ity and inaccessibility by individuals within wards , but the overall percentage in the study area was the same as that obtained for the pooled sampled farmers , that is, 71% by 29% (Table 4 4; Figure 4 6 ). Mean Characteristics of Food A ccessible and Inacces sible H ousehold B ased on Wheat Income Further analysis of the data gathered reveals several important insights with regards to farmer s with and without contracts and their ability to access food from income generated from wheat production. First , the data r eveal that among farmers with
172 contracts, there is a significant (at the 1% level) mean difference between those characterized as food accessible (wheat income sufficient to cover minimum requirement) and food inaccessible (wheat income insufficient to cov er minimum food requirement) . In contrast , this difference is not observed in the case of farmers without contract s (Table 4 5 ). Second , the analysis shows that, in general, food inaccessib il i ty is higher for older individuals (significant at the 10% level ), indicating the negative influence on food accessibil i ty as age increases. As demonstrated in a previous chapter , younger farmers tend to be more efficient than older farmers and therefore the net income obtained from their wheat outputs enabl es them to better satisfy the minimum food requirement s . Although not unexpected, it was observed from the data that households with higher food inaccessib il i ty have more family members, more family members below 18 years old, more family members between 18 years old and 50 years old, and more older family members i.e. older than 50 years old (all at the 1% significan ce level). This is to be expected since invariably the size of the farm does not increase proportionally with the size of the households , and the higher nu mber of household members implies a higher value for the food poverty line , thus making it difficult for wheat income to cross above the household food poverty threshold. An implication of this finding is that household s with a higher number of family memb ers will have to expand wheat production significantly through increase d productivity and farm expansion if income from wheat is to be a major driver toward crossing the poverty line. On the other hand, households with food accessibility are characterized as having more contracts (1% significan ce level), larger land holdings (1%), attending more village meetings (10%), obtaining higher quantity of output per acre ( 1% ) , using more
173 seeds (5%) and, applying higher quantit ies of fertilizer (1%) and insecticides (1%) per acre. In addition , their technical and allocative efficiency scores are higher compared to households with food inaccessibility at the 1% significan ce level in both cases. Factors Influencing F ood A ccessibilit y through Wheat Income Farming experie nce, meetings attendance, farm equipment ownership, the land size used for wheat production, and allocative efficiency were the main variables found to have a significantly positive impact on the likelihood that income generated from wheat production would be sufficient to cover min imum food requirement (Table 4 6 ). However, the results suggest that farmers who had leased land are less likely (at the 5% significant level) of acquiring food accessibility from the income generated from wheat production. This find ing could be due to the fact that the leased land is not necessarily being used for wheat production but for other crops such as b arley, maize, and beans which are also commonly grown in the study area. Somewhat s urprising was the result tha t food acce ssibility decreased with farm equipment ownership . Farmers experience and attendance at agricultural and village meetings increase the probability that the revenue generated from wheat production would be sufficient to meet the minimum food expenditure ne eded. This result seems plausible as farmers could be learning various farming techniques by attending meetings that enable them to produce more output to generate more income to cover the food accessibility gap. As was expected, the larger the land size a llocated for wheat production , the higher (significant at 1% level) the probability of accessing food by wheat income. This fin ding reveals that grater wheat production generates more income to ensure food accessibility at the household without having to d epend on other sources. Allocative efficiency (AE) increases the probability of food accessibility using wheat income. That is the best use
174 of the least cost combination of input s increases the profit gain , thus lessen ing the gap between the generated whea t income and the food poverty line. The results also indicate that several other variables have a positive influence on food access ibility indicator , although not statistically significant. These variables include contracted farmers (vertical coordination ), the level farm extension visits, farmers located in the Mbulumbulu area, means of transport ownership, livestock keeping, and off farm income. Although such variables were not significant at 10% level , w e should not ignore their potential contributions towards improving positively the level of food accessibility . Food A ccessibil i ty b ased on T otal Crop I ncome ( Restricted Sample A nalysis ) In order to gain further insight of the contribution that income from wheat production makes t o food accessibility status, additional analysis was carried out on the restricted sampled household s . In particular , the analysis focussed on the food accessibility of farmers for whom wheat production is the core farm activity. The total crop income cons idered was generated from ma r ginal income per unit of land by assuming that the wheat income per acre is optimally equal across other crops per acre . Thus , the wheat income per unit of land was mu l t iplie d by land sizes allocated to barley, maize, and beans to obtain subtotal crop income s from each crop and later added to obtain total crop income . The f ood poverty line generated in the previous section was the n compared to this amount of farm income . The finding sho ws that 56% of wheat farmers have food inacc essibility, implying that their annual income is below the national food poverty line , and therefore they cannot afford the ir minimum food expenditure s based on the current cost of goods (Table 4 7 and Figure 4 7 ). We further analyze d the likely det erminants of household food accessibility employing a binary
175 model , whereby the dependent variable takes the value of 1 if the income is above the food poverty line, and 0 otherwise. Table 4 8 shows that education level, farming experience, frequency of ex tension visit s , farmers located in Mbulumbulu and Monduli juu , farm land size, and technical efficiency are the main factors influe n cing the food accessibil i ty of wheat households. Food A ccessibil i ty b ased on Total Household I ncome ( Unrestricted Sample A na lysis) As pointed earlier , a survey of selected farmers was conducted to assess the general food security status of the farmers . The total household income was obtained by summing up income derived from crops, livestock, and off farm activities for each me mber of the household . The total income was then compared with food poverty line of the households as shown in previous sections. The households with income below the food poverty line are generally considered poor because cannot afford to purchase (have a ccess to) The result s reveal that majority of wheat farmers (78%) have food accessib ility while only 22% have food inaccessib ility through total household income. Again this implies that 78 percent of them had suffi cient income from all sources to enable them to satisfy the minimum food requirement as reflected in the poverty line. A comparison by Wards reveal that Monduli juu wheat farmers (87%) are the most food accessible households; followed by Mbulumbulu (76%), Ngarenairobi (68%), and Rhotia (67%) wards (Table 4 9 ; Figure 4 8 ). Further analysis using probit model shows factors most influential in enabling farmers to achieved level of food security/food accessibility included: nce, and being in Mbulumbulu and Monduli juu wards . These factors have positive influence and are significant at 10%,
176 5%, 1%, 10%, and 1% level, respectively (Table 4 10) . However , as shown in the previous analysis , the probability of food accessibility decrea ses with the age and is significant at the 5% level. One possible explanation could be the fact that as farmers get older , they become inactive thus reducing their ability to conduct strenuous activities which ultimately adversely affects earnings and abili ty to fulfill their economic and social responsibilities . Conclu ding R emarks The aim of th is study was to examine the extent to which wheat income . Specifically, the research focused on the extent to which i ncome obtained from wheat production by a selected group of smallholders in Tanzania contributed to improved levels of food security, as reflected in an improvement in their purchasing power/food accessibility . In order to conduct such an analysis, the agr icultural household model framework was applied. Use was made of a probit model to analyze the factors influencing the con tribution of wheat production on food accessibility/food security. Among other things, our investigation reveal s that for the study ar ea, 31% of the total l and is allocated to wheat, followed by 28 % to barley, 22 % to maize, and 20 % of beans. Wheat and barley compete against each other for factors of production , including land. The decision to plant either one depends on prices and servic es offer e d by the buyers. Breweries companies were found to offer slight ly higher price incentives to barley farmers and provide them with inputs. This situation lowers wheat production as many of the pr e vious wheat farmers were switching to barley product ion. The lesser the land allocated for wheat production , the lesser the domestic wheat production, and the lower income share on food expenditure. The share of wheat income on household
177 food expenditure was found to be 38.8% , implying that fa r mers have oth er sources of income that contribute to household food expenses . Since this analysis provides the actual spending of food items that arises with the level of income for the poor ( Engel' s law), we used the national food income threshold to assess whether th e wheat income crosses the threshold. This study found that 71 % of wheat farmers c ould not cross the national food poverty line based solely on the income generated from wheat. It could be that the price received by farmers is too low which discourages the e xpansion of wheat land to produce more wheat for more income generation. The remaining 2 9 % who have food accessib il i ty differ positively and significantly from the 71 % who have food inaccessib il i ty in terms of their characteristics such as having contracts, frequency of meetings attendance , farm size , production level, amount of seed used, quantity of fertilizer used, insecticides applied, and TE and AE scores. Using the probit model analysis , i t was found that farming experience, meeting s at tendance, land si ze, and allocative efficiency positively and significantly influence the probability of food accessibility. Farmers are therefore advised to increase their level of efficiency by using the least cost combination of inputs to close the gap between the food poverty threshold wheat, government , and other stakeholders might consider subs i dizing inputs for wheat production as in the case of barley by private brewery companies in the study area. S ubs i dies for wheat would likely encourage farmers to allocate more of their land to wheat production , thus increasing their output level and generating more income at
178 prevailing prices. Farmers are further advised to increase the ir land s ize for wheat production as the realized output from the increased land acreage would increase the volume of wheat sold and thus higher income would be generated . Farmers should also attend agricultu ral and village meetings because participation in meeting s would improve their production skills, leading to higher levels of wheat output, increased revenue, and a greater chance that income from the sale of wheat would be sufficient to cross the poverty line.
179 Table 4 1. Crop land allocation Total sample (N= 310) Restricted sample (N=298) Crop Land (acre) Output (kgs/acre) in "000" Land share (%) Land (acre) Output(kgs/acre) Land share (%) Wheat 5.121 0.707 31 5.171 0.704 47 Barley 4.75 2 0.516 28 1.619 0.524 15 Maize 3.614 0.616 2 2 2.215 0.608 20 Beans 3.272 0.265 20 2.027 0.257 18 Table 4 2. Crop price variation Minimum Maximum Average Std. Dev . Barley 700 1200 852 61.46 Beans 600 2200 1337.13 421.63 Wheat 580 1600 722 115 Maize 100 850 336.75 89.47
180 Table 4 3 . Staple food expen ditures on various food items per week Quantity Expenditure (Tsh "000") Expenditure share (%) Rice (Kgs) 2.697 5.184 7 Maize flour (Kgs) 9.594 10.289 14 Bread (Pieces) 0.648 0.929 1 Buns (Pieces) 7.965 0.960 1 Sugar (Kgs) 2.521 5.023 7 Beans (Kgs ) 4.142 6.072 8 Cooking oil (Lts) 1.573 4.643 6 Tomatoes (Kgs) 3.084 3.788 5 Onion (Kgs) 1.489 1.524 2 Vegetables (Pieces/portion) 7.139 1.505 2 Banana (Pieces) 2.568 0.465 1 Orange (Pieces) 2.803 0.345 0 Goat meat (Kgs) 1.335 7.494 10 Beef (Kgs) 1 .292 7.500 10 Mutton (Kgs) 0.371 2.135 3 Chicken (Pieces) 0.177 1.594 2 Milk (Lts) 11.287 10.443 14 Yogurt (Lts) 0.323 0.411 1 Salt (Kgs) 1.016 0.423 1 Tea (Cups away from home) 2.081 1.349 2 Total expenditure Tsh per week 72.075 100 Total expendi ture Tsh per year (a) 3747.900 Wheat income Tsh per year (b) 1455.898 Barley income Tsh per year (c ) 1105.402 Maize income Tsh per year (d ) 422.912 Beans income Tsh per year (e ) 987.768 Total crop income (b+c+d+e) 3971.98 Wheat income e xp enditure r atio (b/a) 0.388 Total crop income e xpenditure r atio 1 . 060 Wheat income share on total crop income 0.37 Total income (crops, l/stock, off farm) 5356.937
181 Table 4 4 . Food (in)accessibility by wards through wheat income Pooled sample (N= 310) Restricted sample (N= 298) Ward Food inaccessible Food accessible Total Food inaccessible Food accessible Total Mbulumbulu 80 (69 ) 36 (31 ) 116 77 (69) 35 (31) 112 Monduli juu 85 (75 ) 29 (25 ) 114 83 (74) 29 (26) 112 Ngarenairobi 14 (64 ) 8 (36 ) 22 12 (63) 7 (37) 19 Rhotia 42 (72 ) 16 (28 ) 58 40 (73) 15 (27) 55 Total 221 (71 ) 89 (29 ) 310 212 (71) 86 (29) 298 ~ Numbers in brackets are percentage
182 Table 4 5 . Mean characteristics of food accessible and inaccessible farmers Food inaccess Food acc ess t value p value Contract With 0.120 0.324 3.347 0.001*** Without 0.880 0.676 1.635 0.949 Membership With 0.384 0.426 0.620 0.537 Without 0.616 0.574 0.385 0.650 Age 44.306 40.780 1.956 0.053* Education 7.149 7.118 0.069 0.946 Experi ence 14.240 12.353 1.363 0.176 H ousehold composition 6.851 5.353 4.592 0.000*** Agebelow18 3.347 2.397 3.970 0.000*** Age18to50 3.103 2.515 3.361 0.000*** Age50up 0.442 0.426 0.156 0.877 Land lease Lease 0.603 0.676 1.121 0.265 Not lease 0.397 0 .324 0.860 0.805 Ext ension visit 1.030 1.412 1.070 0.287 Meeting 0.388 0.721 1.79 0.077* Transport ownership Own 0.070 0.088 0.469 0.640 Not own 0.930 0.912 0.133 0.553 Farm equipment Own 0.091 0.088 0.068 0.946 Not own 0.909 0.912 0.02 2 0.491 Livestock Keep 0.946 0.897 1.235 0.220 Not keep 0.054 0.103 1.237 0.108 Hybrid seed Use 0.091 0.147 1.193 0.236 Not use 0.909 0.853 0.422 0.663 Off farm income Have 0.153 0.191 0.718 0.475 0.847 0.809 0.295 0.616 Land size used 3.426 11.154 6.225 0.000*** Output/acre 643.236 923.843 8.682 0.000*** Seed/acre 78.781 83.960 2371 0.019** Fertilizer/acre 20.550 35.839 4.406 0.000*** Herbicides/acre 0.682 0.753 1.478 0.142 Insecticides/acre 0.525 0.723 3.719 0.000*** Pesticides/acre 0.391 0.351 1.021 0.309 TE scores 0.779 0.824 2.892 0.004*** AE scores 0.790 0.837 3.457 0.000***
183 Table 4 6 . Probit mo del estimation for food accessibility by wheat income using unrestricted sample Estimate Std. Error z val ue Pr(>|z|) (Intercept) 6.998 2.032 3.443 0.001*** C ontract 0.049 0.361 0.134 0.893 M embership 0.214 0.306 0.699 0.485 Age 0.008 0.020 0.401 0.688 E ducation 0.059 0.044 1.324 0.185 Experience 0.046 0.021 2.220 0.026** Household composition 0.1 49 1.104 0.135 0.892 Age 18 below 0.843 1.122 0.752 0.452 A ge 18 to 50 0.843 1.114 0.757 0.449 A ge 50 up 1.102 1.151 0.957 0.339 Land leased 0.631 0.354 1.782 0.075* E xt ension visit 0.016 0.074 0.209 0.834 M eeting 0.216 0.124 1.740 0.082* Mb ulumbulu 0.339 0.572 0.593 0.553 Rhotia 1.070 0.706 1.516 0.130 Monduli juu 0.688 0.624 1.102 0.270 Transport ownership 0.113 0.514 0.219 0.827 Farm equipment 0.880 0.525 1.675 0.094* Livestock 0.123 0.595 0.207 0.836 Hybrid seed 0.195 0. 434 0.449 0.654 Off farm income 0.020 0.390 0.052 0.958 Land size used 0.574 0.087 6.607 0.000*** TE scores 0.311 2.216 0.140 0.888 AE scores 9.524 3.108 3.064 0.002*** Table 4 7 . Food (in)accessibility by wards through total crop income for restri cted sample Ward Food inaccessible Food accessible Total Mbulumbulu 59 (53) 53 (47) 112 Monduli juu 60 (54) 52 (46) 112 Ngarenairobi 12 (63) 7 (37) 19 Rhotia 35 (64) 20 (36) 55 Total 166 (56) 132 (44) 298 ~Values in brackets are percentages
184 Table 4 8 . Probit model estimation for food accessibility through total crop income for restricted sample Estimate Std. Error z value Pr(>|z|) (Intercept) 5.273 1.257 4.194 0.000*** C ontract 0.374 0.326 1.147 0.251 M embership 0.299 0.227 1.315 0.189 Age 0.00 5 0.013 0.370 0.711 E ducation 0.064 0.034 1.843 0.065* Experience 0.032 0.015 2.141 0.032** Household composition 0.425 0.444 0.958 0.338 Age 18 below 0.163 0.453 0.360 0.719 A ge 18 to 50 0.159 0.445 0.357 0.721 A ge 50 up 0.648 0.489 1.325 0.18 5 Land leased 0.173 0.236 0.734 0.463 E xt ension visit 0.160 0.061 2.610 0.009*** M eeting 0.005 0.108 0.046 0.963 Mbulumbulu 0.905 0.525 1.723 0.085* Rhotia 0.579 0.551 1.049 0.294 Monduli juu 0.890 0.535 1.666 0.096* Transport ownership 0.498 0.4 42 1.127 0.260 Farm equipment 0.600 0.394 1.525 0.127 Livestock 0.322 0.474 0.680 0.496 Off farm income 0.310 0.319 0.970 0.332 Land size used 0.567 0.075 7.546 0.000*** TE scores 4.925 1.140 4.318 0.000*** Table 4 9 . Food (in)accessibility by wa rds through total household income ( unrestricted sample) Ward Food inaccessible Food accessible Total Mbulumbulu 28 (24) 88 (76) 116 Monduli juu 15 (13) 99 (87) 114 Ngarenairobi 7 (32) 15 (68) 22 Rhotia 19 (33) 39 (67) 58 Total 69 (22) 241 (78) 310 ~ Values in brackets are percentages
185 Table 4 10. Probit model estimation for food accessibility with total household income for unrestricted sample Estimate Std. Error z value Pr(>|z|) (Intercept) 1.115* 0.647 1.722 0.085 Contract 0.311 0.266 1.169 0.24 3 Membership 0.378** 0.191 1.981 0.048 Age 0.023** 0.009 2.448 0.014 Education 0.015 0.028 0.532 0.595 Experience 0.031*** 0.011 2.843 0.004 H ousehold composition 0.184 0.227 0.811 0.417 Age below 18 0.031 0.232 0.133 0.894 Age btn 18 and 50 0.2 06 0.233 0.882 0.378 Age above 50 0.038 0.258 0.147 0.883 Land leased 0.240 0.186 1.289 0.197 Extension visits 0.026 0.043 0.613 0.540 Village meeting 0.081 0.084 0.961 0.337 Mbulumbulu 0.640* 0.335 1.912 0.056 Rhotia 0.402 0.356 1.127 0.260 Mon duli juu 1.011*** 0.345 2.925 0.003 Transport ownership 0.049 0.335 0.146 0.884 Farm equipment 0.038 0.314 0.120 0.904 Livestock keeping 0.335 0.378 0.886 0.376
186 Figure 4 1. The average land allocation across crops and their respective average o utput per acre for unrestricted sample (N=310) Figure 4 2 . The average land allocation across crops and their respective average output per acre for restricted sample (N=298) 0 1 2 3 4 5 6 Wheat Barley Maize Beans Land (acre) Output (kgs/acre) in "000" 0 1 2 3 4 5 6 Wheat Barley Maize Beans Land (acre) Output (kg) in '000'
187 Figure 4 3. Crop price variations in the study area for unrestricted sample Figure 4 4 for unrestricted sample 0 500 1000 1500 2000 2500 Minimum Maximum Average Std.Dev Barley Beans Wheat Maize 0.000 2.000 4.000 6.000 8.000 10.000 12.000 Rice Bread Buns Sugar Beans Tomatoes Onion Vegitables Banana Orange Goatmeat Beef Mutton Chicken Milk Yogurt Salt Tea Expenditure (Tsh "000")
188 Figure 4 5 . Food (in)accessibility by wards through wheat income for unrestricted sample Figure 4 6 . Food (in)accessibility by wards through w heat income for restricted sample 0% 10% 20% 30% 40% 50% 60% 70% 80% Mbulumbulu Mondulijuu Ngarenairobi Rhotia Pooled sample Food inaccess Food access 0% 10% 20% 30% 40% 50% 60% 70% 80% Mbulumbulu Monduli juu Ngarenairobi Rhotia Pooled sample Food inaccess Food access
189 Figure 4 7 . Food (in)acce ssibility by wards through total crop income of restricted sample Figure 4 8 . Food (in)acce ssibility by wards through total household income of unrestricted sample 0% 10% 20% 30% 40% 50% 60% 70% Mbulumbulu Monduli juu Ngarenairobi Rhotia Pooled sample Food inaccess Food access 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% Mbulumbulu Mondulijuu Ngarenairobi Rhotia Pooled sample Food inaccess Food access
190 CHAPTER 5 CONCLUSION A griculture is and will continue to be of great importance to the economy of Tanzania. An important aspect of food security is the ability of a country to satisfy a substantial portion of its food requirements from domestic production. Wheat has emerged as one of the preferred cereals in the Tanzanian diet. Yet, despite favorable agronomic conditions , there is a widening gap between consumption and domestic supply, resulting in a greater reliance on imports and a divergence of scarce foreign exchange toward satisfying this nee d. Statistics show wheat grain i s one of the major imported cereals, accounting for over 30% of the tota l food import bill (URT, 2013). Given the potential contribution this crop could make toward mitigating food insecurity and poverty i n Tanzania, it i s important to study some of the factors that could positively influence the development of this industry. Among other things , such assessment would facilitate the development of critical policy strategies to propel the industry forward. T his study consists of three essays on chain participation, and wheat contribution to food accessibility to help policy makers and other stakeholders strategize how to improv e wheat production in Tanzania. U nderstandi ng of policy variables that would help boost wheat production in Tanzania is important . It i s hypothesized that the lack luster response to a seemingly good market opportunity might stem from low farm efficiency , nonparticipation of farmers in the profitab le wheat value chains , and low /insufficient net wheat returns . In the first essay o n farm efficiency , translog production and cost functions were utilized in the stochastic frontier to examine the technical, allocative , and economic efficiencies of 310 sam pled wheat farmers in n orthern Tanzania, where wheat
191 production is a major crop. Further, we utilize d the propensity score matching procedure and specifically the nearest neighbor and caliper radius matching procedure s to analyze the impact of value chain Th is study confirmed our hypothesis and found that the average TE, AE, and EE scores are 79%, 80%, and 64% , respectively , over the pooled sample in the study area , thus implying there is room for improvemen t . This study also found that vertical participation in the value chain could improve scores by 6.8%, 5.7%, and 8.7% , respectively , while horizontal participation in the value chain could improve TE, AE, and EE scores by 6. 3%, 9.5%, and 11.6%, respectively. On the basis of these findings , we ma de several recommendations for farmers to improve their levels of efficiencies and ultimately their outputs. For instance, fa rmers should take steps to increase the size of their whea t plots, apply fertilizer, and use appropriate insecticides to increase wheat production in the study area . T o improve farm unit efficiencies , farmers need to increase their access to farm equipment, hybrid seeds, extension services, and formal education f or the young er generation s . Farmers are also encouraged, where the opportunity presents itself, to become engaged in off farm activities to supplement their incomes since results show ed a noticeable level of improvement of farm efficiencies associated with off farm income. Finally, farmers should participate in the wheat value chain through contract s with buyers and through members hip in association s to improve their farm efficiency levels. The second essay explore d in greater depth the link participation in the value chain and its impact on their welfare. Specifically , it was hypothesized that, ceteris paribus, participation in the value chain would increase
192 in terms of profit per unit of wheat sold. T o expl ore t his broad objective, we traced the wheat grain flows from production point to ultimate consumption. We then describe d in detail the coordination among wheat value chain actors with the main focus on farmers. Next, we analyze d factors that influence fa rmers chain. Finally, we examine d We foun d that t he whe at value chain in the study area consist s of four main chains : the wheat input chain , the wheat grain chain, th e wheat flour chain , and the wheat products chain. For the purpose of this study, we focus ed wheat grain chain because farmers seem to be the most vulnerable group in the chain . Farmers sell wheat grain to local retailers , consumers , bro kers, and wholesalers at the farm gate. Brokers sell to wholesalers and millers , while wholesalers sell to urban retailers and millers who process wheat grain into flour. The urban retailers sell wheat to urban consumers. The wheat brokers in the study are a are the major/dominant players in the wheat grain value chain because they are involved in organizing most transactions between traders and farmers. One disadvantage that farmers face in selling their wheat t hrough brokers is that the brokers buy bags of wheat value d at 100 kg per bag when in reality the bags weigh 110 kg to 130 kg, depending on how much extra (overflow) the bags can hold. In the Swahili language, the overflow bags are called lumbesa . Most farmers sell through brokers be cause they are poorly linked to traders. Only a few farmers are vertically (only ~17%) and horizontally (only ~39%) coordinated through contract s and member ships in groups/ association s, respectively. Farmers with contracts (vertically coordinated) have cha racteristics that a re significantly different from farmers without contract s. Such differences can be
193 observed in terms of wheat land size, technical efficiency, allocative efficien cy , output per acre, frequency of extension services ( visits ) , frequency of village meeting s attendance, and off farm income. For horizontal c o ordination , similar findings a re obtained in which farmers who have memberships in groups/associations differ positive ly and significantly from nonmembers in terms of level of education, f requency of meetings attendance , output per acre, technical efficiency, and allocative efficiency. Following the propensity score technique, a logistic model was use d to explore the factors influencing farmers participation in the value chain and the sco res used to match the covariates of participants and nonparticipants. The results indicate that participation in vertical coordination i s influenced by the age of the farmer, land leased, frequency of extension visits , frequency of agricultural and village meetings attend ance , off farm income, land size allocated for wheat , and technical efficiency. Likewise, p articipation in horizontal coordination i s influenced by frequency of meetings attendance , farm location, farm equipment ownership , technical efficie ncy, and allocative efficiency. The fitted v alues from the logit model were used to establish the propensity scores for m atch ing participants and nonpa rticipants of the value chain. The overlapping and unconfoundedness assumptions were fulfilled by applyin g the nearest neighbor and caliper radius matching algorithms. The t test on covariates after matc hing revealed no significant differences between matched participants and nonparticipants in the wheat value chain. V ertical coordination participants receive d an extra net profit of Tsh 107 per kg of wheat more than did nonparticipants , which was positive and significant at the 1% level. On the other hand, horizontal coordination participants received an extra net profit of Tsh 54 per kg of wheat more than non participants , which
194 was significant at the 5% level. The sensitivity analysis reveal ed that our statistical generally insensitive to unobserved covariates . H owever , we cannot ignore the fact that ho rizontal coordination is somewhat more sensitive to hidden bias compa red to vertical coordination. Given the above findings, t his study calls for policy makers and other beneficiaries to consider providing upfront investments to wheat farmers to facilitate production through binding c ontracts . The evidence suggest s that even with weak oral contracts, farmers can secure more income than those without contract s . T o facilitate farmers to participate in contracts , more attention should be given to increasing fa accessibility to extension services and, agricultural related meetings, and assisting farmers with increas ing the ir land size for wheat production , off farm work opportunities, and level s of technical efficiency. Further, the outcome from horizontal coordination participation suggests the need to improve the efficiency of existing farmers groups/ associations that specifically address concerns of wheat producers . The third essay examine d the contribution of wheat income on household food accessibilit y. The underlying premise was that wheat growers can improve their income levels as a result of improvement s in wheat productivity and value chain participation , thus leading to an improvement in their level of food security. T o conduct such analysis, the agricultural household model framework i s applied to formulate a food accessibility equation; also a pro bit model is used to analyze fa ctors influencing food accessibility . We also analyze how land is allocated to various crops that are considered a source of income and are own produced food. O ur investigation reveal ed that 31% of the total land in the study area is allocated to barley , followed by 30% to wheat, 21% to maize,
195 and 19% to beans. Wheat and barley compete for the same factors of production , inc luding land. A f amers decision to plant wheat or barley depends mainly on prices and services offer e d by the buyers. Also, b reweries offer slightly better price incentives to barley farmers and provide them with inputs. This situ ation lowers wheat product ion because farmers are switch ing to more incentivized crop, barley . L ess land allocated for wheat production lowers domestic wheat production and the income share o f food expenditure s . The share of wheat income on household food expenditure is 38.8% , whic h suggests that fa r mers have other sources of income . W e also found that 78% of wheat farmers can not cross the national food poverty line based solely on income generated from wheat production . A possible explanation could be that the price for wheat recei ved by farmers is too low and discourage s expanding wheat production to generate more income. The 22% of households that have food accessibility differ positively and significantly from the 78% who have food inaccessibility in terms of contracts, meeting a ttendance, land size used for wheat production, production level, amount of seed used, quantity of fertilizer used, insecticides applied, and technical and allocative efficiency scores. Using probit model analysis , it is shown that farming experience, meet ing s attendance, land size, and allocative efficiency positively and significantly influence the probability of food accessibility through wheat income . On the basis of the findings of this investigation, farmers are advised to increase their level of eff iciency by using least cost inputs to close the gap between the food poverty threshold and wheat income. In light of the push to reduce dependence on imported wheat, the government and other stakeholders might consider subs i dizing inputs for whe at production as is done for barley by private brew eries in the
196 study area. S ubs i dies will encourage farmers to allo cate more of their land to wheat production , thus generating more output and more income at prevailing prices. Farmers are further advised t o increase land size for wheat production because the realized output from increased production would increase both the volume of wheat sold and income generated . Farmers should attend agricultural and village meetings to learn new techn ologies in wheat pr oduction that would improve output and income to rise above both the food poverty line and the nonfood poverty line. In summary, the overall conclusion of this dissertation is that a systemic and sustained program aimed at assisting wheat farmers to improv e their levels of efficiency and become better integrated in the value chain will go far in helping such farmers to increase their agricultural income (welfare) and escape poverty. Policies that address some of the major constraints identified in this stud y and focus on strengthing local institutions, including extension service s for better access to modern technology, technical know how , and up to date information, will enable the local wheat industry to ent level of food insecurity.
197 LIST OF REFERENCES Aakvik, A. (2001). Bounding a matching estimator: The case of a Norwegian training program. Oxford Bulletin of Economics & Statistics , 63 , 115. Abadie, A., & Imbens, G. W. (2006). Large sample properties of matching estimators. Econometrica , 74 (1), 235 267. Abate, G. T., Francesconi, G. N., & Getnet, K. (2014). Impact of agricultural Ethiopia. Euricse Working Paper , 50 (13), 1 32. Abebaw, D., & Haile, M. G. (2013). The impact of coope ratives on agricultural technology adoption: Empirical evidence from Ethiopia. Food Policy , 38 (1), 82 91. Abebe, G. K., Bijman, J., Kemp, R., Omta, O., & Tsegaye, A. (2013). Contract farming ibutes. Food Policy , 40 , 14 24. Alemu, A. E., & Adesina, J. (2015). Effects of co operatives and contracts on rural income and production in the dairy su African Journal of Agricultural Economics , 10 (4), 312 327. Alemu, A. E., Maertens, M., Deckers, J., Bauer, H., & Mathijs, E. (2016). Impact of opia. Agricultural and Food Economics , 4 (9), 21. Amor, T., & Muller, C. (2010). Application of stachastic production frontier in the estimation of technical efficiency of irrigated agriculture in Tunisia. Agricultural Journal , 5 (2), 50 56. Asravor, J., Onu mah, E. E., & Osei asare, Y. B. (2016). Efficiency of chili pepper production in the volta region of Ghana. Journal of Agricultural Extension and Rural Development , 8 (June), 99 110. Babatunde, R. O., Omotesho, A., & Sholotan, O. S. (2007). Socio economics characteristics and food security status of farming households in Kware State, North Central Nigeria. Pakistan Journal of Nutrition , 6 (1), 47 58. Barnum, H. N., & Squire, L. (1979). An econometric application of the theory of the farm household. Journal of Development Economics , 6 , 79 102. Barrett, C. B., Bachke, M. E., Bellemare, M. F., Michelson, H. C., Narayanan, S., & Walker, T. F. (2012). Smallholder participation in contract farming: Comparative evidence from five countries. World Development , 40 (4), 715 730. Battese, G. E., & Coelli, T. J. (1988). Prediction of firm level technical efficiencies with a generalized frontier production function and panel data. Journal of Econometrics ,
198 38 (3), 387 399. http://doi.org/10.1016/0304 4076(88)90053 X Battese, G . E., & Corra, G. S. (1977). Estimation of a production function model: Application to the pastoral zone of Eastern Australia. Australian Journal of Agricultural Economics , 21 (3), 169 179. Becker, G. S. (1965). A Theory of the allocation of time. The Econ omic Journal , 75 (299), 493 517. Becker, S., & Caliendo, M. (2007). Sensitivity analysis for average treatment effects. The Stata Journal , 7 (1), 71 83. Becker, S., & Ichino, A. (2002). Estimation of average treatment effects based on propensity scores. Stat a Journal , 2 (4), 358 377. Benjamin, D. (1992). Agricultural household models, Hosehold composition , Labor Question. Econometrica , 60 (2), 287 322. Birthal, P. S., Joshi, P. K., & Gulati, A. (2005). Vertical coordination in high value food commodities: Implication for smallholders. Markets, Trade and Institutions Division Discussion Paper, Washington DC, IFPRI , (85), 63. Blundell, R., Dearden, L., & Sianesi, B. (2005). Evaluating the Effect of Education on Ea rnings: Models, Methods and Results from the National Child Development Survey. Journal of the Royal Statistical Society , 168 (3), 473 512. Boger, S. (2001). Quality and contractual choice: A transaction cost approach to the Polish hog market. European Revi ew of Agriculture Economics , 28 (3), 241 261. Bogetoft, P., & Otto, L. (2010). Benchmarking with DEA, SFA, and R . (F. S. Hillier, C. C. Price, & S. F. Austin, Eds.) (Vol. 157). New York: Springer Science and Business Media, LLC 2011. Bogetoft, P., & Otto, L Bolwig, S., Gibbon, P., & Jones, S. (2009). The economics of smallholder organic contract farming in Tropical Africa. World Development , 37 (6), 1094 1104. Boselie, D., Henson, S. pencer, & Weatherspoon, D. (2003). Superm arket procurement practices in developing countries: Redefining the roles of the Public and Private Sectors. American Journal Agricultural Economic Association , 85 (5), 1155 1161. Bowers, J., & Hansen, B. (2005). Attributing effects to a get out the vote ca mpaign using full matching and randomization inference. Political Science , (September 2004), 1 46. Bravo Ureta, B. E., Greene, W., & Solis, D. (2012). Technical efficiency analysis
199 correcting for biases from observed and unobserved variables: An applicatio n to a natural resource management project. Empirical Economics , 43 (1), 55 72. Bravo Ureta, B. E., & Rieger, L. (1991). Dairy farm efficiency measurement using stochastic frontiers and neoclassical duality. American Journal of Agricultural Economics , 73 (2) , 421 428. Retrieved from http://www.jstor.org/stable/1242726 Caliendo, C., Guida, M., & Parisi, A. (2007). A crash prediction model for multilane roads. Accident Analysis and Prevention , 39 (4), 657 670. Caliendo, M., & Kopeinig, S. (2008). Some practical guidance for the implementation of propensity score matching. Journal of Economic Survey , 22 (1), 31 72. Chavas, J. P., Petrie, R., & Roth, M. (2005). Farm household production efficiency: Evidence from the Gambia. American Journal of Agricultural Economics , 87 (1), 160 179. pesticide use in Bt cotton production in china. China Economic Review , 27 , 15 24. Chen, S., & Ravallion, M. (2007). Absolute poverty measures for the develo ping world, 1981 2004. Proceedings of the National Academy of Sciences of the United States of America , 104 (43), 16757 16762. Cochran, W. . (1968). The effectiveness of adjustment by subclassification in removing bias in observational studies. Biometrics , 24 (2), 295 313. Coelli, T. J., & Battese, G. E. (1996). Identification of factors which influence the technical inefficiency of Indian farmers. Australian Journal of Agricultural Economics , 40 (2), 103 128. Dehejia, R. H., & Wahba, S. (1999). Causal effects in nonexperimental studies: Reevaluating the evaluation of training programs. Journal of the American Statistical Association , 94 (448), 1053 1062. Dehejia, R. H., & Wahba, S. (2002). Propensity score matching methods for nonexperimental causal studies. Re view Economics and Statistics , 84 (1), 151 161. DiPrete, T. A., & Gangl, M. (2004). Assesing bias in the estimation of causal effects: Rosenbaum bounds on matching estimators and instrument variables estimation with imperfect instrument. Econstor , 1 57. Don ovan, J., Franzel, S., Cunha, M., Gyau, A., & MithÃ¶fer, D. (2015). Guides for value chain development: A comparative review. Journal of Agribusiness in Developing and Emerging Economies , 5 (1), 2 23. Dries, L., Germenji, E., Noev, N., & Swinnen, J. F. M. (2 009). Farmers, vertical coordination, and the restructuring of dairy supply chains in central and eastern
200 Europe. World Development , 37 (11), 1742 1758. Fafchamps, M., & Hill, R. V. (2005). Selling at the farmgate or travelling to market. American Journal A gricultural Economic Association , 87 (3), 717 734. Fare, R., & Knox Lovell, C. A. (1978). Measuring the technical efficiency of production. Journal of Economic Theory , 19 (1), 150 162. Foster, J. E. (1989). What is poverty and who are the poor? Redefinition for the United Absolute versus Relative Poverty. The American Economic Review , 88 (2), 335 341. Greene, W. (2010). A stochastic frontier model with correction for sample selection. Journal of Productivity Analysis , 34 (1), 15 24. Greer , J., & Thorbecke, E. (1986). A methodology for measuring food poverty applied to Kenya. Journal of Development Economics , 24 , 59 74. Gulati, A., Minot, N., Delgado, C., & Bora, S. (2007). Growth in high value agriculture in Asia and the emergence of verti cal links with farmers. Food Policy , (March), 91 108. Hagenaars, A. J. M., & van Praag, B. M. S. (1985). A synthesis of poverty line definitions. Review of Income and Wealth , 31 (2), 139 154. Hansen, B. B., & Bowers, J. (2008). Covariate balance in simple, stratified and clustered comparative studies. Statistical Science , 23 (2), 219 236. Heckman, J. ., Ichimura, H., & Todd, P. . (1997). Matching as an econometric evaluation estimator: Evidence from evaluating ajob training programme. The Review of Economic S tudies , 64 , 605 654. Henningsen, A. (2010). Estimating censored regression models in R using the censReg Package. R Package Vignettes Collection , 5 (2), 12. 9. Hernandez, R., Reardon, T., & Berdegue, J. (2007). Supermarkets, wholesalers, and tomato growers in Guatemala. Agricultural Economics , 36 (3), 281 290. Herron, M. C., & Wand, J. (2007). Assessing partisan bias in voting technology: The case of the 2004 New Hampshire recount. Electoral Studies , 26 (2), 247 26 1. Ho, D. E., Imai, K., King, G., & Stuart, E. A. (2007). Matching as nonparametric preprocessing for reducing model dependence in parametric causal inference. Political Analysis , 15 (3), 199 236. Hoken, H., & Su, Q. (2015). Measuring the effect of agricult ural cooperatives on
201 household income using PSM producing cooperative in China. IDE Discussion Paper , (539), 30. Holloway, G., Nicholson, C., Delgado, C., Staal, S., Ehui, S. (2000). Agro industrialization through institutional innovation: transactions costs, cooperatives, and milk market development in the Ethiopian Highlands. Agricultural Economics , 23 (2000), 279 288. Ilembo, B., & Kuzilwa, J. (2014). Technical efficiency analysis of tobacco production in Tanzania. Interdiscpl inary Journal of Contemporary Research in Business , 6 (3), 246 265. Imbens, G. W. (2004). Nonparametric estimation of average treatment effects under exogeneity: a review. Review of Economics & Statistics , 86 (1), 4 29. Jacoby, H. G. (1993). Shadow wages and peasant family labour supply: An econometric application to the Peruvian Sierra. The Review of Economic Studies , 60 (4), 903 921. Jondrow, J., Knox Lovell, C. a., Materov, I. S., & Schmidt, P. (1982). On the estimation of technical inefficiency in the stoc hastic frontier production function model. Journal of Econometrics , 19 (2 3), 233 238. Kalirajan, K. P., & Shand, R. T. (1999). Frontier production functions and technical efficiency measures. Journal of Economic Surveys , 13 (2), 149 172. http://doi.org/10.1 111/1467 6419.00080 Kaplinsky, R., & Morris, M. (2001). A HANDBOOK FOR VALUE CHAIN RESEARCH . Institute for Development Studies: Brighton, UK . sensitivity analysis with matched dat a . Citeseer , (2002), 1 15. Lau, L. J., Lin, W. L., & Yotopoulos, P. A. (1978). The linear logarithmic expenditure system: An application to consumption leisure choice. Econometrica , 46 (4), 843 868. Lechner, M. (19 99). An evaluation of public sector sponsored continuous vocational training programs in east Germany. IZA Discussion Papers , (93), 1 43. Lobell, D. B., & Field, C. B. (2007). Global scale climate crop yield relationships and the impacts of recent warming. Environmental Research Letters , 2 (1), 014002. Lobell, D. B., Schlenker, W., & Coster Robert, J. (2011). Climate trends and global crop production since 1980. Science , 333 (November), 616 621.
202 Madden, D. (2000). Relative or absolute poverty lines: A new app roach. Review of Income and Wealth , 46 (2), 181 199. Maertens, M., & Swinnen, J. F. M. (2009). Trade, standards, and poverty: Evidence from Senegal. World Development , 37 (1), 161 178. Martey, E., Dogbe, W., Etwire, P. M., & Wiredu, A. N. (2015). Impact of f armer mentorship project on farm efficiency and income in rural Ghana. Journal of Agricultural Science , 7 (10), 79 93. Masakure, O., & Henson, S. (2005). Why do small scale producers choose to produce under contract? Lessons from nontraditional vegetable ex ports from Zimbabwe. World Development , 33 (10), 1721 1733. Mason, N. M., Jayne, T. S., & Shiferaw, B. (2012). Wheat consumption in Sub Saharan Africa: Trends, drivers, and policy implication. MSU International Development Working Paper , (127), 1 29. Mayen, C. D., Balagtas, J. V., & Alexander, C. E. (2010). Technology adoption and technical efficiency: Organic and conventional dairy farms in the United States. American Journal of Agricultural Economics , 92 (1), 181 195. Mburu, S., Ackello ogutu, C., & Mulwa, R. (2014). Analysis of economic efficiency and Economics Research International , 2014 . Minten, B., Randrianarison, L., & Swinnen, J. (2007). Spillovers from high value agriculture for ex ports on land use in developing countries: Evidence from Madagascar. Agricultural Economics , 37 (2 3), 265 275. Miyata, S., Minot, N., & Hu, D. (2009). Impact of contract farming on income: Linking small farmers, packers, and supermarkets in China. World De velopment , 37 (11), 1781 1790. Mlote, S., Mdoe, N., Isinika, A., & Mtenga, L. (2013). Estimating technical efficiency of small scale beef cattle fattening in the Lake Zone in Tanzania. Journal of Development and Agricultural Economics , 5 (5), 197 207. Morgan , S. L., & Harding, D. J. (2006). Matching estimators of causal effects. Sociological Methods & Research , 35 , 3 60. Msuya, E., & Ashimogo, G. (2007). Estimation of technical efficiency in Tanzanian sugarcane production: A case study of Mtibwa Sugar Estate Outgrowers Scheme. MPRA , (3747), 0 29. Retrieved from mpra.ub.uni muenchen.de/3747/ Narrod, C., Roy, D., Okello, J., AvendaÃ±o, B., Rich, K., & Thorat, A. (2009). Public private partnerships and collective action in high value fruit and vegetable supply cha ins. Food Policy , 34 (1), 8 15.
203 Neven, D., Odera, M. M., Reardon, T., & Wang, H. (2009). Kenyan supermarkets, emerging middle class horticultural farmers, and employment impacts on the rural poor. World Development , 37 (11), 1802 1811. Neven, D., & Reardon, T. (2004). The rise of Kenyan supermarkets and the evolution of their horticulture product procurement systems. Development Policy Review , 22 (6), 669 699. Ogundari, K., & Ojo, S. O. (2006). An examination of technical , economic and allocative efficiency o Nigeria. Central European Agriculture Journal , 7 (3), 423 432. Okoye, B. ., Onyenweaku, C. ., & Asumugha, G. . (2006). Allocative efficiency of small holder cocoyam farmers in Anambra State, Nigeria. MPRA , 9 (17362), 1 10. Oleke, J. ., & Isinika, A. . (2011). Assessing the technical efficiency of commercial egg production in Tanzania for improved livelihoods. Journal of Development and Agricultural Economics , 3 (8), 343 352. Olukunle, T. O. (20 13). Evaluation of income and employment generation from cassava value chain in the Nigerian agricultural sector. Asian Journal of Agriculture and Rural Development , 3 (3), 79 92. Omonona, B. T., & Agoi, G. A. (2007). An analysis of food security situation among Nigerian urban households: Evidence from Lagos state, Nigeria. Journal of Central European Agriculture , 8 (3), 397 406. Osborne, S., & Trueblood, M. A. (2006). An examination of economic efficiency of Russian crop production in the reform period. Agri cultural Economics , 34 (1), 25 38. Pitt, M. M., & Lee, L. (1981). The measurement and sources of technical efficiency in the Indonesian weaving industry. Journal of Development Economics , 9 , 43 64. Pradhan, M., Suryahadi, A., SUmarto, S., & Pritchett, L. (2 001). Eating like which Review of Income and Wealth , 84 (4), 473 487. PromarConsulting. (2012). 2011 Agricuture, Forestry and Fisheries of Tanzania. Retrieved from http: //www.promarconsulting.com/site/wp content/uploads/2012/05/Tanzania_2012.pdf Rajendran, S., Afari Sefa, V., Karanja, D. K., Musebe, R., Romney, D., Makaranga, M. A., & Kessy, R. F. (2015). Technical efficiency of traditional African vegetable production: A case study of smallholders in Tanzania. Journal of Development and Agricultural Economics , 7 (3), 92 99. Rao, E. J. O., & Qaim, M. (2011). Supermarkets, farm household income, and poverty: Insights from Kenya. World Development , 39 (5), 784 796.
204 Ravallion, M., & Bidani, B. (1994). How robust is a poverty profile? The World Bank Economic Review , 8 (1), 75 102. Ravallion, M., Datt, G., & van de Walle, D. (1991). Quantifying absolute poverty in the developing world. Review of Income and Wealth , 37 (4), 345 361. R eardon, T., Timmer, P., & Berdegue, J. (2004). The rapid rise of supermarkets in developing countries: Induced organizational, institutional, and technological change in agrifood systems. Journal of Agricultural and Development Economics , 1 (2), 168 183. Ro senbaum, P. ., & Rubin, D. . (1983). Assessing sensitivity to an unobserved binary covariate in an observational study with binary outcome. Journal of the Royal Statistical Society , 45 (2), 212 218. Rosenbaum, P. R. (2002). Design of observational studies . NY: Springer Science and Business Media. Biometrika , 70 (1), 41 55. Rosenbaum, P. R., & Rubin, D. B. (1984). Reducing bias in observational studies using subclassification on the propensity score. Journal of the American Statistical Association , 79 (387), 516 524. Roy, D., & Thorat, A. (2008). Success in high value horticultural export markets for the small farmers: The case of M ahagrapes in India. World Development , 36 (10), 1874 1890. Rubin, D. B., & Waterman, R. P. (2006). Estimating the causal effects of marketing interventions using propensity score methodology. Statistical Science , 21 (2), 206 222. SAGCOT. (2009). Value chain and market analysis. Retrieved from http://www.sagcot.com/uploads/media/Appendix_IV _Value_Chain_and_Market_Analysis_03.pdf Saigenji, Y., & Zeller, M. (2009). Effect of contract farming on productivity and income of small holders: The case of tea productio n in north western Vietnam. International Association of Agricultural Economists, Beijing China, August 16 22, 2009 , 1 18. Retrieved from http://ageconsearch.umn.edu/bitstream/51681/2/669.pdf Sarris, A., Savastano, S., & Christiaensen, L. (2006). The role of agriculture in reducing poverty in Tanzania: A household perspective from rural Kilimanjaro and Ruvuma. FAO Commodity and Trade Policy Research Working Paper , (19), 1 37. Retrieved from www.fao.org/es/ESC/ Schlenker, W., & Roberts, M. J. (2009). Nonline ar temperature effects indicate severe
205 damages to U.S. crop yields under climate change. Proceedings of the National Academy of Sciences , 106 (37), 15594 15598. Sekhon, M. K., Mahal, A. K., Kaur, M., & Sidhu, M. S. (2010). Technical efficiency in crop produ wise analysis. Agricultural Economic Review , 23 (December), 367 374. The Quarterly Journal of Economics , 96 (3), 433 464. Seyoum, E. T., Battese, G. E., & Flemin g, E. M. (1998). Technical efficiency and productivity of maize producers in eastern Ethiopia: A study of farmers within and outside the Sasakawa Global 2000 project. Agricultural Economics , 19 (3), 341 348. Shapiro, K., & Muller, J. (1977). Sources of tech nical efficiency: The roles of modernazation and information. Economic Development and Cultural Change , 25 (2), 293 310. Shiferaw, F., Kilmer, R. L., & Gladwin, C. (2003). Determinants of food security in southern Ethiopia. A Selected Paper Presented at the 2003 American Agricultural Economics Association Meetings in Montreal, Canada , 1 , 1 29. Siamwalla, A., & ValdÃ©s, A. (1980). Food insecurity in developing countries. Food Policy , 5 (4), 258 272. Sianesi, B. (2004). An evaluation of of the Swedish sytem of a ctive labor market in the 1990s, 86 (February), 133 155. Singh, I., Squire, L., & Strauss, J. (1986). Agricultural Household Models: Extension, Application, and Policy . Sipilainen, T., & Lansink, A. O. (2005). Learning in organic farming An application on f innish dairy farms. XIth Congress of the EAAE , 1 22. Skoufias, E. (1994). Using shadow wages to estimate labor supply of agricultural households. American Journal of Agricultural Economics , 76 (2), 215 227. Smith, J. A., & Todd, P. E. (2005). Does matching nonexperimental estimators? Journal of Econometrics (Vol. 125). SolÃs, D., Bravo Ureta, B. E., & Quiroga, R. E. (2007). Soil conservation and technical efficiency among hillside farmers in central America: A switching regress ion model. Australian Journal of Agricultural and Resource Economics , 51 (4), 491 510. Strauss, J. (1984). Joint determination of food consumption and production in rural Sierra Leone. Estimates of a household firm model. Journal of Development Economics , 1 4 (1), 77 103.
20 6 Swinnen, J. F. M., & Maertens, M. (2006). Globalization, privatization, and vertical coordination in food value chains in developing and transition countries. LICOS Center for Institutions and Economic Performance, K.U. , 89 102. Taylor, J. E. and extensions. Review of Economics of the Household , 1 (1 2), 33 58. Temu, A. . (2006). Aid for trade and agro based private sector development in Tanzania . Thorp, R., Stewart, F ., & Heyer, A. (2005). When and how far is group formation a route out of chronic poverty? World Development , 33 (6), 907 920. UNIDO. (2011). Pro poor value chain development . United Nations Industrial Development Organization, Vienna, Austria . USAID. (2010 ). Staple foods value Chain analysis: Country report Tanzania, (June). Valkila, J., Haaparanta, P., & Niemi, N. (2010). Empowering coffee traders? The coffee value chain from Nicaraguan fair trade farmers to Finnish consumers. Journal of Business Ethics , 9 7 (2), 257 270. http://doi.org/10.1007/s10551 010 0508 z Wainaina, W. P., Okello, J, J., & Nzuma, J. (2012). Impact of contract farming on International Association of Agricultural Economists, Triennial Conferan ce, Brazil, August 18 24, 2012 , 1 30. Warsanga, W. B. (2014). Coordination and structure of agri of banana value chain strands in Tanzania. Journal of Economics and Sustainable Development , 5 (7), 71 79. Wollni, M., & Zeller, M. (2007). Do farmers benefit from participating in specialty markets and cooperatives? The case of coffee marketing in Costa Rica. Agricultural Economics , 37 (2007), 243 248. World Bank. (2015). TANZANIA MAINLAND POVERTY ASSESMENT. Retrieved from http://www. worldbank.org/content/dam/Worldbank/document/Africa/Tanzania/Repor t/tanzania poverty assessment 05.2015.pdf Zheng, B. (2001). Statistical inference for poverty measures with relative poverty lines. Journal of Econometrics , 101 (2), 337 356.
207 BIOGRAPHICAL SKETCH William Barnos Warsanga was born and raised in Tanzania. In 2003, he enrolled at Sokoine University of Agriculture (SUA) where he obtained his Bachelor of Science degree in agribusiness and agricultural economics in 2006. After two years of working , he went to the same university for his Master of Science degree in agricultural economics and graduated in 2011. The following year (2012), he joined University of Florida in US and pursued his Doctor of Philosophy in the Department of Food and Resour ce E conomics and graduated in 2016 with specialization in production and consumer economics, trade and policy analysis, economic development, and resource economics.