FARMER VALUATION OF SEED QUALITY: A COMPARATIVE ANALYSIS OF TWO PREFERENCE ELICITATION METHODS IN NICARAGUA By Sean Posey A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Agricult ure, Food, and Resource Economics - Master of Science 2018 ABSTRACT FARMER VALUATION OF SEED QUALITY: A COMPARATIVE ANALYSIS OF TWO PREFERENCE ELICITATION METHODS IN NICARAGUA By Sean Posey s in Nicaragua, I surveyed 242 bean producing households in the northwestern region of Nicaragua. The majority of farmers stated that they had heard of or had knowledge of certified seed, which is considered the highest quality seed potentially available to farmers. Despite this knowledge of higher quality seeds, 56% of households stated they had never purchased seed. In this study, I evaluate two non - hypothetical preference elicitation methods quality seeds : a B ecker - DeGroot - Marschak (BDM) auction and a Real Choice Experiment (RCE). The BDM auction and RCE were conducted after farmers examined the experimental plots on which three types of quality bean seeds of the same variety were planted certified seeds, quali ty declared seed, and recycled grain . The se were blind experiments and farmers were unaware of the quality types beyond what they observed in the experimental plots. The results of the BDM auction show that farmers are willing to pay a premium of C$5/lb (U S$0.16/lb) for the plot planted with their highest rated seed compared to the plot planted with lowest rated seed . In the RCE, t he average premium farmers are willing to pay for the highest rated seed was C$30/lb (US$1 /lb). The average premium farmers wer e willing to pay for the lowest rated seed was C$ - 9.16/lb (US$ - 0.31 / lb) and was statistically different from C$0, which is not consistent with the BDM results. Potential reasons for the discrepancies in the results of the two methods are discussed . i ii ACKNOWL EDGMENTS I would like to thank my major advisor Mywish Maredia for her help and support throughout my research. I would like to acknowledge my committee members, Robert Shupp, David Ortega, and Eduardo Nak a sone for their insight s and help , which i s great ly appreciated. I would like to also thank Byron Reyes and all the staff in the Managua office of the International Center for Tropical Agriculture (CIAT) that helped me in the design and implement ation of the experiments in Nicaragua and their support dur ing the field work. Lastly, I would like to acknowledge the financial support towards this research provided by the Feed the Future Innovation Lab for Grain Legumes under the terms of Cooperative Agreement EDH - A - 00 - 07 - 00005 - 00 between Michigan State Unive rsity and the USAID Bureau for Food Security, Office of Agriculture, Research, and Technology ; and the Department of Agricultural, Food, and Resource Economics at Michigan State Universit y iv TABLE OF CONTENTS 1. Introduction 1 1.1 Problem Statemen t and Research Focus 1 1.2 Bean Seed System in Nicaragua 4 2. Review of the Literature 9 2.1 Determinants of Willingness to Pay for Quality Seed 9 2.2 Comparison of Preference Elicitation Methods 1 0 2.3 Preference Elicitatio n Method Comparison in Developing Countries 1 2 3. Conceptual Framework 15 3.1 Technology Adoption Decision Framework 1 5 3.2 Theory of Demand 17 3.3 Choice Experiment 17 3.4 Becker - DeGroot - Marschak (BDM) Auction Experiment s 2 1 3.5 Comparison of Two Methods 2 3 4. Methods and Data 24 4.1 Study Site and Field Experiments 24 4.2 Source of the Three Seed Types Used in the Experiment 27 4.3 Field Days and Farmer Ranking of Seed Plots 29 4.4 WTP Elicitation Experiments 3 0 4.5 Data 34 5. Empirical Data 35 5.1 Analytical Sample 3 5 5.2 Sample Characteristics 37 6. Results 40 6.1 Field Day Rankings 4 0 6.2 nt quality seed based on BDM mechanism 4 3 6.3 Real Choice Experiments 4 5 6.4 Comparing Elicitation Methods for WTP 4 8 6. 5 WTP and Plot Rankings 49 6.6 Comparison of Methods 5 1 6.7 Determinants of WTP Premium for Quality Seed 52 7. Discussion 55 55 7.2 Limitations of the Study 59 v 8. Conclusion 61 8 . Implications for Future WTP Studies 6 1 A PPENDI CES APPENDIX A : Field Day 1: Farmer Ranking Sheet 6 4 APPENDIX B : Field Day 2 : Farmer Evaluation Sheet 6 5 APPENDIX C : Script for Seed WTP Experiment - Nicaragua 2017 6 6 APPENDIX D: Survey 7 8 APPENDIX E : BDM Bidding Sheet (Spanish) 8 4 APPENDIX F : Choice Tasks in RCE 8 5 APPENDIX G : RPL with Error Component Results 8 6 WORKS CITED 8 7 vi i LIST OF TABLES Table 4.1 Choice Experiment Attributes and Level s 3 2 Table 5.1 Sample Size by Villages 3 6 Table 5.2 T - test for Attrition 37 Table 5.3 Descriptive Statistics 39 Table 6.1 Blind Experimental Plot Rankings 4 1 Table 6.2 Best and Worst Ranked Plots by Village 4 1 Table 6.3 Plot Performance Data 4 2 rom BDM Experiments 4 3 Table Different Seed Types: Results of Fixed Effect Regression 4 4 Table 6.6 Mean WTP for Different Seed Types Based on RCE: RPL Model Results 4 5 Table 6. 7 Stated WTP Relative to Seed 3 Based on RCE: Result Disaggregated by Villages 4 7 Table 6.8 Comparison of Mean Premiums Relative to Seed 3 Derived from RCE: Fixed Effects Model Results 47 Table 6. 9 WTP for Seed 1 and Seed 2 Relative to Seed 3: Comparison of Results of RCE a nd BDM 49 Table 6.10 Differences in Preference Rankings by Seed Type Based on BDM and RCE Experiments 5 0 Table 6.1 1 Preference Ranking Compared with Plot Ranking 5 0 Table 6.1 2 Comparison of Estimated Premiums for Seed 1 and 2 between RCE and BDM (base method): FE Regression Results 5 2 Table 6.13 Determinants of Differences in Premiums Estimated from Two Methods (BDM and RCE) 5 4 vi i Table F .1 Choice Tasks in RCE 84 Table G .1 RPL with Error Comp onent Results 85 vii i LIST OF FIGURES Figure 4.1 Map of Nicaragua 2 4 Figure 4.2 Demonstration Plots 2 6 Figure 7.1 Premiums for Seed Type by Method 57 Figure 7.2 WTP for Best Quality 58 iv KEY TO ABBREVIATIONS BDM Becker - De Groot - Marschak C$ - Nicaraguan Cordoba (C$30 = USD$1) CIAT International Center for Tropical Agriculture CSB Community Seed Banks DGS General Directorate of Seeds HH - Household INTA Nicaraguan National Institute of Agriculture Technology KM - Kil ometers MAGFOR Ministry of Agriculture and Forestry QDS Quality Declared Seed RCE Real Choice Experiment RPL Random Parameter Logit WTA Willingness To Accept WTP Willingness To Pay 1 1. Introduction 1.1 Problem Statement and Research Focus As pop ulations in developing countries have increased, so has the need to intensify food production to meet the growing demand. Improving the quality of seeds and the frequency at which farmers purchase these seeds can help farmers intensify production without t he need to learn a new technology. Farmers in developing countries obtain seeds from both the formal seed system and the informal seed system. The formal seed system is defined by a vertically organized production of different generation of seeds, and dist ribution of commercial seed using a strict level of quality control (Douglas, 1980). The informal seed system, which is the most common source of seed in developing countries, consists of farmer - produced grain exchanged between farmers and recycled as seed or purchased from grain traders and vendors in the market for the purpose of planting it as seed. The formal seed system aims to produce high quality seeds of improved varieties. High quality seeds are characterized by higher germination rate, higher pur ity, and higher resistance to pest and diseases, which leads to higher yields. This is achieved by training seed producers in the proper agronomic practices, crop maintenance, and seed storing techniques to reduce diseases, and increase purity and germinat ion rates. But this also increases the cost of seed produced by the formal system relative to the grain price. Thus, farmers mostly obtain seed from the informal seed system where the price differential between seed and grain is close to zero. Reusing g rain as seed leads to lower yields as the quality of the seed declines after each harvest. Crops may intermix with seed of other varieties leading to deterioration of the original varietal traits as well as loss of germination potential from improper stora ge practices. The purchase of new seed of a given variety from the formal system as a replacement of recycled 2 grain of the same variety can be seen as a technology adoption behavior . The decision to purchase new seed is based on the expected utility of pro fits and the deterioration in the and Brennan 1991). Theoretically, farmers will purchase new seed when expected profits are higher than the recycled seed for a given time period. Quality seeds have a significantly lower adoption rate than similar technologies like improved variety due largely to the competition of grain being used as seed. Self - pollinated crops, such as legumes, are at lower risk of losing the genetic trait s of improved varieties (Rubyogo 2007). This is because self - pollination does not allow for cross breeding from other varieties and thus has low risk of genetic deterioration . However, legume crops are still susceptible to reduced germination potential cau sed by improper post - harvest handling and storage , and are more prone to seed borne diseases. These deteriorations increase the need for replenishing the seed stock with new seeds to restore yield potential for small holder farmers. The informal seed sect or is estimated to supply more than 90% of all the legume seed s planted by farmers in developing countries (Rubyogo 2007). With such high use of seed s from the informal seed sector, the need to increase the quality of seed s used by farmers is very high. A major factor in increasing the overall quality of seeds is increasing the volume and frequency of quality seed purchased by farmers through the formal seed sector. Two significant problems with formal seed systems in developing countries are unreliable and inconsistent quality seeds and high prices, making the risk of purchasing new seeds too high for many farmers. Individuals are unable to observe the quality of seed s before planting. Therefore, a third - party certification is needed to create standard prac tices across seed producers to ensure more consistent quality and create trust between producers. A trusted third - party certifier could reduce the risk of purchasing 3 new seed. There is a large disconnect between the demand and supply of quality seeds in de information on quality seeds, unknown demand for quality seeds making larger quantity production too risky, lack of trust in quality seeds, and other problems of asym metric information. These disconnections and market failure lead to high prices and higher risk of adoption of quality seeds. Market studies to measure farmers willingness to pay (WTP) for higher quality seeds can help better understand how consumers valu e different products and make decisions. Such information can in turn help producers make production decisions and in in a developing country context, and may lead to vastly different estimates of WTP. This research was undertaken to highlight the potential issues of implementing WTP studies in a developing country context by comparing two methods of eliciting Nicaraguan ies of seed of a red bean variety. Farmers in this study state their WTP for three different qualities of seed based on three experimental plots grown in each village. The plots were labeled with the shapes Square (QDS), Triangle (recycled grain), and Circ le (certified). Therefore, Farmers are unaware of the names of the three qualities and the variety used. The two preference elicitation methods compared and evaluated are: the Becker - DeGroot - Marschak (BDM) auction and the Real Choice Experiment (RCE ). Theo retically both methods will result in equal or statistically equivalent results. Many studies focus on between individual comparison of methods, however I conduct a within individual comparison to directly compare the impact of method on estimated WTP. I b riefly discuss the bean seed system in Nicaragua and describe the three seed quality types evaluated in this study, before I outline the research objectives and research questions addressed by this study. 4 1.2 Bean Seed System in Nicaragua Dry beans are the second most important crop after maize in Nicaragua . Dry bean was the primary crop on 24 % of the average 1 million hectares planted annually from 2000 to 2016 in Nicaragua. This makes dry bean the second largest crop in Nicaragua behind maize (34%) in terms of area planted (FAOSTAT, 2018) . Increasing bean yields for farmers therefore has the of seeds that farmers plant can rai se yields, reduce the impact of pests, and potentially increase profits. Yet many farmers do not have access or knowledge of the quality of seeds available due to incomplete or nonexistent seed markets. At the same time, many seed producers are not aware o seeds and uncertainty in prices. In this study, I same variety ( INTA Ferroso ). The three se ed quality types evaluated are recycled grain, Quality Declared Seed ( QDS ) , and certified seed, and they represent the informal, semi - formal, and formal seed sector in Nicaragua, respectively. Recycled grain is produced by farmers with no technical supervi sion or quality assurance and is perceived to have the lowest quality and price. QDS and c ertified seeds are grown using agronomic and processing practices to ensure a higher quality seed. But only c fication agency perceived to have the highest quality and has the highest price. QDS seed (referred in Nicaragua as Apta seed) is mostly produced by village based farmer groups for use within the community. To meet the quality standards, certified seed must be produced under the supervision of authorized experts to maintain high levels of seed health, vigor, and genetic purity (Bash, 5 Bowman, Cha pman, & Blandon, 2002) . In Nicaragua all the seed that is imported or produced by seed companies is certified by the DGS. Certification is a guarantee by the Ministry of Agriculture of high germination, uniformity of genetics, and that seeds are free fr om seed - borne diseases. This is achieved through proper agronomics , inputs, and proper post - harvest processing and storage techniques. Farmers can purchase certified seeds from rural retailers. These rural retailers are small retail outlets in rural commun ities that sell production inputs such as fertilizer, pesticide, and seed. Their locations tend to be in close proximity to bus stations and markets. Rural retailers have a high marketing potential to small farmers through personal relationships, however t hey are often very small retailers with little or no access to credit. This greatly hinders the ability of small retailers to purchase and sell certified seed s . The lack of availability and access to certified seeds has resulted in many farmers not adoptin g this type of quality seed (MAGFOR 2009, Sain 2011, Carter el al. 2012). To increase the use of higher quality seed, in recent years there has been increasing efforts by the government to promote Community Seed Banks (CSB) to produce Apta seed or QDS t o meet the seed needs of the community. QDS are not certified by the Ministry of Agriculture and cannot be declared as such. The seeds are produced by farmers without the supervision from the Ministry of Agriculture, however the farmers have been given tec hnical guidance and training from the National Institute of Agricultural Technology (INTA) to produce seeds of higher quality (DeYoung 2015). Since these seeds are produced within communities, the ability to reach many small farmers is significantly higher than certified seed. The use of CSBs allow communities to have access to higher quality seeds of varieties that grow best in their community. 6 Finally, most farmers in Nicaragua use their own saved seed, trade seed with other farmers, or purchase from the grain market. Trading of bean seed allows farmers more access to different quality and variety of seeds at price closer to grain price. Farmers often use their own harvest as seed for planting which causes the demand for fresh seed (which is produced as their own seed while in Costa Rica and Honduras 79% and 58% respectively of bean farmers used their own seed. There is also trade among other farmers to meet deman d that farmers own saved seed could not suffice (Tripp 1997, Bentley et al. 2011). This trading among farmers increases the use of different varieties, but often these seeds are of lower quality and the variety may be nonuniform or unknown. The use of low quality seed leads to lower yields and higher rates of fertilizer and other inputs to combat poor germination rates and to control for seed borne diseases. Some of the constraints leading to low adoption of quality legume seeds in Nicaragua are : i . ) low access, ii . ) lack of knowledge on the benefits of higher quality seeds, and iii . ) lower varietal deterioration of legume seeds because of self - pollination (Rubyogo 2007). On the supply side, the lack of availability of quality legume seed is the low perce ived profitability of seed production. Public sector seed agencies are often characterized by large overhead costs and inefficiencies and lack the capacity to provide an oversight to the production of large quantity of quality seed. The informal seed syste m does not have the capacity to produce higher quality legume seeds. As mentioned before, farmers base their seed input decisions based on perceived quality of the seed and the total price of obtaining that seed. Higher quality seeds are often indistingui shable from seeds of the same variety and lower quality. This creates an issue with 7 asymmetric information that Nicaragua has attempted to reduce by certifying seeds that fit the requirements set by the Ministry of Agriculture. Farmers therefore must have experience with certified seed in order to assess its quality and production capabilities relative to their own saved seed. This introduces many constraints for farmers such as cash constraints and access constraints. E xperimenting with certified seeds or QDS can be too risky for cash constraint farmers. Moreover, farmers may be unaware of the availability of these higher quality seeds. This lack of knowledge causes many farmers to never participate in the formal seed sector reducing the overall demand for quality seeds. These constraints on the demand side, in turn lead seed producers to be unaware of the potential demand for quality seeds . Thus, market research is needed to bridge this potential gap in demand and supply of quality seeds. E stimat ing the po tential premium farmers are willing to pay for higher quality legume seeds (if available) is a first step in bridging this knowledge gap . The problem with asymmetric information in seed markets creates an opportunit y for economist s to conduct market resear ch using preference elicitation methods to construct market demand curves. However, it is not certain that all preference elicitation methods are equivalent and provide same kind of information. This lack of certainty motivates this study which is designed to compare individual stat ed WTP in the context of an experimental auction and a choice experiment. This study took place in ten villages in four different departments located in the northwestern region of Nicaragua. 1 One farmer in each village was sele cted to grow three demonstration plots (one for each seed type) using their own crop management practices. The plots were established side by side to minimize any effects that soil quality would have on plant 1 Originally, I had chosen 12 villages, however 2 villages were taken out due to late planting or destruction of fields by pests. 8 development. These were blind experiments, whe reby the host farmer and the participating farmers were not told the quality type of the seed but only that the three seed types were of the same bean variety (the name of the variety was also not revealed). The seed and plots were labeled with symbols -- biases from preconceived perceptions that may influence the farmer s preference and WTP as well as the host farmer managing the three plots differently. The research objective of th is study is to show the impact a preference elicitation method can have on estimating an individual WTP through a within individual comparison of a BDM mechanism and an RCE . Two methods compared are both non - hypothetical to reduc e the impact of hypothet ical bias that may not be consistent or have the same magnitude of impact on WTP estimate for both methods. By measuring within - individual differences in WTP I can reduce the impact that individual personality traits (unobservable characteristics) have o n choice and auction behavior and reduces selection bias. Furthermore, using the data from this study I can deduce whether farmers are willing to pay a premium for higher quality seed based on the actual performance of the seed giving seed producers vital market information for the demand of quality seed. My thesis is organized as follow. In C hapter 2 , I present a review of the literature followed by the conceptual framework in Chapter 3. In Chapter 4 , I discuss the experimental design and the sources of t he data. In Chapter 5 , I discuss the sources and give descriptive statistics followed by the analysis of the experiment in Chapter 6. In Chapter 7, I discuss the experiment outcomes and possible shortcomings followed by my conclusion s on major findings in Chapter 8. 9 2. Review of the Literature 2.1 Determinants of Willingness to P ay for Q uality S eed The push for agricultural development in many developing countries has come with the push for increasing technology adoption and intensifying production to ra ise productivity and profits. There are many studies on how farmers decide to adopt technology (Feder et al 1985; Munshi 2004; Conley and Udry , 2010), but only a few capture farmers WTP for new technology ( Magnan et al., 2015 ). . Building on the agricultural technology adoption literature, I know that farmers experiment with new technology, and learn the optimum input bundles to maximize profits over time (Foster and Rosenzweig, 1995; Bardhan and Udry 1999). Therefore, farmers they know that it is more profitable. It is important to know how much a farmer is willing to pay for agricultural technology and what are the determinants of their willingness to pay to develop strategic policy for efficient technology adoption. Level and willingness to pay (Fay and Deininger 2005, Holden and Shiferaw 2002). Specifically, nonfarm al technology as income from non - farm sources may reduce liquidity constraints to access inputs, labor and other materials needed to use a new technology (Pender and Kerr 1998; Hondel and Shiferaw 2002). Also, , it reduces the risk associated with adopting a new process new information (Hol den and Shiferaw 2002, Oladele 2008). Education may increase 10 Kerr 1998). Other factors are age and family size. Age may have a mixed effect, as older farmers tend to have more experience, younger farmers have a longer planning horizon which may cause them to be more likely to adopt technology (Holden and Shiferaw 2002). Agricul tural households have a labor endowment based on the size of their household meaning as more kids come to age there is more labor available within the household possibly increasing the likelihood of adopting labor intensive technologies. The expected deter pay for quality seeds would be education, age, knowledge of this technology, and income. Household size should not factor into the adoption of quality seed as the labor requirements are equivalent with recycled grain. Howe ver, in reality farmers may equate quality seed with a new variety (i.e., genetics), and turn from buying them because of additional labor and cash investments needed to realize the full benefits of new seed variety. 2.2 Comparison of Preference Elicitatio n Methods for a product or service, such as experimental auctions, choice experiments, or contingent valuations. Seed market data in developing countries are often una vailable for the seed producers or economists to analyze. Preference elicitation methods are growing in popularity for estimating preferences for goods with unknown market value such as environmental quality or for market goods such as improved qualities o f food. For both choice experiments and experimental auctions there is evidence supporting the validity of each method (Brookshire, Coursey, and Schulze, 1987; Louviere, Hensher, and Swait,2000). However, there are very few studies comparing the two types of methods ( e.g., Lusk and Schroeder 2006, Ginon et al . 2011, Stuer et. al . 2017) and even fewer comparing them in a developing country setting (e.g , DeGroote et al. 2010). Determining whether there are differences in preference elicitation methods and 11 und erstanding why there may be differences is important not only for methodological advancement and accurate interpretation of results; it is important to know what is causing the differences and understanding behaviors when incentivizing accurat e stated WTP. Each preference method estimates individual preferences in a different way. For example, in a BDM mechanism, individuals state their willingness to pay for a product. Then a random price is drawn, and the individual pays the random price if t heir willingness to pay is above that random price. However, in a choice experiment, individuals make choices through a number of different scenarios. In choice experiment, the experimenter must also put a functional form on the y function and for the stochastic nature of random utility. Therefore, any differences in stated WTP between methods may be contributed in part to the functional form the researcher uses to estimate an individual s utility function. In theory, both choice to pay and lead to equivalent conclusions. However, in practice it has proven much more difficult to replicate those results (Lusk and Schroeder 2006, Ginon e t al . 2011, Stuer et al . 2017) and understanding why that may be the case can lead to more precise designs of both experimental auctions and choice experiments. There have been very few papers that measure the differences in estimation of individual will ingness to pay gathered by different elicitation methods. Lusk and Schroeder (2006) were steaks using an auction and a real choice experiment in the United Stat es. They concluded that auction behavior was significantly different than choice behavior (willingness to pay in choice experiment were significantly higher than in auctions), and auction bids predicted a higher frequency of opt - out choices than what occur red. They state that there are two possible 12 conclusions -- either one or both mechanisms are not incentive compatible, and/or preferences are different a priori or constructed differently in the two mechanisms. Lusk and Schroeder suggest that in auctions , in dividuals may have a non - expected utility function (Horowitz 2005), or that in a BDM mechanism the risk of deviation from a utility maximizing bid is much too weak to reinforce the optimal strategy (Harrison 1989; Lusk, Alexander, and Rousu 2006.) They als o argue that since individuals state WTP directly in dollar amounts for auctions (dollar - space) and in choice experiments individuals choose between a number of different options in a set of choice tasks (choice - space), which options they would prefer to p urchase may drive the differences between elicited WTP. Lusk and Schroeder used a between - individuals approach, however Ginon et. al (2014) used a within - individual willingness to pay for French baguettes allowing them to investigate individual - level beha viors using a real choice experiment and a BDM mechanism. They found that WTP were not equivalent between methods but were also not consistent with their reservation prices. They hypothesize that individuals do not have consistent preferences and may have a different preference structure in different settings. 2.3 Preference Elicitation Method Comparison in Developing Countries The studies mentioned above (i.e., Lusk and Schroeder 2006, Ginon et. al 2014) were conducted in developed countries. As explaine d before , individuals in developing countries face different constraints than in developed countries. Therefore, these studies may not have external validity outside of developed countries. DeGroote et. al . (2010) compared a BDM mechanism, kth price auctio - based food product kenkey with and without biofortification information in Ghana. They found that the WTP elicited from choice experiment were comparable to stated WTP from auction s. They found this 13 by accounting for lexicographic preferences and censoring bids for both auctions. Villages had higher premiums for kenkey that was the same color that they also grew and had much lower WTP for other colors. Farmers that were told of the health benefits of beta carotene enriched maize increased their WTP in all methods. This shows that preference ranking may be similar across methods but may not be similar in magnitude. Hans De Steur et al. (2017) did a meta - regression on the determinants of willingness to pay for biofortified crops and food . His independent variables were types of respondent s (student or adult), method of data collection, whether the WTP was hypothetical or non - hypothetical, method of value elicitation, study environment , participation fee, and if information was given or not, and the type of information. This study found that the type of information, nature of method used (i.e., hypothetical or non - hypothetical ) , and the method of value elicitation had a statistically si gnificant impact on individuals reported WTP for biofortified products. However, WTP for stated value - elicitation methods were not significantly different from was that the impact of the nature of the study (hypoth etical or non - hypothetical) had a much larger impact on the estimated WTP than the method used. This current research builds on the recent trend found in the literature of comparing preference elicitation methods by focusing on two mechanisms -- BDM and rea l choice experiment. The research aims to contribute to our understanding of what factors could possibly explain any experiment . Specifically, this study will e xamine differences in stated WTP for quality seeds from two commonly used methods -- a non - hypothetical BDM mechanism and a non - hypothetical or real choice experiment. By comparing two non - hypothetical elicitation methods I can rule out the impact of hypothe tical bias and focus on the behavior of the individual under two different 14 methods. The perceived quality of seed will be determined individually by the farmer through double - blind experimental plots and therefore the WTP should not be influenced by a prio ri knowledge or personal experiences with the seeds used in the experimental plots . This also provides a unique opportunity to compare two methods of individual WTP for three non - branded attribute bundles , which is novel and a contribution to the literatur e. 15 3 . Conceptual Framework 3.1 Technology Adoption Decision Framework Smallholder farmers in developing countries face many constraints that are (just like the farmers) heterogenous. These constraints are not always observed by the researcher, but influen seed can be viewed as a technology adoption decision that farmers make each planting season. This means that farmers make the decision to use quality seeds each planting season based on whether it maximizes either utility or profitability depending on the characteristics of that farmer and the constraints that farmer faces. Agricultural households are both producers and consumers and their consumption are dependent on the ir production (Singh, Squire, Strauss 1986). The production decisions can be made independent of consumption (referred as separable household model) or production and consumption decisions are made simultaneously (referred as non - separable household model) . In sectors where two or more input markets (often labor and land) are incomplete or missing, farmers are suspected of making production and consumption decision simultaneously (i.e., non - separable household model assumption). Farmers in a non - separable h ousehold are utility maximizers and do not behave as profit maximizing producers. This can be seen by the decision of farmers to produce local varieties for personal consumption even though it may not be profitable to do so. In the context of technology a doption decision - making, there are usually two types of farmers -- risk averse and risk neutral. Risk averse individuals maximize expected utility of profit. 16 Therefore, farmers may not choose the bundle of inputs that maximize profits but maximize their util ity of the expected profits (Cameron, 1999), which can be expressed as: Risk neutral individuals maximize profit and choose the bundle of inputs that maximize their profi t as expressed by, experience or prior knowledge of a given technology (i.e., quality seed in the context of this study) . The main factors influencing adoption decisions by farmers are profitability, access, and knowledge of the new technology. If the price of the new technology is such that farmers cannot make a profit, or the risk is too high than they do not exper iment, and adoption and diffusion will suffer. Market studies help understand what price farmers are willing to pay for this added risk, and potential increase in yields. It is important to know how farmers behave in higher risk environments than in a rela tively low risk environment such as a hypothetical experiment. This may influence their WTP estimates or may not capture their true behavior. This study focuses only on non - hypothetical experiments to reduce any hypothetical bias and create a more realisti c purchasing scenario. I first discuss the theory underlying the WTP elicitation methodology, followed by the conceptual underpinning of the two elicitation mechanisms used in this study choice experiments and BDM auctions. 17 3.2 Theory of Demand The L ancaster theory of demand states that the intrinsic properties of a good define that good. This means that the utility an individual derives from a good is from the attributes contained within that good. Lancaster (1966) put forth an approach that suggeste d that consumers derive utility from the properties of the good and not the good itself. This allows for the demand of a good to be divided into many attributes of that good, and utility is a function of those attributes. Individuals will c hoose the attribute bundle that maximizes his or her utility function. For example, an individual may prefer the color black versus pink. Therefore, it would be assumed that, all else equal, this individual will derive more utility from a black car as oppo sed to a pink car. From this theory, I am able to base the design of our preference elicitation methods to establish WTP for three seed products that differ in two attributes quality and price. 3.3 Choice Experiment Choice Experiments are growing in popul arity as a tool to estimat e willingness to pay for marginal changes in attributes, measuring unknown market values, or estimating individual preference. Choice experiments are designed to replicate a purchasing scenario where individuals most commonly make their purchases. In choice experiments the researcher relies on the repeated choices to distinguish preference ranking and measure marginal utility. Through repeated responses to presented choice situations of dem bundle. There are many models that are used to estimate individual preferences that account for the heterogeneity in individual taste. The mixed logit model assumes a continuous distribution of heterogeneity in taste. Mixed logit assumes that individuals are inherently heterogenous in 18 nature, therefore their tastes and preferences are heterogeneous. Choice models allow the researcher to estimate marginal values for dif ferent attributes encompassed in several goods and services. Furthermore, choice modeling allows for estimations of welfare effects with marginal changes in attributes (Colombo, Hanley, & Louviere, 2009) . Choice experiments have been used widely in the agricultural and environmental economics literature for studying consumer preferences for environmental amenities ( List et al.,2006 ) , food safety attributes and food certification ( Oles e n et al., 20 10 ; Lusk et al., 2004 ) , and measuring farmers WTP for new technology (Ward et al., 2014; Magnan et al., 2015) . three different qualities of a red bean seed, Inta Ferosso . Choice experiments a re most often used in the context of consumer theory, but few studies have used choice experiments to analyze producer welfare effects and technology adoption (Ward et al. 2014). In this study I used the theory presented by Singh, Squire and Strauss (1986) in which producers that are risk adverse make input choices that maximize their utility. However, if farmers are not risk averse , th e n maximizing utility will be equivalent to maximizing profits . Therefore, a choice experiment should theoretically measure averse and risk neutral farmers. Also, a s mentioned before agricultural households that face more than one missing market are non - separable meaning production decisions and consumption dec isions are made simultaneously, I th us view the adoption of quality seeds as a utility maximizing decision. The reasoning for this is that non - separable agricultural households maximize utility of farm production by choosing a set of technology attributes or inputs among a set of obtainable attributes or inputs (Useche, Barhma, and Foltz,2012). 19 According to Random Utility Theory an individual i faces alternatives within a choice set during occasion t. Following the approach used in Ward et al. (2014), I then assume demand. This attribute bundle will be represented by that denotes the value function associated with individual i choosing option during occasion t . An individual, i, facing a fixed budget constraint will choose alternative m so long as . T he researcher cannot directly observe , but instead observes , through the choices the individual makes such that: (1) Where is the option individual is choosing with attribute bundle from the choice task . Using the assumption that indirect utility is linear, I can write individual i ect utility function as: (2) where is a vector of attributes for the m th alternative, is a vector of taste parameters, and is a stochastic component of utility that is independe ntly and identically distributed across individuals and alternative choices , and follows a Gumbel (extreme value type I) distribution with a cumulative distribution function and a probability density functio n . The probability of observing (i.e the farmer chooses option m in choice task t ) can be written as: (3) 20 (4) Then with the assumption made previously that the is identically and independently distributed, the expressio n for the probability of observing can be rewritten as: (5) Which is the basic conditional logit model which I can estimate using maximum likelihood. Since farmers are heterogeneous, their preferences are heterogeneous in nature . A mixed logit model accounts for continuous heterogeneity among individuals and approximates any random utility model and relaxes the limit ations of the traditional multinomial logit by allowing the taste parameters to vary by individual within a sample according to a pre - specified distribution (McFadden and Train, 2000) which is the random parameters logit (RPL). Train (2003) states that the probability that individual i chooses alternative m from the choice set in time or task t is given by the RPL: (7) Where the vector defines the distribution of the random parameters as pre - specified by the researcher. I used the RPL with an error component with utilities specified in WTP - space. The RPL is designed to account for continuous heterogene ity in the error component, which accounts for systematic differences between the experimentally designed alternatives and the status - quo option, beyond what is explained by the attributes (Scarpa et al. 2005). Therefore, the utility 21 that individual d erives from choosing option in choice tasks can be specified as the following: (8) Where is a random positive scalar representing th e price/scale parameter. is a vector of random parameters that are normally distributed as defined by the researcher and measures individual WTP. - each choice task. is the normally distributed error component . 3. 4 Becker - DeGroot - Marschak (BDM) Auction Experiments The BDM mechanism was first introduced by Becker, DeGroot, and Marschak (1964). This mechanism is designed to obtain individuals reservation p rice or willingness to pay for an item. This is obtained by creating a utility optimizing strategy that incentivizes an individual to state their WTP. The theory states that individuals have a monetary equivalent for any good which is when U(M)=U(G) or uti lity of the monetary equivalent (M) is equal to the utility of that good (G). This can be interpreted as how much money they are willing to forego or accept to obtain or sell a good. Individuals bid against a random price, and if their bid is greater than or equal to the random price they receive the good at the random price. If their bid is less than the random price they do not purchase the good. This creates two outcomes: Outcome 1: (9) Outcome 2: 22 Where R is the random price, B is the bid, M is the monetary equivalent of the good , and Y is the participation allowance. However, with these two outcomes the optimal strategy for a participant is to bid where . This will ensure that no matter R, the indi vidual will never receive a utility where . If than and the individual is worse off from winning. This can also be seen by the expected utility (11) Equation 11 s hows the importance of the distribution of the random price R. if R is not uniform ly distributed th e n the optimal strategy is no longer B=M, but instead it is to maximize utility based on the . However, if the utility maximizi ng strategy breaks down into . The distribution of the random price is known in advance by the participant and they are asked to state their maximum WTP or willingness to accept (WTA) for each item. It is assumed that this procedure is incentive compa tible (since deviating from optimal strategy can potentially lead to lower utility than a non - purchase). However, Horowitz (2005) shows that theoretically, since individuals face uncertainty of price, if they experience a non - expected utility function (suc h as disappointment aversion) individuals will not state their true WTP for that good, but instead it will be utility derived from attributes not associated with that good. Mazar et al . (2014) found that in within - individual study, when using two different random price distributions , the differences between the two bids were not significantly different, and that individuals may make mistakes when reporting true WTP when considering only one price distribution. Plott and Zeiler (2005) showed that under certa in controls the WTP - WTA gap can be diminished or completely removed. This shows that the individuals knowledge of the optimal - expected utility function. 23 3.5 Comparison o f Two Methods Since choice models are discrete and auctions are continuous, one of the two methods must , 259 - 281) formula to estimate individual level parameters in WTP space. This uses the nature of the choices in the choice set. Theoretically the individual conditional WTP parameters should be statistically similar to the stated WTP in the BDM mechanism. To test this, I will use an individual fixed effects model with bootstrapped standard errors clustered at the village to test whether the WTP stated in both methods are equivalent as suggested by theory. I assume that underlying utility function does not change between methods and is constant across the two which states that individuals purchase attribute bundles that maximize their utility. 24 4. Methods and Data 4.1 Study Site and Field Experiments This study took place in four Departments -- Jinotega, Esteli, Matagalpa, and Madriz located in the northwest region of Nicaragua, where beans are an important crop (Figure 4.1) . A total of 12 villages were sele cted based on whether they were average bean producing villages for that region and were served by a technician from INTA. Figure 4.1 Map of Nicaragua *Red circle is the location of our experiment Source: https://geology.com/world/nicaragua - satellite - image.shtml 25 based on the performance or perceived quality of three different seed types. To ensure that the performance of the seeds and not biased by his/her previous experiences with quality seeds, it was vital to not reveal the identity of seed types. Towards this goal, double - blind field experiments were set up in each village. One farmer from each selec ted village was chosen with the help of staff from INTA and CIAT to host the field experiments. The host farmers were selected mostly for their status as a village leader and their bean growing experience. Their relationship with the rest of the communit y The host farmers volunteered to grow the experimental plots during the Primera season (May - August 2017) using their own bean growing practices. Host farmers were asked to s et aside a part of their field to grow three types of bean seeds of the same variety in adjacent plots. Farmers were able to keep the harvested grain after the end of the experiment, which incentivized the farmer s to maintain the plots as he or she would m aintain their own. Fligner and Verducci (2012) show that in blind experiments individuals chose items seeds given to host farmers and the plots on which they were planted were identified by symbols of square (for QDS), triangle (for recycled seed), and circle (for certified seed) (see Figure 4.2 ). 26 Figure 4.2 Demonstration Plots 1. Triangle Plot 2. Square Plot. 3. Circle Plot. Source: Pictures take n by Researcher The field experiments were designed to observe the performance of each seed quality type (i.e., certified, QDS, and recycled) while controlling for genetics (all seeds were of the same variety), soil quality (all the plots were grown in adjacent plots to r educe the effect of soil quality and characteristics), and farmers management practices of the plots (the blind nature of the experiment reduced the incentive for farmers to systematically favor one plot over another). Bean growing farmers from the commun ity were invited to participate in two field days organized around flowering and close to harvesting stages. This allowed participating farmers to observe the performance of each quality seed in terms of plant characteristics during the flowering stage (fi eld day 1) and potential yield at maturity (field day 2). Research assistants from CIAT worked with the farmers to ensure that the plots were labeled correctly in each village and that farmers did not know the identity of the seed quality 27 type. The experi mental plots served as a direct comparison of performance of seed quality to plots represented by symbols of square, triangle and circle. The preference elicitatio n methods used were an incentivized non - hypothetical BDM mechanism and an RCE (explained in later sections). 4.2 Source of the Three Seed Types Used in the Experiment Certified Seed: This seed was purchased from a government organization that is license d to sell certified seed. This seed is grown by certified seed producers under a contract with the government organization . These seeds are considered to be the highest quality of seed available to farmers and is the most expensive of the three seed types used in this experiment. Only government agencies can certify seeds after the seed is determined to meet the quality standards. Certified seed is grown from registered seed maintained by the research program to reduce deterioration of genetic traits and is stored in a way to reduce seed borne diseases and to maintain seed germination rates. The certified seed used in this study was produced in Postrera season (October - December 2016) and represents first generation seed after registered seed. Apta s eed (QDS) : This type of seed was purchased from a CSB and is not regulated by any third - party agency. QDS is often grown by farmers who are associated with the CSB and trained by INTA seed producers. The quality of seeds may vary from CSB to CSB. They are not requi red to be grown from registered or certified seed, but are produced, processed and stored using similar practices as certified seed. This seed type is supposedly of higher quality than recycled grain. It is sold at a price higher than grain but lower than certified seed. The Apta seed 28 used in this study was produced in Postrera season (October - December 2016) and represents first generation seed as it was also produced from registered seed. Recycled seed : d that they replant from past harvests. This seed is not grown to any specifications and is often considered of low quality and of unknown genetic identity. However, recycled seed is the lowest cost seed as it is produced as grain rather than seed. T o ensu re that the variety was uniform across all three seed type s , it was not possible to procure this seed type from farmers . The recycled seed used in this study was purchased from the same CSB that produced the QDS. It was also produced in Postrera season (Oc tober - December 2016), but represents second generation seed after registered seed (or one generation older than the QDS and certified seed used in this study) . The seeds used in this study are a small sample of the overall seed system and therefore it cann ot be assumed that the performance of these seeds in our experimental plots are representative of the seed s available to farmers in Nicaragua. In particular, the recycled seed used in this study was produced by a seed producing organization and not a farme r, and represents just a second generation seed. Due to both these characteristics, the recycled seed may not be representative of the quality of seed that farmers usually plant. S ince these seeds are not meant to represent the seed system , I re fer to them as seed 1 (Certified seed), seed 2 (QDS), and seed 3 (recycled) in the remainder of this paper. Since all three of these seeds were of the same variety any differences in the performance of the seeds will be due to the quality of the seed and n ot the genetics. What I am interested in this study is how the observed performance of seeds of three different qualities planted in the field experiments is perceived by the farmers, and how that perception of quality is reflected in their WTP for each se ed type. 29 4.3 Field Days and Farmer Ranking of Seed Plots Two field days were organized in 10 out of the original 12 villages. Two villages were dropped either because farmers planted the plots too late or the crop was destroyed due to pest problem. Durin g field day 1 , farmers from the village met either at the experimental plots or at the farmer s house where they were informed about the purpose of the meeting. Each village chose three attributes to rank the performance of the three plots on an evaluation sheet given to each farmer with their ID number (Appendix 1 ) . These attributes varied across villages, but most common attributes were plant growth, resistance to diseases, resistance to too much rain (as Nicaragua was experiencing a more than usual level of precipitation that year), and amount of foliage. After they chose the three attributes to judge the plots, they were asked to pick the plot that was the best in each category and then chose overall best and worst plot. The first field day was conducted on average 32 days after planting , which corresponded to just before the flowering stage. Farmers were able to see the differences in the germination rates, development of the plants, and the foliage. This ranking mechanism captures which field farmers pe rceived to be the best and worst, but not the magnitude of differences. The second field day took place just before maturity when the bean pods had formed fully on the bean plants but had not fully dried out for harvest. Farmers could see the number of po ds each plant produced and the relative density of pods between plots Farmers were asked to complete the evaluation sheet that asked them to rank the plots on different characteristics, and an overall best and worst plot (Appendix 2 ) . 30 4.4 WTP Elicitatio n Experiments Following the field observation and ranking of the three plots on the second field day, individual farmers then participated in two elicitation experiments -- BDM mechanism and RCE. The experiments were facilitated by a research team member wh o was a native Spanish speaker. The script followed to explain the experiments and how the mechanisms work in included in Appendix 3. The BDM mechanism was always first to prevent price anchoring from the real choice experiment. The RCE was not a full orth ogonal design so not all possible combinations of prices and seeds were present. This may create a bias that one seed is more preferred than another if that seed comes up more often under a higher price. Therefore , to not bias individual s bid levels and p reference ranking in BDM the RCE was always implemented second. This may create a fatigue bias in the RCE results . However, this was considered to be less of a problem. By giving individuals both a choice experiment and a BDM mechanism I can compare within individual WTP estimated by both models. BDM experiments: Farmers were given C$40 or an item of similar value to participate in both the BDM mechanism and the real choice experiment. This amount represented about 18% more than the market price of one poun d of certified seed, which was the highest cost seed type among the three types used in the experiment. Farmers were told that at the end one of these exercises will be selected randomly, and one seed type (if BDM was selected) or choice set (if RCE was se lected) will be chosen as binding to determine whether and which seed type the farmer would end up purchasing. Thus, for each bid (in BDM) or choice set (in RCE), they were reminded that they had C$40 available to purchase a one lb. bag of seed of a given type and that 31 Before farmers bid on the seeds in the BDM mechanism, farmers participated in a practice round to get them familiar with the mechanism. Farmers were asked to bid on an item with a commonly known market price such as a bar of soap or a pen. Farmers were given an endowment of C$10 or an item of similar value to bid with the random price ranging from 0 to 9 by using a 10 - sided die. The practice bid was carried out to hel p farmers understand the BDM mechanism, how the random price is drawn, and how the purchase/no purchase decision is determined. Following the practice auction, farmers were given a bidding sheet (Appendix 4 ) and asked to bid on the three seed types that w ere planted in each experimental plot and reminded that each seed was of the same variety but different quality. The seed quality type was not revealed to the farmers and they were told to bid as if they were purchasing one lb. seed used in the experimenta l plots. The bids were between C$0 and C$39 (approximately US$0 - US$1.30) with a uniform probability of being C$0 or C$39 (1/40) . T his was to prevent farmers from choosing a utility maximizing strategy that is not consistent with the BDM mechanism that in creas es expected utility above and beyond the attributes of the goods farmers are bidding on. Real Choice Experiments: The choice tasks were designed following the method used by Street, Burgess, and Louviere (2005). The product attributes and correspondi ng levels were first used to develop an orthogonal fractional factorial design (Appendix 5) . Following that, the generators described by Street and Burgess (2007) were used to develop 12 choice tasks, each containing two product alternatives (price and see - option with a D - effic i ency of 96% . The price attribute had four levels ranging from C$14/lb. to C$34/lb (Table 4.1) . This is based on the market level prices for the highest cost c ertified seed (C$34/lb.) and lo west cost grain price ($14/lb.). 32 Table 4.1 Choice Experiment Attributes and Levels Attributes Levels Quality 3 Circle ( Seed 1 ); Square (Seed 2); Triangle ( Seed 3 ) Price 4 C$14; C$21; C$28; C$34 The design of the RCE only allowed for the comparison o f WTP based on premiums relative to another attribute. Therefore, we measured premiums relative to seed 3 in for each individual seed. Following the BDM mechanism , farmers were shown an example choice task to familiarize them with how to answer choice tasks and how to understand the options. This choice example and not a binding choice experiment or RCE and used as a way to explain a choice task. Farmers were then put into four similar sized groups where they participated in the RCE. Each group saw a different ordering of the same choice tasks. This is to reduce the systematic impact of fatigue on overall choice, as not all individuals saw the choice tasks in the same order. Farmers were told before beginning the choice experiment that there is an equal probability of a ny one of these choice sets to be selected as binding. This was to reconfirm that the utility maximizing strategy was to answer each question in the choice set truthfully. Farmers were told to treat each choice task as if they were purchasing a pound of se ed in the local market. Each choice task had three options, two were seeds and one no purchase option. Farmers were 33 encouraged to not discuss their answers among each other and were told to hide their answers once they finished selecting the choice task. A fter the end of RCE, o ne method was selected by flipping a coin to determine a binding option. If the choice experiment was chosen than a 12 - sided die was rolled to select the binding choice set. However, if the BDM was chosen, then the seed type was rando mly selected by asking one of the farmers to pick up one of the three cards on which the seed type was written. Then, two dice were used to determine the random price. The first die was a 4 - sided die with numbers 1 - 4, where 4 was treated as 0. The first di e chose the first number in the price, and the second die was the 10 - sided die used in the practice auction with the number 0 - 9. This gave the probabilities a uniform distribution with 0 - 39 having a 1/40 probability of being chosen. The differences in the three bids is interpreted as the premium (discount) a farmer is willing to pay (accept) to obtain a different quality attributes as observed or perceived by the individual farmer in the experimental plots. Purchases only happened after both methods were c ompleted to prevent influences of disappointment aversion. Having only one randomly binding method does not allow individual s utility curve to change between methods. For example, if an individual won seed 2 , they may not want to win a second pound of see d 2 , or if they won a seed type they were not excited about it may shift their preference to prevent such outcome from occurring again that may not be aligned with their true preferences. 34 4.5 Data In addition to the plot ranking data and WTP elicitatio n data from BDM and RCE described in the previous section, this study also uses farmer survey data collected from households that had participated on field day 2 (Appendix 6 ) . The survey was conducted using a structured questionnaire to obtain participatin knowledge of technology, and access to technology. For each experimental plots, data was also collected on t he harvested yields and inputs used by farmers. Plot size, soil quality, and weather variables ( e.g. , precipitation) during the season were also recorded by the INTA agronomist and CIAT researchers collaborating on the field experiments. Host farmers were told to record their input decisions and management practices in order to compare across all three plots and to help control for underlying farming objective data plots. 35 Chapter 5. Data Description 5.1 Analytical Sample A total of 222 farmers had participated in field day 1. However, t he data for this paper comes from farmers wh o attended field day 2 when the preference elicitation experiments were conducted. .Across all 10 villages a total of 219 farmers had participated in the second field day. However, i n the comparative analysis, farmers who chose one option for all 12 choice tasks (6 farmers) were dropped from the study. This is to prevent biases in reported WTP from individuals who may not have understood the mechanism or lacked interest in the experiment. In total, the sample size of farmers included in the analysis present ed in this paper is 213, and its distribution by villages is shown in table 5.1 . Regrettably, 19 farmers who participated in the experiments did not complete a survey. I have survey data for 194 farmers who participated in the preference elicitation expe riments. Therefore , all WTP data I present are from the 213 farmers who participated in the preference elicitation experiment and all the survey data consist of only 194 farmers. Since multiple members of the household could have participated in the field day 2 activities and in the preference elicitation experiments, the number of household level surveys (188) is less than the number of farmer sample (213) . Table 5. 1 also shows the distribution of household sample in each village. 36 Table 5.1 Sample S ize by V illages Out of 213 farmers included in my analysis, 125 farmers had participated in both the field days and 88 farmers had only participated in field day 2. During the preference elicitation experiments on field day 2 , farmers were told to bid based on their evaluations of the experimental plots they had just observed. Therefore, any differences in bidding or choice behavior based on their observation s should be attributed to individual taste and preference urve. To test the hypothesis that individuals WTP does not depend on whether they went to both field days , I used an unpaired t - test to do a mean comparison between farmers who went to both field days and farmers who only went to the second. I am assuming for this study that the differences in Village ID Village N ame Department Number of farmers Number of households A1 Santa Rosa Esteli 49 42 A2 Bramadero Esteli 10 9 A3 El Horno Esteli 11 10 A4 Matapalo Esteli 15 12 A5 Moropoto Madriz 15 13 A6 El Porcal Madriz 38 35 B2 Susul i Matagalpa 22 19 B3 Las Mesas Sur Matagalpa 21 20 B4 Ojo de Agua Matagalpa 16 15 B6 La Chichigua La Concordia 16 13 Total 213 188 37 WTP from two methods is not affected by whether individuals participated on both field days or only the second. The hypothesis being tested is Table 5. 2 T - test for Attrition Seed Quality Mean Differences P - Value Seed 1 - 2.74 0.30 Seed 2 2.45 0.47 Source: Primary Data Collected by Researcher 2017 Table 5. 2 shows that there is no statistically significant difference between stated WTP that attended both field days or just the second field day. Therefore , I assume that the level of attrition should not be a factor driving my results. 5. 2 Sa mple Characteristics Table 5.3 provides descriptive statistics of the household and farmer characteristics of my analytical sample. The survey was designed to collect household demographic and other socio - and experience with quality legume seeds. As reported in Table 5. 3 92% of participant farmers belonged to male headed households. On average, the age of the head of the household was approximately 46 years, and the average years of education for the head of household was 5 years. F ive percent of farmers stated that they regularly purchase bean seed for grain production. This is consistent with the low level of 38 involvement in the formal sector reported by previous studies. For example, Wierma et al, (1993) reported that 72% of bean farmers in Nicaragua used saved seed. The average reported highest price paid for seed in our sample was C$ 21.23 per pound (approximated US$0.70/ lb. ). About 75 % of farmers claimed to know or have heard of CSBs, however only 54% reported knowledge or awareness of quality declared seed (or Apta seed), which is the type of seed produced by CSBs. A reason for this low awareness or recognition of Apta seed could be that CSBs are not labelling their seeds or promoting it as Apta , as n from farmers response to the question -- what was the seed type they last purchased. Twenty eight percent of farmers reported certified seed, 41% stated they were u nlabeled seed, and only 5% claimed they purchased Apta (QDS) seed , and the rest of the farmers unsure of the seed type. About 31% of farmers have no knowledge or have not heard of either QDS or certified seed, the other 69% were aware of these higher qual ity seed types. This shows that most farmers are aware of the quality seed technology but either choose to not purchase these seeds or they are not available for them to purchase. On average farmers in our sample sold approximately half of their bean har vest annually, and about half of their total income came from bean sales. A major share of bean production (close to 50%) is retained for home consumption. This could be a factor for why 72% of farmers do not purchase new seeds from the formal system, whi ch are improved varieties, and may not 39 Table 5.3 Descriptive Statistics VARIABLES N Mean SD Gender of Individual Surveyed (Male=1) 194.00 0.84 Age of Individual Surveyed 194.00 4 3.46 15.09 Years of Education of Individual Surveyed 193.00 5.12 4.38 Head of Household (HH) Gender (Male=1) 194.00 0.91 Head of HH Age 194.00 45.78 14.68 Head of HH Years of Education 192.00 5.11 6.82 Total Members in HH 194.00 4.66 1.95 Males in H H 194.00 2.49 1.27 Females in HH 193.00 2.20 1.19 Have you ever produced bean seed (1=Yes) 193.00 0.29 Heard of CSB (1=Yes) 194.00 0.76 Heard of QDS (1=Yes) 194.00 0.56 Heard of Certified Seed (1=Yes) 194.00 0.70 Heard of INTA (1=Yes) 194.00 0.94 Do you belong to a farmer group/ organization? (1=Yes) 194.00 0.30 Are you a leader of these groups? (1=Yes) 60.00 0.30 Does this organization produce/distribute seeds of any crops? (1=Yes) 60.00 0.82 Distance to paved road from house (in km) 194. 00 6.61 16.20 Distance to nearest road marker from house (in km) 194.00 15.70 17.13 Do you regularly purchase or have you ever purchased bean seed (1=Yes) 192.00 0.28 Highest price per pound you have ever paid for bean seed for grain (C$) 76.00 22.41 5 5.87 Last time you purchased bean seed for grain production (year) 76.00 2,013 4.45 Price per pound you paid for acquiring this seed last time? (C$) 76.00 20.83 55.96 Total quantity of seed purchased last time? (lbs.) 76.00 77.39 72.36 In a normal year , what percent (%) of your bean harvest do you sell? 194.00 50.94 22.63 Annual percent of income from beans 193.00 50.73 28.78 Total amount of land area owned by your HH? (in Manzanas) 193.00 5.35 9.29 What was the total land area in all plots cultivate d (in Manzanas) 193.00 3.99 6.14 Number of fields cultivated by the household 193.00 1.98 0.97 Number of fields planted with beans in the last agricultural year 193.00 1.93 6.28 Did you plant bean validation fields the last agricultural year? (1=Yes) 19 4.00 0.06 Likelihood of being poor at 100% Nicaraguan National Poverty Line 194.00 47.60 30.25 Likelihood of being poor at 150% Nicaraguan National Poverty Line 194.00 76.16 25.53 1 if Farmer states they are an early adopter 194.00 0.67 Source: Prima ry Data Collected by Researcher 40 Chapter 6. Results 6.1 Field Day Rankings Table 6.1 shows overall ranking of the experimental plots across all 10 villages. Farmers were unaware of the quality type of the seed as the plots were labeled ( seed 2 ), and ( seed 3 ). In all the data in the following tables we refer to these seed types as seed 1, seed 2 , and seed 3 as their identities (certified, QDS, and recycled) were unknown to the farmer and instead they bid (BDM) , cho o se (RCE), and ranked the symbols (or seed types 1, 2, and 3) and not the seed quality names (certified, QDS and recycled) . In the first field day 58% of farmers ranked the plot planted with seed 2 as the best plot, 29% ranked the plot with seed 3 as the best, and 13% ranked the plot planted with seed 1 as the best. Approximately 72.5% of farmers chose seed 1 as the worst plot with seed 3 and seed 2 being chosen by 14% and 11% farmers , respectively. In the second field day the rankings changed in favor of seed 2. About 79% of farmers chose seed 2 as the preferred plot, 12% chose seed 1, and 9% chose seed 3 as the best plot. Relative to first field day, seed 1 was rated as the worst plot by fewer farmers in the second field day. However, 56% still chose seed 1 as the worst plot on the second field day, significantly more than for seed 3 (38%) or seed 1 plot (7%). 41 Table 6.1 Blind Experimental Plot Rankings Seed Type First Field Day Second Field Day % That Chose Best Plot (N=222) % That Chose Worst P lot (N=222) % That Chose Best Plot (N=213) % That Chose Worst Plot (N=213) Seed 1 13% 72% 12% 56% Seed 2 58% 11% 79% 7% Seed 3 29% 14% 9% 38% Source: Primary Data Collected by Researcher Given the large variation in participation by villag e the average numbers reported in Table 6.1 may be biased towards villages where there was large participation. Table 6.2 shows which plot was ranked the best and the worst in each village. Table 6. 2 Best and Worst Ranked Plots by Village Village ID First Field Day Second Field Day Best Worst Best Worst A1 Seed 2 Seed 1 Seed 2 Seed 3 A2 Seed 2 Seed 1 Seed 2 Seed 3 A3 Seed 1 Seed 2 & Seed 3 Seed 2 Seed 3 A4 Seed 2 Seed 1 Seed 2 Seed 1 A5 Seed 2 Seed 1 Seed 2 Seed 3 A6 Seed 2 Seed 1 Seed 2 Seed 1 B2 Seed 3 Seed 1 Seed 2 Seed 1 B3 Seed 3 Seed 1 Seed 2 Seed 1 B4 Seed 1 Seed 2 Seed 1 Seed 2 & Seed 3 B6 Seed 3 Seed 1 Seed 2 Seed 3 Source: Primary Data Collected by Researcher As shown here in the first field day seed 2 and seed 3 were the two best l iked , and seed 1 was overwhelmingly disliked. In the second field day seed 2 is ranked best in every village but one, however seed 3 is now ranked worst in every village but four. 42 Researchers had also collected agronomic performance data from the experim ental plots such as yield, number of pods per plant, and number of seeds in pod. This data gives us objective data on the quality of seeds planted as reflected in these plant performance indicators and how y and WTP. Based on Table 6.3 seed 2 plots had more pods per plant, more se eds per pod and higher average yield than both seed 1 and seed 3 . On average, across all 10 experimental plots, a bean plant s from seed 1 produced 53.6 seeds per plant, plant s from seed 2 produced 61.3 seeds per plant, and plant s from seed 3 produced 52.2 seeds per plant. Similar relative pattern is observed for bean yields across the three seed type plots. The relative difference in these objective measures of plant performance by seed type are well - based on subjecti ve measures of performance of the three plots. This gives insight on how farmers perceive quality of bean plants , and how their perceptions match objective measures of seed quality in demonstration plots. Table 6.3 Plot Performance Data Seed Type Pods per Plant Seeds per Pod Average Yield in kg/ha Average S.D Average S.D Average S.D Seed 1 9.85 a 2.63 5.53 a 0 .37 1336.5 a 798.09 Seed 2 10.40 a b 2.99 5.71 ab 0 .51 1542 a 696.44 Seed 3 10.51 b 2.67 5.38 b 0 .30 1431 b 727.15 a = not significantly different than seed 3 at 10% level b = not significantly different than seed 1 at 10% level Source: CIAT Yield data Nicaragua 2017 43 on BDM mechanism The benefits of using an experimental auction such as the BDM mechanism is that needed or parameterization by the researcher that may mis interpret individual WTP. In this mechanism the individual explicitly states their WTP for different seed projects that reveals their preference ranking. I types to match the best and worst plots. In other words, the highest bid would be for the perceived highest performing plot and the lowest bid would be for the perceived lowest performing plot. Table 6. 4 shows results of stated WTP for different seed plots by villages based on th e BDM mechanism. Table 6. 4 S tated WTP: Results from BDM E xperiments Village ID Seed 1 Seed 2 Seed 3 A1 C$18.57 C$22.33 C$14.67 A2 C$21.50 C$23.20 C$18.70 A3 C$10.09 C$18.36 C$9.64 A4 C$21.00 C$27.13 C$23.07 A5 C$14.73 C$18.80 C$18.53 A6 C$13.05 C$19.53 C$12.66 B2 C$17.32 C$25.09 C$19.82 B3 C$14.81 C$19.38 C$16.33 B4 C$18 .19 C$16.13 C$15.13 B6 C$25.38 C$27.31 C$25.38 Total C$17.17 C$21.66 C$16.64 Source: Primary Data Collected by Researcher In every village except B4, farmers stated WTP was highest for seed 2 , which was also the highest ranked plot. This shows that individuals are willing to pay a premium for the perceived quality of seed, irrespective of what is the market signal in terms of what it is called and the 44 underlying belief system (i.e., seed 1 are considered the highest quality and seed 3 the lowest qua lity seed). In all villages except A4, A5, B2, and B3 farmers had lowest WTP for seed 3 . However, when all the villages are aggregated, the average WTP is C$21.66 for seed 2 , C$17.17 for seed 1 , and C$16.64 for seed 3 . This shows a preference ranking on av erage for seed 2 and possibly seed 1 over seed 3 . However, as seen in Table 6. 5 when testing for statistical significance using a fixed effect regression with clustered standard errors at the village level I WTP for seed 2 is statistical ly significant ly different from seed 3 (first column) , also from seed 2 (column 2) . Table 6.5 Testing the M ean D ifference in F B ids for D ifferent S eed T ypes: Results of F ixed E ffect s R egression VARIABLES Mean WTP relative to seed 3 (C$/ lb. ) Mean WT P relative to seed 1 (C$/ lb. ) Seed 1 0.53 -- (0.94) Seed 2 5.02*** 4.48*** (0.92) (0.82) Seed 3 -- - 0.53 (0.94) Mean of excluded s eed type 16.64 17.17 Observations 639 639 Number of farmers 213 213 R - squared 0.255 0.255 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Source: Primary Data Collected by Researcher Based on the results presented in Table 6.5 , I conclude that in the BDM mechanism individual farmers on average are willing to pay a statistically significant premium for highest 45 ranked seed 2 relative to lowest ranked seed 3 and second lowest ranked seed 1. B ut there is no statistically significant premium/discount between the second (seed 1) and third (seed 3) ranked seed types . t quality seed based on Real Choice Experiments The stated WTP in the BDM mechanism does not have to be parameterized and is exactly the number that the individual states on their bids. However, for the RCE, I use a n RPL with an error component to estimat e WTP using random utility theory and derive individual level (conditional) parameters (Appendix 7 ) . Table 6. 6 Mean WTP for D ifferent S eed T ypes B ased o n RCE: RPL Model Results Mean WTP relative to seed 3 (C$/ lb. ) Variables RPL RPL with error componen t Seed 2 29.99*** (5.10) 29.86*** (4.31) Seed 1 - 9.79** (4.01) - 9.07*** (2.81) Log - Likelihood - 2354.87 - 2238.67 AIC 4723.7 4497.3 N 213 *** p<0.01, ** p<0.05, * p<0.1 Source: Primary Data Collected by Researcher I chose the RPL with error compo nent to account for systematic differences between the experimentally designed alternatives and the status - quo option. I believe this model gives me more accurate estimates of WTP as indicated by the log - likelihood and AIC. I used seed 3 as the 46 base and th ese estimated WTP are the premiums (discount) individuals are willing to pay (accept) for these two types of seed. On average farmers are willing to pay approximately C$30 more for seed 2 relative to seed 3 . Meaning that stated WTP in the RCE shows that on average individuals stated that they would pay a premium of C$30 for seed 2 . On the other hand, o n average farmers showed a willingness to accept a discount of C$9. 07 for seed 1 relative to seed 3 . The premiums stated in the RCE show a clear preference ra nking unlike the BDM mechanism, where farmers are willing to pay a statistically significant premium for seed 2 and are willing to accept a statistically significant discount for seed 1. U sing the methods explained in Chapter 11 of Train (2011), I estimate d individual WTP parameters conditional on individual choices they made. The individual WTP parameters were estimated relative to seed 3. U sing these individual estimated parameters, I can look at how premiums stated in the RCE changes in each village. T he stated premiums for seed 1 and seed 2 relative to seed 3 by farmers in different villages are shown in Table 6. 7 : 47 Table 6. 7 Stated WTP Relative to Seed 3 Based on RCE: Results Disaggregated by Villages Village ID Seed 1 S.D Seed 2 S.D A1 C$ - 8.6 1 8.20 C$27.88 21.24 A2 C$ - 5.97 17.54 C$23.20 19.54 A3 C$ - 8.36 22.94 C$27.41 26.69 A4 C$ - 12.02 17.73 C$37.70 21.91 A5 C$1.90 9.34 C$18.00 13.54 A6 C$ - 20.37 20.98 C$40.61 27.67 B2 C$ - 17.23 17.99 C$42.08 25.40 B3 C$ - 8.79 16.46 C$31.88 20.05 B4 C$9.89 1 8.42 C$5.55 22.06 B6 C$ - 2.90 15.57 C$25.59 20.13 All C$ - 9.16 C$30.01 Source: Primary Data Collected by Researcher The RPL estimates in Table 6.7 show that farmers WTP in the RCE for seed 2 and seed 1 are significantly different than seed 3 across several villages . To test whether farmers premiums for seed 1 and seed 2 are statistically significantly different I used a fixed effects regression model to estimate the mean premiums ( Table 6. 8 ) Table 6. 8 Comparison of Mean Premiums Relative to Seed 3 Derived from RCE: Fixed Effects Model Results VARIABLES Premium Mean premium for Seed 2 39.173*** ( 6.178 ) Mean premium f or Seed 1 - 9.1 6 *** ( 2.913 ) Observations 426 Number of id 213 R - squared 0.449 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Source: Primary Data Collected by Researcher 48 Table s 6. 7 and Table 6. 8 show that in RCE method , individuals are willing to pay significantly different amounts for all three qualities of seed they observed in the blind experi ments. F armers are willing to pay a significant premium for seed 2 and willing to accept a significant discount for seed 1 , both relative to seed 3 . 6.4 Comparing Elicitation Methods for WTP When comparing premiums (discounts) that individuals are willi ng to pay (accept) between methods, the BDM mechanism does not yield statistically significant difference in WTP for seed 1 relative to seed 3 , however the RCE shows that there is a statistically significant difference and that the willingness to accept fo r seed 1 is a discount of approximately C$9. Table 6. 9 shows the comparison of WTP or WTA for seed 2 and seed 1 by method and village. This table highlights the vast difference between individual WTPs by village. On average the premium estimates using t he RCE method are approximately five times more than the stated WTP premiums in the BDM mechanism for seed 2 , which was the highest rated seed type . As shown in Table 6. 9 the difference in WTP for seed 1 in RCE is on average 18 times smaller than the state d WTP in BDM. Th e s e differences between the two methods is significantly higher than what was found in Lusk & Schroeder (2006) where stated WTP in a real choice experiment was only twice the magnitude as found in experimental auctions. 49 6.5 WTP and Plot Rankings Individuals preference rankings should not differ depending on the type of preference elicitation method used. Therefore, within an individual both methods should conclude that an indivi dual prefers a certain good or service. Using individual level premiums, we can see whether an preference ranking is exactly the same in both methods. We can test this by comparing directionality of the premiums. This means that i f a farmer prefers seed 1 over seed 2 it should be that both methods would show that this farmer is willing to pay a premium to obtain seed 1 over seed 2. However, in our study we see that in 36% of individuals the preference elicitation method concluded d ifferent preference rankings. Table 6.1 0 breaks down preference elicitation by seed 1 and seed 2. Of the individuals that have different preference ranking, 32 of them have different preference rankings for both seeds. Table 6. 9 WTP for S eed 1 and S eed 2 Relative to Seed 3: Comparison of R esults of RCE and BDM Village RCE Seed 1 BDM Seed 1 RCE Seed 2 BDM Seed 2 A1 C$ - 8.6 C$3.90 C$27.88 C$7.65 A2 C$ - 5.97 C$2.80 C$23.20 C$4.50 A3 C$ - 8.36 C$.45 C$27.41 C$8.73 A4 C$ - 12.02 C$ - 2.07 C$37.7 0 C$4.07 A5 C$1.90 C$ - 3.80 C$18.00 C$0.27 A6 C$ - 20.37 C$0.39 C$40.61 C$6.87 B2 C$ - 17.23 C$ - 2.50 C$42.08 C$5.27 B3 C$ - 8.79 C$ - 1.52 C$31.88 C$3.05 B4 C$9.89 C$3.06 C$5.55 C$1.00 B6 C$ - 2.90 C$0.00 C$25.59 C$1.94 All C$ - 9.16 C$0.53 C$30.01 C$5.02 Sourc e: Primary Data Collected by Researcher 50 Table 6.1 0 Difference s in Preference Ranking s by Seed Type Based on BDM and RCE Experiments Seed Type Same Preference Ranking in both BDM and RCE Different Preference Ranking Seed 1 172 41 Seed 2 146 67 Source: Primary Data Collected by Researcher Farmers were asked to rank the demon stration plots and rank one plot as the best and one plot as the worst. Using farmers stated preferred plot, we can look at which preference elicitation method align with farmers stated preference in these ranking sheets. D ue to the structure of the quest ion farmers could not state that two plots had an equivalent ranking. If there is a stated premium of 0 than it was assumed that the plot ranking s matched unless the seed type was ranked the best and seed 3 (baseline seed) was labeled the worst plot. Table 6.1 1 compares the plot preference in field day 2. Table 6.1 1 Preference Ranking Compared with Plot Ranking Seed Type Percent premiums in the BDM matched Plot Rankin g Sheet Percent premiums in the RCE matched Plot Ranking Sheet Seed1 61% 53% Seed2 74% 73% Source: Primary Data Collected by Researcher Neither BDM nor RCE, are consistently matching the field ranking sheets, I expected the BDM to be more consistent as farmers state their WTP directly and there is no parameterization of their WTP. There are a few possibilities that can explain the disconnect between the two methods in ranking individual preferences, since farmers always took the choice experiment seco nd there could be a fatigue effect that made them less price sensitive. Since there was no 51 incentive for farmers to accurately state the best and worst plot , farmers may not have taken the exercise seriously leading to conflicting answers. However, there i s no way to test these hypotheses in this study . It would be assumed that individual s would bid higher (or equal to the next preferred plot) for plots they preferred over others. This comparison of preference rankings (2006) claim that individual s construct utility preference within each preference elicitation method separately. This hypothesis would account for the differences in preferred seed type in plots across methods within the same individual. 6. 6 Comparison of Methods As mentioned in the conceptual model, both methods theoretically claim to be incentive , this assumes that the stated WTP within individuals should be identical. I test this by comparing th e stated /estimated WTP measures of two methods using a fixed - effects regression. I utility function does not change between methods and is constant. This assumption is based on that individuals purchase attribute bundles that maximize their utility. Therefore, I compare across the two methods (RCE and BDM) the WTP for seed 1 and seed 2 relative to seed 3 (i.e., the premium/discount) holding individual preferences constant. By m easuring within individual differences by methods I can control for heterogeneity of unobserved characteristics of farmer participants . I assume these unobserved characteristics are constant in this experiment, and therefore the differences in WTP is assoc iated Finally, I compare how determinants of WTP differ between the two methods. This can help us or seed quality. Table 6.1 2 presents the results of the fixed effect model comparing the estimated premiums for seed 1 and 2 52 across two methods RCE and BDM. Seed 1 and seed 2 are the dependent variables which are lity relative to seed 3. As shown in Table 6.12, the RCE method has a negative and significant effect on the estimated premium individuals are willing to pay for seed 1; while the method has a very large positive effect for the premium individuals are will ing to pay for seed 2. Therefore, I conclude that preference elicitation method used in this study has a large and significant impact on estimated WTP premiums for different seed types. Table 6.1 2 Comparison of E stimated P remiums for S eed 1 and 2 between R CE and BDM (base method): FE Regression Results VARIABLES Premium relative to seed 3 Estimate for Seed 1 : RCE method - 9.69*** (2.866) Estimate for Seed 2 : RCE method 25.00*** (2.922) Base Mean seed 1 (BDM method ) 0.53 Base Mean seed 2 (BDM met hod ) 5.02 Observations 852 Number of farmers 213 R - squared 0.426 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Source: Primary Data Collected by Researcher 6. 7 Determinants of WTP Premium for Quality Seed As mentioned in Ch apter 2, many factors can determin e farmer s decision to adopt a technology. This includes demographic characteristics such as gender, household size, and household composition (Fay and Deininger 2005, Holden and Shiferaw 2002) , economic factors such as the income potential of a technology ( i.e., how risky is th e adoption), h ousehold income and source of income (i.e., off - farm income ) (Pender and Kerr 1998; Hondel and Shiferaw 2002) , awareness and knowledge of the new technology , and access to that technology (Pende r and Kerr 1998). One of the most important variables, in the context of this study that can 53 perception of quality and WTP. To test how these theoretical pr edictions compare with the WTP premium for different seed types observed in my sample, I r egress these attributes on the differences in farmers premiums predicted by the BDM and the RCE for the overall highest r ated seed a nd the overall lowest rated seed, seed 2 and seed 1 respectively . In both the models the standard errors are cluster ed at the village level . Results are reported in Table 6.1 3 . The model used is: Whe re, is individual i j in method m. is farmer i quality of seed type j , and vector of household characteristics for individual i . Table 6.1 3 shows that the individual characteristics tha t may help explain the differences in premiums in preference elicitation methods for seed 1 and 2 ( relative to seed 3) are plot ranking of the other seed type , likelihood of being poor , and the gender of the head of household. If individuals p urchase bean seeds more often, they are much more likely to have a much larger and statistically significant difference between the two methods. This may be due to their understanding of the advantages of quality seed. The seed type s own plot ranking had no significant effect, but the plot ranking for the other seed type had a negative effect for both seed 1 and seed 2 . This means that if the farmer liked the other seed type more they were more likely to have a smaller difference in premiums between method s. Finally, the likelihood that an individual is poor has a small but significant effect. This coefficient stat e s that poor farmers are 54 more likely to have a higher premium in the RCE compared to the BDM mechanism for seed 1, but a smaller difference in pr emiums for seed 2. Table 6.13 Determinants of Differences in Premiums Estimated from Two Methods (BDM and RCE) VARIABLES Seed 1 Seed 2 Males in HH - 2.55** 1.74 (1.151) (1.942) Females in HH - 2.87** 2.67 (1.346) (1.851) Head of HH Years of Edu cation - 0.73 0.52 (0.558) (0.524) Head of HH Age 0.08 - 0.11 (0.178) (0.180) Head of HH Gender - 9.34* 15.52** (5.675) (7.489) 0.14*** - 0.12* (0.047) (0.073) Ranking of Plot Planted with Seed 1 3.90 - 8.31*** (3.116) (3.134) Ranking of Plot Planted with Seed 2 - 7.01*** 3.83 (2.534) (3.508) Have you ever produced bean seed 0.65 - 2.40 (4.367) (6.020) distance to paved road from house 0.18 - 0.09 (0.151) (0.263) distance to nearest marker from house 0.13 - 0.03 (0.156) (0.239) 1= likely to adopt - 0.55 - 0.01 (2.771) (3.386) 1= They have or Regularly Purchased Seeds - 5.00 7.59** (3.365) (3.778) Annual percent of income from beans 0.00 - 0.04 (0.063) (0.073) Primary Total area planted? 0. 35 - 2.77 (1.185) (2.493) Total amount of land area owned by your HH? 0.11 0.06 (0.159) (0.223) Constant 14.54 17.45 (10.988) (15.481) Observations 185 185 R - squared 0.225 0.186 55 7 . Discussion 7.1 The objective of this study was to evaluate two WTP elicitation methodolog ies through a within individual comparison. The significance of a within individual comparative study of WTP elicitation methods is that i t directly compare s the outcomes of both methods for the same individual. This gets rid of the need to match individuals that can have numerous unobserved characteristics that can be a potential source for differences in the estimated WTP using the two methods. By keeping t hese unobserved characteristics constant be tween methods , I was able to sho w that preference elicitation methods do impact WTP estimates . It would be expected that the WTP estimates from two different methods may differ in magnitude, but in preference elicitation methods should conclude preference rankings due to a similarity in underlying utility curves. However, in this study I find that RCE stated WTP and the BDM mechanism s do not reach similar preference rankings within an individual . This may indicate that individuals construct preferences dif ferently between the two mechanisms as suggested by Lusk and Schroeder (2006) or experimental auctions may be biased due to individuals exhibiting a type of non - expected utility preference function as suggested in Horowitz (2005). The differences in WTP estimates of the two methods could be due to several factors . One of the factors that apply directly to this study is price. It is possible that the prices used in both the RCE and the BDM mechanism did not reach a choke price and were more on a price inel astic point of the demand curve. The range of random prices in the BDM mechanism could have put a bias on the relative prices and lead to a large discount in individual WTP. Another factor is o ne or both methods are not truly incentive compatible and do no t elicit true WTP as suggested in Lusk and Schroeder ( 2006 ) . 56 S ome studies have put forth theoretical reasons why the BDM mechanism may not be incentive compatible (Horowitz 2005). The differences in the marginal effects on WTP by the determinants of WT P put further constraint on the idea that both methods measure equivalent preference ranking or WTP. It follows that the method used to elicit preference ranking and overall magnitude of WTP has a significant impact on the estimated WTP. The difference be tween the estimated WTP for different seed types relative to market price of the best quality seed available in the market is surprisingly large in the RCE. The premium was C$24 while the price premium in the market is approximately C$7 per pound. The BDM mechanism was designed in a way that it would be closer to the market prices which explains the closeness to market price premiums in the BDM mechanism for highest rated seed 2 with certified seed, which is considered to be highest market quality . Both met hods showed very low premium farmers were willing to pay for lower rated seeds ( seed 1 and seed 3) (BDM) or even a discount that farmers were willing to accept (RCE). Our results indicate that on average bean farmers in our study area are willing to pay a premium for higher performing seeds based on individual perce ption of seed quality. O n average individuals in our study were WTP more for seed 2 , which received best plot ranking by more farmers, and which also had high est yields . Compared to seed 2 , s eed 3 and seed 1 had lower yields and also received lower preference rankings , b ut the difference in the rankings and yield was not significant. This explains the close estimated WTP for these two seed types in the BDM auctions. However, in the RCE , result s show a significant di fference between the estimated WTP for seed 1 compared to seed 3 . Figure 7.1 shows farmers WTP premiums (discounts) by method and seed quality. 57 Figure 7.1 Premiums for Seed Ty pe by Method Source: Primary Data Collected by Rese archer This study shows that there is demand in Nicaragua for higher quality seed. However, as indicated earlier, the three seed types procured for our study are not a representative sample of the seed qualities of each type. Therefore, it is difficult t WTP for the three seed market classes -- certified, QDS, and recycled. However, this study shows that in both preference elicitation methods, the farmers WTP for their perceived highest quality seed was significantly m ore than the second - best seed and worst rated seed. In our study, the average premium farmers were willing to pay for highest rated seed was C$5.02 (C$30.01) in the BDM (RCE). Currently, in the market the highest priced seed is C$34/lb. and represents a pr emium of approximately C$21/lb. over the grain price. Figure 7.2 compares farmers WTP for their highest ranked plot (perceived highest quality plot) compared to current market premium of most expensive seed (i.e., certified seed). The estimates of premiums are relative to seed 3. -60 -40 -20 0 20 40 60 80 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Premiums relative to seed 3 Precent of Farmers Seed 2 RCE Seed 2 BDM Seed 1 RCE Seed 1 BDM 58 Figure 7.2 WTP for Best Quality Seed Source: Primary Data Collected by Researcher To compare the premiums for the individually highest rated seed estimated in these two elicitation methods, I omitted observations of farmers who had rated the plot with seed 3 as the highest (n=20) since the premiums are relative to seed 3. Th e graph in Figure 7.2 continues to highlight the differences in premiums by method. In the RCE, for approximately 60% of farmers their WTP premium for the ir perceived highest quality seed is more than C$21 . H owever , in the BDM this number is only 3%. the RCE is not bounded by a range of prices and predicts premiums above the highest possible bid in the BDM. The premiums in the RCE could be explained by expe rimental fatigue (as it always came after BDM) , or participants may not have understood the concept and mechanism of the choice experiment, or the marginal price range in the RCE may not have been large enough to dissuade individuals from choosing their pr eferred seed , or a combination of all these factors . -40 -20 0 20 40 60 80 100 0 0.2 0.4 0.6 0.8 1 1.2 Premium relative to seed 3 Percent of Farmers n=193 WTP for Best Quality (excluding seed 3) RCE BDM Premium 59 7. 2 Limitations of the Study The main limitation of this study is the small sample size . A larger sample size would have allowed us to change the range of the random prices in the BDM or the prices in the RCE and test whether the price range had a significant effect on biasing individual bids. Theoretically if the upper range of the random price distribution is too low, then a large number of farmers would be censored at the top bid . H owever , in our stu dy this was not the case , as only 9 out of 639 bids in the BDM method were C$39 , which was the top price. However, Karna and Sarfi (1987) showed that this theory may not be accurate as individuals bid in relation to the distribution of random prices meani ng that farmers may not bid C$39 as the y may not understand the optimal strategy, or they may be taking a strategic bet . Horowitz (2005) showed that under uncertainty o f price the individual s optimal bid may not be the same as their WTP. Horowitz states t hat the nature of the uncertainty is based on the distribution of random prices therefore the bid is dependent on this same distribution of random prices. In Morkbak et al., 2010 shows that increasing only the last price in the price vector of a choice des ign increases individuals WTP. This may explain the much larger premiums estimated in the RCE. Lazo et al. (1992) and Bohm et al. (1997) show the effects of increasing the upper bound of the random price in a BDM mechanism. Lazo et al. found out that incr easing the upper bound of the random price above the value of a redeemable ticket led to 25% of individuals stating their WTA above the redeemable value of the ticket. This study shows that the distribution of the random prices impacts individuals bidding strategy. This can be explained by individuals not following the utility maximizing strategy of the BDM mechanism, but deviating to either maximize their own utility from winning or not understanding the BDM mechanism and bidding in a different 60 strategical matter. The impact of price ranges may also impact WTP in each method differently leading to much larger differences in WTP estimates combined with fatigue effects. 61 8. Conclusion 8. Implications for Future WTP Studies In this study I evaluated two incent ivized WTP elicitation methodologies that are widely used in the literature, namely the Becker - DeGroote - Marschak auction experiment and the Real quality difference s among three types of bean seeds produced under different quality assurance system s main finding of this study is that preference elicitation methods do impact WTP estimate s. I find that the premiums for higher rated seeds relative to the lowest rated seed are not consistent between RCE and BDM, both in magnitude and direction. , there is no benchmark against which to compare the estimated WTP from the two methods. I t is thus impossible to tell which of the two methods evaluated in this study BDM or RCE is more accurate in estimating the true WTP for seed quality . Instead I show that preference elicitation methods are not interchan geable. This study is similar to other studies (i.e., Lusk and Schroeder; 2006), where they find that WTP price premiums estimate d using choice experiments are sometime s double the estimates derived using experimental auctions (in our study it was appro ximately five times more). As explained before , there are many studies that try to explain why this gap may occur. both hypothetical and real experimental auction s and choice experiments. When comparing methodology and accounting for personality traits they see that different personalities contribute to how individuals behave across different mechanisms. Therefore, individuals may state their WTP in a BDM mechanism , and a RCE differently based on certain personality traits that may 62 be difficult to control for in design. This could have also been the case in these experiments I conducted in Nicaragua. However, lack of data on personality traits does not allow me to c onfirm or reject this as a potential explanation of the observed differences in the estimated WTP from the two methods I studied. Many policy decisions and marketing decisions are decided based on the finding of market demand elicit ation using methods si milar to those used in this study . This study has shown that WTP may be dependent on the method used and therefore may not be accurately capturing the underlying unobserved true WTP of individuals. As further studies are done to understand this phenomenon, it would be important to understand how the design of a preference elicitation method leads to changes in differences in WTP elicited by different methods and how individual behavior can be incentivized to report true WTP. 63 APPENDICES 64 APP ENDIX A : Field Day 1: Farmer Ranking Sheet 65 APPENDIX B : Field Day 2: Farmer Evaluation Sheet 66 APPENDIX C : Script for Seed WTP Experiments Nicaragua 2017 NOTE: All text in italics are instructions for the enumerator . All text not in italics must be read to the farmer. This experiment/survey will be performed at field days in 12 villages in Nicaragua. Each village has 1 field experiment (FE) and the field days will be run in all 12 villages. During each field day, farmers (who attended the first field day and were surveyed) will participate in a willingness - to - pay (WTP) auction experiment and a real choice experiment. A FE consists of one field split into 3 plots. All the plots contain the same variety of beans, but were planted using different qu alities of seed Semill a Certificada(CS), Semilla Apta (SA) or Grano Comercial should not be told) which quality of seed was used for which plot. After a brief welcome to the field day and running through the criteria and plot ranking exercise including a question regarding WTP per lb for each seed type. The script below is for the enumerator and helpers running the WTP auctions and RCE. Step 1: Introduction/consent The enumerato r will introduce his or herself and read the consent script to the farmers and record their verbal consents to participate. Step 2: Overall description of Exercise ENUMERATOR: Ok, thank you for being willing to participate. To begin with, let me give y ou an overall description about what we will be doing. We are interested in getting an idea about how much you would be willing to pay for the three types of seed that was used to grow each of the 3 plots that you looked at earlier. To make your decisions more realistic, we are going to give you C$40 that you can use towards the purchase of one lb bag of one of the seed types used to grow these plots. We will be performing two types of exercises today. The first involves you providing the maximum price you are willing to pay for each of 67 these three seed types, and the second involves you selecting one of the seed type to purchase under different scenarios that we will present. In both these exercises, please determine your willingness to pay for one lb of se ed based on your observations of the performance of these three types of seed qualities in the field you just visited. [how we will decide which exercise becomes binding] At the end of both these exercises, one of these two types of exercises will be cho sen to give you a chance to actually purchase the seed. . This will be decided by flipping a coin so that either exercise will be chosen at random and will have equal probability of being chosen. We will flip a coin to decide whether the first exercise or second exercise will be chosen to be carried out through to purchase. We will let you all decide which Exercise is heads and which will be tails. This will be explained further on. Before we begin with the two primary exercises, we would like to do a pra ctice of the first exercise. For this practice exercise, we will give each of you C$10 to bid on purchasing a bar of soap like this one. Hold up bar of soap ok ? Do you have any questions? (answer questions) Should we begin? 68 Step 3: Practice Auction The enumerator will begin explaining the practice auction. ENUMERATOR: Ok, so one of the exercises you will participate in today is a seed auction. We w ant you to understand how the seed auction will work, so we want to run a practice auction first. For this practice auction each of you will be given C$10 to bid on one bar of soap. Unlike in most auctions, or in auctions you may have participated in the past, in this type of auction, it is possible for everyone to win and thus everyone might purchase a bar of soap using part or all of their C$10. Let me explain how you bid and how we determine who wins and buys a bar of this soap. First, we will hand out a bidding sheet like this one. Hold up bidding sheet. On this bidding sheet you will write down the maximum amount you would be willing to pay for this bar of soap, which could range from C$0 to C$9 . Please note down this amount in increments of C $1. For example, the amount you are willing to pay for this bar of soap could be C$ 0, 1, 2, 3, 4, 5, 6, 7, 8, or 9. The maximum bid you can make is C$9. Once everyone has done this, we will collect the bidding sheets and move on to determine how many of y ou win and will buy a bar of soap. To determine who wins we will simply choose a random price between 0 and C$9 we will explain how this will be done in a moment. 69 If the price you bid is greater than or equal to this random price, then you win, BUT yo u pay the random price not what you bid. This means that if you win, you pay a lower price for the soap than your bid (unless the random price is the same as your bid). On the other hand, if the price you bid is less than this random price, then you do not purchase the soap and you can keep the money (C$10). If you win, we will give you a bar of soap and the balance amount of your C$10; that is, C$10 minus the random price. For example, if you bid C$5 and the random price is C$4, then you would pay C$ 4 for the soap and get it, along with the remaining C$6. If you do not end up buying the soap, you do not spend any of your C$10 buying soap and we will give you C$10. name an enumerator1 in the room n ame an enumerator2 in the room this case, enumerator1 would buy the soap, but would pay C$3, not his/her bid of C$6. He/she would get a bar of soap and 10 - 3=C$7. I would also buy the soap and pay C$3 (my b id was C$4) so I would also get a bar of soap and 10 - 3=C$7. Enumerator2 would not buy the soap since his/her bid of 1 is less than 3 so he/she would just get C$10. [Best strategy explanation] Before we hand out the practice round bidding sheets, let m e explain the best strategy in this type of auction. The BEST thing to do is to bid the MAXIMUM amount you are willing to pay. This is because it is very likely you will actually pay LESS if you win. However, bidding less than what you would be willing t o pay might mean that you miss out on buying the soap at a price lower than what you would be willing to pay. For example if you are willing to pay C$7 and you only bid C$3 and the random price is C$4 then you will not purchase the soap at C$4. 70 Similarly, bidding more than what you would be willing to pay might mean that you end up having to pay more for the soap than you really want to. For example, if you are willing to pay a maximum of C$3, but you bid C$5 and the random price ends up being C$4, then y ou would pay C$1 more than you were willing to! Overall, your best strategy is to bid the MAXIMUM amount you are willing to pay. Are there any questions? We will determine the random price as fo llows: There are 10 possible options for the random price of this soap to be. The random price can be either a 0, 1, 2, 3, 4, 5, 6, 7, 8, or 9. We will thus roll this 10 - sided die to 9 that will be the price of the soap, if we roll a 10 than that will be C$0 Overall, we will end up with one of the following numbers: 0, 1, 2, 3, 4, 5, 6, 7, 8, or 9; and that number will determine the random price. If your bid on the sheet is higher o r equal to that random price, you will purchase the soap at the random price, if it is lower, then you will not purchase the soap and can keep the C$10. . Are there any questions so far? 71 [Hand out bid ding sheets] Ok, go ahead and write down your ID number from your name tag this helps us keep track of who to pay how much) and your bid for a bar of soap. Please do not talk with others until we have collected the bids. [Collect bidding sheets, making sure that bids and numbers are entered and legible and that the bid is in C$1 increments (i.e., 1.35 is not a valid bid).] [Determine random price as outlined above while writing it down on board. A helper should record this number on one of the bidding sheets so we have this information. We can allow farmers to flip coin/role die as long as it is tossed sufficiently to make it random.] Ok, so this is the price ( say the random price ) if you bid m ore than or equal to this price, you buy a bar of soap at this price ( say the random price ). If your bid was less than ( say the random price ) you will not buy a bar of soap, but will receive the C$10. omfortable sharing, raise your hand if bid.] Ok, so we will pay you and give you the soap, if you bought one, after the other exercises have completed. Step 4: See d Auction 72 The enumerator will begin explaining the seed auction. ENUMERATOR: Ok, so hopefully you have a better idea about how this seed auction will operate. It will be very similar to the practice auction you just did, except for a few things: Firs t, you will be bidding to purchase a one lb bag of the seed that was used to plant the plots in the field Exercise you just looked at. Specifically, you will be making 3 bids one fo ). HOWEVER, even though you are bidding for each type, ONLY ONE type will actually have a random price determined and will be bought/sold. You will not know which type is available until after you bid, so you should bid as if each one might be the one chosen. Second, instead of C$10, we are now giving you C$40 to use to bid. Just as before, your bid should be in increment of C$1. And as before, any amount you do not use to purchase seed, will be given to you after we are done. Third, the random price can be between 0 and C$39 and will be determined as follows: Enumerator: Write on a board two spaces _____ ______ the first digit can be a 0, 1, 2, or 3. We will roll a 4 sided die to determine this first digit. If the die lands on 1, 2, 3 the first digit will be that number, but if it comes up 4, it will be a 0. Then a second 10 sided die will be rolled, the second digit can be 0, 1, 2, 3, 4, 5, 6, 7, 8, or 9. If the die rolls a 10 than the second digit will be 0. O verall, we will end up with a number between 0 and C$39 in C$1 increments. As before, each number between 0 and 39 is equally likely. 73 Are there any questions? [remind them of the strategy] Ok, before we hand out the bidding sheets, let me just remind you that your best strategy is to bid the MAXIMUM amount you are willing to pay for each seed type represented by a symbol that was used in the field plots your just saw. Remember, since we are only going to determine a random price for ONE of the seed type, y ou do NOT need to try and spread your C$40 across the three seed types in fact you can bid C$40 for each quality and not have to worry about spending more than C$40. Any final questions? As before, please do not talk with others until we have collect ed the bids. [Hand out seed bidding sheets.] Ok, go ahead and write down your ID number (from the card) and bids for all three seed qualities. Remember that this is for a 1 lb bag of the seed type used to plant the indicated ease keep bids to C$1 increments ranging from 0 to 39. [Collect bidding sheets, making sure that bids and numbers are entered and legible and that all bids are in C$1 increments.(i.e., C$10.48 is not a valid bid) This ends our first Exercise. We will n ow begin the second Exercise. Step 5: RCE In this second exercise we will split into four groups. Your group leader will show you 12 different ellecions with different types of seed available in the market at different prices. Based on this information, you will be asked to select one option in each scenario. You will record your selection on a the sheet we will hand out to each one of you. Please fill out the information required as per the instruction of your group leader (e.g., Id code, Village code, name and choice set letter) 74 (Example RCE) Each option will look like this one (show example sheet of RCE), but with seeds and different prices. As you can see, there are two different products with two different prices as well as a ninguna option. (Ex plain Strategy) When you are shown each set, we want you to choose which of the two seeds you would ninguna. It is in your best interest to choose the option you would choose in a market setting. If you are not willing to pay for either but you choose one of them, you may have to pay that price for that seed later. If you choose ninguna, then you will keep the C$40 but purchase neither seed. Again, your best strategy i s to choose the seed you are willing to pay for. For example, if spend C$7 more than you would be willing to pay. So therefore your best option would be to choose the seed and the price as if you were purchasing between these seeds at a market. Are there any questions? Before we begin, it is important to note that if this second exercise is chosen to count, only one of the 12 set of options will be chosen for payment and pu rchase. We will decide by rolling a 12 - sided die like this one (hold it up). Since only one set of options will be chosen, there is no need to spread your C$40 over all your choices in fact, you could choose the highest price seed in every choice and not worry about spending more than your C$40. We would like you to not talk to others about what choice you are making while we go 75 [Separate participants into 4 groups and] Group Leader s: Ok, please go ahead and fill out the Id code, Village code, your name and the choice set letter. [group leaders should provide the codes and the choice set letter] Ok, so here is the first choice [group leaders show the first choice and make sure ev eryone fills in their choice. Continue for all 12 choices.] [Collect all the choice sheets] Step 6: Choosing the binding choice Ok, now we will flip a coin to decide whether the first or second exercise will be chosen to be carried out through to purch ase. We will let you decide which one will be heads and which one will be tails. [flip coin] (if BDM) Ok, since the first exercise was selecedwe will now reveal which seed quality was determined.] 76 [Determine random price as outlined above while writing it down on board. A helper should record this number on one of the bidding sheets so we have this information. We can allow farmers to flip coin/role die as long as it is tossed sufficiently to make it random.] Ok, so this is the price if you bid more than or equal to this price, you won and will buy a 1 lb bag of this quality seed at this price. If your bid was less than this price you will not buy seed, but will receive the C$40. (If RCE) Ok, since the second Exercise was selected, we will now roll a 12 - sided die to decide which Ellecion will be chosen. For example, if 5 is rolled than Ellecion 5 will be selected, and you will purchase the seed as per the choice you made in Ellecion# 5 . If you chose a seed then you wil l purchase that seed type at the price stated under that seed. If you had selected ninguna, then you will keep C$40. The choice will be random and each Ellecion has equal chance to be selected [Roll the die] This is the choice that is chosen you will no w purchase the seed type you chose in that Elleccion number at that given price. If you chose niguna for this choice you will not purchase anything and you will keep the C$40. [Closing Statements] Ok, so we will call you up one or two at a time to give you the seed/soap if you bought them and however much we owe you in Cordobas. Thank you and please do not discuss this with the other group of farmers until they have completed the exercises. Thank you! 77 APPENDIX D : Survey 78 79 80 81 82 83 APPENDIX E: BDM Bidding Sheet (Spanish) 84 APPENDIX F : Choice Tasks in RCE Table F.1 Choice Tasks in RCE Choice Task Seed1 Seed 2 Seed3 1 - C$21/lb. C$14/lb. 2 C$34/lb. C$28/lb. - 3 C$28/lb. - C$34/lb. 4 C$21/lb. - C$28/lb. 5 - C$28/lb. C$21/lb. 6 C$14/lb. C$34/lb. - 7 C$21/lb. C$14/lb. - 8 C$14/lb. - C$21/lb. 9 - C$34/lb. C$28/lb. 10 C$28/lb. C$21/lb. - 11 C$34/lb. - C$14/lb. 12 C$14/lb. - C$34/lb. 85 APPENDIX G: RPL with Error Component Results Table G .1 RPL with Error Component Results Random Parameter Coefficient S.E P - Value S.D Price 1.0 Seed2 29.86 4.31 <.0 1 29.92 Seed1 9.07 2.81 <.01 24.08 Error Component 0 78.42 Log - Likelihood - 2238.67 AIC 4497.3 86 WORKS CITED 87 WORKS CITED Bardhan, P., & Udry, C. (1999). Development Economics. New York: Oxfrord University Press Inc. Bash, K., Bowman, J., Chapman, J., & Blandon, J. A. (2002). Assessment of Nicaragua's Seed Sector. Maryland: Development Alternatives, Inc. Becker, G. M., DeGroot, M. H., & Marschak, J. (1964). Measuring utility by a single - response sequential method. Systems Research and Behavioral Science , 226 - 232. Benzing, A. (1989). Andean potato farmers are 'seed bankers'. ILEIA , 1 - 3. Brennan, P. W. (1991). An Analytical Model of Farmers' Demand for Replacement Seed. American Journal of Agricultural Economics , 1044 - 1052. Brookshire, D. S., Coursey, D. L., & Schulze, W. D. (1987). The external validity of experimental economics techniques: analysis of demand behavior. Economic Inquiry 25.2 , 239 - 250. Brush, S. (1991). A farmer - based approach to conserving crop germpla sm. Economic Botany , 153 - 165. Cameron, L. A. (1999). The Importance of Learning in the Adoption of High - Yielding variety seeds. American Journal of Agricultural Economics , 83 - 94. Carter, M. R., Toledo, P., & Tjernstrom, E. (2010). Investing in Small - Fram P roductivity: Impact Dynamics and Hetergeneity in Nicaragua. Working Paper . Colombo, S., & Louviere, N. H. (2003). Modeling preference hetergeneity in stated choice data: an analysis for public good generated by agriculture. Agricultural Economics , 307 - 322. Colombo, S., Hanley, N., & Louviere, J. (2009). Modeling preference heterogeneity in stated choice data: an analysis for public goods generated by agriculture. Agricultural Economics , 307 - 322. Conley, T. G., & Udry, C. R. (2010). Learning about a New Tech nology: Pineapple in Ghana. American Economic Review , 35 - 69. Dennis, J. V. (1987). Farmer management of rice variety diversity in Northern Thailand. Ann Arbor, MI: Cornell University, Univ. Microfilms. DeYoung, D. J. (2015). Determinants of Sustainability of Community Seed Banks in Nicaragua: A Duration Analysis Approach, Master Thesis. East Lansing, MI: Michigan State University. Douglas, J. (1980). Successful seed programs. A planning and management guide. Boulder: Westview Press. 88 FAOSTAT. (2018, January 9). Country . Faye, I., & Deininger, K. W. (2005, July 24 - 27). Do New Delivery Systems Improve Extension Access? Evidence From Rural Uganda. Paper presented at the American Agricultural Economics Association Annual Meeting . Providence, RI, US. Feder, G., Ju st, R. E., & Zilberman, D. (1985). Adoption of Agricultural Innovations in Developing Countries: A survey. Economic Development and Cultural Change , 255 - 298. Fligner, M. A., & Verducci, J. S. (1993). Probability Models and Statistical Analyses for Ranking Data. New York: Springer - Verlag. Foster, A. D., & Rosenzweig, M. R. (1995). Learning by Doing and Learning from Others: Human Capital and Technical Change in Agriculture. Journal of Political Economy , 1176 - 1209. Ginon, E., Chabanet, C., Cobris, P., & Issan chou, S. (2014). Are decisions in a real choice French baguettes. Food Quality and Preference , 173 - 180. Grebitus, C., Lusk, J. L., & Jr., R. M. (2013). Effect of distanc e of transportation on willingness to pay for food. Ecological Economics , 67 - 75. Groote, H. D., Tomlins, K., Haleegoah, J., Awool, M., Frimpong, B., Banerji, A., . . . Meenakshi, J. (2010). hana with experimental auctions and a simulated radio message. Cape Town: CYMMIT. Harrison, G. (1989). Theory and Misbehavior of First - Price Auctions. American Economic Review , 749 - 762. Holden, S. T., & Shiferaw, B. (2002). Poverty and Land Degredation: Pe asants' Willingness to Py to Sustain Land Productivity. Natural Resource in African Agriculture: Understanding and Improving Current Practices . Horowitz, J. K. (2005). The Becker DeGroot Marshak Mechanism is Not Always Incentive Compatible, Even for Non - Ra ndom Goods. Department of Agricultural and Resource Economics, University of Maryland. Ingrid Olesen, e. a. (2010). Eliciting consumers' willingness to pay for organic and welfare - labelled salmon in a non - hypothetical choice experiment. Livestock Science . Jayson Lusk, T. F. (2004). Experimental auction procedure: impact on valuation of quality differentiated goods. American Journal of Agricultural Economics , 389 - 405. Jean Claude Rubyogo, L. S. (2007). A new approach for facilitating farmers' access to bean seed. LIESA - LEUSDEN . Jeffrey K. Lazo, G. H. (1992). What is the Future Worth: An Experimental Examination of Rates of Time Preference. Unpublished Manuscript . 89 John A List, P. S. (2006). Using choice experiments to value non - market goods and services: evid ence from field experiments. Advances in economic analysis & policy . Karni, E., & Safra, Z. (1987). "Preference Reversal" and the Observability of Preferences by Experimental Methods. Economtrica: Journal of the Econometric Society , 675 - 685. Kender, J. L., & Kerr, J. M. (1998). Determinants of Farmers' Indigenous Soil and Water Conservation Investments in Semi - arid India. Agricultural Economics , 113 - 125. Lancaster, K. J. (1966). A new approach to consumer theory. Journal of Political Economy , 132 - 157. Lovie re, J. J., Hnsher, D. A., & Swait, J. D. (2000). Stated choice methods: analysis and applications. Cambridge university press. Lusk, J. L., & Schroeder, T. C. (2006). Auction bids and shopping choices. Advances in Economic Analysis & Policy . Lusk, J. L., A lexander, C., & Rousu, M. (2006). Disgning Experimental Auctions for Marking Research: Effect of Values, Distributions, and Mechanisms on Incentives for Truthful Bidding. Department of Agricultural Economics, Oklahoma State University. MAGFOR. (2009). Fort alecimiento al Sistema Nacional de Semilla, Evaluacion Social de Territorios, Ampliacion Proyecto de Tecnologia Agropecuaria II. Retrieved June 2, 2015, from Ministerio Agropecuario y Forestal: http://www.magfor.gob.ni/programas/pea/salva/Evaluacion%20Soci al%20de%20Territor ios%20Ampliacion%20PTA%20II.pdf Maredia, M., Reyes, B., & DeYoung, D. (2014). Farmer perspective on the use of and demand for seeds of improved bean varieties: Results of beneficiary survey in Guatemala, Honduras, and Nicaragua. Departmen t of Agricultural, Food, and Resource Economics, Michigan State University , Staff Paper No. 2014 - 04. Mazar, N., Koszegi, B., & Ariely, D. (2014). True context - dependent preferences? The causes of market - dependent valuations. Journal of Behavioral Decision Making , 200 - 208. McFadden, D., & Train, K. (2000). Mixed MNL Models for discrete response. Journal of Applied Econometrics , 447 - 470. Munshi, K. (2004). Social learning in a heterogeneous population: technology diffusion in the Indian Green Revolution. Jour nal of Development Economics , 185 213. Nicholas Magnan, D. J. (2015). Leveling with Friends: Social Networks and Indian Farmers' Demand for a Technology with Heterogeneous Benefits. Journal of Development Economics . Oladele, O. I. (2008). Factors determini Oyo State, Nigeria. Agricultura tropica et subtropica , 165 - 170. 90 Opoku, E. (2017). Willingness to pay for quality cowpea seeds: Experimental evidence from northern Ghana; Master's Thesis. East Lansing Michigan: Michigan State University. Peter Bohm, J. L. (1997). Eliciting reservation prices: Becker DeGroot Marschak mechanisms vs. markets. The Economic Journal . Plott, C. R., & Zeiler, K. (2005). The willingness to pay - willingness to accept gap, the " endowment effect," subject misconceptions, and experimental procedures for eliciting valuations. American Economic Review , 530 - 545. R. Tripp, N. L. (1997). Seed regulation: choices on the road to reform. Food Policy , 433 - 446. Sain, G. (2011). Un análisis d e la evaluación campesina (aceptabilidad y satisfacción) de algunas innovaciones promovidas por el Proyecto Red SICTA. Retrieved March 29, 2018, from http://repiica.iica.int/DOCS/B3425E/B3425E.PDF Scarpa, R., & Willis, S. F. (2005). Perfomance of error com ponent models for status - quo effects in choice experiments. Applications of simulation methods in environmental and resource economics , 247 - 273. Seeds of peasant subsistence: agrarian structure, crop ecology and Quechua agriculture in reference to the loss of biological diversity in the souther Peruvian Andes. Ph.D. Thesis. (1989). Berkeley: University of California. Singh, I., Squire, L., & Strauss, J. (1986). Agricultural household models: Extensions, applications, and policy. The World Bank. Steur, H. D. , Wesana, J., Blancquaert, D., Straeten, D. V., & Gellynck, X. (2017). Methods matter: a mete - regression on the determinants of willingness - to - pay studies on biofortified foods. Annals of the New York Academy of Sciences , pp. 34 - 46. Street, D. J., Burgess, L., & Louviere, J. J. (2005). Quick and easy choice sets: constructing optimal and nearly optimal stated choice experriments. International Journal of Research in Marketing , 459 - 470. Streets, D. J., & Burgess, L. (2007). The construction of optimal stated choice experiments: Theory and methods. John Wiley & Sons. Train, K. (2003). Discrete choice modles with simulation. Cambridge: Cambride University Press. Useche, P., Barham, B. L., & Foltz, J. D. (2012). Trait - based adoption models using ex - ante and ex - p ost approaches. American Journal of Agricultural Economics , 332 - 338. Ward, P. S., Ortega, D. L., & Singh, D. J. (2014). Heterogenous demand for drought - tolerant rice: Evidence from Bihar, India. World Development 64 , 125 - 139. Wierema, H., Almekinders, C., Keune, L., & Vermeer, R. (1993). La Produccion campesina en Centroamerica: los sistemas locales de semilla. Tilburg, Netherlands: IVO P.O. Box 90153.