|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 THREE ESSAYS ON THE ELICITATION OF W ILLINGNESS T O P AY By LIJIA SHI A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2012
2 2012 Lijia Shi
3 To my parents
4 ACKNOWLEDGMENTS I would like to express my gratitude to my committee members: Dr. Lisa A. House, Dr. Zhifeng Gao, Dr. Kim Morgan and Dr. Chunrong Ai for their generous help and support in the completion of this dissertation. Most gratefully, I thank Dr. Lisa House and Dr. Zhifeng Gao for their generous help and continuous encouragement. They have been among the best mentors in the department and a ll my achievement during the last four years would not be possible without their support. Moreover, I want to thank Dr. Kim Morgan and Dr. Chunrong Ai for their valu able comments and suggestions. They are both respected professionals. I believe their help will also benefit me in my future studies. I would also appreciate the support from my fellow graduate students. They made my graduate life full of wonderful memories. Last but not least, I would like to take this opportunity to thank my family, especia lly my parents for their endless supports and love. Without them, my study would not be possible.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 LIST OF ABBREV IATIONS ................................ ................................ ............................. 9 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 12 1.1 Blueberry Market Review ................................ ................................ .............. 12 1.2 Consumer Willingness to Pay for Fruit Attributes ................................ ......... 13 1.3 Value Elicitation Method ................................ ................................ ............... 14 2 CONSUMER WILLINGNESS TO PAY FOR BLUEBERRY ATTRIBUTES: A HIERARCHICAL BAYESIAN APPROACH IN THE WILLINGNESS TO PAY SPACE ................................ ................................ ................................ .................... 18 2.1 Introduction ................................ ................................ ................................ ... 18 2.2 Consumer Attitudes for Food Attributes ................................ ........................ 21 2.3 The Choice Experiment and the Data ................................ ........................... 22 2.4 The Model ................................ ................................ ................................ ..... 24 2.5 Results ................................ ................................ ................................ .......... 26 2.5.1 Mix ed Logit Estimation Result ................................ .............................. 27 ................................ .......... 27 2.5.3 The Reliability of the Individual Level Estimates ................................ .. 30 2.5.4 Willingness to pay and Demographics ................................ ................. 31 2.6 Conclusion and Discussion ................................ ................................ ........... 31 3 ON MODEL SPECIFICATION OF THE MIXED LOGIT: THE CASE OF CONSUMER PERCEPTION ON BLUEBERRY ATTRIBUTES .............................. 41 3.1 Introduction ................................ ................................ ................................ ... 41 3. 2 The Model ................................ ................................ ................................ ..... 45 3.3 Case Study: Consumer Preference for Blueberry Attributes ......................... 48 3.4 Results ................................ ................................ ................................ .......... 50 3.5 Conc lusion and Discussion ................................ ................................ ........... 53
6 4 TO PAY FOR ORGANIC AND LOCAL BLUEBERRIES: A MULTI STORE BDM AUCTION CONTROLLING FOR PURCHASE INTENTIONS ................................ ................................ ..................... 58 4.1 Introduction ................................ ................................ ................................ ... 58 4.2 Literature Review ................................ ................................ .......................... 61 4.3 Auction Procedure ................................ ................................ ........................ 64 4.4 Results ................................ ................................ ................................ .......... 65 4.4.1 Partial Bids for Organic and Local Blueberries at Different Marketing Outlets ................................ ................................ ................................ ........... 66 4.4.2 Impact of Purchase Intention ................................ ................................ 67 4.4.3 The Tobit Model for Full Bids ................................ ................................ 68 4.4.4 The OLS Regression for Partial Bids ................................ .................... 70 4.5 Conclusion ................................ ................................ ................................ .... 71 5 CONCLUSION ................................ ................................ ................................ ........ 77 APPENDIX A CHOICE SITUATIONS FOR THE CHOICE EXPERIMENT ................................ .... 81 B BDM AUCTION QUESTIONNAIRE ................................ ................................ ........ 84 C BDM AUCTION INSTRUCTIONS AND PROCEDURE ................................ ........... 89 LIST OF REFERENCES ................................ ................................ ............................... 91 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 98
7 LIST OF TABLES Table page 2 1 Blueberry attributes and levels in the choice experiment ................................ ....... 34 2 2 Demographics of choice experiment participants ................................ ................... 34 2 3 Estimation results of the mixed logit model in the WTP space ............................... 35 2 4 ................. 36 2 5 ................................ ........... 36 2 6 WTPs of three respond ents ................................ ................................ ................... 37 3 1 Literature comparing models from preference space and WTP space ................... 55 3 2 Blueberry attributes and levels in the stated preference experiment ...................... 56 3 3 Estimation results of the three models ................................ ................................ ... 56 3 4 Means and variances of the individual level WTP estimates ................................ 57 3 5 Correlation between individual level WTP estimates among the three models. ..... 57 4 1 Demographics of participants at each marketing outlet ................................ .......... 73 4 2 Summary statistics of bids ................................ ................................ ..................... 73 4 3 Difference of mean bids across purchase intentions ................................ .............. 74 4 4 WTPs at different locations ................................ ................................ .................... 74 4 5 Tobit for reported bids ................................ ................................ ............................ 75 4 6 Robust OLS for WTPs ................................ ................................ ............................ 76
8 LIST OF FIGURES Figure page 1 1 Total imports of fresh blueberries in the United States from 1993 to 2011. ............ 16 1 2 Per capita consumption of fresh plus frozen blueberries in the United States from 1992 to 2010. ................................ ................................ ............................. 16 1 3 Percentage of fresh use among total utilized production of blueberries from 2000 to 2011. ................................ ................................ ................................ ...... 17 2 1 Scatterplots of WTP estimates v ersus demographic information. ......................... 38 2 2 Comparison of scale parameters for participants and non participants. ................. 40
9 LIST OF ABBREVIATION S HB Hierarchical Bayesian WTP Willingness to pay
10 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements fo r the Degree of Doctor of Philosophy THREE ESSAYS ON THE ELICITATION OF WILLINGNESS TO PAY By Lijia Shi August 2012 Chair: Lisa A. House Cochair: Zhifeng Gao Major: Food and Resource Economics behavior. A large amount of information is collected, including consumption habits, willingness to pay ( WTP) for several blueb erry attributes was also included in the survey. Additionally, a non hypothetical method: experimental auction was conducted to data from the two val ue elicitation methods: choice experiment and experimental auction for empirical model comparisons and studies of consumer behavior. In the first essay, a stated preference experiment is conducted to elicit consumer WTP for various blueberry attributes. A mixed logit model estimated by the hierarchical Bayesian approach (HB) is employed to account for consumer heterogeneity and the distributions of WTPs are directly specified. The results show that locally produced blueberries are preferred over U.S. pro duced blueberries by most respondents. By contrast, less than 50 percent of the respondents demonstrate positive premium s for organic blueberries. Additionally, hardly any relationship between demographics and WTPs is detected.
11 In the second essay, three specifications of the mixed logit model are compared. The first is specified in the preference space with all parameters random. The second is specified in the preference space with the coefficient of price fixed. The third is speci fied in the WTP space wi th all parameters random. The data is from the choice experiment organic. The purpose is to see whether fixing the price coefficient in the preference space can narrow do wn the gap (i.e., difference in model fits and estimated coefficients) between the preference space model and the WTP space model and how such constraint affects the individual level WTP estimates the BDM auctions. The Additionally, the auction was conducted at multiple types of stores to capture a more representative sample of consumers
12 CHAPTER 1 INTRODUCTION 1.1 Blueberry Market Review Blueberry is widely known as a healthy fruit. The consumption of blueberries has experienced a dramatic change during the last few decades in the United States. In the late 1990s, scientific research revealed special health benefits of blueberries Accordi ng to the USDA, the total imports of fresh blueberries increased from 17.5 million pounds in 1993 to 192.5 milli on pounds in 2008 (Figure 1 1). At the same time, per capita consumption of fresh and frozen blueberries in the U.S. increased (Figure 1 2), fro m 0.62 pounds in 1992 to 1.71 pounds in 2010 (USDA, Economics, Statistics, and Market Information System). We can see that per capita consumption of fresh blueberries has been increasing dramatically since 2000, but that of frozen blueberries has not demon strated any obvious increasing trend since 1992. While per capita consumption consumption of fresh blueberries has been dominating the non processed blueberry market since 2002 can reach 44 ounces by 2015 (U.S. Highbush Blueberry Council). In spite of the overall growing consumption level, there is still quite large undeveloped domestic market Consumption may be further motivated by the education of the health benefits of blueberries among consumers and effective promotion strategies. As a leading blueberry producer in the world, the blueberry segment in the United States will encounter thriving commercial opportunities with the expanding national consumption level. At this critical time of development, retailers should set up corresponding
13 promotion plan based on a comprehensive understanding of the structure of blueberry market. The quantity o f literature on the consumption of blueberries is quite limited, especially on the consumption of non processed (fresh+ frozen) blueberries. To our knowledge, the only literature on consumer preference for blueberries is from Hu, Woods, and Bastin (2009), which focused on processed blueberry products. According to the data from USDA, the percentage of the fresh use among total utilized production of blueberries increased from 27 .1 % in 2000 to 5 4.6 % in 2011 (Figure 1 3). The percentage is expected to keep ri sing as more people have realized the health benefits of blueberries. Also, non processed blueberries usually possess higher market values and consumer acceptance. Therefore, we will conduct a systematic study about the consumption behavior of non processe d blueberries, such as how do consumers tradeoff different blueberry attributes and how much they would pay for the credence attributes like organic. Our results will serve as reference for the setup of effective marketing strategies, which is crucial for potential expanding of the blueberry market. 1.2 Consumer Willingness to Pay for Fruit Attributes With the rising concern about health and environment among consumers, the production method or production location of food has received plenty of attention. to pay (WTP) for val ue added food attributes, such as organic and local. In economics, WTP is known as the maximum amount of money people would be willing to pay for a good or WTP can be measured with some economic methods (i.e. choice experiment or experimental auctions). Thus, WTP enables researchers to quantify and compare
14 WTP has been one of the hottest discussed topics in agric ultural marketing literature. Research includes WTP for a product or a product attribute, such as organic or genetically modified. A review of the literature on WTP is conducted in Chapter 2 and Chapter 4. 1.3 Value Elicitation Method Value Elicitation methods, such as contingent valuations and experimental auctions, are critical for the valuation of non market goods and new product development. Choice experiments are one of the most widely used contingent valuation approach in social sciences (Hanley, Wright, and Adamowicz 1998; Birol, Karousakis, and Koundouri 2006; Alfnes et al. 2006 etc.). They are easy to implement and can be used to value products that may not exist in the market. However, in spite of the numerous merits of choice experiments, lit erature has revealed the exaggerated WTP estimates due to their hypothetical valuation context (Lusk and Schroeder 2004; List and Shogren 1998). Therefore, researchers began to rely on the non hypothetical method, such as experimental auction (EA), to elic it and Willinger. 2004; Huffman et al. 2003 etc.). Some commonly used experimental auction mechanisms include 2 nd price auction, BDM auction, random N th price auction et c., among which BDM is conducted at point of purchase locations while the others are usually conducted in labs. and evaluate market share, various findings that violate the econom ic theory are observed and some methodological problems are discussed in literature. Additionally, even though the value elicitation methods are proved to be incentive compatible,
15 problems can still arise in the data analysis process. For the same dataset, different models can yield significantly different results. The objective of the dissertation will be to conduct a systematic study on U.S. producers and retailers important m essages about the efficient promotion of blueberries. Meanwhile, some problems in the value elicitation methods are discussed. Specifically, several estimation methods for the choice experiment data will be compared empirically and how to obtain a better i nterpretation of the results will be discussed. In addition to the hypothetical method, the dissertation will also discuss some issues in conducting an on site auction. The second and third chapters will focus on the model estimation method with the data f rom a choice experiment. In the second chapter, we hypothesize that the mixed logit model estimated in the WTP space for the choice experiment data may help researchers obtain more reasonable WTP estimates and the estimation results should be interpreted a t the individual level. In the third chapter, several estimation methods are compared and the hypothesis is that imposing some constraints on the utility coefficients might narrow the gap between the preference space model and the WTP space model (Hensher and Greene 2011). Different from Chapter 2 and Chapter 3, which focus on the hypothetical value elicitation methods, Chapter 4 will address some issues in the implementation of a non hypothetical value elicitation method: BDM auction (Becker, Degroot, and Marschak 1964). We intend to discover the potential influence of purchase intention and the choice of auction location on the auction results.
16 Figure 1 1 T otal i mports of fresh blueberries in the United States from 1993 to 2011 F igure 1 2. Per capita consumption o f fresh plus frozen blueberries i n th e United States from 1992 to 2010 50,000,000 100,000,000 150,000,000 200,000,000 250,000,000 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Pounds Year 0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60 1.80 Pounds Year Fresh Frozen Fresh+Frozen
17 Figure 1 3. P ercentage of fresh use among total utilized production of blueberries from 2000 to 2011 0 0.1 0.2 0.3 0.4 0.5 0.6 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 Percentage Year
18 CHAPTER 2 CONSUMER WILLINGNESS TO PAY FOR BLUEBERR Y ATTRIBUTES: A HIERARCHICAL BAYESIAN APPROACH IN THE WILLINGNESS TO PAY SPACE 2.1 Introduction Per capita consumption of fresh blueberries has increased dramatically since 2000 (United States Department of Agriculture, Economic Research Service [USDA/ ERS]) This growth reflects both increased consumer awareness of the importance of healthy diets and the proactive effort by the U.S. blueberry industry in publicizing the benefi ts from blueberry consumption. Faced with such rapid growth of this new market, a systematic study about consumer behavior in blueberry consumption is critical. Consumer choice of fruit has become complicated, as even the same type of fruit, for example, blueberries, can have multiple attributes. Market segmentation is used to reach different consumer segments. In a market rapidly growing market such as the one for blueberries understanding consumer choice of these attribute combinations is perhaps even more critical. Production method origin of production, and form of the fruit (i.e., frozen versus fresh) are among a number of attributes (appearance, flavor, price, etc.) that consumers c onsider when purchasing fruit. on their preferences. Some consumer s may prefer fresh blueberries over frozen ones, while other consumers may prefer frozen blueberries because of their long shelf life. As for the credence attributes, such as production method (i.e., organic) and production location (country of origin), co variance Some consumers may consider country of origin of blueberries a more important attribute than whether the product is fresh or frozen, while to others, country of origin may not be important. Some consume rs prefer organic production and thus are willing to pay more for organic blueberries. Though this preference choice has been shown to
19 exist for other fruits (i.e., Lin, Smith, and Huang, 2008; Batte et al., 2007; Yue and Tong, 2009; Loureiro, McCluskey, a nd Mittelhammer, 2001), it is important to explore the size of the consumer segments specifically for blueberries. Different consumers appreciate different attributes, thus, marketing strategies that fail to take consu mer heterogeneity into account are des tined to be less efficient. The explanatory power of demographics for consumption behavior is limited (Frank, Massy, and Boyd, 1967; Yankelovich, 1964), especially for small purchases such as one pint of blueberries in a highly competitive fruit market Th is study will compare the importance of different attributes on consumer willingness to pay (WTP) for blueberries at the individual level. If consumers have a diversity of opinions regarding fruit attributes, especially given the large amount of substitute s in the fruit market, a single marketing strategy might not be ideal. Studying the impact of attributes on different consumers will aid blueberry producers to target different consumers with different marketing strategies. The objective of this study is t blueberry attributes and differentiate consumers in terms of their individual level WTP estimates (i.e., WTP e stimates for each respondent). Our work contributes to the literature in two aspects. First, we examine non processed (fresh) and frozen blueberries. To our knowledge, the literature on blueberry consumption (Hu, Woods, and Bastin, 2009; Hu et al., 2011; etc.) mainly focused on the attributes of processed blueberry products. Non processed blueberries possess much higher market values and consumer recognition. Additionally, the percentage of fresh use among utilized production of blueberries increased from 27 .1 %
20 in 2000 to almost 5 4.6 % in 2009 (USDA). 1 The percentage is expected to keep rising as more people realize the health benefits of fresh blueberries. Second, our results are based on the individual level estimates, which provide valuable information about the variety of consumer attitudes, such as diversified a ttribute importance ranking. Such information is extremely valuable when demographic information is not available or the explanatory power of demographics is marginal. Most previous literature reported only aggregate WTP estimates (i.e., the average WTP) o r the distribution of WTPs across consumers, which are much less informative for the implementation of differential marketing. I ndividual level estimates provide us with valuable information about individual consumption behavior, which is indispensable for differential marketing strategies. For example, price cut strategies are expected to be more effective for consumers who are more sensitive to price. Organic labeling may only attract those who prefer organic production. In this light, supermarkets can is sue different types of coupons to different consumers based on their individual preferences (Rossi, McCulloch, and Allenby, 1996). In addition, since all kinds of WTP elicitation methods have some shortcomings, the accuracy of the WTP estimate cannot be gu aranteed. The comparison of the relative importance of various attributes might be of more practical value. The individual level estimates enable us to calculate the proportion of consumers that pref er one attribute over another. Such information cannot be obtained from the two estimated distributions of WTPs for the two attributes. 1 The number is based on the production and utilization data of Maine, Michigan, New Jersey, North Carolina, Oregon, Washington, Alabama, Arkansas, Florida, Georgia, Indiana, New York, California and Mississippi.
21 2.2 Consumer Attitudes for Food Attributes There is a large amount of literature on the valuation of food attributes. For example, country of origin is among the most popularly discussed food attributes in Umberger et al. (2003) showed that most consumers were willing to pay premiums for quality of U S products are generally believed to be one of the main reasons for ition of U S products. In addition to the country of origin label, the et purchasing behavior for apples. In the study of Hu, Woods, and Bastin (2009), (within t free attributes of processed blueberry products. Darby et al. (2006) also concluded that consumers were willing to lab el. In addition to country of origin, there is extensive literature on the choice between organic and conventional products. Wang and Sun (2003) concluded that the organic market had a large consumer base Batte et al. (2007) considered multi ingredient processed organic foods with four levels of organic content under the National Organic Program (100% organic, 95% organic, 70 95% organic, <70%organic). Their results indicated that customers were willing to pay a premium for fo od with organic content, even those that were not totally organic. Loureiro, McCluskey, and Mittelhammer (2001) showed that consumers with similar perceptions
22 of food security and environmental issues tended to choose the organic apples over the eco labele d apples when these two types of apples were both available and sold at the same price. Yue and Tong (2009) found that the WTP for organic and that for local were almost the same. However, Hu, Woods, and Bastin (2009) concluded that a lthough organic foods were believed to be one of the solutions for food safety, supporters of organic foods were not so broad and were limited to a certain proportion of consumers. The reason might be that the price of organic foods is generally much higher than that of convent attribute can be found in Brooks and Lusk (2010); Bond, Thilmany, and Bond (2008); Lin, Smith, and Huang (2008); Janssen, Heid, and Hamm(2009); Managi et al. (2008); etc. 2.3 The Ch oice Experiment and the Data A stated preference experiment was conducted and distributed online to random consumers living in the Northeast and Southeast regions of the United States. Each respondent was asked to make choices over a series of choice situa tions. Since multiple observations per respondent were collected, individual level information can be inferred. The stated preference experiment approach has been widely used by researchers (i.e., Brooks and Lusk, 2010; Gao and Schroeder, 2007; Alfnes, 20 04; Carlsson, Frykblom, and Lagerkvist, 2004). Although there might be some potential problems associated with the stated preference approach, such as exaggerated WTP estimates (Brooks and Lusk, 2010; Johansson Stenman and Svedster, 2008), there are four main reasons for it to be the suitable approach for the study of consumer purchase of blueberries. First, it is not usual for consumers to have access to a great variety of
23 blueberries (all possible sources, forms, or prices) in the same section of the gro cery store or other places when they are shopping for blueberries, so this gives reasons to question the dependability of the WTP estimates even if we have real transaction data. For example, people might actually prefer frozen blueberries, but end up buyi ng fresh ones because they fail to find the frozen ones. Second, one of the attributes included in this experiment is organic, but organic blueberries may not be very common in most marketplaces. Third, the stated preference experiment is relatively simple to conduct, and i ts convenience in obtaining measurement of the tradeoffs between product attributes is noticeable. Fourth, multiple observations for each respondent are obtained in the stated preference experiment, so estimates for each individual are ma de possible. Both main effects and first order interaction effects are included in the model. The purpose is to see whether the WTP for an attribute will be affected by the existence of other attributes We set freshness (fresh and frozen), production met hod (organic and conventionally produced), price and place of origin (whether they are produced locally, in the United States or imported from other countries) as attributes of blueberries. Four different prices are used for the price attribute: $1.50/pin t, $2.50/pint, $3.99/pint, and in which the respondent resides. Different combinations of the levels of the four attributes form a choice alternative. Each choice situ experiment more similar to real purchase situations where consumers can simply
24 choose not to buy any of the described blueberries. A constant to be estimated is two kinds of blueberry specified in each choice situation (no matter whether they exist in reality o r not). Respondents choose one kind from these two or they can choose in the experiment to achieve the maximum D efficiency Each respondent in the survey was asked to make choices in these 12 hypothetical situations. The prices are made negative in the estimation process so that the coefficients of the prices will have positive signs. This ensure s that the estimated scale parameters also have positive signs A ll the a ttributes and their corresponding levels included in the experiment are listed in Table 2 1 For identification purpose, we set frozen, imported, and conventional as the base and omit them in the model estimation. Data collection began in December 2010 an d ended in January 2011, with approximately 400 participants recruited for a total of 772 responses each month Since the choice experiment is based on hypothetical comparison, we do not expect the time of year (seasonality) to impact answers. 2.4 The Model Previous research in the field of food marketing has taken into account the diversity of consumer preference. However, m ost of the literature reported only aggregate estimates. The traditional aggregate estimates (population level estimates) of W TPs are not very informative as they cannot provide the information of each respondent. In this study, a mixed logit model is estimated by the hierarchical Bayesian (HB) approach. The Bayesian method, which is well suited to update estimates based on each he HB
25 approach also has irreplaceable advantages in finite sample inference (Rossi, Allenby, and McCulloch, 2005) and can generate the individual level estimates as byproducts (Allenby and Rossi, 1998). To obtain more sensible and accurate WTP estimates, we estimate the mixed logit model in the WTP space. 2 The laten t utility in the WTP space is: (2 1) W here P nit is the price of alternative i in s ituation t face by individual n X nit is the vector of other attributes, except price. W n is the vector of WTPs for attributes of individual n and is the error term. If respondents demonstrate a certain degree of randomness (i.e., the attention they paid to the task varied) in the decision process over the choice situations, it is necessary to divide both sides of the equation by the scale parameter s n ( s n >0 ) to account for this level of variation (Train and Weeks, 2005). The joint posterior for can be expressed as follows according to (2 2) W is assumed to be multivariate normally distributed and assuming that is distributed as log normal ensures positive values of the scale parameters. The prior distribution for the population mean ( ) of the multivariate normal distribution is 2 In the WTP space, WTP is directly estimate d instead of derived by the ratio of the coefficients of the attribute and price, as is done in the preference space.
26 assumed to be diffuse multivariate normal, and that for the population variance ( ) is assume d diffuse inverted Wishart. Nonzero covariance is allowed between the elements of W and The HB approach relies on the Gibbs sampler to obtain the draws of based on which population level parameter estimates are calculated. Thirty thousand iterations are taken during the burn in period (before convergence) and 5, 000 every other tenth draw 3 are retained after burn in to calculate the parameter estimates. Detail of the Gibbs sampler can be found in Robert and Casella (2004) or Casella and Geo rge (1992). The HB method is used because it gives out the individual level estimates as byproducts, so no additional procedures are needed (Allenby and Rossi, 1998). Details of the Bayesian method can be found in Rossi, Allenby, and McCulloch (2005) and T rain (2003). 2.5 Results Table 2 2 lists the demographic information of the choice experiment respondents. Among the respondents, 65 .0 % of them are females. The average age is approximately 47 and the average hous ehold income is about $53,403 approxim ately. 83.7 % of the respondents are Caucasian, 3.2% are Hispanic and 1 1.7 % are Black, or African American. The r est are Asian, American Indian, native Hawaiian or Pacific Islander, etc. 23.6 % of the respondents have a high school degree or equivalent, 58 .0 % have a four year college degree or some college, and 11 .1 % have attained a postgraduate degree. 3 In this way, correlation between subsequent retained draws can be reduced.
27 69.6 4.9 % of them have one or two children. 2.5.1 Mixed Logit Estimation Result The estimation result s of the mixed logit model are presented in Table 2 3 Most of the estimated population means of the WTPs are significant at the 5% level. The variances of the WTP distributions of all attributes are significantly different from zero at the 5% level. Thus, we conclude that consumer heterogeneity does exist. The estimated population means of the WTPs are significantly positive for fresh, local ly produced, and U .S. produced. This is consistent with our expectation that U.S. consumers generally pay more for fresh blueberries than for frozen ones and pay more for local and U.S. blueberries than for imported ones. However, the mean WTP estimate of organic is not significantly different from zero, which indicates that consumers in general might be indifferent about whether the blueberries are organic or not. The estimated means for all interaction terms are negative and are significant except for th e interaction between fresh and organic. The negative signs of the interaction terms can be explained as the outcome of the concavity of the utility function. 2.5.2 The proportions of respondents who fall in a certain range of WTP values for each attribute are shown in Tables 2 4 and Table 2 5 This method of summarizing individual level WTPs is used because of its robustness to extreme values. The comparison of esti mated WTPs for local and U.S. produced blueberries are shown in Table 2 4 While the proportions of positive WTPs for local and U.S. produced do not show much difference, the proportions of WTP estimates above $3/pint are all bigger for local than for U.S. produced. These figures indicate that most U.S.
28 consumers hold the identical attitude that U.S produced blueberries are superior to imported ones and, at the same time, some consumers are willing to pay more for local than for simply U.S. produced. The la st column of Table 2 4 lists the percentage of respondents whose WTPs for local are bigger than that for U.S. produced. All of the percentages exceed 65.7 %, with the largest reaching nearly 89.1 %. Therefore, the majority of the respondents prefer local ove r simply U.S produced blueberries. We conclude that locally produced blueberries can attract a larger price premium than si mply U.S. produced blueberries. This is also indicated by the larger population mean of W local compared to W U.S. in Table 2 3 Altho ugh the population mean of W org (Table 2 3 ) is not significantly different from zero, 24.6 % to 45.6 % of the respondents are willing to pay a positive premium for organic blueberries, depending on the other attributes (Table 2 5 ). The largest proportion (45.6 %) is for frozen and imported blueberries. Thus, the organic blueberries are most attractive when the other favorable blueberry attributes (fresh and U.S. produced) are not available. Although a certain proportion of respondents are demonstrating posi tive WTPs for organic attribute, only a small proportion are willing to pay more than $2/pint. In addition to the stated preference questions, participants answered questions designed to elicit their attitudes towards organic fruits and vegetables. Only 51 .9% of ic
29 WTPs for organic. The result from the stated preference experiment is consistent with the result from the attitude statement questions. Based on the average W TP for each attribute across the other attribute combinations, we find that 95. 1 % of the respondents are willing to pay more for U.S. produced blueberries than for organic ones and the proportion is 96. 4 % when we compare local and organic blueberries. Thus the results indicate that consumers place more emphasis on the country of origin attribute than they do on organic. One finding from the Bayesian estimates worth attention is that there are a large proportion of negative individual level WTP estimates fo r organic. One of the reasons might be the assumption of normal distribution for W which does not impose any restrictions on the signs of the WTP estimates. The other reason might be that some respondents ignored the attribute of organic because of indiff erence, especially when they saw other more favorable attributes in the choice sets. Therefore, we might judgments in the choice experiment. A final example of the individua l differences found is shown in Table 2 6 WTP values of three respondents (numbers 9, 65, and 618) are shown. In this case, the differences among the participants are clearly seen. For example, participant 65 is willing to pay more for fresh blueberries a nd respondent 618 is willing to pay more for organic blueberries, while respondent 9 places the highest value on the country of origin attribute. From this simple illustration, differences among consumers and the weights they place on each attribute are de monstrated. Although the average WTP for organic
30 blueberries across all the respondents is nearly zero, there are still respondents who demonstrate substantial positive WTPs for organic blueberries. 2.5.3 The Reliability of the Individual Level Estimate s To check the reliability of the individual level estimates, w e compute the mean absolute error (MAE) for the within sample prediction using the population WTP mean and individual level WTPs separately. The MAE 4 calculated from the population WTP mean is 0.2959 and that from the in dividual level WTPs is 0.0592. T he in sample fit improves dramatically by the use of disaggregate information. Therefore, the individual level estimates have the better performance in terms of in sample fit. Moreover, in the exp eriment, w e find that there are 57 respondents who always magnitude of the scale parameters of the nonparticipants are smaller t han those of the participants. Figure 2 1 compares the individual scale parameter estimates for non participants and participants We can see that the scale parameters of non participants are much smaller than those of participants and there is not much variation in the non by an almost horizontal line). In addition, from the choices of th organic blueberries whenever the choice situation included both conventional blueberries and organic blueberries, no matter what other attribute combinations were included. Their averages ac ross the six conditional WTPs for organic, which range from 4 A MAE value of 0 indicates that the choices are perfectly predicted.
31 $3.69/pint to $4.00/pint, are also the biggest among all the respondents. Therefore, although the number of observations for each respondent is not big enough (t=12), the updated Bayesian individu al level estimates are fairly informative. 2.5.4 Willingness to pay and Demographics To investigate the relationship of demographics with individual level WTP estimates, averages of the conditional WTPs across all types of blueberries are calculated for each respondent. A series of scatterplots that relate income, education, and age to the individual level WTP estimates for organic and locally produced blueberries are shown in Figure 2 1. The magnitude of the numbers on the horizontal axis stands for the level of income, education, and age and the magnitude of numbers on the vertical axi s s tands for the WTP estimates. The relationship between demographic information and WTP for organic blueberries is marginal, so is the relationship between demographics and WTP for locally produced blueberries. Upon further investigation, other demographic i nformation, such as race and gender, is also not found to have a significant relationship. Overall, demographics provide little 2.6 Conclusion and Discussion In this study, we applied a mixed logit model to the data from a stated preference to pay (WTP) for different blueberry attributes. Our results show that locally produced blueberries elicited the largest WTP values among all the attributes considered. Somewhat surprising is the result that, generally, consumers were indifferent to the production method (organic versus convention al) for blueberries. Less than 45.6 % of the respondents in the experiment had a positive WTP for organic blueberries and more than 95 .0 % of the respondents
32 placed more emphasis on the origin attribute than they did on organic. Though potentially surprising when compared to other fruits, this result is supported by the responses to the attitude statement ques tions. It can also be interpreted as the result of a quickly growing market, in which consumers have not adapted fast enough to demand organic blueberries yet. As a validation method, we estimated the model using additional data (data from the same experim ent but different sample of respondents) and the results (i.e., overall ranking of the importance of attributes, estimated parameters) are almost identical. Consumer preferences and attitudes are highly diversified. The traditional method of relying on de mographic information to explain consumer behavior may not always be effective, especially for small purchases like fruits or vegetables. While the purchases of choices of fruits might not be well differentiated by demographic characteristics. Our results show that although demographic information makes little contribution in explaining tradeoffs for blueberry attributes. The individual level WTP estimates provides us with valuable dis aggregate information that can help retailers differentiate consumers and set up more effective marketing strategies. With the rapid growth in information technology supermarkets can issue d ifferent coupons or leaflets to different consumers based on their individual attitudes instead of distributing them indiscriminately (Rossi, McCulloch, and Allenby, 1996). Coupons or brochures with sales featuring imported blueberries might only work with consumers who do not have a strong bias toward imported fruits (i.e., respondent
33 65 if compared with the other two respondents in Table 2 6 ). Retailers can also issue coupons of different values, depending on the pri ce sensitivities of consumers. Such dif profits. Moreover, consumer preference may not always be stable. The dynamic change in consumer preference or perception can also be captured by Bayesian estimation. There are several limitations of our study. First, our experiment was constrained by the limited space of the survey and concern on respondent burn out, so the number of observations per respondent was not enough to make an unbiased estimation of the individual level WTP estimates, though the estimates already enabled us to compare sample prediction criterion was used to compare the performances of individual level estimates and population level estimates. Out of sample prediction should also be conducted for a more comprehensive performance comparison.
34 Table 2 1. Blueberry attributes and levels in the choice experiment Attribute Level 1 Level 2 Level 3 Level 4 Freshness Fresh Frozen Place of origin Locally produced U.S. produced Imported Production method Organic Conventional Price ($/pint) $1.50 $2.50 $3.99 $5.99 Table 2 2. Demographics of choice experiment participants Independent Variables Percentage Female 65.0% Age (18 34) 29.2 % Age(35 60) 44.2 % Age(above 60) 26.7 % Caucasian 83.7 % Hispanic 3.2% Asian 1. 8 % Black or African American 11.7 % Other races 3.0 % Income($34,999 or below) 37.4% Income($35,000 $99,999) 54.4 % Income($100,000 or above) 2.7 % 5. 4% Postgraduate 11.1% College 58.0 % High school 23.6 % Other education level 7.3 % No child at home 69.6 % One or two children 24.9 % More than two children 5.6 % Note: The median of age is used.
35 Table 2 3 Estimation results of the mixed logit model in the WTP space WTP( W ) Distribution Mean Variance Fresh Normal 2.78 ** (0.20 ) 9.42 ** (1.34) Organic Normal 0.01 (0.20 ) 1.72 ** (0.4 4 ) Local Normal 3.94 ** (0.19) 12.43** (1.52) U.S. Normal 3.02 ** (0.17) 8.47** (1.17) Fresh_org Normal 0.38 (0.28) 1.10 ** (0.35) Fresh_local Normal 0.69 ** (0.29) 17.23** (2.65 ) Fresh_us Normal 0.73** (0.33) 15.21** (2.23 ) Org_local Normal 0.85 ** (0.14) 1.52 ** (0.38 ) Org_us Normal 0.58 ** (0.22 ) 2.81 ** (0.51 ) Constant Normal 1.63 ** (0.20 ) 15.31 ** ( 2.56 ) 1/s Lognormal 4.48 53.77 Simulated Log likelihood 5801.7 Notes: ** and indicate that the parameter is significant at 5% level and 10% level respectively. Figures in the parenthesis are standard errors. The mean and variance of 1/s are for the simulated distribution of 5000 exponen tiated random draws from N (0. 84 1.32), which is the estimated distribution of log(1/s).
36 Table 2 Other Attributes Positive WTP for Local (%) Positive WTP for U.S. Produced (%) Respondents who Prefer over >3 $/pint >0 $/pint >3 $/pint >0 $/pint Fresh and Conventional 49.35 83.63 35.81 81.15 89.12 Frozen and Conventional 49.47 84.92 44.40 84.57 77.20 Fresh and Organic 38.16 79.03 30.39 66.55 65.67 Frozen and Organic 37.81 80.80 29.45 89.28 66.84 Table 2 5. Summary s tatistics of r o rganic Other Attributes Positive WTP for Organic (%) (%) (%) Total (%) Fresh and Imported 28.27 5.54 33.80 Fresh and Local 18.37 6.24 24.62 Fresh and U.S. 22.50 7.66 30.15 Frozen and Imported 42.64 2.94 45.58 Frozen and Local 21.55 3.53 25.09 Frozen and U.S. 29.56 7.18 36.75 Note: The unit of the ranges is dollars/pint.
37 Table 2 6. WTPs of three respondents WTP($/pint) Respondent # Organic Local U.S. Produced Fresh 65 0.04 1.90 0.56 5.04 9 0.97 4.14 3.36 1.19 618 3.69 2.98 2.17 1.51 Note: The WTP values are the averages across the corresponding conditional WTPs.
38 Figure 2 1. Scatterplots of WTP estimates versus demographic information A) Organic versus income B) Organic versus education C) Organic versus age D ) Local versus income E ) Local versus education F ) Local versus age -10 -5 0 5 10 0 2 4 6 8 10 WTP for organic Income A -10 -5 0 5 10 0 2 4 6 8 10 WTP for organic Education B -10 -5 0 5 10 0 2 4 6 8 10 WTP for organic Age C
39 Figure 2 1. Continued -10 -5 0 5 10 0 2 4 6 8 10 WTP for local Income D -10 -5 0 5 10 0 2 4 6 8 10 WTP for local Education E -10 -5 0 5 10 0 2 4 6 8 10 WTP for local Age F
40 Figure 2 2. Comparison of scale parameters for participants and non participants 0 1 2 3 4 5 6 7 8 9 10 participants nonparticipants
41 CHAPTER 3 ON MODEL SPECIFICATION OF THE MIXED LOGIT: THE CASE OF CONSUMER PERCEPTION ON BLUEBERRY ATTRIBUTES 3.1 Introduction The mixed logit model, which can account f or consumer heterogeneity, has been used by many researchers in the study of consumer preference. The specifications of the mixed logit model h ave also been widely discussed. Train and Weeks (2005) and Sonnier, Ainslie, and Otter (2007) discussed two ways of placing distributional distributional assumptions on the coefficients in the utility function and then derive the distributions of willingness to pay (WTP). The other is to p lace distributional assumptions directly on WTP by transforming the utility function. Numerous studies have found that the WTP space model provides more plausible WTP estimates than the preference space model (Train and Weeks, 2005; Scarpa, Thiene, and Tra in, 2008 etc.). As a results, the WTP space specification is now used more frequently due to its Willis, 2010; Thiene and Scarpa, 2009; Balcombe, Fraser, and Harris, 2009 etc.) Se veral studies have compared the preference space and the WTP space models. The five major studies are listed in Table 3 1. Train and Weeks (2005) used lognormal distribution for the price coefficient and lognormal or normal distribution for the other attri bute coefficients depending on whether the attribute is favorable or not. Sonnier, Ainslie, and Otter (2007) and Scarpa, Thiene, and Train (2008) also assumed lognormal distribution for the price coefficient but only normal distributions for the other coef ficients. They all concluded that the WTP estimates from the WTP space models were more reasonable in terms of magnitude and dispersion than those from the
42 preference space because additional transformation led to excessive extreme values. Balcombe, Chalak and Fraser (2009) estimated 20 models, which included models with fixed parameters models with parameters with various random distributions, models in the WTP space or in the preference space and models with and without misreporting. They compared the models based on logged marginal likelihoods and concluded that the WTP space model with lognormal distribution for the price coefficient and normal distribution for all the other parameters is the best among the 20 models. The best model in the preference space is lognormal for price coefficient and censored normal for the other parameters. They found no support for fixing the price coefficient. Hensher and Greene (2011) compared four models: preference space model and WTP space model both with unconstraine d triangular distributions for the random parameters, preference space model with constrained triangular distributed random parameters and a generalized mixed logit in the preference space that also used unconstrained triangular distributions for all the r andom parameters. They concluded that imposing constrained distributions on the random parameters in the preference space model reduced the gap between the WTP estimates from the preference space model and the WTP space model. Among the five studies, Train and Weeks (2005), Sonnier, Ainslie, and Otter (2007) and Hensher and Greene (2011) found better statistical fit for the WTP space model while Balcombe, Chalak, and Fraser (2009) and Scarpa, Thiene, and Train (2008) reached an opposite conclusion in terms of model fit. Ruud (1996) argued that a fully random specification of th e model was barely identified. Revelt and Train (2000) suggested that the randomness of the price
43 coefficient would make the distribution of the WTPs hard to evaluate (e.g. Normal dis tribution for attribute coefficients and lognormal for the price coefficient will make the calculation of WTPs the ratio of normal and lognormal random variables). They also proposed that the distributions commonly specified for the price coefficient had v arious problems, such as positive price coefficient estimates from normal distributions and extremely small price coefficients from log normal distributions, which made the derived WTP estimates poorly behaved. Since additional transformation of the estim ated parameters in the preference space models with a fully random specification is one of the reasons for the relatively poor performance of the estimates in WTP estimation, one of the solutions might be to fix the price coefficient in the preference spac e. In this way, the WTP estimates will have the same distribution in these two spaces and differ just in their estimation methods. Many researchers have made such assumptions in their studies (Hensher, Shore, and Train, 2005; Layton and Brown, 2000; Lusk, Nilsson, and Foster, 2007; Lusk and Schroeder, 2004; Provencher and Bishop, 2004; Revelt and Train, 1998, etc.). However, although fixing the price coefficient might be a good alternative s o that no additional transformation would be needed in calculating WTPs, such constraint may lack face validity since the marginal utility of price is not always constant (Scarpa, Thiene, and Train, 2008) Although the WTP space model has advantages over the traditional preference space model, the estimation of WTP space models is not as straightforward as that of the preference space model. Additional transformation is needed and the scale parameter needs to be introduced to avoid fixing the price coefficient as one.
44 Balcombe, Chalak and Fraser compared twenty widely use d specifications of mixed logit models and ranked the models based on the values of maximum likelihood. According to this criteria, the model with fixed price coefficient and the one with random log normal price coefficient in preference space ranked numbe r eight and sixteen, respectively and the model with random lognormal price coefficient in WTP space ranked the first. However, economists are more interested in the welfare measures such as WTP derived from the econometric models rather than the maximum l ikelihood values. In this article we attempt to explore the impact of mixed logit model specification on estimation results with a focus on their efficiency in estimating WTP. We compare three specifications of the mixed logit model: fixed price coefficien t preference space model (all other coefficients are random) ; fully random preference space model ; and fully random WTP space model in terms of statistical fit, aggregate WTP estimates and individual level WTP estimates using the data from a stated prefer ence experiment. According to Hensher and Green (2011), constrained distributions for the random parameters in the preference space can help reduce the gap between the WTP estimates from the preference space and the WTP space models. Our study differs from the study of Hensher and Greene (2011) by placing a constraint only on the price coefficient instead of on all the parameters. The preference space models with fixed price coefficient have been widely used by many applied researchers, but the efficiency o f WTP estimates from suc h models has not been evaluated comprehensively. If this constraint of fixing the price coefficient does narrow the gap, using the prefrence space model with fixed price coefficient instead of using the WTP space model is suggested
45 because using preference space models help avoid the additional steps in WTP space models. 3.2 The Model The latent utility in the WTP space can be express ed as (Train and Weeks, 2005): (3 1) P nit is the price and X nit is the vector of attributes, except price, of alternative i in situation t face by individual n n1 and n2 are the utility coefficients of individual n In the mixed logit model, the utility coefficients are assumed random while in the tr aditional conditional logit model, the coefficients are specified a s constant over the consumers. For the mixed logit model in the preference space, the probability that individual n would choose alternative j in choice situation t is: (3 2) is the dis tributional assumption for the utility coefficients 1 is called the population level parameter (hyper parameter) as it describes the distribution of over the w hole population (Train, 2003). Many distributional assumptions can be used for depending on the behavioral assumption on consumers. The mixed logit model in the WTP space is obta ined by dividing both sides of E quation ( 3 1) by the coefficient of price: (3 3)
46 W n is the vector of WTPs for attributes of individual n If respondents demonstrate a certain degree of randomness (i.e., the attention they paid to the task varied) in the decision process over the choice situations, it is necessary to divide both sides of the equation by the scale parameter s n ( s n >0 ) to account for this level of variation for each individual (Train and Weeks, 2005). The probability that individual n would choose alternative j in choice situation t in the WTP space is as follows: (3 4) where is the population parameter and is the joint distributional assumption for W n and Therefore, WTP estimates do not need to be derived by the ratio of two random parameters as is done in the preference space. It is directly specified in the WTP space. In E quation ( 3 2), n1 is assumed multivariate normal and n2 is assumed log normal to ensure the positive sign of the coefficient of ( P nit ) In E quation ( 3 4 ), W n is assumed to be multivariate normally distributed and assuming that is distributed as log normal ensures positive values of the scale parameters and thus the coefficient of ( P nit ) The Hierarchical Bayesian method is used to estimate the mixed logit model because the HB approach has irreplaceable advantages in finite sample inference (Rossi, Allenby, and McCulloch, 2005) and can generate the individual level estimates
47 as byproducts (Allenby and Rossi, 1998). The prior distributions for the population mean of the parameters n1 W n log(1/s n ) and n2 ) are assumed to be dif fuse multivariate normal, and that for their population variance is assumed diffuse inverted Wishart. Nonzero covariance is allowed between the coefficients in each model. The joint posterior for in the preference space can be expressed as follows accor (3 5) The joint posterior for in the WTP space can be expressed as follows: (3 6) If the price coefficient is assumed fixed, n2 in E quation ( 3 5 ) is constant over n so no distributional assumption is needed. In this study, 3 0,000 iterations were taken during the burn in period (before convergence) and 5, 000 every other tenth draw 5 were retained after burn in to calculate the parameter estimates with the Gibbs sampling. Detail of the Gibbs sampling can be found in Robert and C asella (2004) or Casella and George (1992). The HB method is used because it gives out the individual level estimates as byproducts, so no additional procedures are needed (Allenby and Rossi, 1998). Details of the Bayesian method can be found in Rossi, All enby, and McCulloch (2005) and Train (2003). 5 In this way, correlation between subsequent retained draws can be reduced.
48 3.3 Case Study: Consumer Preference for Blueberry Attributes In the late 1990s, scientific research revealed special health benefits of blueberries, including high levels of antioxidant properties. Likely as a result of this, and of other factors, p er capita consumption of fresh blueberries has increased dramatically since 2000 (USDA, 2010). Consumer choice of fruit has become complicated, as even the same type of fruit, for example, blueberries, can have multiple attributes. Production method, origin of products, and form of the fruit (i.e., frozen versus fresh) are among a number of attributes (appearance, flavor, price, etc.) that consumers consider when purchasing on their preferences. Some consumers may prefer fresh blueberries over frozen ones, while other consumers may prefer frozen blueberries because of their long shelf life. As for the credence attributes, such as production method (i.e., organic) and product preference also demonstrates large variance Some consumers may consider country of origin of blueberries an important attribute, while others may think organic attributes as the most important. A stated prefere nce experiment was conducted to explore ption of blueberry attributes. Although there might be some potential problems associated with the stated preference approach, such as exaggerated WTP estimates (Brooks and Lusk, 2010; Johansson Stenm an and Svedster, 2008), it has been widely used by researchers (e.g., Brooks and Lusk, 2010; Gao and Schroeder, 2007; Alfnes, 2004; Carlsson, Frykblom, and Lagerkvist, 2004). There are several reasons for it to be the suitable approach for the study of co nsumer choices of blueberries. For example, it is not usual for consumers to have access to a great variety
49 of blueberries (all possible sources, forms, or prices) in the same section of grocery stores when they are shopping for blueberries. Therefore, re al transaction data may not be reliable. Additionally, using non hypothetical valuation methods such as experimental auction is difficult because some products are hard to find in the market and non hypothetical methods are too costly for collecting large sample data. Table 3 2 lists all the attributes and their corresponding levels included in the choice experiment. We set freshness (fresh and frozen), production method (organic and conventionally produced), price and place of origin (whether they are pro duced locally, in the United States or imported from other countries) as attributes of blueberries. Per capita consumption of fresh blueberries in the U.S. has increased substantially since 2000 (USDA). However, the increasing trend for per capita consumpt ion of frozen blueberries is marginal over the years (USDA, Economic Research Service Calculations). Therefore, fresh and frozen blueberries are included in the stated preference experiment to detect the price premiums of fresh blueberries over frozen ones Organic and local are two widely discussed fruit attributes in agricultural economics literature (Batte et al., 2007; Loureiro, McCluskey, and Mittelhammer, 2001; Dentoni et al., 2009 etc.) we include them in the experiment for comparison purposes. Four commonly seen blueberry prices in grocery stores$1.50/pint, $2.50/pint, $3.99/pint, are defined as those produced within the state in which the respondent resides. A main effect fractional factorial design is used to generate 12 choice sets in the choice experiment. Each respondent in the survey was asked to make one choice from
50 each of these 12 hypothetical choice sets. In the choice experiment, different combinations of the levels of the blueberry attributes form a choice alternative. Those alternatives are used to construct a choice set. Each choice set consists of two bluebe rry products and a NONE option. The purpose of including NONE option is to make the choice ex periment more similar to real purchase situations where consumers can simply choose not to buy any of the described blueberries. When estimating the three models previously discussed, a constant is assigned to the utility associated with the NONE option in the models. The prices are made negative in the estimation process to ensure that the results have positive signs for the estimated scale parameters. For identification pur pose, we set frozen, imported, and conventional as the base and omit them in the model estimation. 3.4 Results A survey containing the stated preference experiment was distributed online to consumers living in the Northeastern and Southeastern regions of the United States. Data collection began in December 2010 and ended in January 2011 with approximately 400 participants recruited on a monthly basis. After deleting unqualified or incomplete responses, 772 responses were used for the analysis. The demo graphic information of the respondents is the same as it in Chapter 2 since the same data is used. E stimates of coefficients for the three models are presented in Table 3 3 In the first model (Model 1), all the coefficients are random in the preference s pace. The coefficient of price is assumed to have a lognormal distribution and all the other coefficients are normally distributed. The second model (Model 2) is also specified in the preference space. The coefficient of price is assumed fixed while all th e other
51 coefficients are r andom and normally distributed. The third model (Model 3) is the WTP space model with all the parameters being random. The coefficient of the price conforms to lognormal distribution as the scale parameter is positive and all the other parameters are normally distributed. The estimated coefficients of fresh, U.S. produced and local are significantly positive and the relative attribute importance rankings are consistent in all the three models. Surprisingly, the coefficients of org anic are significantly negative in all three models. The reason might be that consumers are still unfamiliar with organic blueberries and the impact of the local attribute surpasses the impact of being organic. The variances of the parameter distributions for all the attributes are significantly different from zero at the 5% level in all the three models. Thus, the models reach the same conclusion that heterogeneous preference among consumers exist. According to the results, consumers regard the local produ ction of blueberries the most favorable attribute. The next two favorable attributes are U.S. produced and fresh. Based on the results of Model 3, on average, consumers are willing to pay $3.45/pint and $2.49/pint to exchange imported blueberries for local ly produced and U.S. produced ones, respectively. The price premium of fresh blueberries over frozen ones is $2.24/pint. However, these numbers may be exaggerated as previous studies have found exaggerated WTP estimates from hypothetical contexts ( List and Gallet, 2001; Lusk and S c hr o eder, 2004; List and Shogren, 1998; Lusk and F o x 2003 etc.). The simulated log likelihood value is the highest ( 5722.0) for Model 1 and is the lowest ( 6111.6) for Model 2. These results are consistent with Train and Weeks (2005) and Sonnier, Ainslie and Otter (2007) in that the fully random preference space model
52 fits the data better than the fully random WTP space model with stated preference data. In addition, our results show that the fully random WTP space model fits the data better than the fixed price preference space model, which indicates that imposing cons traints on parameters decreases the model fit more than transforming the model from the preference space to the WTP space. However, Model 2 and Model 3 give more efficient estimates of parameters because the standard errors of the parameters from those two models are smaller than those from Model 1. Table 3 4 gives the means and variances of the individual level WTP estimates for Model 1 are much larger than those from Model 2 and Model 3. The mean WTP estimates from Model 2 and Model 3 are quite similar. In addition the variances of the individual level WTP estimates from Model 1 are much larger than those from the other two models, indicating more efficient estimates of WTPs from Model 2 and Model 3. The estimates of variances for WTPs from Model 2 and Model 3 do not demonstrate much difference. Therefore, our results indicate that Model 2 and Model 3 produce quite similar results regarding WTP estimates. This is consi stent with the conclusion of Hensher and Green (2011) that the difference between the estimates from the preference space and the WTP space narrows when a constraint is imposed on the preference space model (Hensher and Greene assumed a constrained triangu lar distribution for the random coefficients in the preference space). Wedel et al. (1999) and Rossi, McCulloch, and Allenby (1996) both suggested the importance of individual level purchase information in marketing, so the comparison of WTP at the indivi dual level becomes critical. Table 3 5 shows the correlation of the
5 3 individual level WTP estimates from the three models. The individual level WTP estimates for each attribute in the preference space are obtained by taking the ratio of n1 and n2 where n2 is the coefficient of price, for each individual in each iteration. Such method helps reduce the occurrence of extreme values of the derived WTP estimates (Hensher, Green and Rose, 2006). Clearly, all the correlations are significant and the correlatio n between WTP from Model 2 and Model 3 is the strongest. Almost all the correlations a coefficient of 0.893. Therefore, the gaps of WTP estimates between the preference space model with fixed price parameter and the WTP space model with fully random parameters are not only small in terms of population level estimates, but also in terms of individual level estimates. As a result, Model 2 might serve as a good substitute for Model 3 in estimating WTPs in stated preference experiments. The advantage of such practice is that Model 2 is simpler to estimate and is not subject to the identifiability concerns (Ruud, 1996). 3.5 Conclusion and Discussion In this study, we compare three different mixed logit models using the data from a stated preference experiment. We find that when the price coefficient is fixed in the preference space, the preference space model (all the other parameters are random) and the fully random WTP space model produce quite similar results, at both population level and individual level WTP estimates. However, there are large differences between the WTP estimates from the two spaces when the price coefficient is assumed random. Our result is consistent with previous literature in that the WTP estimates in the WTP space are more reasonable than those from the preference space in terms of magnitude and dispersion when all the parameters are random. The narrowed gap between results
54 from the preference space with fixed price parame ter and WTP space indicates that the fixed price preference space model might be a better option than the fully random preference space model. In addition, our results indicate that the fixed price preference space model might be used as an alternative to the WTP space model since the model specification in the WTP space needs extra steps. Further justification of using a fixed price preference space model are that this model produces more accurate parameter and WTP estimates, and the mean WTPs for all attr ibutes are lower than those from a fully random preference space model. Considering that stated preference surveys usually suffer from the problem of overestimation of consumer WTP, the lower estimates of WTP from fixed price preference space and fully ran dom WTP space models imply that those two models can, to some extent, alleviate the problem of hypothetical bias. Although our model is estimated by HB method, Huber and Train (2001) concluded that the classical and Bayesian methods resulted in almost equ ivalent individual level estimates. Therefore, we expect that our results will also hold when the model is estimated with the classical procedure (Maximum simulated likelihood method). However, whether our results can be extended to other cases is still su bject to more empirical validation.
55 Table 3 1. Literature comparing models from preference space and WTP space Literature WTP space Preference space Better Fit Price Others Price Others Hensher and Greene (2011) unconstrained triangular unconstrained triangular unconstrained, constrained triangular unconstrained, constrained triangular P Balcombe, Chalak, and Fraser (2009) lognormal, triangular normal, Censored normal fixed, normal, lognormal, triangular fixed, normal, censored normal W Scarpa, Thiene, and Train (2008) lognormal normal lognormal normal W Sonnier, Ainslie, and Otter (2007) lognormal normal lognormal normal P Train and Weeks (2005) lognormal lognormal normal lognormal lognormal, normal P Note: P represents preference space. W represents WTP space.
56 Table 3 2. Blueberry attributes and levels in the stated preference experiment Attribute Level 1 Level 2 Level 3 Level 4 Freshness Fresh Frozen Place of origin Locally produced U.S. produced Imported Production method Organic Conventional Price ($/pint) $1.50 $2.50 $3.99 $5.99 Table 3 3. Estimation results of the three models Parameter Distribution Model 1 Model 2 Model 3 Mean StDv Mean StDv Mean StDv Price lognormal 0.09 (0.08) 1.71 ** (0.24) 0.90 ** (0.03) 0.42 ** (0.07) 1.17 ** (0.16) 2.13 4.15 2.67 3.91 Fresh Normal 2.86 ** (0.17) 8.82 ** (1.02) 2.21 ** (0.12) 5.56 ** (0.58) 2.24 ** (0.14) 5.76 ** (0.59) Organic Normal 0.70 ** (0.12) 5.87 ** (0.65) 0.67 ** (0.09) 3.88 ** (0.39) 0.64 ** (0.10) 3.50 ** (0.37) Local Normal 4.32 ** (0.20) 10.48 ** (1.22) 3.23 ** (0.15) 7.21 ** (0.73) 3.45 ** (0.16) 8.90 ** (0.92) U.S Normal 3.18 ** (0.17) 7.26 ** (0.94) 2.42 ** (0.12) 4.04 ** (0.49) 2.49 ** (0.13) 5.19 ** (0.57) Constant Normal 3.69 ** (0.51) 65.40 ** (7.86) 1.59 ** (0.19) 17.16 ** (1.71) 2.11 ** (0.18) 14.40 ** (1.64) Simulated Log likelihood 5722.0 6111.6 5993.0 Note: The numbers in bold are the mean and variance of price coefficients implied by the estimated distribution parameter. Use 5000 draws from the corresponding normal distributions, exponentiate each of them and then calculate the mean and variance. Model 1: All random (Preference space); Model 2: Fixed price (Preference space); Model 3: All Random (WTP space); ** indicates significance at 5% level. The numbers in the parenthesis are the standard errors.
57 Table 3 4. M eans and variances of the individual level WTP estimates Attribute Model 1 Model 2 Model 3 Mean Med ian Vari ance Mean Median Vari ance Mean Median Vari ance Fresh 3.33 1.79 5.29 2.46 2.15 2.08 2.24 2.02 1.84 Organic 1.05 0.32 4.2 0.74 0.86 1.77 0.64 0.62 1.46 Local 5.1 2.73 6.13 3.59 3.19 2.33 3.45 3.16 2.35 USA Produ ced 3.79 1.97 4.84 2.69 2.42 1.68 2.49 2.28 1.76 Note: Model 1: All random (Preference space); Model 2: Fixed price (Preference space); Model 3: All Random (WTP space) Table 3 5. Correlation between individual level WTP estimates among the three models. WTP Model 1 and Model 2 Model 2 and Model 3 Model 1 and Model 3 Fresh 0.70 ** (0.00) 0.94 ** (0.00) 0.73 ** (0.00) Organic 0.76 ** (0.00) 0.98 ** (0.00) 0.75 ** (0.00) Local 0.66 ** (0.00) 0.90 ** (0.00) 0.66 ** (0.00) USA produced 0.64 ** (0.00) 0.89 ** (0.00) 0.63 ** (0.00) Note: The numbers in the parentheses are p values of the correlation coefficients. ** indicates significance at 5% level.
58 CHAPTER 4 WILLINGNESS TO PAY FOR ORGANIC AND LOCAL BLUEBERRIES: A MULTI STORE BDM AUCTION CONTROLLING FOR PURCHASE INTENTIONS 4.1 Introduction The Becker DeGroot Marschak method (BDM) is well suited for field experimental auctions (Becker, Degroot, and Marschak, 1964) and as a result, has been increasingly food (Wertenbroch and Skiera, 2002; Carrigan and Rousu, 2008; Silva et al., 2007; Rozan et al., 2004; Rousu et al., 2005 a nd Lusk et al., 2001 etc.). The most common grocery stores and the procedure is relatively simple compared with other auction formats. In a typical BDM auction, the participa nt places a bid for the auctioned product and then draws a random price. If the randomly drawn price is greater than the bid, the the product and pays the price he/she draws. The bidding behavior of participants is affected by many factors. cognitive structure is complex and a lot of questions still remain unanswered. Many researchers have accepted the theory that instead of h a ving fixed or previously formed values, consumers often construct differe n t assessme n t co n tinge n t on differe n t s p ots or c hoice co n texts (T v ersky and Simonson 1993; Pa yne, Bettman, and Johnson 1992; Bettma n Luce, and P a yne 1998). Horowitz (2006) valuation of an item is correlated with the circumstances in which he/she is going to pay for it. People go to grocery stores to buy food, so whether they intend to buy the auctioned products when the auction is conducted might be a fa ctor that would affect
59 Since real money is usually involved in the auction process, if auction participants do not plan to purchase the auctioned products, they may bid less than they usually pay for the product and use the money fo r the products that they plan to purchase. Corrigan and Rousu (2008) argued that purchase intention critically to pay (WTP) for perishable goods. They found that auction participants who planned to buy the auctioned produc ts that day submitted bids equal to their perceived market prices of the products while participants who did not have such plan submitted bids less than the market prices. People report valuation according to their goals at that time, task environment, des cription of the task and more factors, though many of which might be difficult to measure, that can influence Another important point wo rth attention for BDM auctions is the choice of auction locations. One of the biggest advantages of BDM auctions over the other auction formats is its point of purchase locations, which offer participants a more realistic choice situation (Carson et al. 19 94). However, the locations should be chosen with caution and the potential impact of location choices on auction results should be considered. There are many types of marketing outlets that feature different types of food (e.g. organic, local etc.). Food quality, availability and price expectations may differ depending on the location of purchase. The co existence of different marketing outlets indicates that consumers have different demands and expectations for food. These different attitudes toward food may drive a consumer to select an outlet they believe to satisfy their demand for specific attributes. For example, people who prefer organic food
60 may go to a high markets. People wh Differences in prices, food quality or availability of organic or local fruit across different locations. Sin ce there are numerous types of marketing outlets that attract different consumer groups, the generalizability of a BDM auction result conducted at only one store, or one store type, is questionable. For example, if one were to investigate the WTP for a loc al food product, but conducted BDM auctions at a grocery store, they might be missing a segment of the population that values local higher and purchases their good from farmers markets. On the other hand, if the auction is only conducted at rket, the valuation might represent a subset that values the attribute more than the average consumer. Thus, i f a broader, more complete picture of consumer demand for attributes is wanted, reaching more consumer types than found at one type of location ma y be needed In this study, we extend previous literature in two directions. First, we extend the study by Corrigan and Rousu to explore the effect of purchase intentions on partial bids (e.g. for a value added attribute of the products) in addition to f ull bids (for the whole instead of the whole product. As such, it is important to understand the impact of butes in experimental auctions to provide a recommendation for the need to control for this variable in studies that focus on attributes over whole products.
61 Second, we conduct BDM auctions in three different types of marketing outlets: grocery stores tha t focus on delivering products at the lowest price (price conscious), grocery stores that focus on high quality (quality Although previous literature did explore the impact of conducting auctions at the point of purchase, su ch as grocery stores (Lust and Fox 2003; Shogren et al. 1999 etc.), the experiment was usually conducted only in one type of location. Although there are attributes in diffe rent countries (Lusk et al. 2004; Lusk, Roosen, and Fox 2003), little research has been conducted using auctions to consider whether such differences also exist across different types of marketing outlets. If such difference does exist, it may be recommend ed that field auctions be conducted in multiple locations or locations with consumers who are more representative of the population being studied so that the auction results can be generalized. The study seeks to answer the following questions: 1. How does partial bids and full bids? 2. Will the BDM auction yield different results at different marketing outlets? 4.2 Literature Review Literature has re v ealed exaggerated willingness to p a y (WTP) estimates due to the h y p othetical nature of some v aluation methods. For example, List and Gallet ( 2001) suggested that on av erage, preference in the h y p othetical setting w as inflated b y a factor of 3. Therefore, researchers began to rely on non h y p othetical meth o ds, su c h as ex p erime n tal auctions (EA), to elicit WTP for pr o ducts or additional
62 attributes ( Rozan et al. 2004; Jaeger et al. 2004; Melton et al. 1996; Huffman et al. 2003, H a y es et al. 2002, Bernard and Bernard 2009, H ay es et al. 1995, U m b erger et al. 2002, Alfnes and Ri ck ertsen 2003 etc.). Over time, locations of experimental auctions were no longer limited to labs and have been extended to more realistic contexts, such as grocery stores, or other point of purchase locations ( Rozan et al. 2004; Rousu et al. 2005; Lusk and Fox, 2003; Lusk et al. 2001; Silva et al. (2007); List 2001 and List and Lucking Reiley 2000 etc.). While lab market goods (i.e., environm ent conservation, food safety and novel good), field auctions have been gaining popularity in the valuation of market goods and empirical consumer behavior studies. Auctions conducted in a retail context have a number of advantages over lab auctions, such as a more realistic circumstance for participants, a better capture of the target population and reduced compensation and recruiting fees for the investigator (Lust and Fox 2003; Lusk et al. 2001). Carson et al. (1994) argued that the choice context of val ue elicitation should be as close as possible to the real purchase situation. Wertenbroch and Skiera (2002) found that different from hypothetical methods, there was no overbidding in BDM auctions. Carrigan and Rousu (2008) concluded that participants in t he BDM auction understood that the auction was demand revealing, which indicated that their bids were unbiased. However, literature has reported value elicitation results that are inconsistent with basic preference theories. For example, in a sports card auction conducted in a sports card show, List (2002) found that the superior bundle was preferred when juxtaposed with the inferior bundle. However, when in isolation, the inferior bundle was valued
63 much higher. The paper argued that the dichotomous choice questions might be challenged and various other nonmarket valuation methods should be examined. Thaler dramatically across different points of purchase (a run down grocery st ore and a fancy hotel). Additionally, despite many studies that concluded with exaggerated WTP estimates in hypothetical context (Lusk and S c hr o eder 2004; List and Shogren 1998; Shogren et al. 1999 etc.) Lusk and Fox (2003) found that auctions conducted in stores yielded larger bids than lab auctions, given that participants were happy to participate and were not constrained by cash (not having enough cash to pay for the auctioned product if they win). Mo re research is needed to understand the inconsistencies that have been found. Irwin et al. (1993) argued that people actually do not have an exact monetary value for goods, except for those very simple and familiar. As a way to explain the gap between theo ry and action, numerous studies have been done to test the generalizability of previous findings by controlling more consumer side, environment side, or methodology side factors that might deviate from theoretical assumptions and affect consumer behavior i n research experiments. For example, on the consumer side, Lusk Corrigan and Rousu (2008) analyzed the effect of purchase intenti value) on reducing hypothetical bias in elicitation procedures. The results were shown to depend on
64 (2001) argued that the extent of over statement of preferences was related to the distinction between public and private goods, WTP and willingness to accept (WTA) and differen t elicitation methods. Such empirical findings serve as useful warnings for products, and should be controlled for when possible in experimental auctions. 4.3 Auction Procedure A series of BDM auctions were conducted in July and August 2011 in Pittsburgh, Pennsylvania and Orlando, Florida. In each city, two grocery stores (one price conscious and one quality Approximately 70 observations were collected in each location, with a total number of qualified observations of 356. The auctions were set up at the entrance to the stores randomly stopped and a sked whether they would like to participate in an experiment about food consumption. Qualified consumers (adult, blueberry consumers without food allergies) were asked to fill out a questionnaire about purchase intention, demographics etc. Four types of b lueberries ( organic and locally produced blueberries, conventional and locally produced blueberries, organic blueberries produced in the U.S. and conventional blueberries produced in the U.S. ) were auctioned. Before the auction, each participant was given a sheet explaining the auction procedure. The auctioneer then explained the auction procedure to the participants and answered questions from the participants about the procedure. The auction procedure had four steps:
65 1. Each participant was endowed with $7 cash, which they were told could be used to purchase the blueberries if they won the auction or was theirs to keep if they did not win. The participants were asked to bid simultaneously for four types of blueberries in one pint clamshell packages. The a uctioneer explained to the participant that it was his/her best interest to bid exactly what he/she was willing to pay for each blueberry package. 2. After the participants placed the bids, they randomly drew a letter to determine which blueberry was actua lly auctioned. Therefore, they only had the opportunity to buy one type of blueberries. 3. Once the auctioned blueberry type was determined, participants were asked to draw a random price. 4. rries was higher than the price drawn for those blueberries, the participant purchased the blueberries at the price, the participant did not purchase the blueberries. 4.4 Results A demographic summary of participants at each marketing outlet is shown in Table 4 1. Demographics does differ by location, with the price conscious marketing outlet featuring consumers who tended to be younger, had lower income, had a lower education level, and were more likely to be Black or African American. They also tended to have m ore children than consumers at the other two marketing outlets. The level. The variation in demographics at different marketing outlets indicates that the auctions may produ ce different full bids and partial bids at different locations.
66 Summary statistics for the four bids for the four types of blueberries at each marketing outlet are shown in Table 4 2. The bids for organic, local blueberries (bid1); conventional, local blu eberries (bid2); and conventional, U.S. produced blueberries conscious stores. The bids for organic, U.S. produced blueberries (bid3) are the highest in the quality focused stores a nd the lowest in the price conscious stores. 4.4.1 Partial Bids for Organic and Local Blueberries at Different Marketing Outlets In the survey before the auction, information on attitudes toward organic and local fruits was collected. Significant differ ences are found among the answers from participants at different locations. Specifically, participants at the quality focused grocery stores demonstrate more trust in organic fruits than participants at the price conscious grocery stores. They are more lik ely to agree to a statement that they will pay more for fruits with an organic label than participants at the price conscious grocery stores (on a 5 likely to indicate that they will purchase local blueberries than those from the other two locations (on a 5 point Likert scale). The means and standard deviations of bids for organic and local blueberries at different marke ting outlets are shown in Table 4 3 According to the B onferroni multiple comparison test, participants at the quality focused stores, which are known for selling organic food, have higher bids for organic blueberries than participants at the price markets have the highest bids among the three marketing outlets. The results are consistent with our expectation that the experiment will yield higher bids for organic blueberries in quality focused
67 grocery stores and higher bids for local blueberries in f bids for organic blueberries are larger than the bids for local blueberries, no matter where the auction was conducted. This indicates that consumers generally consider the organic production of blueberries a more importan t attribute than production location. However, the standard deviations of bids for organic blueberries are all bigger than toward the organic production of blueberries. 4.4.2 Imp a ct of Purchase Intention people will bid differently if they plan to buy blueberries on the day of auction. To control for such impact, participants were asked to indicate if they planned on purchasing fresh blueberries at the store/market. Participants from different marketing outlets but with the same purchase intentions are grouped together to test if purchase intentions impact bids. The Bonferroni multiple c omparison test is used to detect pairwise differences for full bids as well as partial bids between different groups. The results are summarized in Table 4 4 Significant differences exist between the full bids of participants with purchase intentions and the full bids of participants without purchase intension for all the four types of blueberries. The differences in the full bids between participants without p urchase intention and those who are not sure if they planned to purchase are not all significant (Only the differences for bid3 and bid4 are significant). This result is consistent with previous findings that consumers who plan to buy the auctioned product s tend to have higher bids than those who do not intend to buy the auctioned products. However, the partial bids for local or organic blueberries are not significantly
68 different across groups with different purchase intentions. Therefore, although purchase intention has effects on the full bids in the auction, it does not affect the revelation of the relative values among multiple auctioned products. This indicates that the impacts of purchase intention on products are in same direction and with similar sca le such that the partial bids calculated as the difference between the full bids for various products do not vary among consumers with different purchase intentions. 4.4.3 The Tobit Model for Full Bids To investigate the relationship between reported full bids an demographics, attitudes, auction location and auctioned blueberry type, a Tobit model is used since the dependent variable (reported bids) is left censored at zero. The model is specified as follows: If ( 4 1) If is the latent dependent variable and is the observed dependent variable. is the vector of demographic and attitude independent variables that include age, gender, income, race, education level, number of children in the household, how well participants like fresh blueberries, whether they had purchased fresh blueberries before and whether they intended to buy fresh blueberries that day. and are focused stores respectively. is the dummy variable for Florida auction participants. and indicate the blueberry typ es. The dummy variables for price
69 conscious stores and Pennsylvania are omitted for identification purpose. is the error term. The regression results are shown in Table 4 5. For the demographic independent variables, age, Asian and college education are significant. Age and Asian are significantly negative, indicating that older participants bid less than younger participants and Asian participants bid less than African American participants (African American is omitted for identification purpose). Partic ipants with college degrees bid significantly higher than those without a college degree. Surprisingly, no significant difference is found between the bids of postgraduates and participants without college education. The coefficients of the variables indi cating blueberry types are all significantly positive. Therefore, both value added attributes (organic and local) of blueberries attract some price premiums from the auction participants. The result also shows that consumers are willing to pay more for org anic blueberries than for locally produced blueberries. Dummy variables for the marketing outlets are significant with the bids at the focused stores significantly higher than those at the price conscious stores. This resul t is as expected since these two stores are intended to attract more consumers who seek high quality food and thus are relatively less price conscious. Importantly, this holds after controlling for the demographic differences among marketing outlets and bl differ among different types of stores. In addition to marketing outlets, the location of the auctions was significant. Participants from Florida bid significantly lower than those from Pennsylvania.
70 As expected, how well the participants like fresh blueberries has a significant positive effect on the reported full bids. However, the purchase history (whether the participant had purchased fresh blueberries before) is not significant, which indicates t intention, participants planning to buy fresh blueberries on the day of auction bid significantly higher than those who did not have such plans. This is also consistent with the results in Table 4 2. 4.4.4 The OLS Regression for Partial Bids To investigate how the independent variables in Equation (1) affect the partial bids, the same set of independent variables, with the exception of the dummy variables for blueberry types are regressed on partial bids for organic and local blueberries. The partial bids for organic are calculated as the difference between bids for organic, U.S. produced blueberries and conventional, U.S. produced blueberries and the partial bids for local are calculated as the difference between bids for conventional, local blueberries and conventional, U.S. produced blueberries. In both cases, there are both positive and negative signs, thus, we use a robust OLS regression. The models are specified as follow: ( 4 2) ( 4 3) and are observed dependent variables. The regression results are illustrated in Table 4 6. Many variables that were significant in the model for full bids become insignificant in the model for partial bids. In the model for local blueberries, only age and the two store dummy variables are significant. As expected, the coefficients for
71 the two store dummy variables are significantly positive, which is a reasonable result since participants from these two types of marketing outlets a re more quality focused or more likely to care about place of origin of food. This result also holds when demographic differences among the stores are controlled for. For organic blueberries, only age, female and the dummy for quality focused store are sig nificant. Female consumers are more willing to pay premiums for organic blueberries than males, which suggests that females are more interested in organic food products t han males. The significance of the coefficient for the quality focused stores indicates that consumers in this type of store place more emphasis on the organic production of blueberries than consumers from the other two types of stores. Purchase intentions are insignificant in both models. This also confirms our previous results tha t purchase intentions have no effect on partial bids. Additionally, added attribute of 4.5 Conclusion In this study, a series of BDM auctions were conducte valuation of organic and locally produced blueberries. We argue that the purchase intention of auction participants and the choice of auction location could impact ousu (2008) by analyzing the effect of purchase intentions on partial bids as well as full bids. In addition, the auctions were conducted at three types of marketing outlets: price conscious focused grocery stor es to capture a more diverse sample of consumers.
72 Our results show that purchase intention on the day of the auction only affects full bids, but not partial bids (e.g. bids for a value added attribute). Auction participants who planned to purchase the auct ioned product reported higher bids than those who did not plan to purchase. However, the partial bids were consistent among participants with different purchase intentions. Therefore, future research on partial bids might be less purchase intentions. However, future research is needed to determine if these results hold only for perishable goods or for all types of products. It was also found that there are differences in demographics among consumers, as well as their attitudes to ward organic and local production of fruits at different marketing outlets. Even after controlling for the demographics and attitudes difference, Specifically, cons blueberries while consumers in the quality focused stores had the highest partial bids for organic blueberries. Therefore, if the auction was conducted at only one type of marketing outle t, the bids might be underestimated or overestimated even if demographics and intentions are controlled for Our study also indicates that variation than their attitudes toward locally produced blueberries.
73 Table 4 1 Demographics of participants at each marketing outlet Independent Variables Price conscious Quality focused Female 79.4 % 69.1 % 78.1 % Age 40 45 50 Caucasian 38.2 % 80.2 % 82.8% Hispanic 3.9 % 2.4 % 4.7 % Asian 2.0 % 3.2 % 2.3 % Black or African American 49.0% 5.6 % 7.8 % Other races 6.9 % 8.7 % 2.3 % Income($34,999 or below) 46.1 % 23.0% 18.8 % Income($35,000 $99,999) 44.1 % 43.7 % 47.7 % Income($100,000 or above) 6.9 % 31.0 % 28.9% 2.9% 2.4 % 4.7 % Postgraduate 4.9% 26.2 % 20.3% C ollege 50 .0 % 56.4 % 60.2 % High school 33.3 % 9.5 % 12.5% Other education level 11.8 % 7.9 % 7.0 % No child at home 43.1 % 69.8% 69.5% One or two children 42.2 % 25.4% 25.0% More than two children 14.7 % 4.8 % 5.5 % Note: The median of age is used. Table 4 2. Summary statistics of bids Bid1($) Bid2($) Bid3($) Bid4($) Price conscious stores 2.93 (1.71) 2.64 (1.25) 2.85 (1.67) 2.59 (1.27) 4.13 (1.73) 3.39 (1.35) 3.43 (1.54) 2.79 (1.30) Quality focused stores 3.89 (1.51) 3.04 (1.26) 3.63 (1.50) 2.61 (1.29) Note: Bid1: organic, local blueberries; Bid2: conventional, local blueberries; Bid3: organic, U.S. produced blueberries; Bid4: conventional, U.S. produced blueberries. The numbers in the parentheses are the standard deviations of the bids.
74 Table 4 3 D ifference of mean bids across purchase intentions Bids Purchase Intention Mean Difference Yes No Not sure Yes No Yes Not sure No Not sure Bid1($): organic, locally produced 4.26 (1.79) 3.46 (1.84) 3.81 (1.45) 0.80** (0.25) 0.45 (0.27) 0.35 (0.20) Bid2($): conventional, locally produced 3.41 (1.42) 2.94 (1.39) 3.04 (1.16) 0.47** (0.20) 0.37 (0.21) 0.10 (0.16) Bid3($): organic, U.S. produced 3.94 (1.49) 3.12 (1.70) 3.36 (1.44) 0.82** (0.23) 0.59** (0.25) 0.23 (0.19) Bid4($): conventional, U.S. produced 3.07 (1.17) 2.60 (1.33) 2.59 (1.26) 0.48** (0.19) 0.49** (0.20) 0.01 (0.15) Partial bids for local($) (Bid2 Bid4) 0.33 (1.05) 0.34 (1.04) 0.45 (0.88) 0.01 (0.15) 0.12 (0.15) 0.11 (0.12) Partial bids for organic($) (Bid3 Bid4) 0.87 (1.25) 0.53 (1.63) 0.77 (1.52) 0.35 (0.23) 0.10 (0.24) 0.24 (0.18) Note: Tukey multiple comparison test is used. The numbers in the parentheses are standard errors. ** indicates significant results at 5% level. Table 4 4. WTPs at different locations Store Type WTP for Organic WTP for Local Mean($) Std. Dev Mean($) Std. Dev Price conscious (1) 0.26 1.57 0.05 0.92 Quality focused (2) 1.02 1.64 0.42 0.91 0.64 1.34 0.60 1.06 Significant difference of means (Bonferroni test) 2>1 2>1,3>1
75 Table 4 5. Tobit for reported bids Independent Variables Coefficient Std. Err. t P>t Age 0.01** 0.00 2.55 0.01 Female 0.17 0.10 1.69 0.09 Income($35,000 $99,999) 0.01 0.11 0.10 0.92 Income($100,000 or above) 0.18 0.12 1.41 0.16 Caucasian 0.01 0.13 0.09 0.93 Hispanic 0.03 0.25 0.14 0.89 Asian 0.67** 0.28 2.36 0.02 Others races 0.230 0.21 1.40 0.16 College 0.36** 0.10 3.41 0.00 Postgraduate 0.19 0.14 1.32 0.19 Number of Children 0.07 0.04 1.71 0.09 0.48** 0.13 3.67 0.00 Quality focused 0.38** 0.13 2.92 0.00 Florida 0.44** 0.09 4.82 0.00 Organic 0.64** 0.12 5.48 0.00 Local 0.42** 0.12 3.61 0.00 Organic_local 1.01** 0.12 8.69 0.00 Like 0.32** 0.06 5.00 0.00 History 0.13 0.17 0.76 0.45 Intention 0.43** 0.11 3.73 0.00 Intercept 0.70** 0.34 2.09 0.04 /sigma 1.52 0.03 Model Fitting Statistics Number of Observation 1376 Log likelihood 2470.92 LR chi2(20) 214.25 Prob > chi2 0.00 Note: Dummies for income ($34,999 or below), Black, education level below college and the price conscious stores are omitted for identification; Organic=1: the auctioned blueberry is organic; Local=1: the auctioned blueberry is local; Organic_local=1: the auctioned blueberry is both organic and local; Like: How well the participant likes fresh blueberries (1=dislike very much; 5=like very much); History=1: The participant has purchased fresh blueberries before; Intention=1: The participant intended to pur chase fresh blueberries that day. ** indicates significance at 5% level.
76 Table 4 6. Robust OLS for WTPs Independent Variables Local Organic Coef. Std. Err. P>t Coef. Std. Err. P>t Age 0.02** 0.00 0.00 0.02** 0.01 0.00 Female 0.18 0.13 0.18 0.45** 0.21 0.03 Income($35,000 $99,999) 0.04 0.12 0.71 0.16 0.20 0.42 Income($100,000 or above) 0.17 0.16 0.29 0.12 0.22 0.60 Caucasian 0.16 0.20 0.42 0.04 0.30 0.89 Hispanic 0.03 0.30 0.91 0.36 0.54 0.50 Asian 0.13 0.29 0.67 0.31 0.43 0.48 Others races 0.22 0.24 0.35 0.07 0.35 0.83 College 0.19 0.13 0.14 0.26 0.20 0.19 Postgraduate 0.28 0.18 0.13 0.03 0.27 0.92 Number of Children 0.01 0.05 0.82 0.12 0.08 0.17 0.59** 0.17 0.00 0.58** 0.26 0.03 Quality focused 0.51** 0.17 0.00 0.95** 0.28 0.00 Florida 0.09 0.12 0.46 0.03 0.18 0.89 Like 0.02 0.06 0.71 0.05 0.10 0.66 History 0.25 0.21 0.22 0.25 0.28 0.37 Intention 0.09 0.15 0.52 0.28 0.21 0.19 Intercept 0.57 0.39 0.14 0.74 0.52 0.16 Model Fitting Statistics Number of Observation 344 0.13 0.00 344 0.11 0.03 R squared Prob > F Note: Dummies for income ($34,999 or below), Black, education level below college and the price conscious stores are omitted for identification; Like: How well the participant likes fresh blueberries (1=dislike very much; 5=like very much); History=1: The participant has purchased fresh blueberries before; Intention=1: The participant intended to purchase fresh blueberries that day. ** indicates significance at 5% level.
77 CHAPTER 5 CONCLUSION preference s and attitudes is critical for new product development, labeling policy formulation or the prediction of product market shares. However, the measurement of preferences and attitudes could be difficult as they are both abstract concepts. Willingn ess to pay (WTP), as a measurable concept, has enabled people to have a one to one correspondence between preference and monetary values. Studies that used different methodologies could yield different WTP values. Therefore, r esearchers have been explorin g ways to get an unbiased WTP values Some commonly used value elicitation methods include contingent valuation, experimental auctions etc. Although these methods have been used and discussed for a long time, research is still ongoing about methods that wi ll give us more accurate measurement. In this dissertation, attempts have been made to discover some problems and methodological improvement in the process of eliciting WTPs. Two value elicitation methods: choice experiment and experimental auction are di scussed. The research nutrients, the blueberry has an expanding consumer group but studies about For the indirect methods to calculate WTPs, such as the choice experiment, the models used to estimate WTPs are among the hottest discussed topics. Researchers have found a lot of inconsistencies between various value elicitation methods. However, different models can also result in different results with the same data. In Chapter 2 of
78 this dissertation, a hypothetical method: a choice experiment is used to compare place of production (local, U.S. produced or imported). The all random WTP space mixed logit model is used for parameter estimation. The results indicate that consumers in the U.S. regard the local production of blueberries a more important attribute than whether they are organic or not. Even though there is a fair amount of consumers that are willing to pay positive premiums for organic blueberries, the m have positive WTPs for the organic attribute. Additionally, we conclude that disaggregate information, which can be derived from the Bayesian estimation method, is more useful for target marketing In Chapter 3, three widely used specifications for mixed logit are estimated and compared. The results show that the divergence found in previous literature between the preference space model and the WTP space model can be narrowed when the price coefficient is assumed fixed in the prefe rence space T he results suggest that the fixed price preference space model might be used as a substitute for the all random WTP space model, which is used in Chapter 2. This is an interesting finding as the all random WTP space model need extra steps in estimation and may also suffer from the identification problems suggested by Ruud (1996). This chapter reaches the same conclusion as Chapter 2 in terms of the attribute importance rankings of blueberries. Although the choice experiment is relatively easy for researchers to implement and straightforward for consumers to understand, hypothetical bias has been one of its most questioned problems. For direct methods, such as experimental auctions, the problem of hypothetical bias has been dealt with, but the c hoice s of the auction format, auction
79 locations and specific auction procedu res are other problems that need serious consideration for the experiment to be incentive compatible. In Chapter 4, a non hypothetical method: experimental auction is conducted fo r experimental auction are much smaller than those from the choice experiment. Therefore, our results are consistent with previous studies in that the choice experiment tends to p rovide exaggerated WTP estimates. Additionally, we find that auction locations have significant impact on p Specifically, purchase intentions af fect full bids, but not partial bids while the choice of auction locations affects both full bids and partial bids. Even though the experiment auctions might provide more plausible WTP estimates than choice experiments, the reliability of the results from experimental auctions has not been verified. A variety of observable or More research should be done to discover and evaluate these factors so that a better reflection of Apart from the findings regarding methodologies, Chapter s 2 and 3 and Chapter 4 Although most literature also found different W TP values with different value elicitation methods, little of them have found inconsistent attribute importance rankings. One of the reason s might be that in the choice experiment, the organic attribute was ignored because there were too many attributes to be compared in each choice situation. In the auction process, however, there were only two attributes: organic and local Another
80 reason might be that the comparison is based on WTP means, which are subject to the influence of extreme values. This con tradiction might also indicate that the design of choice experiment s needs extra caution or more studies should be conducted on the reliability of the estimates from choice experiments.
81 APPENDIX A CHOICE SITUATIONS FOR THE CHOICE EXPERIMENT Which of the following two choices for blueberries would you pick? Q97 Which of the following two choices for blueberries would you pick? Q98 Which of the following two choices for blueberries would you pick? Q99 Which of the following two choices for blueberries would you pick? Q100 Which of the following two choices for blueberries would you pick?
82 Q101 Which of the following two choices for blueberries would you pick? Q102 Which of the following two choices for blueberries would you pick? Q103 Which of the following two choices for blueberries would you pick? Q104 Which of the following two choices for blueberries would you pick?
83 Q105 Which of the following two choices for blueberries would you pick? Q96 Which of the following two choices for blueberries would you pick? Q106 Which of the following two choices for blueberries would you pick?
84 APPENDIX B BDM AUCTION QUESTIONNAIRE Have you ever purchased fresh blueberries? Yes No Approximately how much money did you spend on fresh blueberries in the LAST MONTH? Nothing, I did not purchase fresh blueberries last month Less than $4 $4.00 $9.99 $10.00 $14.99 $15.00 $19.99 $20.00 $29.99 $30 or more Is this the location where you typically purchase most of your frui t from? Yes No Did you intend to purchase fresh blueberries here today? Yes No Not sure Do you currently have any fresh blueberries at home? Yes No What state do you currently live in (at least 6 months out of the year)? Florida Other __________
85 Please indicate how well you like the following types of blueberries: Dislike Very Much Dislike Neither Like nor Dislike Like Like Very Much Fresh Frozen Dried Canned As an ingredient (e.g. blueberry muffin, blueberry smoothie) If you wanted to purchase fresh blueberries, how likely are you to purchase them from the following sources, assuming the berries look the same, are the same size packages and are available at the same time, place, and price (e.g. the only difference is location where they are produced): Very Unlikely Unlikely Undecided Likely Very Likely Produced in your state Produced in the US, but not in your state Imported (not from the US) Please indicate how strongly you agree with the following statements: I eat blueberries for the... Strongly Disagree Disagree Neither Agree nor Disagree Agree Strongly Agree Health benefits Taste Convenience (easy to eat) Diversity they add to my diet Price Please select Agree for this row Other: __________________
86 Please indicate what percent of FRUIT you purchase is: Percent: organic 0% (None) 10% or less 10 50% 50 99% 100% (all) How important are the following factors when you make a decision to purchase fresh blueberries? Not at all Important Somewhat Unimportant Neither Important nor Unimportant Somewhat Important Extremely Important Color Size of the blueberries Freshness Dryness/no mold on blueberries Price Firmness Container size Where the blueberries were grown Availability of samples
87 Please indicate how strongly you agree with the following statements. Strongly Disagree Disagree Neither Agree nor Disagree Agree Strongly Agree I trust fruits labeled as organic Organic blueberries are healthier than other blueberries I will pay more for fruits with an organic label I can regularly find organic blueberries The following questions are used to help us understand how different types of people have different tastes and preferences. Your responses are anonymous; we do not have anything that ties your names to your responses. We appreciate your honest answers to the following questions. What is your gender? Male Female What is your age? _______ What is your ethnicity/race (please select all that apply)? Caucasian Black or African American Hispanic Native Hawaiian or Pacific Islander Asian American Indian or Alaska Native Other ____________________ Please indicate you estimated annual household income: Less than $14,999 $75,000 $99,999 $15,000 $24,999 $100,000 $149,999 $25,000 $34,999 $150,000 $199,999 $35,000 $49,999 $200,000 or above $50,000 $74,999 Do not know What is your edu cation level? Less than high school Postgraduate High school degree or equivalent Trade/technical school Some college Other ____________________
88 Four year college degree Don't know What is your current employment status? Full time Part time Currently not working Retired Student Other How many people including yourself live in your household? 1 2 3 4 5 6 7 8 9 or above How many children age 18 or younger currently live in your h ousehold? None 1 2 3 4 or more Approximately how much per week does your family spend on food? Less than $49 $200 $249 $50 $99 $250 $299 $100 $149 $300 or more $150 $199 Please take this survey to an interviewer to complete the last portion of the interview and receive your gift card.
89 APPENDIX C BDM AUCTION INSTRUCTIONS AND PROCEDURE We are now interested in your preference for four packages of blueberries typical ly found at a grocery store. Now, we will conduct you will have the opportunity to purchase one package of blueberries. In case you do not have cash with you, we are providing you with $7 cash. This cash is yours to keep and is in addition to the $10 gift card you will receive at the end. In a moment, you will be asked to write the most you are willing to pay for each package of blueberries in the spaces below. After you write these amounts, we will auction the blueberries like this: 1. We will randomly draw a let ter to determine which blueberry package to actually auction. Even though you will bid on four products, only one will be auctioned. If we for organic, produced realize that even though you will bid on four products, you will only have the opportunity to buy one package of blueberries. 2. Once we determine which blueberry package will be auctioned, you will randomly bowl 3. If your bid for the randomly selected blueberry package is higher than the price from the bowl, then you will purchase the blueberries from me for a price equal to the price from the bowl. Note: payment will actually take place. 4. If your bid for the package is less than the price from the bowl, then you will not purchase the blueberries and will walk away with no blueberries. If you do not wish to purchase one of the bluebe rry packages, simply bid $0.00 for that package. It is in your best interest to bid exactly what you are willing to pay. For example, if you bid less than it is worth to you to have the blueberries, you might end up not winning the auction even though you could have bought the blueberries at a price you were actually willing to pay. Or, if you bid more than it is worth to you to purchase the blueberries, you may end up having to buy them at a price higher than you really wanted to. Thus it is in your best interest to bid exactly what you are willing to pay. Do you have any questions? A. Organic, Locally produced Blueberries is $________ B. Locally produced Blueberries is $________ C. Organic, produced in the US Blueberries is $________
90 D. Blueberries produced in the US is $________
91 LIST OF REFERENCES Pay for the Color of Salmon: A Choice Experiment with Real Economic Ameri can Journal of Agricultural Economics 88(2006):1 050 1061. treated Beef: Application of a European Review of Agric ultural Economics 31 (2004):19 37. Alfnes, F. and K. Rickertsen. American Journal of Agricultural Economics 85 (2003):396 405. Journal of Econometrics 89 (1998):57 78. Balcombe, K., Journal of Environmental Economics and M anagement 57 (2009):226 237. flight Journal of Air Transport Management 15 (2009):221 226. Batte, M. T., N. H. Hooker, T. C. Haab, a Their Mouths are: Consumer Willingness to Pay for Multi ingredient, Processed Food Policy 32 (2007):145 159. U tility by a S ingle re sponse S equential M Behavioral Science 9 (1964):226 232. for Pesticide Free, Non Journal of Agricultural and Resource Eco nomics 35 (2010):457 475. The Journal of Consumer Research 25 (1998):187 217. Prefer ence Heterogeneity in Wetland Attributes: The Case of Cheimaditida Wetland in Gr Ecological Economics 60 (2006):145 156. Journal of Agricultural and Resource Economics 33 (2008):402 427.
92 American Journal of Agriculture Economics 92 (2010):1229 1241. on Genetically Modified Food 2004. Working Paper in Economics no. 129, Department of Economics, Gothenburg Unive rsity. Carson, R. T., J. J. Louviere, D. A. Anderson, P. Arabie, D. S. Bunch, D. A. Hensher, R. M. Johnson, W. F. Kuhfeld, D. Steinberg, J. Swait, H. Timmermans, and J. B. Wiley. "Ex perimental Analysis of Choice." Marketing Letters 5 (1994):351 368. The American Statistician 46 (1992):167 174. Corrigan, J. R. and M. C. Rousu. Journal of Agricultural and Resource Ec onomics 33 (2008):290 301. Foods: A Customer Intercept Study of Direct Market and Grocery Store g Beach, California, July 23 26, 2006. Agricultural and Resource Economics Review 38 (2 009):384 396. CVM X: Calibrating American Journal of Agricultural Economics 80 (1998):455 465. Journal of Marketing Research 4 (1967):184 190. Willingness to pay for Food Lables. Meeting, Portland, Oregon, July 29 August 1, 2007. Environmental and Resource Economics 11 (1998):413 428. Hayes, D. J., J. American Journal of Agricultural Economics 77 (1995): 40 53
93 to pay Estimates of Travel time Savings from Individual Environment and Planning 38(2006):2365 2376. Journal of Trans port Economics and Policy 45 (2011):505 525. Environmental and Resource Economics 32 (2005):509 531. Horowitz, J. K. "The Becker DeGroot Marschak Mechanism is not Necessarily Incentive Compatible, even for Non random Goods." Economic Letters 93 (2006):6 11 Blueberry Products Journal of Agricultural and Applied Economics 41 (2009):47 60. Pay for Value Journa l of Agricultural and Applied Economics 43 (2011):243 258. Marketing Letters 12 (2001):259 269. Huffman, W. E., J. F. Shogren, M. Rousu, and A. Teg ene. "Consumer Willingness to Pay for Genetically Modified Food Labels in a Market with Diverse Information: Evidence from Experimental Auctions." Journal of Agricultural and Resource Economics 28 (2003):481 502. Irwin, J. R., P. Slovic, S. Lichtenstein, an d G. H. McClelland."Preference Reversals and the Measurement of Environmental Values." Journal of Risk and Uncertainty 6 (1993):5 18. Jaeger, S. R., J. L. Lusk, L. O. House, C. Valli, M. Moore, B. Morrow, and W. B. Traill. on hypothetical E xper imental Markets for Measuring the A cceptance of G enetically Modified F Food Quality and Preference 15(2004):701 714. Renewable Agriculture and Food Systems 24 (2009):205 213. Johansson etical Bias in Choice The B.E. Journal of Economic Analysis & Policy 8 (2008).
94 The Review of Economics and Statis tics 82 (2000):616 624. Renewable Agriculture and Food Systems 23(2008):208 216 Procedures? Evid The American Economic Review 91 (2001):1498 1507. The American Economic Review 92 (2002):1636 1643. List, J. A. and C. A. Gallet Environmental and Resource Economics 20 (2001):241 254. List, J. A. and D. Lucking from a Sportscar The American Economic Review 90 (2000):961 972. Journal of Economic Behavior & Organization 37 (1998):193 20 5. Preferences for Organic, Eco Journal of Agricultural and Resource Economics 26 (2001):404 416. Lus licitation in Retail and Laboratory E Economics Letters 79 (2003):27 34. Lusk, J. L., J. A. Fox, T. C. Schroeder, J. Mintert, and M. Koohmaraie. "In store Valuation of Steak Tenderness." American Journal of Agricultural Economics 83 (2001):539 550. Lusk, J. L., L. O. House, C. Valli, S. R. Jaeger, M. Moore, J. L. Morrow, and W. B. Traill. of Information about Benefits of Biotechnology on Consumer Acceptance of Genetically Modified Food: Evidence from Experimental Auctions in the United States England and European Review of Agricultural Economics 31 (2004):179 204. Environme ntal and Resource Economics 36 (2007):499 521.
95 Based Conjoint American Journal of Agricultural Economics 87 (2005):771 785. mand for Beef from Cattle Administered Growth Hormones or Fed Genetically Modified Corn: A Comparison of American Journal of Agricultural Economics 85 (2003):16 29. Lusk, J. L. and T American Journal of Agricultural Economics 86 (2004):467 482. Underl Agricultural Economics 39 (2008):339 348. Journal of Applied Econometrics 15 (2000):447 470. Melton, B. E., W. E. Huffman, J. F Fresh Food Items with Multiple Quality Attributes: Evidence from an Experimental American Journal of Agricultural Economics 78 (1996):916 923. Journal of Health Economics 28(2009):894 901. Payne, J. W., J. R. Bettman, and E. J. Johnson. "Behavioral Decision Research: A Con structive Processing Perspective." Annual Review of Psychology 43 (1992):87 131. Journal of Environmental Econo mics and Management 48 (2004):793 810. Review of Economics and Statistics 80 (1998):647 6 57. Specific Taste Pa rameters and Mixed Logit: 274, Department of Economics, University of California at Berkeley. Robert, C. P. and G. Casella. Monte Carlo Statistical Methods Springer, 2004.
96 Rossi, P. Marketing Science 22 (2003):304 328. Rossi, P. E., G. M. Allenby, and R. E. McCulloch. Bayesian Statistics and Marketing Wiley, 2005. f Purchase History Data Marketing Science 15 (1996):321 340. Marketing Inf Journal of Agricultural and Applied Economics 37 (2005):647 657. to Pay for Food Safety: An European Review of Agric ultural Economics 31 (2004):409 425. California. n Willingness to Pay Space: A Tool to American Journal of Agricultural Economics 90 (2008):994 1010. to Pay for Renewable Energy: Primary and Discretionary Choice of British Households' for Micro Energy Economics 32 (2010):129 136. American J ournal of Agricultural Economics 81 (1999):1192 1199. Comparison of Hypothetical and Non Hypothet International Food and Agribusiness Management Review 10 (2007):165 180. to pay in Quantitative Marketing and Economics 5 (2007):313 331. Thaler. R. "Mental Accounting and Consumer Choice." Marketing Science 4 (1 985): 199 214. Environmental and Resource Economics 44 (2009):379 395.
97 Train, K. Discrete Choice Methods with Simulation Cambridge: Cambridge University Press, 2003. to Resource Economics. R. Scarpa and A. Alberini, eds. Dordrecht, The Netherlands: Springer Publisher, 2005, Chapter 1, pp. 1 16. Tversky, A. and I. Simonson. "Context dependent Preferences." Management Science 39 (1993): 1179 1189. Umberger, W. J., D. M. Feuz, C. R. Calkins, and K. Killinger preference and Willingness to Pay for Domestic Corn Fed Beef versus International Grass Agribusiness 18 (2002):491 504. Umberger, W. J., D. M. Feuz, C. Of Origin Labeling Journal of Food Distribution Research 34 (2003):103 116. USDA, Economics, Statistics, and Market Information System, Economic Research Ser vice, U.S. Blu eberry Industry. Available online at : http://usda.mannlib.cornell.edu/MannUsda/viewDocumentInfo.do?documentID=1 765 (Accessed June 02, 2010). 2003 Paper presented at American Agricultural Economics Association 2003 Annual Meeting s, Montreal, Canada Wedel, M., W. Kamakura, N. Arora, A. Bemmao r, J. Chiang, T. Elrod, R. Johnson, P. Marketing Letters 10 (1999):219 232. lingness to Pay at the Point Journal of Marketing Research 39 (2002):228 241. Harvard Business Review 42(1964):83 90. Yue, C. ference for Fresh Hortscience 44 (2009):366 371.
98 BIOGRAPHICAL SKETCH Lijia Shi received her Ph.D. in food and resource economics at the University of Shanghai Jiao consumer behavior, survey design and quantitative methodologies.