THREE ESSAYS ON CONSUMER PREFERENCES AND BEHAVIOR ANALYSIS By JING XIE 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 2014
Â© 2014 Jing Xie
To my parents and my husband
4 ACKNOWLEDGMENTS I would like to express my sincere gratitude to my committee members, Dr. Zhifeng Gao, Dr. Lisa A. House, Dr. Sherry Larkin, Dr. Chunrong Ai, and Dr. Xin Zhao for their generous help and support in completion of this dissertation. Most grat efully, I thank Dr. Zhifeng Gao and Dr. Lisa A. House for their encouragement and support duri ng the last four years. They were always available for my questions, and they were always positive and help me overcome a lot of difficult times. Their support an d continuous guidance enabled me to complete my dissertation. Moreover, I want to thank Dr. Sherry Larkin, Dr. Chunrong Ai, and Dr. Xin Zhao for their valuable comments and suggestions. These valuable suggestions will greatly benefit me in my future studie s. I would also appreciate the support from my fellow graduate students. Beca use of them, my graduate life was full of happiness memories. A special thanks to my family. Words cannot expres s how grateful I am to my parents, Deming Xie and Jieping Wu, for all of the sacrifices that they have made on my behalf. I would also want to thank my beloved husband, Leming Lin , who has supported me all the time to strive towards my goal , and encouraged me to be a tough person .
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 LIST OF ABBREVIATIONS ................................ ................................ ............................. 9 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 13 1.1 Organic Label and Country of origin Label ................................ .................... 13 1.2 Comparing Valuation Methods Role of Bargaining Power .......................... 15 1.3 Further Explanation of the Valuation Differences Purchase Intention ......... 17 2 OF ORIGIN MATTER? ................................ ................................ ........................... 20 2.1 Organic Label vs. COOL ................................ ................................ ................... 20 2.2 Literature Review ................................ ................................ .............................. 23 2.3 Survey Design and Data ................................ ................................ ................... 27 2.3.1 Survey Design ................................ ................................ ......................... 27 2.3.2 In formation Treatment ................................ ................................ ............. 29 2.4 Analysis ................................ ................................ ................................ ............ 31 2.4.1 Conditional Logit Model ................................ ................................ ........... 31 2.4.2 Generalized Mixed Logit Model ................................ ............................... 32 2.4.3 Estimation in the WTP Space ................................ ................................ .. 34 2.5 Empirical Results ................................ ................................ .............................. 36 2.5.1 Data Collection and Descriptive Statistics ................................ ............... 36 2.5.2 Shopping Behaviors and Perceptions ................................ ...................... 37 2.5.3 Conditional Logit Model ................................ ................................ ........... 38 2.5.4 Generalized Mixed Logit Model ................................ ............................... 39 2.5.5 WTP for Broccoli and Impact of Information ................................ ............ 40 2.5.6 Value of Information ................................ ................................ ................ 43 2. 6 Concluding Remarks ................................ ................................ ......................... 45 3 BARGAINING ENVIRONMENTS IN NON HYPOTHETICAL EXPERIMENTS ....... 55 3.1 Valuation Methods ................................ ................................ ............................ 55 3.2 Ba rgaining Environment in RCE, RCVM, and EA ................................ ............. 58
6 3.3 Experimental Design ................................ ................................ ......................... 60 3.3.1 General Experimental Design ................................ ................................ .. 60 3.3.2 Measure Aggressiveness in Price Bargaining ................................ ......... 61 3.3.3 RCE Design ................................ ................................ ............................. 62 3.3.4 RCVM Design ................................ ................................ .......................... 63 3.3.5 EA Design ................................ ................................ ............................... 65 3.4 Models and Specifications ................................ ................................ ................ 65 3.4.1 WTP Estimation from RCE ................................ ................................ ...... 66 3.4.2 WTP Estimation from RCVM ................................ ................................ ... 67 3.4.3 WTP Estimation from EA ................................ ................................ ......... 68 3.5 Results ................................ ................................ ................................ .............. 69 3.5.1 Regression Results from RCE data ................................ ......................... 70 3.5.2 Regression Results from RCVM data ................................ ...................... 72 3.5.3 Auction Bids Results ................................ ................................ ................ 72 3.5.4 Comparing WTP Values from Three Experiments ................................ ... 73 3.6 Concluding Remarks ................................ ................................ ......................... 75 4 PURC HASE INTENTION EFFECTS IN CHOICE EXPERIMENTS AND EXPERIMENTAL AUCTIONS ................................ ................................ ................. 87 4.1 Purchase Intention Effects ................................ ................................ ................ 87 4.2 Literature Review ................................ ................................ .............................. 90 4.3 Experimental Design ................................ ................................ ......................... 91 4.4 Models and Specifications ................................ ................................ ................ 92 4.4.1 MNL Model for RCE Data ................................ ................................ ........ 92 4.4.2 Multivariate Tobit Model for EA Data ................................ ....................... 93 4.5 Results ................................ ................................ ................................ .............. 94 4.5.1 Results of MNL Regression in the RCE ................................ ................... 95 4.5.2 Results of Multivariate Tobit Regression in the EA ................................ .. 97 4.5.3 Endowment Effect ................................ ................................ ................... 98 4.6 Conclusion Remarks ................................ ................................ ......................... 99 5 CONCLUSION ................................ ................................ ................................ ...... 107 APPENDIX A INFORMATION PROVIDED TO RESPONDENTS ................................ ............... 110 B INFORMATION IN EXPERIMENTAL DESIGN ................................ ..................... 112 LIST OF REFERENCES ................................ ................................ ............................. 113 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 122
7 LIST OF TABLES Table page 2 1 Attributes and levels for fresh broccoli in the choice experiment. ....................... 47 2 2 Summary statistics for survey respondents. ................................ ....................... 48 2 3 Comparison of information treatment and control groups: Conditional Logit and G MIXL model. ................................ ................................ ............................ 50 2 4 WTP for broccoli attributes ................................ ................................ ................. 51 3 1 Bargaining attitudes, intention, and competitiveness ................................ .......... 78 3 2 Variable definitions on consumer specific characteristics ................................ ... 79 3 3 Statistical summary of aggressiveness measurement ................................ ........ 80 3 4 Es timates of models for choice experiments ................................ ...................... 81 3 5 Double bounded CVM estimation results ................................ ........................... 83 3 6 Summary statistics for the bids ................................ ................................ ........... 84 3 7 Multivariate Tobit model results from EA data ................................ .................... 85 3 8 WTP values and auction bids grouped by aggressiveness level in price bargaining ................................ ................................ ................................ ........... 86 4 1 How familiar with different types of orange juice ................................ ............... 101 4 2 Purchase experience ................................ ................................ ........................ 101 4 3 Summary statistics for the answer to purchase intention question ................... 102 4 4 Summary statistics for the answer to purchase experience .............................. 102 4 5 Summary statistics for how familiar with orange juice ................................ ...... 102 4 6 MNL estimates for choice experiments ................................ ............................. 103 4 7 Multivariate Tobit model estimates for experimenta l auctions .......................... 104 4 8 Endowment e ffect ................................ ................................ ............................. 105
8 LIST OF FIGURES Figure page 1 1 USDA organic seal ................................ ................................ ............................. 19 2 1 Annual imported fresh broccoli (pounds) ................................ ............................ 52 2 2 Annual imported broccoli (pounds) . ................................ ................................ .... 53 2 3 A s ample choice set in the choice experiment. ................................ ................... 54 4 1 Purchase intention question ................................ ................................ ............. 106
9 LIST OF ABBREVIATIONS AMS Agricultural Marketing Services BDM The Becker, DeGroot, and Marschak CL Conditional Logit COOL C ountry of origin L abeling EA Experimental Auction EU European Union FCOJ Frozen Concentrated Orange Juice FDA Food and Drug Administration G MIXL Generalized Mixed Logit Model MIXL Mixed Logit Model MNL Multinomial Logit NFC Not from Concentrated NOP National Organic Program NOS National Organic Standards OJD Orange Juice Drink OTA The Organic Trade Association PDO Protec ted Designation of Origin PGI Protected Geographical Indications RCE Real Choice Experiments RCVM Real Contingent Valuation Methods S MNL Scale Heterogeneity Multinomial L ogit S CL Scale Conditional Logit USDA United States Department of Agriculture
10 WFM Whole Foods Market WTP Willingness to pay
11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy THREE ESSAYS ON CONSUMER PREFERENCES AND BEHAVIOR ANALYSIS By Jing Xie August 2014 Chair: Zhifeng Gao Cochair: Lisa House Major: Food and Resource Economics This dissertation discusses consumer preferences and willingness to pay (WTP) for organic produce with difference country of origin, compares three methods that are commonly used estimate WTP, and discusses two possible reasons why different methods cannot provide consistent estimates of The first essay examines the impact of country of origin label ing on US consumers choice s of organic foods . Results show that consumer valuation of domestically produced organic broccoli was significantly higher than that of imported organic broccoli. Adding information about USDA organic certificat ion standards/rules for imported products mildly increases consumer valuation of imported organic broccoli in some cases. The second essay preferences studies: do different valuation m ethodologies give consistent willingness to pay values? It examines the preferences revealed by thr ee non hypothetical experiments: real choice experiment s , real contingent valuation method s , and experimental auction s . The results sho w that WTP estimates from real choice experiment s are the larg est, followed by real contingent valuation methods, and then
12 experimental auctions. By comparing respondents with different bargaining preferences, t he results suggest that the discrepancies amo ng expe riments may come from the The last essay in this dissertation extends the studies in the second essay by testing is another reason why respondents behave differently in different non hypothetical experiments. Since we have found significant estimated gaps between real choice experiments and experimental auction, the third essay only focuses on these two experiments. Whe purchase intention in the analysis, the results show that purchase intention had different effects experimental auctions and real choice e xperiments. R esults show that in experimental auctions purchase intention only affect of non novel goods , but in choice experiments purchase intention only affect respondents choice of novel good . The three essays significantly contribute to the literatures in consumer valuation both emp irically and theoretically. The first essay implies that country of origin purchase intention can affect the experiments results, and the impacts vary by type of experiments.
13 CHAPTER 1 INTRODUCTION 1.1 Or ganic L abel and Country of origin L abel Because of the increasing consumer demand for healthier and more environ mentally friendly food pro ducts, food labels are becoming a more important information source for consumers when purchasing food products. Th e first essay of this dissertation focus es on food labels , specifically organic and co untry of origin (COOL) labels , and the interactive effect between these two labels . The implementation of the National Organic Program (NOP) in October 2002 with the USDA organic labeling g uidelines and uniform USDA organic seal (Figure 1 1) provide US consumers a standard recognition of organic food products. Based on the definition from NOP 1 agricultural product has been pr oduced through approved methods , This regulation requires that all organic growers and handlers (including food processors) be certified by a certifying agen t , unless the famers or handlers sell less than $5,000 organic products a year. Under this regulation, org anic food is viewed as safer and more environmental friendly than conventional food by consumers , and previous research reaches to a general conclusio n that consumers are willi ng to pay more for organic food ( Tsakiridou et al. 2008; Urena, Bernabeu, and Olmeda 2008 ) . 1 Accessed February 21, 2014, available at http://www.ams.usda.gov/AMSv1.0/nop
14 On the other hand, COOL is a food label that requires retailers notify their customers with information regarding the source of certain f oods. Mandatory COOL laws in the United States require country of origin information for muscle cut and ground meats; wild and farm raised fish and shellfish; fresh and frozen fruits and vegetables; peanuts, pecans, and macadamia nuts; and ginseng (USDA 2012 ) . Because of this label, US consumers can easily obtain the country of origin information of the food products they purchase . Much research has been conducted on consumers preference for COOL, and most findings confirm that consumers prefer food produ ced domestically to those imported from other countries ( e.g., Gao and Schroder 2009; Kim et al. 2012; Peterson et al. 2013 ). However, most previous studies overlook the interactive effects between/among food labels. Because of the labeling regulation, wh en consumers are shopping in the grocery stores, many food labels would show up on the package at the same time, and some of them may have interactive effects on each other. For example, a lot of consumers may think a food product is safe and environmental friendly when they see the USDA organic seal on the package. But when they also observe another food label such as COOL (e.g., Products of China), on the same package, they may change their opinion about this product, and ask why the food from China can h ave the USDA organic seal on it. T he first essay tries to determine whether COOL information on the food has an interactive effect with the USDA organic label. This essay provides an information treatment to random respondents to test whether consumers fully
15 understand UDSA organic seal regulations. Some r espondents were randomly selected to see the information that USDA organic food is produced under the same requirement n o matter where it is produced to test whether t his group of respondents would make choice decision differently comparing to those who did not receive this information. 1.2 Comparing Valuation M ethods Role of Bargaining Power The first essay provide s an example of a choice experiment that is commonly used by researchers to estimate WTP . There are different valuation meth ods that can be used to estimate consumers WTP . Based on whether real money transactions are involved, these methods can be categorized into two groups , hypothetical and non hypothetical valuation methods . The main difference between these two is that products offered and price chosen are hypothetical in hypothetical methods , but in the latter case the price is actually paid by participants . The valuation method used in the first essay is a hypo thetical choice experiment in online surveys. One advantage of conducting hypothetical choice experiments online is that researchers can collect vast information within very short period of time , with very low cost, and in multiple locations or even in multiple countries. In addition, respondents in a hypothetical ch oice experiment face products with different attribute combinations, which closely reflect actual shopping conditions (Lusk and Schroeder 2004) . H owever, hypothetical experiments have a well known drawback that respondents tend to exaggerate what they real ly want to pay in the experiments. This difference between consumers true WTP and the estimated WTP from hyp othetical experiments is called hypothetical bias. Because a lot of previous literature confirms the existence of this bias ( Champ and Bishop 2001; Harrison and RutstrÃ¶m 2008; List 2001; List and Shogren
16 1998; Murphy et al. 2005 ) , more studies are now using non hypothetical experiments in order to The second essay in this dissertation focuses on thre e commonly used non hypothetical experiments : real choice experiments, real contingent valuation methods, and experimental auction s . These non hypothetical experiments are viewed as incentive compatible in that at the end of the experiment the real transaction of money and product s happens, therefore respondents have the incentive to reveal their true WTP. Lusk and Schroeder ( 2006 ) and Gr acia, Loureiro, and Nayga (2011), however, found that non hypothetical experiments cannot provide similar WTP. Ho wever, they the differences across methods is important. Because it will help choose more appropriate valuation methods if the factors that cause the gaps between methods can be controlled. By recruiting the same type of respondents (demographic characteristics were statistically indifferent), t he second essay conducted three commonly used non hypothetical experiments on the same products in the same location dur ing the sa me period of time. The purpose of this essay is to find out whether these no n hypothetical experiments lead to different estimates of if they do, what are the reasons. Starting from the bargaining environment in each non hypothetical ex periments, this essay argues that experimental auctions provide more bargaining power to participants compared to other t wo non hypothetical experiments. T hat is, the process
17 of experimental auctions provide participants the chance to bargain the price of the product, but real choice experiments do not. To test this hypothesis , this essay with low aggressiveness, middle aggressiveness, and high aggressiveness price bargaining preferences. The objective is to determine whether highly aggressive people behave differently compared to low aggressive people in the same experiment. If an experiment provide little space for respondents to bargain, these two groups would behave similar ly; and if an experiment provide huge bargaining power to respondents, these two groups would behave differently. 1.3 Further E xplanation of the Valuation D ifferences Purchase I ntention Bargaining environment differences may not be the only reason that respondents behave differently in different non hypothetical experiments. Another important factor purchase intention before they participate in the experiments. Since more non hypothetical experiments are now conducted in or near the grocery stores, most respondents in the experiments may have had certain purchasing plan before they go to grocery stores. Th is purchase intention could affect the value they bid or choose in the experiments. Shi, House, and Gao (2013) have found this effect in the experimental auctions, however, few studies look at the purchase intention effects in real choice experiments. As sta ted in the second essay, the process of the real choice experiments and the experimental auction is different, especially their focus on the price attribute. The bidding process in the experimental auction emphasize s the price that respondents want to pay and they can name their own price of the product; however, in the choice
18 experiments, respondents are told to choose the product they like the most among several choice options, and the price attribute is presented along with all the other attributes. As a resul ts, in the experimental auction when price is placed in the center of the whole process, a respondent who happen s to have a plan to purchase that product omes to the choice experiment in which respondents are asked to pick one product they li ke the most compared to other similar products with different attribute levels, they may just focus on picking their favorite product and the purchase intention may not be as important anymore. If this argument holds, purchase intention may be another potential reason why different non hypothetical e xperiments provide inconsistent WTP values. The third essay of this dissertation discusses this e ffect in great detail .
19 Figure 1 1 . USDA organic seal .
20 CHAP TER 2 ORIGIN MATTER? 2.1 Organic Label vs. COOL Organic farming has constituted one of the fastest growth segments of agriculture worldwide in the last decade. In the United States, total sales of organic products were 9.5% higher in 2011 than that in 2010, reaching a total value of US $31.5 billion. In contrast, total sales of comparable conventionally produced food and non food items grew by only 4.7% in the same period 1 . Although organic foods only constituted 4.2% of all US food sales in 2011 2 , US consumer demand for organic foods has surpassed domestic supply . Th e United States imported approximately $1.0 $1.5 billion worth of organic food s in 2002, 3 or about 12 18% of the US organic retail sales in that year . The United States and Canada reached an organic equivalency agreement in 2009, and the United States and the European Union reached an agreement in 2012. 4 These arrangements are expec ted to boost trade in organic foods and non food products in all three markets. Implementation of the National Organic Standards (NOS) by the United States Department of Agriculture (USDA) in 2002 may have played a key role in boosting 1 The Organic Trade Association (OTA) Annual report 2011 2012 Accessed April 2, 2014, available at http://ipaperus.ipaperus.com/OTA/OTAAnnualReport/OTAAnnualReportFall2012/ . 2 Accessed February 21, 2014, available at http://www.foodbusinessnews.net/News/News%20Home/Consumer%20Trends/2012/4/Organic%20food %20sales%20increase%20more%20than%209.aspx?cck=1 . 3 Accessed February 21, 2014, available at http://www.ers.usda.gov/Briefing/Organic/Trade.htm. The United Sates does not have consistent data on organic trade because organic product codes have not yet been added to the US and int ernational harmonized system of trade codes. Therefore, the data in 2002 are the latest we could find. 4 Information about these agreements was a ccessed on February 21, 2014, available at http://www.ota.com/GlobalMarkets/Trade and Equivalency Agreements.html .
21 organic sales. These rules and regulations govern organic production practices and any standards. It is hoped that consumers will gain confidence from the government oversight of organic p roduction, and recognize organic foods from the voluntarily displayed USDA organic label. The standardized requirements on organic labels appear to have encouraged increased supply as well as demand for organic foods (Durham 2007). On the other hand, count ry of origin labeling (COOL) became mandatory for of Origin Labeling Final Rule. 5 This labeling law requires that retailers, such as supermarkets, provide information about the source of meats; fish and shellfish; fresh and frozen fruits and vegetables; peanuts, pecans, and macadamia nuts; and ginseng (AMS 2013). 6 COOL makes it relatively easy for consumers to know where most of the food products are produced. Since both the USDA organic label and COOL provide useful information about the products, consumers may use both labels as proxies for food quality and safety. A widely viewed televised investigative report in 2008 gives some indication of how consumers ma y react to an organic product displaying COOL on it. The report revealed that some frozen organic vegetables sold by Whole Foods Market (WFM) in the United 5 Information available about the COOL Final Rule Agricultural Marketing Services (AMS) . Accessed February 21, 2014, av ailable at http://www.ams.usda.gov/A MSv1.0/ams.fetchTemplateData.do?template=TemplateN&navID=FinalRule&r ightNav1=FinalRule&topNav=&leftNav=CommodityAreas&page=COOLRulesandRegulations&resultType =&acct=cntryoforgnlbl . 6 Agricultural Marketing Service (2013) Country of Origin Lab eling rules and regulations. Accessed February 21, 2014, available at http://www.ams.usda.gov/AMSv1.0/COOL .
22 States were imported from China. In a short time period, the sales of organic foods at WFMs dropped as consumers began questioning the quality and safety of organic foods any food products from China since 2010 except for frozen edamame (WFM Blog 2012 7 ). Consumer reactions to imported organic foods at WFM raise many interesting questions. The organic designation can reduce information asymmetry for products and provide consumers with more information about how the products were produced. However, in the WFM case, consumers hesitated to buy organic products from China, product certified by a USDA agency must comply with all US organic standards, no matter where it is produced (hereafter referred a s the Equal Organic Standard Rule). This raises several questions: (1) Are consumers aware that USDA has the same requirements for all organic foods, regardless of country of origin?; (2) Would knowing rted organic foods?; and (3) Do consumers have the same concerns with the food quality of imported organic foods, no matter in which country they are produced? about multiple at tributes of a food product. There are many studies about consumer preferences for organic foods or COOL, but few have studied the interaction between these two attributes. In addition, no research has examined the impact of consumer knowledge of the Equal Organic Standard Rule on consumer attitudes about imported organic foods. Our research fills the gap in current research by determining the impact 7 Accessed February 21, 2014, available at http://www.wholefoodsmarket.com/blog/whole story/dispelling rumors organics china
23 WTP for organic foods. We further examined whether providing information about the Equa l Organic Standard Rule would reduce consumer concerns about imported organic foods. To determine whether consumers are concerned about organic foods from all foreign countries or specific countries, we consider COOL in both developed and developing countr ies in our analysis. This study helps to understand consumers purchasing behavior when they are face d with foods with multiple attributes at the same time. This information helps growers to provide organic products that are most suitable to consumer preferences. For domestic growers, a better understanding of consumer preferences will allow them to seize the best marketing opportunities, rather than competin g with foreign producers based on the simple low about imported organic products also provides policy makers useful in sights on how consumers perceive government certification standards, as well as their increasingly polarized attitudes with respect to globalization. 2.2 Literature Review gene ral conclusion that consumer are willing to pay a higher price premium for organic foods than conventional food (Thompson 1998; Tho mpson and Kidwell 1998; Michelse n et al. 1999; Worner and Meier Ploeger 1999; Browne et al. 2000; Magnusson et al. 2001; Jone s and Clarke Hill 2001; Zanoli and Naspetti 2001; Wier and Calverley 2002; Batte et al. 2007). For instance Wier and Calverley (2002) find that over half of consumers would pay a price premium of 5 10% for organic foods in European markets. Based on a cust omer intercept survey in six stores of a US national grocery
24 chain, Batte et al. (2007) find that consumers are willing to pay premium prices for organic foods, even those with less than 100% organic ingredients. Thompson (1998) summarizes that the demogr aphic variables, such as age, marital status, number and age of children, and education, are possible indicators in studies show that women, people with higher levels o f income and education, and families with more children have more positive attitudes towards organic foods and are more willing to purchase organic products, and that younger consumers tend to have a more positive attitude towards buying organic foods, alt hough their purchase frequency is low because of the price premium (Davies et al. 1995; Magnusson et al. 2001, 2003; Yiridoe et al. 2005; Krystallis , Fotopoulos, and Zotos 2006; Roitner Schobesberger et al. 2008; Tsakiridou et al. 2008; Urena, Bernabeu, a nd Olmeda 2008). Studies also demonstrate that health conce rns and environmental concerns are important reasons for consumers to purchase organic foods (Wilkins and Hillers 1994; Wandel and Bugge 1997; Schiffer stein and Oude Ophuis 1998 ; Magnusson et al. 2001; Lea and Worsley 2005; Shepard et al. 2005; Hoefkens et al. 2009). In addition, Briggeman and Lusk (2011) show that about 15% of the WTP premium for organic foods is derived from antipathy toward inequality. At the same time, a large body of research has studied country of origin effects Most confirm that consumers prefer foods produced in their own country or region (Verlegh and Steenkamp 1999; Loureiro and Umberger 2003; Loureiro and Umberger 2005; Gao and Schroder 2009; Kim et al. 2012; Peterson et al. 2013). For example, Loureiro and Umberger (2003) show that US consumers are willing
25 to pay $1.53 and $0.70 per pound more for steak and hamburger, respec tively, produced in the United States. Lim et al. (2013 ) show that consumers would be willing to pay $1.09 to $35.12 per pound more for Canadian steak than for imported ones in Canada. In addition, Peterson et al. (2013) find that Japanese consumers are wi lling to pay more for rice and pork from their own country than those imported from the United States even though they think the tastes are very similar. Tonsor, Shroeder, and Lusk Product of North America and Product of United States are similar, and the Product of Canada, Mexico, and US label is the least preferred. Country of origin has an impact on consumer preference because consumers use COOL to infer the quality of a product based on t heir shopping experiences (e.g., Bilkey and Nes 1982; Steenkamp 1990; Loureiro and Umberger 2005) . Other studies show that country of origin is not merely a cognitive cue, but also has sy mbolic and emotional meaning. COO L affects consumer behavior through country image, along 8 ethnocentrism, 9 and trust in the certification processes and agencies (Agra wal and Kamakura 1999; Verlegh and Steenkamp 1999; Ehmke, Lusk, and Tyner 2008) . In recent years, more and more studies include both organic and geographic/location labels in their experiments to study consumers preferences. Sackett, Shupp, and Tonsor (2012) have conducted a choice experiment to study 8 Animosity is defined as anger related to previous or ongoing political, economic, or diplomatic events. 9 Ethnocentrism is defined as a belief that it is inappropriate, or even immoral, to purchase foreign products because to do so is damaging to the domestic economy, domestic jobs, and is unpatriotic.
26 locally grown labels are included. Their results show that consumers are willing to pay higher premiums for locally grown products, and they also find little market differentiation between organic and sustainably labeled products. In addition, Aprile, Caputo, and Nayga (2012) find that respondents are willing to pay the highest pr emium price for one origin cue, Protec ted Des ignation of Origin (PDO) , followed by organic farming labe l, and then another origin cue, Protected Geographical Indications (PGI ). Most of the previous studies, however, do not discuss the interactive effects between geographic/location labels and organi c labels. Dentoni et al. (2009) is one of local ly ha s a positive effect in combination with other crede nce attributes . The most relevant study is by Onozaka and Thilmany (2011) , which investigates the interactive effects between claims of sustainable production practices ( organic, fair trade, and low c arbon footprint ), and production location ( local, domest ic , or international). T hey found that the negative valuation of import ed foods could be mitigated by USDA organic certification standards. Their results show that USDA organic certification standards helps consumers assess the quality of the food product and may reduce the information asymmetry between the international producers and US consumers. Nonetheless, the extent to which consumers understand the USDA organic certification standards, especially for imported foods, remains unclear. Hutchins and Greenhalgh (1995) find that consumers are confused about the meaning of the term Friedland (2005) points out that some consumers may assume that organic foods prod uced in foreign countries follow less stringent standards than those produced
27 in the United States, even though they are certified organic by the USDA. With increased demand for organic foods and globalization of the organic food industry, this research im plies the importance of examining the interactive effects between COOL and USDA organic label when they are informed about the Equal Organic Standard Rule may help explain the interactions betwe en COOL and organic labels. 2.3 Survey Design and Data 2.3.1 Survey Design To study the interactive effects between COOL and USDA organic labels, we chose a vegetable that is both domestically produced and imported from foreign countries. In this case, fresh broccoli was chosen for this study because it is a very popular vegetable that is both domestically produced and imported and has similar ex port and import levels (Figure 2 1 ). US per capita fresh broccoli consumption increased by 5.1 pounds from 198 0 to 2011, partially due to consumer awareness of its high fiber, vitamin, and mineral content (USDA/ERS 1999). In our study, we included two developing countries (China and Mexico), and two developed countries (United States and Canada), to analyze the COOL attribute. Research shows that whether the country of origin is a developed or a developing country does matter when consumers have little information about an imported product (Erickson, Johansson, and Chao 1984; Hulland, TodiÃ±o, and Lecraw 1996). Me xico and Canada were chosen because they each share a border with the United States. Imported fresh broccoli from Mexico has increased dramatically since 2000. In 2010, United States imported around 250 million pounds fresh broccoli from Mexico (Figure 2 2 ). Canada has very similar food regulations to the United States, and it has exported
28 fresh broccoli to the United States for decades (Figure 2 2 ). We also included products of China in the COOL attribute because China is one of the top fresh vegetable exp orters and one of the largest organic food producers in the world. In 2010, China had 1.39 million hectares of organic agricultural land and the largest organic aquaculture in the world (Willer and Kilcher 2012). However, while China has the potential to b ecome a large organic food exporter to the United States and elsewhere, the incident in the 2008 WFM case shows that US consumers have great concerns with food imported from China. Adding China in the COOL scenario along with Mexico would help researchers understand whether consumer s value fresh broccoli imported from other countries differently than domestically produced broccoli, even if they carry the same USDA organic labels. This research used choice experiments to analyze the interactive effects betw een USDA organic certification and COOL. In a choice experiment, respondents face products with different attribute combinations, which closely reflect actual shopping conditions; in addition, choice experiments are compatible with random utility theory an d particular market with different types of attributes. Therefore, compared to other survey methods, such as contingent valuation methods, choice experiments are frequently us incentive comparable (Lusk and Schroeder 2004). Other than the COOL and organic attributes, price attribute with four levels was also included (Table 2 1). We used the median mark et price ($1.65/lb) of fresh broccoli as the baseline and generated four levels of price by increasing/decreasing this price by about 30%. When constructing the
29 choice sets in the choice experiment, we considered both the main effects and the interactive e ffects of COOL and organic attributes. The D Optimal criterion was used to identify the best combination of choice sets, resulting in a choice experiment with 13 random order as Savage and Waldman (2008) recommend. In each choice set, survey participants were presented with three alternatives two fresh broccoli options and a no purchase option. In the survey, respondents answered a series of discrete choice questions rega rding which option they would choose. The levels of all attributes are shown in Table 2 1 and an example of the choic e questions is shown in Figure 2 3 . 2.3.2 Information Treatment As is the case, sometimes consumers are confused about the meaning of the t less stringent standards than those from the United States ( Hutchins and Greenhalgh 1995; Friedland 2005) . Therefore lack of information may be a key factor that makes consumers hesitate to purchas e imported organic foods. The acceptance of imported organic foods could vary among consumers with different understandings of USDA organic certif ication requirements. In order to determine whether the lack of the information on the Equal Organic Standard Rule is the key factor that incurs the negative attitudes toward imported organic food, we constructed two ver sions of the choice experiments . One version of the choice experiment gave respondents information about the USDA Equal Organic Standard Rule by presenting the following statement before the choice experiment questions:
30 Organic 10 : The USDA requirements for organic certifi cation are extensive . The can only be used to describe an agricultural product sold in the United States if it meets all of these requirements. No matter where a product is produced, the same rules and procedures apply. To make sure that products labeled "organ ic" do meet the USDA requirements, one of the 85 (49 in the United States and 36 abroad) certifying agencies that the USDA has accredited must verify that all of the products and procedures used in production, processing, packaging and transportation compl y with the USDA requirements and that the inspe ction revealed no exceptions. Conventional: In this study the te refers to agricultural products that do not meet the USDA standards for organic products. The other version of the choice ex periment did not provide this information. The two versions of the choice experiments were randomly assigned to respondents. We treatment control assigned to the control group may have already known and understood the equal requirement certification standards/rules before completing the experiment. In this case, WTP between the information treatment and control groups constitutes the lower bound of information effects. 10 The information treatment is designed based on the info rmation from the USDA website. Accessed February 21, 2014, available at http://www.ams.usda.gov/AMSv1.0/ams.fetchTemplateData.do?template=TemplateJ&navID=NationalOrg anicProgram&leftNav=NationalOrganicProgram&page=NOPACAs&description=USDA%20Accredited%2 0Certifying%20Agents&acct=n opgeninfo .
31 2.4 Analysis The consumer choice decision process is modeled within a random utility theory framework (Thurstone 1927; Bockstael, Hanemann, and Kling 1987; Morey, Rowe, and Watson 1993) . In this case, a decision maker , is facing choice scenarios and each choice scenario has alternatives. The utility level of the th product on scenario for the th decision maker can be expressed as: . (2 1) where is the deterministic component and is the random component. The decision maker chooses the alternative that provides the greatest utility (i.e., he/she will choose alternative if and only if , ). Following McFadden (1974), the probability that decision maker chooses alternative on th choice scenario is (2 2) In this chapter we assume a usual linear in parameter u tility function al form of the deterministic component, , such that (2 3) where represents the price of product in th choice scenario, vector represents the dummy vector of the country of origin (US is the base), and represents the dummy variable of O rganic . 2.4.1 Conditional Logit Model Under the assumption that is independent and identically distributed (i.i.d.) with extreme value distribution , and if all individuals are assumed to have homogenous
32 preferences for each attribute, the probability of consumer choosing alternative can be estimated by Equation ( 2 4) with a conditional logit (CL) model (McFadden 1974; Louviere , Hensher, and Swait 2000) : for (2 4) Let , if respondent chooses option at occasion , the lik elihood function then can be expressed as (2 5) where if alternative is chosen by the th individual, and otherwise. 2.4.2 Generalized Mixed Logit Model The generalized mixed logit (G MIXL) model relaxes the independence of irrelevant alternatives (IIA) assumption of the conditional logit model and can also capture the heterogeneou s preference among consumers. The G MIXL model is a generalized model nested with the mixed logit model, scale heterogeneity multinomial (or conditional) logit model, and conditional logit model (Fiebig et al. 2010). Steckel and Vanhonacker (1988) and Train ( 2003 ) show that heterogeneity of preference could exist due to differences in unobservable attitudinal characteristics across individuals. Therefore a model such as the mixed logit (MIXL) model that can capture the heterogeneity amon g consumer s is currently quite popular. Louviere et al. (1999, 2002, 2008), Louviere and Eagle (2006), and Louviere and Meyer (2007) have argued that MIXL model can be misspecified and the taste heterogeneity in most choice experimen
33 the scale of their idiosyncratic error terms is greater or less than others. This leads to the development of the scale heterogeneity multinomial logit (S MNL) model or scale hetero geneity conditional logit model (S CL). To embrace both MIXL and S MNL, Fiebig et al. (2010) developed the G MIXL model. 11 Using G MIXL would allow us to capture both preference and scale heterogeneity. Following Fiebig et al. (2010), G MIXL that nests both MIXL and S CL models in the utility function to person from choosing alternative on purchase occasion (or choice scenario) is given by (2 6) vector is usually assumed as MVN (0, ) . vector is all the attributes and interaction terms to person from choosing alternative on choice scenario . The random scaling factor, has mean 1 and variance . , and are parameters, and some special value of , and can imply specific models: implies the original conditional logit model ; implies the scale d conditional logit model ; implies the mixed logit model as and ; 11 Fiebig et al. (2010) call it the G MNL model, which combines the MIXL and S MNL models. In our logit model. Thus in this paper, we call it G MIXL, which is a generalized model nested with the MIXL and S CL model.
34 implies a hybrid model where , Fiebig et al. (2010) call it G MNL I ; implies a scaled mixed logit model as , Fiebig et al. (2010) call it G MNL II . Since we have panel data that each respondent has multiple choice tasks, let if respondent chooses option at occasion , then the simulated choice probabilities that this person choosing a sequence of choice becomes: (2 7) where . 2.4.3 Estim ation in the WTP S pace With choice for a certain attribute is as the negative ratio of the attribute coefficient to the price coefficient. 12 In other words, parameters in the model are first estimated in the and then WTP is calculated as a function of the preference parameters. However, with conveniently tractable distributions for preference parameters, estimating WTP in the preference space often results in counterintuitive distribution of the WTP (Scarpa, Thiene, and Train 2008). For instance, if the value of the denominator of this ratio is close to zero, the ratio would be extremely large; thus, the distribution of the WTP will have an infinite upper tail that artificially skews the mean and the variance of WTP estimates. 12 This is true when the attribute levels are coded with dummy coding. Other coding of the attribute levels may result in similar but slightly different formulae for the WTP calculati on.
35 One solution for this problem is to re parameterize the estimation model, such that the WTP is estimated directly from the models instead of from the preference coefficients of each attribute Weeks 2005; Sonnier, Ainslie, and Otter 2007). Scarpa, Thiene, and Train (2008) found that models within the WTP space fitted their data better, and reduced the incidence of extremely large WTP estimates. Following Train and Weeks (2005), d ividing the utility function in Equation ( 2 6) by a scale parameter, , will result in a new error term with a different variance without . (2 8) where . If , which is the negative of the WTP vector for all attributes facing by individual , then Equation ( 2 8) becomes , (2 9) o r where is the price parameter in preference space, and is a vector of negative of WTP for product attributes. By adopting a WTP space in G MIXL in the WTP space, this study uses the same simulation method as stated in Equation ( 2 7), but now we directly parameterize and instead of . The WTP esti mates, therefore, do not need to be derived by the ratio of two random parameters as is done in the preference space models. In the
36 meanwhile, we convert the prices to their negative in the regression analysis so that the WTP for attributes, , can be directly estimated. 2.5 Empirical Results 2.5.1 Data Collection and Descriptive Statistics Toluna, an online survey company, delivered the survey s to its representative consumer panels in the eastern half of the United States i n November 2010. A total of 508 individuals were invited to participate in the online survey and 348 completed questionnaires were collected. The 348 respondents were over 18 years of ag e, primary grocery shoppers in their households, and had purchased fres h produce within the last month. The choice experiment section gave us 4,524 13 valid observations (choice sets) in total. The demographic characteristics of the respondents (gender, education, household income, employment status, family size, and number of children in the family) are reported in Table 2 2 . Compared to the 2010 American Community Survey, the distribution of males and females in our sample is very close to that of the US population. The median household income is $50,000~$74,999, consistent w ith the national median household income of $50,046. The percentage of households with children under 18 years old in our sample is also consistent with the national average. The proportion of high income respondents in our sample is slightly larger than t hat of the US population and our sample has a higher education level than the national 13 Each respondent was presented with 13 choice sets, thus the number of valid observation of choice sets is 348 x 13 = 4,524.
37 average, due in part to the restriction of recruiting adults only and that the survey was conducted online. T he demographic characteristics of the information treatment and control groups are also reported in Table 2 2 . The chi square tests for each demographic characteristic cannot reject the null hypothesis that these two groups share the same demographic characteristics . 2.5.2 Shopping Behaviors and Perceptions R espon dents were asked about their shopping behaviors at the beginning of the survey. Among all respondents, read labels regarding labels do play an important role in consumer p urchas e decision s . W e also asked how often they purchased organic foods in the categories of fresh fruits/vegetables, processed food s, meat/poultry/eggs/milk, other dairy products, bread/bakery items, and grains. Respondents indicated that they purchased organic fresh fruit s and vegetables more often than any other category. This result is consistent with our expectation and previous r esearch that organic fresh fruits and vegetables are the most common ly purchased organic products. About 70% and 67% of respondents believed that foods from China and Mexico, respectively, posed median or high food safety risks. All factors combined, 59% and 44% of the respondents stated that they were unlikely to purchase food from China and Mexico, re spectively. Only 14% of respondents said they were unlikely to purchase food Mexico illustrates the impact of country of origin in their purchasing decision, and that
38 they treat fo od from different countries differently. The reason that more consumers think foods from China and Mexico have higher food safety risks than foods from Canada might be because of the food safety incidents in China and Mexico, or that consumers may have the impression that the standard of food producing process i s lower in these countries than in Canada and the United States. questions related to their feelings about purchasing imported p roducts. The majority (71%) of results imp ly that another reason why consumers prefer domestic products is to support the US economy. In total, over 40% ng implies that when consumers purchase organic food, their purchase decisions are affected by the COOL information. 2.5.3 Conditional Logit Model domestically produced conventiona l fresh broccoli, were estimated using STATA 13. The first three columns of Table 2 3 report the conditional logit in WTP space estimates for the pooling sample, the information treatment group, and the control group. Domestically produced conventional fre sh broccoli is the base in analysis. All price (negative) coefficients are restricted to one in the WTP space . The coefficients of Organic are in significant in all three regression models , but some interacti ve terms between COOL and Organic are significant ly positive. All the COOL coefficients (imported from foreign countries) are significantly negative, showing that consumers
39 prefer domestically produced fresh broccoli to imported broccoli. T he ranking order from most preferred to least preferred fresh br occoli is US domestically produced, imported from Canada, imported from Mexico, and imported from China. The significant negative coefficients of COOL and the significant positive coefficient of the interaction term an mitigate the negative effects of COOL to some extent, consistent with the conclusion reached by Onozaka and Thilmany ( 2011) . 2.5.4 Generalized Mixed Logit Model Hausman McFadden tests (Hausman and McFadden 1984) reject the IIA assumption in the conditional logit model, therefore we use the G MIXL model in the WTP space to relax the IIA assumption and to capture heterogeneous preferences among consumers. Further discussion in this study will focus on the results from the G MIXL models. In the G MIXL models, all the attribute coefficients (all main and interaction terms) are specified as random parameters with normal distributions , and the mean coefficient of the price attribute is restricted as one. Because we convert the prices to their negati ve in the analysis, the coefficient estimate of each attribute is The estimations were conducted using STATA 13, with 1,000 Halton draws to simulate random parameters. 14 Comparing the log likelihood value between the condi tional logit model and the G MIXL model, the G MIXL model shows significant improvement in fit. Results of the G MIXL regressions are reported in the last three columns in Table 2 3. The significant negative coefficients of COOL labels in both groups prov ide 14 The stability of the results was verified by using various numbers of Halton draws and different starting values.
40 evidence that consumers would prefer US domestically produced to imported broccoli from Canada, Mexico, and China. The coefficient of the Organic attribute in the information treatment group is significantly positive, but this coefficient is insignific ant in the control group. This implies that consumers in the information treatment group differentiate more between organic and conventional products than do those in the control group. The reason may be that the information treatment raises consumer aware ness about the US organic label . The coefficients of the interaction term Organic China in both information treatment and control label mitigates the negative effects of Products of China in the control group, consistent with the conclusion s of Onozaka and Thilmany ( 2011) . However, the interactive term Organic Mexico in the information treatment group is insignificant, implying that meeting USDA organic standards to improve product co untry image is not a panacea for all countries. In addition, most standard deviations of the random parameters are significant, same attribute, consumers have different attitudes and WTP. Most of the standard deviation estimates of the coefficients in the information treatment group are smaller than that of the control group, indicating that having information might reduce the 2. 5.5 WTP for Broccoli and Impact of Information Using the means and standard deviations of the random parameters from the G MIXL we applied the Krinsky and Robb (1986) bootstrap method to simulate 1,000 relative WTP estimates for each type of broccoli. The means and 95% confidence
41 intervals of WTP for broccoli with various attribute combinations are reported in the first two columns of Table 2 4. The differences in WTP values between the information treatment group and the control group ( ) are reported in the last column . Panel 1 of Table 2 4 compares organic broccoli with conventional broccoli produced in the same country. Results show that most of the marginal WTP value s for Organic (compared with conventionally produced) are significantly positive, except organic broccoli imported from Mexico in the control group. Comparing the conventional counterparts, Chinese organic broccoli receives a relatively higher premium than Canadian and Mexico organic broccoli. For example, in the control group, consumers are willing to pay $0.54 per pound more for organic versus conventional broccoli from China, while they are willing to pay $0.26 per pound more for organic versus conventional broccoli from Canada. When the WTP es timates for imported organic broccoli are compared with WTP for US organic broccoli (Panel 2 of Table 2 4), all the marginal WTP estimates are significantly negative , indicating that US organic broccoli has a significant price premium over imported organic broccoli from these three countries. This is consistent with previous research that consumers prefer domestic foods to imported foods (Verlegh and Steenkamp 1999; Lourei ro and Umberger 2003) . US organic broccoli receives a $0.31 and $0.32 per pound premium over Canadian grown organic broccoli for the information treatment group and control group, respectively. This premium reaches over $1 when comparing US organic broc coli over organic broccoli from China and Mexico. It shows that consumers place similar values on domestically produced organic broccoli and
42 organic broccoli from Canada, but treat organic broccoli from China and Mexico quite differently. A similar pattern is found in Panel 3 of Table 2 4, which shows that imported organic broccoli from China and Mexico cannot compete on any level with US conventional broccoli. This is consistent with previous research showing that consumers have a better image of products from developed countries than those from developing countries (Erikson, Johansson, and Chao, 1984; Hulland, TodiÃ±o, and Lecraw 1996). The impact of the information treatment on consumer behavior can be determined by comparing WTP estimates between the info rmation and control groups (third column in Table 2 4). First, information about the Equal Organic Standard Rule significantly increases the premium of organic over conventional broccoli for products from Mexico, but it has no impact on the premium of orga nic over conventional broccoli for products from Canada and the United States (third column in panel 1 of Table 2 4). Second, Information about USDA organic certification standards/rules adds $0. 19 per pound to the value of organic broccoli from China when it is compared to US organic broccoli. That is, the information brings additional value for organic food from China. However, imported broccoli from Canada and Mexico does not benefit from the information (third column in panel 2 of Table 2 4). Panels 2 a nd 3 of Table 2 4 show that imported organic broccoli is unable to compete with organic broccoli from the United States even when consumers are given the information that imported organic broccoli is subject to the same organic standards as domestically pr oduced broccoli. However, presenting the information reduces the premium of US conventional broccoli over Chinese organic broccoli ($1.47 without information vs $1.31 with information). In general this implies that sourcing organic
43 broccoli from foreign co untries to satisfy the increasing demand in the United States is only profitable when the production costs of organic foods in foreign countries are significantly lower than those in the United States. The information about the Equal Organic Standard Rule imported organic broccoli. 2.5 .6 Value of Information The estimates of consumer acceptance of imported organic broccoli could help address policy questions, such as whether or not importing organic broccoli from foreign countries can improve social welfare, and whether providing information about USDA organic certification standards/rules would further improve social welfare. To answer these questions, we propose a simplified before/after sce nario. In Scenario 1 (pre importing) there is no imported organic broccoli, and in Scenario 2 (post importing) imported organic broccoli is allowed. We assume that imported conventional broccoli exists in both scenarios because the United States has import ed large amounts of conventional fresh produce for many years. In the Pre importing scenario consumers have the following choices: domestic organic broccoli; domestic conventional broccoli; imported conventional broccoli from Canada, China, and Mexico; and no purchase option. In the Post importing scenario the number of choices increases to nine , the previous six choices plus organic broccoli from Canada, China, and Mexico. Consumer surplus is first compared between the two scenarios, after which we compare the welfare changes between the pre importing and post importing scenarios with and without the information treatment. The difference in welfare changes with and without information determines the value of the information. Following Leggett
44 (2002), Train (2003), and Brooks and Lusk (2010) , the change in consumer surplus pre and post importing organic broccoli is calculated by (2 10) Because we estimate the coefficients in WTP space, the coefficient of price (negative), , is assumed to be one, Equation ( 2 10) becomes (2 11) If importing organic broccoli brings a positive consumer surplus, the value of the welfare change between Scenarios 1 and 2 should be positive. By comparing the welfare change of the informati on treatment group to that of the control group, the value of the information is determined. If the difference between these two groups is positive, it could be worthwhile to give consumers more information about USDA organic certification standards/rules from the perspective of improving consumer surplus. We found that consumer surplus changes by 1 cent per choice (one pound) from Pre importing to Post importing without the information. With the information, consumer surplus changes by 2.2 cents per choice . In 2011, broccoli consumption in the United States was 1,850 million pounds (USDA/ERS 2011). Given this volume, the annual welfare change of allowing imports of organic broccoli would be approximately $40.7 million if consumers have information on the US DA organic certification standards/rules, understanding of the USDA organic certification standards/rules is approximately $2 2 . 2 million. Nonetheless, we cannot reach conclusions about the net impact of this
45 information because a complete cost benefit analysis that also takes into account the changes in social costs is needed, which is beyond the scope of this study. 2.6 Conclud ing Remarks This study determines consumer preference for organic foods when COOL wa s present ed as well as the impact of the information on the USDA Equal Organic Standard Rule . We found that consumers preferred domestic organic broccoli to imported orga nic broccoli, and preferred organic broccoli imported from Canada to organic broccoli from Mexico and China. In general, the USDA organic label could mitigate the lower value of imported food to some extent . Information on Equal Organic Standard Rule signi ficantly increased WTP estimates for organic broccoli from China, compared to domestic conventionally produced broccoli. However, few impacts were found for organic broccoli from Canada and Mexico. This essay contributes to the literature by introducing C OOL in the analysis of ing an information treatment to test whether consumers are willing to pay more for imported organic food if they have better knowledge of the USDA organic certification standards/rule s . This essay also studies the welfare changes due to new information about Equal Organic Standard Rule, filling a gap in the literature since studies that include both COOL and organic labeling are rare and no research determinants the impact of informat ion on the Equal Organic Standard Rule. This essay primarily focuses on whether the information has an effect on such as why information creates an impact, or not. Cons umer trust in the information or in the foreign certification agencies or even their trust in the USDA organic certification
46 process can be introduced in future studies. Additionally, a complete cost benefit analysis is needed to assess the net social welf are impacts of increasing the amount of imported organic foods and providing additional information on USDA Equal Organic Standard Rule. The effects of the equal requirement information, however, can be underestimated. In the questionnaire, respondents wer e not asked about their previous knowledge about organic label as well as the understanding of imported organic food labels. It is possible that some of the respondents had the equal requirement knowledge before they participated the survey, therefore the real WTP gap between people with this knowledge and without this knowledge can be higher than the WTP gap we estimated in this chapte r. The results of this chapter indicate that in general consumers would want to pay higher premium for domestically produce d organic food than for imported organic food. Even if they know that the same USDA organic standard is used for food carrying USDA organic label no matter where it is produced, this premium for domestically produced organic food still exists. Food safety issues in developing countries significantly affect their impression about imported food from China and Mexico, and USDA organic label cannot make consumers feel the same about the food from these countries as they feel about domestically produced food in the United States.
47 Table 2 1. Attributes and levels for fresh b roccoli in the c hoice e xperiment . Attributes Levels Unit Price $1.18/lb., $1.52/lb., $1.85/lb., $2.20/lb. Organic USDA organic Conventional COOL Produced in the United States Produced in Canada Produced in China Produced in Mexico
48 Table 2 2. Summary statistics for survey r espondents . Variable Description Information Information free Joint U. S Population 15 Gender 0 if female 53.37 47.19 50.28 49 (%) 1 if male 46.63 52.81 49.72 51 Pearson chi2(1) = 2 P value = 0.157 Education Less than high school 1.12 1.12 1.12 High School or (%) High school degree or equivalent 19.66 13.48 16.57 Higher 86 Trade/technical school 5.62 5.06 5.34 degree Some college 33.71 28.65 31.18 and higher Four year college degree 19.66 29.21 24.44 28 Post graduate (MS, PhD . , MD etc.) 20.22 22.47 21.35 Pearson chi2(20) = 24 P value = 0.242 Household Less than $14,999 7.30 6.18 6.74 13.4 income $15,000~$24,999 7.30 11.24 9.27 11.5 (per year) $25,000~$34,999 8.43 10.67 9.55 10.8 (%) $35,000~$49,999 24.72 17.42 21.07 14.2 $50,000~$74,999 23.03 17.98 20.51 18.3 $75,000~$99,999 11.80 13.48 12.64 11.8 $100,000~$149,999 12.92 14.61 13.79 11.8 $150,000~$199,000 2.81 4.49 3.65 4.2 $200,000 or above 1.69 3.93 2.81 3.9 Pearson chi2(56) = 63 P value = 0.243 Employment Full time 48.02 49.44 48.73 In labor force: status Part time 9.04 14.61 11.83 Employed (%) Current not working 15.82 11.80 13.80 50.7 Retired 22.03 18.54 20.28 Unemployed Student 2.26 2.81 2.54 6.9 Other 2.82 2.81 2.82 Pearson chi2(20) = 24 P value = 0.242 Family 1~2 55.06 55.06 55.06 Average Family size (%) 3~4 37.08 37.08 37.08 Size 5 or above 7.86 7.86 7.86 2.6 Pearson chi2(4) = 6 P value = 0.199 Number of None 65.34 67.80 66.57 Households with children 1 18.75 15.25 17.00 One or More in the family 2 11.36 13.56 12.46 Children under 18 15 Source: US Census Bureau, 2010 American Community Survey.
49 Table 2 2. Continued Variable Description Information Information free Joint U. S Population (%) 3 or more 4.54 3.39 3.97 33.1 Pearson chi2(9) = 12 P value = 0.213
50 Table 2 3. Comparison of information treatment and control groups: C onditional Logit and G MIXL model . Conditional Logit Model WTP Space G MIXL Model WTP Space Pooling Data Information Treatment Control Group Pooling Data Information Treatment Control Group Mean Estimates Main Effects Organic 0.07 0.09 0.07 0.17 0.11 * 0.17 Canada 0.48 *** 0.47 *** 0.48 *** 0.94 *** 0.49 *** 0.43 *** China 1.61 *** 1.45 *** 1.80 *** 4.62 *** 1.57 *** 2.05 *** Mexico 1.10 *** 1.10 *** 1.10 *** 2.95 *** 1.25 *** 1.03 *** Price (negative) 1 (fixed) 1 (fixed) 1 (fixed) 1 (fixed) 1 (fixed) 1 (fixed) Label interaction Organic Canada 0.20 * 0.23 * 0.13 0.71 *** 0.17 * 0.13 Organic China 0.35 *** 0.22 ** 0.51 *** 0.71 *** 0.19 ** 0.41 *** Organic Mexico 0.15 0.19 0.06 0.28 0.14 0.15 Constant 2.64 *** 2.44 *** 2.89 *** 4.88 *** 2.43 *** 2.69 *** Standard Deviation Estimates s.d. of Price 1.09 *** 0.33 *** 0.33 *** s.d. of Organic 2 .23 *** 0.58 *** 0.99 *** s.d. of Canada 1.30 *** 0.47 *** 0.53 *** s.d. of China 3.65 *** 1.23 *** 1.42 *** s.d. of Mexico 2.53 *** 0.94 *** 1.12 *** s.d. of Organic Canada 0.12 0.10 * 0.05 s.d. of Organic China 0.26 0.23 * 0.03 s.d. of Organic Mexico 0.34 * 0.13 * 0.16 ** Tau 0.14 * 0.72 0.71 *** Observations 4,524 2,275 2,25 4,524 2,275 2,25 Log likelihood(LL) 4218.62 2094.01 2097.63 3224.74 1572.26 1472.90 Note: ***, **, * = Significance at 1%, 5%, 10% level; Halton draws=1,000.
51 Table 2 4. WTP for broccoli a ttributes (G MIXL) (dollar/lb) Information Treatment Group Control Group WTP c Panel 1. Organic (Org) vs. Conventional (Con) US Org vs. US Con 0. 10 b 0.13 0.03 [0.06 , 0.13 ] a [0.07, 0.19 ] [ 0.10, 0.04 ] Canada Org vs. Canada Con 0.27 0.26 0.01 [0.24, 0.31 ] [0.20, 0.32 ] [ 0.06, 0.09 ] China Org vs. China Con 0.30 0.54 0.24 [0.26, 0.34 ] [0.47, 0.60 ] [ 0.31, 0.16 ] Mexico Org vs. Mexico Con 0.22 0.01 0.24 [0.19, 0.260 ] [ 0.01, 0.05 ] [0.16, 0.31 ] Panel 2. Imported Organic vs. U.S. Organic Canada Org vs. US Org 0.31 0.32 0.01 [ 0.34, 0.28 ] [ 0.35, 0.29 ] [ 0.04, 0.05 ] China Org vs. US Org 1.41 1.59 0.19 [ 1.48, 1.33 ] [ 1.68, 1.51 ] [0.08, 0.30 ] Mexico Org vs. US Org 1.08 1.13 0.05 [ 1.14, 1.02 ] [ 1.20, 1.06 ] [ 0.04, 0.14 ] Panel 3. Imported Organic vs. US Conventional Canada Org vs. US Con 0.22 0.19 0.03 [ 0.26, 0.17 ] [ 0.26, 0.12 ] [ 0.11, 0.06 ] China Org vs. US Con 1.31 1.47 0.16 [ 1.39, 1.23 ] [ 1.57, 1.36 ] [0.02, 0.29 ] Mexico Org vs. US Con 0.98 1.00 0.02 [ 1.05 , 0.91] [ 1.09, 0.91 ] [ 0.10, 0.13 ] Note: a Numbers in the square brackets are the 95% confidence interval of the WTP. b Mean and confidence intervals of WTP were determined by 1,000 bootstrapped WTP estimates calculated using the Krinsky Robb bootstrapping method. c WTP was calculated by the difference between the 1,000 bootstrapped WTP estimates in information trea tment group and the 1,000 bootstrapped WTP estimates in control group.
52 Figure 2 1. Annual imported fresh broccoli (p ounds) 17 . 17 Resource: USDA (US Broccoli Statistics) http://usda.mannlib.cornell.edu/MannUsda/viewDocumentInfo.do?documentID=1816 0 50000000 100000000 150000000 200000000 250000000 300000000 350000000 400000000 450000000 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Broccoli Imports and Exports Imported to US Exported from US
53 A B Figure 2 2 . Annual imported b roccoli (pounds) 18 . A) Annual broccoli i mports from Mexico. B) Annual broccoli i mports from Canada . 18 Resource: USDA (US Broccoli Statistics) http://usda.mannlib.cornell.edu/MannUsda/vie wDocumentInfo.do?documentID=1816 0 50,000,000 100,000,000 150,000,000 200,000,000 250,000,000 300,000,000 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Fresh Broccoli imported from Mexico 0 2,000,000 4,000,000 6,000,000 8,000,000 10,000,000 12,000,000 14,000,000 16,000,000 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 Fresh Broccoli imported from Canada
54 Which of the following choices for fresh broccoli would you pick? $1.52/lb $1.18/lb I would not choose either product Produced in Mexico Produced in United States USDA Organic Conventional Figure 2 3. A s ample choice set in the choice e xperiment .
55 CHAPTER 3 BARGAINING ENVIRONMENTS IN NON HYPOTHETICAL EXPERIMENTS 3 .1 Valuation Methods The inconsistency in product valuations between hypothetical and non hypothetical experiments has been examined t horoughly in the literature. A wealth of evidence has indicated that individuals tend to over state the amount of mone y they are willing to pay in state d preference survey (Champ and Bishop 2001; Harrison and RutstrÃ¶m 2008; List 2001; List and Shogre n 1998 ; Murphy et al. 2005). Because of this hypothetical bias in state d preference methods, more researchers have recommended and switched to non hypothetical experiments in order to obtain more incentive compatible WTP estimates. Non hypothetical experiments are viewed as incentive compatible because each is equal to his or her valuation of the offered product. In other words, these experiments provide respondents with in centive to reveal their true preferences. In recent years, however, several studies have found significant disparities between two popular non hypothetical experiments: experimental auction s (EA) and real choice experiments (RCE) (Lusk and Schroeder 2006; Gracia, Loureiro, and Nayga 2011; Grebitus, Lusk, and Nayga 2013). These studies show that the auction bids in EA were significantly lower than the WTP estimates from RCE, and the demand elasticities calculated from the two methods were also inconsistent. While Lusk and Schroeder (2006) and Gracia, Loureiro, and Nayga (2011) suggested that the disparity in these two experiments might come from mechanism differences; where exactly the differences originated was not specified. Since non -
56 hypothetical experime nts are increasingly used in consumer preference studies, it is important to both quantify the differences in WTP estimates from different valuation methods and determine the reasons that lead to this discrepancy. This can help researchers better interpret their WTP estimates, as well as improve the current valuation methods. characteristics play different roles in RCE and EA. After measuring six types of they found that personality has a larger role in explaining behavior in choice experiments than in auctions, and different traits have different impacts in these two experiments. While this study also attempts to explore the potential reasons for the dispa rities in WTP estimates across different non hypothetical experiments, our perspective differs significantly from Grebitus, Lusk and Nayga (2013). estimates in EA and RCE, this study focuses on the differences in the bargaining environments provided by valuation methods, and tests whether this can explain the WTP estimates gap. Because the price attribute is presented differently in these experiments, the bargaining environment (bargai ning power granted to respondents) varies across experiments. That is, EA may give respondents more bargaining power than RCE during the valuation process. Rather than only focusing on EA and RCE, this study also contributes to the body of literature by in cluding real contingent valuation methods (RCVM) in the comparison. The contingent valuation method (CVM) is an important valuation method in both marketable and non marketable good valuations. Within a double bounded dichotomous choice CVM framework, this study designed an
57 effective way to provide respondents incentives to reveal their WTP. Specifically, the design prevents the scenario that respondents can always choose the lowest price level even if their true WTP is higher than lowest price in the exper iment . 1 Compared to RCE, respondents in RCVM can choose to accept or reject a given price level, but they still cannot name their own price as they do in EA. To summarize, the main objectives of this study are to test: (1) whether the WTP estimates from the three non hypothetical valuation methods: EA, RCVM, and RCE; differ significantly, and (2) whether the bargaining power of experiments can be a significant factor that explains the differences in WTP estimates from these valuation methods. The first hypothesis of this study is that the WTP estimates from EA, RCE, and RCVM are the same . In addition, because RCE, RCVM, and EA grant different levels of bargaining power to respondents, the discrepancy between respon dents who are more aggressive and these who are not in price bargaining varies in different valuation methods. Specifically, the hypotheses are: Hypothesis 1: WTP estimates from RCE, RCVM, and EA are equal. Hypot hesis 2: In each experiment, WTP estimates for respondents in low, middle, and high aggressiveness groups are equal. In RCE: In RCVM: In EA: 1 If respondents figure out the price structure of double
58 The results show that in EA, the WTP estimates from respondents who are very aggressive in price bargaining are significantly lower than the WTP estimates from respondents who are not aggressive. While significant diffe rences in WTP estimates exist for some but not for all products in RCVM, the WTP estimates are not significantly different between respondents with different bargaining preferences in RCE. This implies that EA may grant participants more bargaining power, while RCE provides fewer chances for participants to bargain and RCVM falls between EA and RCE with bargaining opportunities. 3 .2 Bargaining E nvironment in RCE, RCVM, and EA Evans and Beltramini (1987) define bargaining as an exchange activity that promotes mutual benefits between the bargainers. Most studies on bargaining behavior focus on the determination of important factors that influence behavior (Buchan, Croson, and J ohnson 2004; Brucks and Schurr 1990; Campbell et al. 1998; Graham 1985; Schurr and Ozanne 1985). Compared to these papers, Lee (2000) intention model 2 to conduct a cross cult ural study of bargaining behavior between American and Chinese customers. Particularly, he develops a sequence of questions to ttitude, intention, and action. decreasing, and this can be an important factor for the inconsistent WTP estimates that 2 Fishbein behavioral intent ion model (Fishbein and Ajzen 1975; Ajzen and Fishbein 1980; Fishbein 1980) was constructed and developed by Fishbein and Ajzen. It postulates a set of relations between attitudes, intentions, and behavior.
59 previous research has discovered. The decreasing bargaining power of respondents in EA, RCVM, and RCE, as well as the potential of bargaining environments in explaining the disparity, can be explored by checking the roles of price in these methods. In EA, although involved in monetary transactions to ensure that the experim ent is incentive compatible, participants are asked to name their own prices in the experiments. In discrete choice RCVM, such as single bounded or double bounded RCVM, participants are not given the option of naming their own prices; however, they can dir ectly show their preferences for price by accepting or rejecting the prices offered by experiment hosts. In RCE, participants are asked to choose their most preferred products based on combinations of both price and non price attributes, and they do not ha ve the chance to bargain about the price directly. Therefore may be the highest in EA, followed by that in RCVM , and respondents in RCE may have the lowest bargaining power. We conjecture that the different bargaining powers respondents are granted could be one of the important factors for the inconsistent WTP estimates among the three experiments. However, bargaining power is not easy to measure directly in regular marketing valuation experiment al designs. Therefore, we use t he behavior of consumers with different bargaining preferences as an indirect indicator of bargaining power. The idea is that, ceteris paribus, if an experiment (e.g., EA) provides respondents greater bargaining power, those who are more aggressive in pric e bargaining are more likely to offer a lower price than those who are less aggressive in price bargaining. However, in an experiment (e.g., RCVM or RCE) in which respondents have lesser bargaining power, the difference between respondents who are aggressi ve
60 in price bargaining and who are not should not differ significantly. This study measures difference in WTP estimates between aggressive and unaggressive respondents to be highest in EA and lowest in RCE. 3 .3 Experimental D esign 3 .3.1 General Experimental D esign To test the consistency of the WTP estimates among the three methods, this study carefully designed RCE, RCVM, and EA to be comparable. The features of the differe nt experiments were kept very similar by using the same products, maintaining the same flow of the experiments, and setting the range of price s as close as possible. 3 Moreover, we recruited the experiment participants randomly so that their demographic cha racteristics are similar across all groups. In July of 2012, experiment respondents were recruited in front of local grocery 4 Grocery shoppers who were randomly stopped and invited to participate in a consumer survey involving orange juice. Those who had consumed orange juice in the past qualified for the survey and received 20 (about US$3.20) 5 a s compensation for his/her time. Respondents then, were randomly assigned to one of the three experiments. The orange juice products used in the experiments were 100% Not From 3 EA, but we can control the price range in RCE and RCVM. 4 Participants in Changsha may not represent the whole population in China, but this should not be a problem because the focus of this study is the difference among three non hypothetical experiment s. 5 The exchange rate of US dollars to Chinese Yuan was around 6.3/$ in July of 2012. Accessed August 20, 2012, available at http://www.bloomberg.com/quote/USDCNY:CUR/chart .
61 Concentrate (NFC) orange juice, 100% Fr om Concentrated Orange Juice (FCOJ), and 10% Orange Juice Drink (OJD). The 100% NFC orange juice is fairly new in the Chinese market and is not widely available in grocery stores yet. The 10% OJD is the most popular orange juice product in China, followed by 100% FCOJ. In the real market, the product size of NFC is usually larger than FCOJ and O JD, but the size was equalized (500ml) in all experiments because consumers may have different size preferences. The information about each type of orange juice was introduced to participants at the beginning of the experiments (Appendix A ). Each type of o range juice was served in small cups, and respondents were asked to taste and rank them from 1 (the best) to 3 (the worst). Then they were given time to finish the experiment and complete a survey about their socioeconomic and demographic characteristics. In the last section of the and behaviors in order to measure their bargaining preferences. Respondents were told that they should evaluate each product carefully and stat e their true WTP , as each one had an equal chance of being binding and an actual payment might occur at the end of the interview . 3 .3.2 Measure Aggressiveness in Price B argaining (2 000) to ask a sequence of questions (Table 3 1 ) about their bargaining attitude, intention, and competitiveness. 6 For each question, a five point Likert scale was used 6 The subjective norm of bargaining as included by Lee (2000) was left out as this study does not compare the culture difference.
62 and respondents were asked to choose from 1 (Strongly Agree) to 5 (Strongly Disagree). T he lower the score, the more aggressive a respondent was in price bargaining. Since bargaining attitude, intention, and competitiveness can present different bargaining prefer ence. Based on eight measurements (see Table 3 1 ), the K means cluster analysis (MacQueen 1967) is used to categorize the respondents into low, middle, and high aggressiveness groups. Respondents in the same group sha red similar bargaining attitude, intent ion , and competitiveness. The K means cluster analysis classifies individuals into a certain number ( ) of clusters by finding is equal to 3 in this study) centroids that minimize the distance between the points and the centroid within a cluster, s uch as: ( 3 1) where is a chosen distance measured between point and the cluster center 3 .3.3 RCE D esign The choice experiment was designed with four levels of pricing and three types of orange juice. 7 The average price levels of the orange juice products were chosen to in which each type of orange juice was treated as a factor and was varied at four p rice levels. The choice experiment was designed to maximize the D efficiency of the 7 The details of the attribute levels and an example of the RCE question are reported in Appendix A.
63 attribute matrix. In total each respondent was presented with ten choice scenarios, and each scenario had three different types of orange juice with certain price levels as well To start, respondents were given step by step instructions on the process of RCE. To ensure the elicitation mechanism was incentive compa tible, respondents were told that after finishing the survey, a binding shopping sce nario would be randomly determine d , and they should purchase the product they chose in that scenario. If they juice. 3 .3.4 RCVM D esign The double bounded dichotomous choice CVM was used because several studies pointed out that it can get more efficient estimates of WTP compared to the single bounded dichotomous choice CVM with the same number of survey respondents (Hanemann, Loomis, and K anninen 1991; Cameron and Quiggin 1991). 8 The challenge with designing an incentive compatible double bounded CVM is providing individuals with an incentive to reveal their true value. In a traditional hypothetical double bounded dichotomous choice CVM, th e respondents are first asked price. In the RCVM with the real purchasing re quirement, the best strategy for 8 The details of the price levels and an example of the RCVM question are reported in Appendix B.
64 product at the lowest price without revealing their true WTP. One way to prevent chance that they will figure out the price structure after the first or second products are offered. In this case, the conventional CVM design provides no incentive for respondents to reveal their true value of the product. a R espondents were told that they would draw a rand om price from a bowl to determine the market price at the end of the experiments. A nd they could only allowed to 9 was lower than the price they accepted in the RCVM. their true preference, and thus make the RCVM incentive compatible. The procedures in the RCVM were explicitly explained to the participants and were conducted accordin g to the following steps: First, the experiment structure was similar to the conventional double bounded CVM with two layers of questions: respondents were asked if they were willing to purchase a type of orange juice under a certain price level, then a fo llow up question was asked depending on their answer to the first question. Next, the respondents randomly drew a number to determine the type of orange juice as the binding product. ice. If the market price was higher than the accepted price for the binding orange juice, the 9
65 respondent did not purchase it; if the market price was lower than the accepted price, 3 .3.5 EA D esign The Becker, DeGroot, and Marschak (BDM) auction method (Becker, DeGroot, and Marschak 1964) was used in the experimental auction. The BDM is a widely used EA in field studies because it is easy for participants to understand, and easy to conduct as t he respondents can be randomly picked in front of a grocery store without being gathered together at the same time. The BDM auction was conducted a ccording to the following steps 10 : First, respondents wrote down the price they were willing to pay for each t ype of orange juice, using 0 for orange juice that they did not want to purchase. Next, respondents randomly drew a number to determine which type of orange juice was the nge juice. If the bid price for the binding orange juice was equal or higher than the if the bid price was lower than the , the respondent finished the EA w ithout purchasing any orange juice. 3 .4 Models and S pecifications All relevant independent va riables are reported in Table 3 2 . Demographic variables include age, gender, income, income square, and the number of children under age 18 in the household. Tast e Ranking represents the ranking that respondents 10 An example of an EA is shown in Appendix B.
66 gave for each type of orange juice. The remai ning three variables in Table 3 2 are the group dummies in terms of price bargaining preferences. 3 .4.1 WTP E stimation from RCE pr ocess from RCE data is the same as that in Section 2.4. , and both CL model and G MIXL model in the WTP space are used to analyze the data from RCE. A decision maker , is facing choice scenarios and each choice scenario has alternatives. The utility level of the th product on scenario for the th decision maker can be expressed as . ( 3 2) where is the deterministic component and is the rando m component, which is assumed to have an independent identical extreme value distribution. We a ssume a usual linear in parameter utility functional of , such that . ( 3 3) where represents the price of product in th choice scenario, and the vector represents all the non price juice attributes and interaction terms between these attributes and the demographic variables as well as price bargaining preferences group indicators . The decision maker chooses the alternative that provides the largest utility, i.e., the alternative is chosen if and only if , . Following McFadden (1974) the proba bility that decision maker chooses alternative on th choice scenario is ( 3 4)
67 The coefficient of an attribute in the C L model coefficient is specified as a constant term, all individuals are assumed to have homogenous preferences for this attribute. The probability of consumer choosing alternative is estimated by for ( 3 5) This is the basis for CL model and the likelihood function then can be expressed as ( 3 6) where if alternative is chosen by the th individual, and otherw ise. The details of G MIXL model in the WTP space has been illustrated in Section 2.4.2 and Section 2.4.3. 3 .4.2 WTP E stimation from RCVM Following Hanemann, Loomis, and Kanninen (1991), we let be the index for each respondent in the sample, , be a vector of determinants of WTP, and assume a linear function for the WTP equation, then the true WTP can be written as ( 3 7 ) where is a vector of parameters for . Vector includes all the attribute variables and other control variables. The last term is a random error assumed to be normally distributed with mean 0 and standard deviation . Let be the initial price, be the higher second price when the subj possible outcomes exist for each respondent: ( ) both
68 and such that ( ( 3 8 ) ( th ( ( The log likelihood function of the RCVM model is ( 3 9 ) We estimate and by maximizing the log likelihood function , and the WTP estimates for product is . 3 .4.3 WTP E stimation from EA bargaining behavior, a Tobit model (Tobin 1958) is usually used because auction bids are left censored at zero. The Tobit model is specified as follows: ( 3 10 ) The latent dependent variable for product is and the observed dependent variable for product is . X represents the demogra phic and other control variables desc ribed in Table 3 2 . Since respondents bid for three different types of orange juice, we can construct three Tobit models separately; however, the error terms of these three equations could be correlated with each other. The Seemingly Unrelated Regressions (SUR) (Zellner
69 1962) can estimate these three Tobit models simultaneously and improve the efficiency of the estimates. The SUR system involves observations on each of the dependent variables ( equations). A SUR constructed with several Tobit models is the called a Multivariate Tobit model, and its structure is: ( 3 11 ) where , , , and . Following Cornick, Cox, and Gould (1994), the likelihood function for obtaining the par ameter estimates of Equation (4 12 ) is the Multivariate normal density function : ( 3 12 ) This model assumes that the error term of each individual Tobit regression is correlated with each other with a variance covariance matrix After estimating the coefficients and , the pre dicted biding value of product could be obtained by . 3. 5 Results In total, 321 individuals agreed to participate in the experiments, of which 289 (90.3%) completed the experiments. Among these individuals, 76 completed the RCE, 1 06 completed the RCVM, and 107 completed the EA. The null hypothesis of equality of means for demographic variables among RCE, RCVM, and EA data cannot be rejected at any standard significance levels. The mean and standard deviation of each control variabl e are reporte d in the last column of Table 3 2 .
70 measurement is reported in Table 3 3 . The low aggressiveness group has the highest average total score, 24.5, and the high aggressivenes s group has the lowest average total score, 16.0. 3 .5.1 Regression Results from RCE data This section presents the results from the CL and G MIXL models estimated in a WTP space. The coefficients ar e reported in Table 3 4 . The estimates of the CL model are reported in t he first two columns of Table 3 4 . The WTP estimates for NFC and FCOJ are significantly positive, indicating that order of the WTP estimates is NFC > FCOJ > O taste ranking of an orange juice have significant impacts on their choices. Younger people are more likely to purchase NFC or FC orange juice. The income effect is a reverse U shape, as the coefficients of interaction term s with Income are positive, and the coefficients of interaction terms with Income Square are negative. An orange juice with a higher ranking (smaller number) has a higher chance of being chosen by respondents. The coefficients of all the interaction terms with the bargaining bargaining behavior does not affect their choices in RCE.
71 The study estimates a G MIXL II model where is fixed at zero, 11 using simulated maximum likelihood with 500 Halton draws. 12 The results are reported in the last two columns of Table 3 4 . This model significantly increases the value of the log likelihood function from 694.97 (in the CL model) to 591.58. To account for the heterogeneous preferences, the coef ficients of the main attributes ( N egative Price, NFC, FC, and OJD) are assumed to be normally distributed. 13 All the estimat ed standard deviations of the main attributes are significant, indicating preference heterogeneities. The estimate of parameter is significant, showing that the scale parameter is significantly different from one. The magnitudes and the signs of the WTP estimates in the G MIXL model are close to those in the CL model. Respondents preferred NFC the most, followed by FCOJ and OJD. For FCOJ and OJD, age, income, and taste ranking still significantly aggressiveness level affects the WTP estimates. This is consistent with the analysis at the beginning of this study, that is, in RCE, the experiment environment provides little have little impact on their choice decisions. 11 We have tried different model specifications. For the most generalized case without specifying the value of and It was found that the estimate of is insignificantly different from zero. To improve the efficiency of the estimation, we set . 12 See Train (2003) for details on simulated maximum likelihood and Halton draws. 13 It is theoretically plausible to specify parameters of interaction terms as random parameters; however, the estimate results of the G MIXL model with random interaction parameters showed that the estimate standard deviations of the interaction parameters were not statistically significant.
72 3 .5.2 Regression R esults from RCVM data All the coefficients estimated from the R CVM data are reported in Table 3 5 . The first three column s show the estimated coefficients for NFC, FCOJ, and OJD , respectively . Contrary to the findings with RCE, age has a significant positive impact on the probability t hat consumers will choose NFC at the 1% significance level, but has no impact on other orange juice products. Education level also significantly affects likelihood of acce pting NFC compared to those who do not. The acceptance of OJD has a significant gender difference in that OJD is more acceptable by females than by FCOJ and OJD: the higher the ranking (the smaller the number) is, the higher the chance the consumers would accept the product. RCVM. For NFC and OJD, the coefficient of the high aggressiveness group d ummy is significantly negative at the 10% significance level; however, no aggressiveness variable has significant impacts for FCOJ. This finding shows that the heterogeneous of orange juice. 3 .5.3 Auction Bids R esults The summary statistics of the auction bids are reported in Table 3 6 . Both mean and median auction bids for all three orange juice types present in the same order: NFC > FCOJ > OJD. Zero bids exist for all three products, indicating that the bidding value is left censored at zero.
73 The results of the Multivariate Tobit model are shown in Table 3 7 . Older respondents are willing to pay less for NFC and OJD. Male respondents bid higher for FCOJ than do female respo ndents. The higher the education level is, the lower the WTP value for FCOJ and OJD. Taste ranking significantly affects the predicted auction bids for NFC and OJD. bidding val ue for all three types of orange juice in EA. All the coefficients of High are significantly negative, implying that the high aggressiveness group bids significantly lower than the low aggressiveness group. For OJD, the middle aggressiveness group bids sig nificantly lower than the low aggressiveness group as well. This result is consistent with our expectation that because respondents are given relatively large bargaining power in EA, the bargaining preferences significantly affect the predicted auction bid s. 3 .5.4 Comparing WTP Values from Three E xperiments To calculate the WTP estimates for each type of orange juice and make them comparable across experiments, first, we simulate 1,000 samples, whose demographic characteristics have the same mean and standard deviation as the original sample. The WTP estimate s for a certain type of orange juice in certain experiments then is calculated by the simulated coefficients from each experiment, accordingly. The mean of these WTP estimates from all three experiments are reported in Table 3 8 , and the numbers in the squ are brackets are at the 95% confidence interval for these estimates. The last column of Table 3 8 also reports the WTP difference between the low and high aggressiveness groups.
74 WTP estimates from RCE are significantly higher than WTP estimates from EA fo r all types of orange juice. This is consistent with the results in Lusk and Schroeder (2006) and Gracia, Loureiro, and Nayga (2011) that some of the WTP estimates in RCE even doubled the WTP estimates in CE. We can easily reject Hypothesis 1 that these th ree non hypothetical experiments give us the same WTP estimates. However, these three non hypothetical experiments provide the same order that consumers prefer: NFC > FCOJ > OJD. When we compare WTP estimates across aggressiveness groups, for all three typ es of orange juice in RCE, WTP estimates from the high aggressiveness group are not significantly lower than that from the low aggressiveness group. In RCVM, the high aggressiveness group is predicted to pay more than 3 higher for NFC than the low aggressi veness group would pay, and this difference is statistically significant. But for FCOJ and OJD, the difference between the low and high aggressiveness groups is not significantly different from zero. On the other hand, in EA, the WTP estimates show a clear order across the low, middle, and high aggressiveness groups for all three types of orange juice. The predicted auction bids from the low aggressiveness group are always the largest, and the predicted auction bids from the high aggressiveness group are al ways the smallest . In addition, all the differences in the last column are significantly different from zero in EA. This result implies that bargaining preference is an important determinant in EA, where respondents could have relatively more bargaining po wer. To summarize, the comparison of the WTP estimate among aggressiveness groups confirms the expectation that the bargaining environments of EA, RCVM, and
75 RCE are significantly different, and more aggressive respondents would pay less than less aggressiv e respondents in experiments that give them more bargaining power. 3 .6 Concluding Remarks This essay compares the WTP estimates from three different incentive compatible experiments: RCE, RCVM, and EA and analyzes the behavior of individuals with price ba rgaining preferences in these experiments. By testing whether more aggressive respondents behave differently from less aggressive respondents, this study aims to illustrate that EA and RCVM provide some bargaining power to experimental participants, but RC E does not, and this leads to disparities of WTP estimates. We find that WTP estimates from RCE are the highest, followed by WTP estimates from RCVM, and the predicted auction bids from EA. After categorizing respondents into low, middle, and high aggressi veness groups, we find that heterogeneities in price bargaining preference lead to WTP estimates disparities in EA, some cases of disparities in RCVM, and no cases in RCE. This finding confirms our hypothesis that the mechanism designs of EA and RCVM provi de room for respondents to bargain on price; therefore, the WTP estimates from RCVM and EA are significantly lower than the WTP estimates from RCE. All of the estimated WTP difference in EA are huge in magnitude, which indicates that respondents behave dif ferently in EA by their price bargaining preferences. The results in this essay indicate that non hypothetical experiments are not WTP values, they need to consider differences in the bargaining environments provided by the experiment as well as the characteristics of the respondents, such as their preference in price bargaining. This
76 chapter has not, however, made a conclusion on wh ich non hypothetical experiment is the most accurate in terms of estimating true WTP values. Additional research is needed to explore the efficiency of different experiments so that the most accurate estimates of consumer preferences can be obtained. The results in this chapter also preferences or evaluating the benefit of a policy or a program using experiments, re searchers have to pay more attention on choosing appropriate experimental methodologies. Since differe nt experiments provide different bargaining environments, ing one experiment, e ven if the hypothetical bias can be controlled by non hypothetical methods , we may still obtain WTP estima tes that are significantly different from other alternative methods . Moreover, the three methods discussed in this chapter are used for market goods valuation, and sometimes for non market goods valuation. For example, RCVM is very popular in valuing non market goods. For non market goods, it is not easy for respondents forming their own price since this type of goods don't have a clear market value. The process of forming non market goods in their mind could be totally different from those market goods. T herefore, a possible extension of this chapter can be stud y ing whether our conclusion regarding the disparity in WTP between different methods and whether the role of bargaining environment in explaining the disparity still hold for non market goods evaluations. One more interesting implication is the different reactions towards these three types of orange juice in RCVM. To consumers in China, compare to other two types of
77 orange juice, NFC orange juice i s very new and they are rarely found in common grocery stores. We call this type of orange juice novel good, and those orange juice consumers are familiar with are non novel goods. In the RCVM, those respondents who are very aggressive in price bargaining are willing to pay significantly lower price for NFC orange juice than those who are not, but no significant difference were found in other non novel orange juice products. This is an indication that whether consumers are familiar with the product in the e xperiments may affect their behaviors. The next essay in the dissertation discusses more details about this effect.
78 Table 3 1 . Bargaining attitudes, intention, and competitiveness Bargaining Preference Measurements Panel 1 Bargaining Attitudes Indicate how you agree (or disagree) with the following statements as 1) strongly agree, 2) agree, 3) neutral, 4) disagree, or 5) strongly disagree. Attitude_1. Bargaining makes shopping pleasurable. Attitude_2. Bargaining makes my life interesting. Attitude_3. Sometimes it is not about money, if I can get discount by bargaining, I enjoy it and feel happy. Attitude_4. I feel comfortable when I bargain. Panel 2 Bargaining Intention Indicate how you agree (or disagree) with the following statements as 1) strongly agree, 2) agree, 3) neutral, 4) disagree, or 5) strongly disagree. Intention_1. Whenever I go shopping I try to bargain if bargaining is possible. Intention_2. I will bargain during my next shopping trip if bargaining is possible. Panel 3 Bargaining Competitiveness Suppose you want to buy the following products. If bargaining is possible, please indicate the price level that closest to your bargaining price: Competitiveness_1. A simple cotton T shirt, price 20 1) 10. 2) 13. 3) 15. 4) 17. 5) 20. Competitiveness_2. A regular desk computer with all the common features you need, price 4000 1) 3000. 2) 3300. 3) 3500. 4) 3700. 5) 4000.
79 Table 3 2 . Variable definitions on consumer specific characteristics Variable Unit Definition Mean (st. dev.) Age Year Age of a r espondents in years 35.81 (15.10) Gender Dummy 1, female 0, male 0.85 (0.36) Income Scale Monthly household income Scale from 1 as less than 1,000 to 12 as over 15,000 7.80 (2.50) Income 2 Scale Income square 67.01 (38.20) Education Dummy 1, if college or equivalent degree, and above 0, otherwise 0.81 (0.95) # of Child Persons How many children under 18 in the household 0.68 (0.72) Taste ranking Scale Rank this orange juice as 1, best; 2, middle; 3, worst NFC 2.00 (0.83) FC 1.89 (0.72) OJD 2.11 (0.88) Aggressiveness Indicator Low Dummy 1, belong to low aggressiveness group 0, not belong to low aggressiveness group 0.30 (0.46) Middle Dummy 1, belong to middle aggressiveness group 0, not belong to middle aggressiveness group 0.37 (0.48) High Dummy 1, belong to high aggressiveness group 0, not belong to high aggressiveness group 0.33 (0.47)
80 Table 3 3 . Statistical summary of aggressiveness measurement Pooling data Low Middle High Attitude_1 2.48 3.21 2.23 2.11 (0.91) a (0.90) (0.68) (0.75) Attitude_2 2.69 3.53 2.35 2.30 (0.88) (0.59) (0.65) (0.81) Attitude_3 2.53 3.22 2.23 2.47 (0.88) (0.85) (0.69) (0.75) Attitude_4 2.66 3.17 2.40 2.49 (0.88) (0.90) (0.78) (0.78) Intention_1 2.17 2.57 2.07 1.94 (0.82) (0.97) (0.68) (0.69) Intention_2 1.95 2.20 1.92 1.77 (0.62) (0.79) (0.50) (0.47) Competitiveness_1 2.62 3.23 2.82 1.86 (1.13) (1.21) (0.92) (0.82) Competitiveness_2 2.49 3.36 2.88 1.30 (1.25) (1.18) (0.83) (0.65) Total Score 19.59 24.50 18.88 16.02 (4.32) (3.02) (2.11) (2.85) Note a The numbers in the parentheses are standard errors calculated in the conventional manner.
81 Table 3 4 . Estimates of models for choice experiments Conditional Logit Model WTP Space G MIXL WTP Space Mean St. Err. Mean St. Err. NFC 28.58 *** 9.95 21.79 * 13.02 FCOJ 21.04 *** 6.43 19.75 ** 10.00 OJD 5.17 6.78 7.41 7.68 NFC : Age 0.17 ** 0.07 0.06 0.090 NFC : Income 4.78 ** 2.20 4.39 3.11 NFC : Income 2 0.39 *** 0.14 0.35 * 0.20 NFC : Gender 3.75 2.85 2.02 4.08 NFC : Number of Children 1.10 1.22 0.34 1.53 NFC : Education 1.85 2.05 2.86 3.15 NFC : Ranking a 4.01 *** 0.86 2.89 *** 0.78 NFC : Middle 0.69 2.17 2.66 2.96 NFC : High 2.42 2.15 4.61 3.20 FCOJ : Age 0.13 ** 0.06 0.130 * 0.07 FCOJ : Income 2.46 * 1.45 2.30 2.29 FCOJ : Income 2 0.24 ** 0.10 0.23 0.15 FCOJ : Gender 1.06 2.35 0.98 2.87 FCOJ : Number of Children 0.52 1.06 0.44 1.22 FCOJ : Education 0.57 1.74 1.29 2.59 FCOJ : Ranking 0.25 0.51 0.66 0.66 FCOJ : Middle 0.84 1.87 2.36 2.73 FCOJ : High 1.60 1.76 3.69 2.65 OJD : Age 0.09 0.06 0.10 * 0.06 OJD : Income 4.48 *** 1.62 3.78 ** 1.77 OJD : Income 2 0.35 *** 0.11 0.31 *** 0.12 OJD : Gender 2.12 2.53 3.36 2.34 OJD : Number of Children 0.82 1.85 0.37 1.09 OJD : Education 0.06 1.87 0.83 2.37 OJD : Ranking 2.83 *** 0.54 2.26 *** 0.62 OJD : Middle 1.16 1.95 2.43 2.67 OJD : High 0.15 1.81 1.98 2.46 Price (Negative) 1 (Constrained) 1 (Constrained) Standard Deviation Estimates s.d. of Price (Negative) 0.46 *** 0.05 s.d. of NFC 1.40 ** 1.11 s.d. of FCOJ 1.50 * 0.43 s.d. of OJD 1.61 *** 0.45 Other Parameter (Tau) 0.84 *** 0.17 Number of Obs. = 3040 = 3040
82 Table 3 4. Continued Conditional Logit Model WTP Space G MIXL WTP Space Wald Chi2 = 1095.01 = 358.64 Prob>Chi2 = 0.000 = 0.000 Log likelihood = 694.97 = 591.58 Note: a juice
83 Table 3 5 . Double bounded CVM estimation results NFC FCOJ OJD Age 0.20*** 0.00 0.01 (0.07) (0.03) (0.01) Gender 3.76 0.28 1.53*** (2.88) (1.16) (0.57) Income 1.08 0.55 0.43 (1.56) (0.59) (0.27) Income 2 0.06 0.03 0.03 (0.11) (0.04) (0.02) Education 6.41** 0.10 0.47 (2.62) (0.95) (0.43) # Child 1.26 0.02 0.29 (1.30) (0.52) (0.24) Ranking 1.08 1.24** 0.48** (1.21) (0.58) (0.21) Middle 2.82 0.64 0.13 (2.32) (0.91) (0.43) High 4.82* 1.33 0.73* (2.51) (0.94) (0.44) Constant 11.51 8.99*** 1.82 (8.04) (3.05) (1.22) 8.12*** 3.31*** 1.36*** (1.27) (0.43) (0.25) Log likelihood 116.97 134.46 110.24 Number of obs. 106 106 106 Note : The numbers in the parentheses are standard errors calculated in the conventional manner.
84 Table 3 6 . Summary statistics for the bids Alternatives Values Mean NFC 11.85 FCOJ 8.09 OJD 3.56 Median NFC 10 FCOJ 8 OJD 3 Standard deviation NFC 7.40 FCOJ 3.70 OJD 1.83 Percentage of NFC 7.5% zero bid FCOJ 1.9% OJD 6.5%
85 Table 3 7 . Multivariate Tobit model results from EA data NFC FCOJ OJD Age 0.05* 0.01 0.04*** (0.03) a (0.01) (0.01) Gender 0.32 1.53*** 0.05 (0.98) (0.48) (0.24) Income 0.25 0.16 0.06 (0.94) (0.46) (0.24) Income 2 0.03 0.08 0.01 (0.06) (0.03) (0.02) Education 0.94 0.85* 0.64*** (0.90) (0.44) (0.22) # Child 0.40 0.23 0.19 (0.65) (0.31) (0.15) Ranking 3.13*** 0.05 0.48*** (0.45) (0.23) (0.12) Middle 0.81 0.82 0.66** (1.06) (0.52) (0.27) High 1.96** 1.68*** 1.09*** (1.05) (0.52) (0.26) Constant 16.58*** 8.62*** 6.79*** Sigma1 7.00*** Rho12 0.53*** (0.29) (0.04) Sigma2 3.46*** Rho13 0.09 (0.14) (0.06) Sigma3 1.74*** Rho23 0.10* (0.07) (0.06) Number of observation = 321 Log likelihood = 2442.92 Wald chi2(21) = 197.42 Prob. > chi2 = 0.00 Note a The numbers in the parentheses are standard errors calculated in the conventional manner.
86 Table 3 8 . WTP values and auction bids grouped by aggressiveness level in price bargaining Note: a b c d Low High e Numbers in the square brackets are the 95% confidence int erval for the WTP estimates and the WTP difference estimates. WTP and WTP Differences Panel 1 Real Choice Experiment Pooling data Low a Middle b High c Low High d NFC 25.53 28.05 25.33 23.21 4.85 [22.42, 28.64] e [22.86, 33.25] [19.77, 30.88] [17.76, 28.65] [ 2.66, 12.35] FCOJ 18.67 20.73 16.97 18.32 2.41 [16.88, 20.47] [17.86, 23.60] [13.67, 20.28] [15.17, 21.46] [ 1.84, 6.66] OJD 10.90 11.86 7.79 13.06 1.19 [9.31, 12.49] [9.01, 14.72] [5.14, 10.43] [10.31, 15.81] [ 5.15, 2.76] Panel 2 Real Contingent Valuation Method Pooling data Low Middle High Low High NFC 20.03 21.42 20.46 18.22 3.19 [18.71, 21.35] [19.20, 23.62] [18.10, 22.82] [15.95, 20.50] [0.02, 6.36] FCOJ 8.88 8.69 9.40 8.54 0.16 [8.38, 9.37] [7.88, 9.51] [8.53, 10.27] [7.63, 9.44] [ 1.06, 1.37] OJD 3.13 3.15 3.50 2.75 0.40 [2.90, 3.367] [2.76, 3.55] [3.10, 3.90] [2.32, 3.18] [ 0.17, 0.97] Panel 3 Experimental Auction Pooling data Low Middle High Low High NFC 11.28 12.21 12.07 9.58 2.63 [10.60, 11.97] [11.02, 13.39] [10.88, 13.25] [8.38, 10.77] [0.95, 4.31] FCOJ 7.47 8.35 7.27 6.79 1.56 [7.13, 7.87] [7.78, 8.92] [6.69, 7.85] [6.16, 7.42] [0.71, 2.40] OJD 3.59 4.27 3.59 2.92 1.35 [3.42, 3.77] [3.99, 4.55] [3.30, 3.88] [2.60, 3.23] [0.93, 1.77]
87 CHAPTER 4 PURCHASE INTENTION EFFECTS IN CHOICE EXPERIMENTS AND EXPERIMENTAL AUCTIONS 4.1 Purchase Intention Effects Chapter 3 provides one possible explanation why RCE, RCVM, and EA would environments that each experiments would give respondent different level of bargaining power, therefore respondents with d ifferent price bargaining preferences would behave differently in those experiments. Purchase intention can be another potential factor that results in WTP discrepancies among non hypothetical experiments. In this dissertation, purchase intention is define d as a plan to purchase the subjects in the experiment that day 1 . In subjects in the experiments before they participant in the experiments would affect their decision in the experiments, and whether this purchase intention effect would be different in the RCE and the EA. Purchase intention effects have been examined in previous literature, and it has the EA ( Lusk and Fox 2003 ; Shi, House, and Gao 2013 ) . Examining purchase intention effects in laboratory and field valuation experiments. Lusk and Fox (2003) have compared results from laboratory and field valuation experiments and found that field valuations 1 future. In this dissertation, however, t he study only focuses on whether respondents have plan to purchase the subjects before they participant in the experiment that day.
88 were greater than laboratory valuations. One possible reason is the purchase intention effects. Respondents who are recruited in the grocery store may have hig her purchase intention for the subjects since they have planned to purchase it compared to those respondents in the lab experiments . T herefore the estimated WTP value is higher if the experiment is conducted in the grocery store s . However , purchase intent ion may have different effect s in EA and RCE because the experimental process of these two experiments are so different. Most purchase intention effect studies are conducted in the EA but not in the RCE. The experimental settings in the RCE simulate the re al shopping scenarios and present multiple choices to the respondents with very similar products. People in the RCE would focus on the multi attributes of the subjects in the experiment, therefore purchase intention m ay play a different role in this experi mental setting comparing to EA . A better understanding on purchase intention effects in both experiments may help us determine whether one method is more appropriate to use to estimate WTP than the other one. Price attributes plays quite different roles i n RCE and EA when eliciting , the bidding process are draw s full attention to the price. In the choice experiment, however, participants are facing the choices with price and non price att ributes at the same time . As a result, in EA consumers are focusing on price levels and more likely to bid or choose a lower price while participants in RCE are more likely to consider all the attribute levels at the same time and accept a higher price cho ice . Compared to EA, the price attribute is no longer the main focus but is part of a group along with other non
89 T herefore c onsumers with less purchase intention may behavior differently compared to consumers with strong purchase intention in EA rather than that in RCE. In addition, whether the target product in the experiment is novel good or not would also matter. Purchase i ntention may have more impact on the valuation of non novel good than on novel good. The non novel good may be the exact product that a consumer plans to buy, therefor a consumer with high purchase intention is more likely to bid more than those with low p urchase intention who may try to avoid buying the product in the experiment. However, the novel good is not something a consumer usually buy in the market, therefor a consumer with high purchase intention may not have more desire than those with low purcha se intention to buy the product in the experiment to satisfy their shopping plan. Studying the impact of purchase intention on novel and non novel products is critical because it help determine whether the valuation of a product is susceptible to the novit y of a product. In this study, most Chinese consumers are not familiar with the NFC orange juice . E ven if consumers may have purchase intention to buy a bottle of orange juice that day, their purchase intention may not be a strong determinate on how muc h they are willing to pay for this new orange juice The third essay of this dissertation uses the same samples in the second essay, but provide s another possible explanation why non hypothetical experiments provide inconsistent estimations. Since the WTP gaps has been confirmed in Chapter 3, and RCE and EA provide the largest discrepancies amo ng all types of orange juices, t he third essay will only compare the purchase intention effects in RCE and EA .
90 4.2 Literature Review Chapter 3 has reviewed that Lu sk and Schroeder (2006) and Gracia, Loureiro, and Nayga (2011) have found disparities between the EA and RCE that the auction bids were lower than the choice prices, and the demand elasticites calculated from each experiment were also inconsistent with eac h other. T he results in the second essay suggest that bargaining environments can be one of the reason. There could be more reasons that result in the WTP gaps among different non hypothetical experiments. The process of determining how much consumers want to structed contingent on different spots or choice contexts, or the circumstances in which he/she was going to pay for it (Tversky and Simonson, 1993; Bettman et al.; Horowitz 2006). Therefore a decision s , and their decisions can also change as the circumstance changes. One important element that some literature have already noticed is the purchase intention of respondents in experiments . Corrigan and Rousu (2008) find that purchase intention significantly affects the estimated WTP if the products are perishable. That is, those respondents who intend to purchase that product submitted their bids equal to their perceived market value of th at product, but those who do not intend to purchase it submitted lower bids compared to their perceived market value. Using BDM auctions Shi, House, and Gao (2013) find that purchase intention only has effects on full bids, WTP for cert ain product, but no effects on the inferred partial bids, which is the partial WTP for certain food labels compares to other labels. However, most of the purchase intention effect studies were conducted in the EA, not in the RCE.
91 Moreover, different type o f goods may have different purchase intention effects. Payne, Bettman, and Johnson (1992) and Irwin et al. (1993) show that when it comes to novel goods that respondents are not familiar with or environmental goods that have not being valued, it is hard fo r people to have an exact monetary value for them. Moreover, List and Shogren (1999) find that when a good is novel to subjects, bidder affiliation could exist in the EA. Bidder a ffiliation is the scenario that bidders in the EA may affiliate their own bid Therefore, for novel goods, purchase inte Since respondents have shopping experiences for non novel goods but little for novel goods, the process of forming t he value of them could be different, and the purchase intention effects could be different as well. 4.3 Experimental Design This essay is using the same RCE and EA samples from the second essay . Therefore, the subject products were three type s of orange juice, NFC , FCOJ, and OJD, and each respondent participated one and only one experiment. Each respondent received 20 to compensate their time, and they were told that real purchase would happen at the end of the experiment so their best strategy is to re veal their true WTP value for each product in each question. This essay focuses on one specific independent variable when analyzing the purchase intention of the respondents. During the face to face interview, respo ndents were also asked about whether they were intend to purchase orange juice when they came to the grocery stores, and they can choose (see Figure 4 1).
92 To identify novel and non novel products, respondents were directly asked about if they were familiar with each orange juice product, and have they purchased each one of them. Question examples are shown in Table 4 1 and Table 4 2. 4.4 Models and Specifications 4.4.1 M NL Model for RCE Data RCE is based on random utility theory (Hanemann 1984; Hanley et al. 1998; Hanley, Wright, and Adamowicz 1998). To determine the purchase intention effects for each type of orange juice from the RCE in a simplest way , we use MNL model . The utility level of the th product for the th respondent can be written as: . (4 1) where is the deterministic and is the stochastic portion of utility, is the pr ice in the choice set, is the intrinsic preference of respondent that captures all the non price attributes of product , is the marginal utility of price, and vector is the type of orange juice in this choice , and is the vector of all the other control variables such as demographic variables including gender, income, education level, and number of children in the household, high aggressiveness), and their purchase intention dumm y variable. In this analysis, we WTP value is not marginal WTP but total WTP for each product. Under the CL model assumption that is iid with an extreme value di stribution, the probability of consumer choosing alternative is estimated by the CL model: for (4 2 ) (4 3 )
93 where if alternative is chosen by the th individual, and otherwise. versus the b ase , consuming nothing , can be calculated as the negative value of the ratio of the coefficient of all the to the price coefficient. 4.4.2 Multivariate Tobit Model for EA Data T he same as in the second essay, purchase intention effect in EA is estimated by Multivariate Tobit Model. purchase intention, a Tobit model is usually used because auction bids are left censored at zero. The Tobit model is specified as follows: (4 4 ) The latent dependent variable for product is and the observed dependent variable for product is . X is same the control vector as T in Equation (4 1) . Same as in Chapter 3, s ince respondents bid s for three different types of orange juice, the error terms of these three equations can be correlated with each other. Therefore, using SUR model to estimate these three Tobit model s simultaneously could improve t he e fficiency of the estimates. W hen SUR is constructed with several Tob it model s , it is called Multivariate Tobit Model. This model is the same as Equation (3 11) in Chapter 3 except now we include purchase intention dummy in the control vector X : ( 4 5 )
94 where , , , and 4.5 Results This essay uses the same RCE and EA data as these in Chapter 3 with 76 valid sample in the RCE and 107 valid sample in the EA. The null hypothesis of equality of means for demographic variables in the RCE and EA data cannot be rejected at any standard significance levels as shown in Table 3 2 . Respondents purc hase intention is shown in Table 4 3 . For respondents participated in RCE, about 24% of them had plan to purchase orange juice that day in the grocery stores. About 17% participant s participated in EA claimed they planned to purchase orange juice on their trip to the grocery stores . NFC ora nge juice was fairly new to these participant s as most of them claimed that they were not familiar with it , nor had purchased it before. Table 4 4 reports the percentage of respondents who have purchased each type of orange juice in each experiment, and Table 4 5 summarizes how respondents were familiar with each type of orange juice. From these two tables , only 14% respondents in the RCE have pur chased NFC orange juice, and about 95% of them said that they were unfamiliar or very unf amiliar with NFC orange juice. And i n the EA, only about 20% respondents said that have purchased NFC orange juice before , and most of them (93%) claimed they were not familiar with it. Compared to NFC orange juice, respondents were more familiar with FCOJ and OJD and most of them have purchased both types of orange juice in both experiments. Therefore NFC orange juice can be tre ated as a novel good and FCOJ and OJD can be treated as non novel goods . Besides purchase intention, all the other control variables in Chapter 3 are controlled in the analysis, such as age, income, gender, number of children under age
95 18 in the family, e ducation, their ranking for each orange juice product, and their price bargaining preference (middle and high) 2 . 4.5 .1 Results of MNL R egression in the R CE In RCE, MNL price bargaining, we can control dem ographic s and other determinants choices. The estima te results are presented Table 4 6 . the coef ficients of product dummies NFC and FCOJ are significa ntly positive, means when other conditions are equal , consumers would prefer consuming these two orange juice rather than nothing. NFC has the highest coefficient, followed by FCOJ, and then OJD, indicating that consumers would willing to pay more for NFC, compared to FCOJ and OJD. Price has negative coefficient, means the higher the price, the less probability that consumers would choose that option. This is consistent with the estimated results in Chapter 3. Demographic variables have similar effects as in Chapter 3. Younger people with higher income were more likely to buy NFC orange juice, and most demographic the other two types of orange juice. Also from Table 4 6 we can see that most demographic variables have impacts in RCE similar to these in EA except income. Ranking of the products 2 All the controlled variables (besides purchase intention dummy) are reported in Chapter 3 Table 3 2 .
96 such that the highe r ranking (lower number) of a type of orange juice, the more likely it will be ch ose n . limited impact in RCE, and the results rep orted in Table 4 6 also confirm this proposition. Only for NFC orange juice, respondents in the middle aggressiveness group would want to pay lower price comparing to those respondents in low aggressiveness group. All the coefficients of other price bargaining preference variables are all insignificant, in dicating that respondents did not behave differently by their price bargaining preferen ces. Most importantly, the results from this MNL regression suggest that there is no significant purchase intention effects in the RCE for non novel goods . For novel goods, NFC orange juice, however, purchase intention has significantly negative effect on This confirms the analysis in the introduction of this chapter that in the choice experiment, as respondents are focusing on choosing one product they prefer most and the price is not the focus in the experiment, their purchas e plan before they participate in the experiments has little impact on their choices for non novel goods . However, this purchase intention significantly reduced their WTP value for non novel goods. That is, for those respondents who has purchase intention to buy have purchase plan that day. The finding is a bid surprising that in the choice experiments, purchase intention actually have significant impacts on novel goods. Possible explanation could be that, respondents with purchase intention have clear plan to buy their familiar products, thus compared to people without purchase intention, they
97 are more likely to avoid choosing the NFC orange option so that they have more chance to get the FOCJ or OJD as they have planned . 4.5 .2 Results of Multivariate T obit R egression in the EA The results of Multivariate Tobit model in Table 4 7 sh ow that younger respondents are tend to bid more on NFC and OJD, females bid more on FCOJ than males, and people with less education levels bid more on FCOJ and OJF but not on NFC orange juice. The same effect of Ranking has been found here as the higher the ranking, the more people are willing to pay for NFC and OJD. Moreover, the significance of the coefficients of price bargaining preference s indicates that in the EA, respondents behavior could be affect by their preference in pric e bargaining, and those who are very aggressive are willing to pay less than those who are not. Most importantl y, purchase intention has differently effects in the EA. For novel good , NFC, the coefficient of purchase intention is insignificant, indicating that purchase intention has little impact on the bids for NFC, in the EA. For FCOJ and OJD, however, purchase i ntention has interesting oppos it e impacts: the coefficient of purchase intention for FCOJ is significantly negative, but it is significantly positive for OJD. These the experiment could affect their bidding behavior. One possible explanation to the opposite signs could be, when respondents have plan to purchase orange juice, in their mind they may already determine to purchase the most familiar orange juice, OJD, instead of FCOJ. Th erefore those people who have purchase plan would bid more on OJD and less on FCOJ compared to those who do no t have plan . Re sults in S ection 4.5. 2 and 4.5.3 show that purchase intention effects are different in the RCE and EA. This can be a potential explanation that WTP estimates
98 from the RCE is significantly higher than the WTP estimates from the EA. For respondents in the EA, especially those wh o d id not have plan to purchase orange juice that day, since the non hypothetical experiments would result in purchasing a bottle of orange juice, they would want to bid a very low price to reduce the probability of wining the bid, therefore the predicted bids in the EA is lower than the estimated WTP in the RCE . 4.5.3 Endowment E ffect Because the market price of NFC orange juice price is higher than 20, the limited compensation biding value. To test if money compensation has impacts on respondents biding values, EA with compensation 10 was also conducted to see if different level of compensation will affect their bidding values . Basic bidding value information in both types of EA is reported in Table 4 8 . The average bid s of respondents with 10 compensation for NFC orange juice is around 13 , 8 for FCOJ , and 3.3 for OJD (second column of Table 4 8) , which are very close to the average bids when respondents got 20 compensation (first column of Table 4 8) . The percentage of zero bids in low compensation group is a little higher, but when comparing the percentage of bids above 10 in both groups , the group with less compensation even has higher percentage of people bid above it. This simple comparison with different level s of compensation sug gests that endowment effect has little impact on res pondents bidding behavior in our EA.
99 4.6 Conclusion Remarks This essay focuses on the purchase intention effect in these two experiments wit h one novel food and two non novel food. Results show that purchase intention has impact on non novel food such as FCOJ and OJD that consumers are familiar with, but not on novel (new) products like NFC for Chinese consumers in the EA. H owever, opposite im pacts have been found in the RCE . This chapter presents another possible explanation why different non hypothetical experiments provide inconsistent estimated WTP values. By including purchase intention in the analysis regressions, the coefficients of purchase intention in the EA are significant for FCOJ and OJD, but not for NFC orange juice, indicating that respondents with no pla n to purchase orange juice that day would bid significantly lower than those who have purchase intention that day. Interestingly, in the RCE, there is no significant purchase intention effect for these two juice products , but significantly negative impact on NFC orange juice . R esult s in this chapter suggest that researcher should be very careful when choosing which experiment to evaluate consumers WTP values , and should also especially when they are using . be the only two reasons that non hypothetical experiments provide different estimating results. Other personality difference such as these s ix indicators in Grebitus, Lusk, and Nayga (2013) can also affect the estimating results. Therefore, the results of this chapter , along with other literature that studies experimental differences, suggest that preferences s hould not simply ask questions such
100 as how much consumers want to pay, but also need to consider the personalities of the respondents.
101 Table 4 1 . How familiar with different types of orange juice 1) 2) 3) Table 4 2 . Purchase experience 1) Yes 2) No 1) Yes 2) No 1) Yes 2) No
102 Table 4 3 . Summary statistics for the answer to purchase intention question Purchase Intention RCE RCE Frequency Percent Frequency Percent Yes 18 23.68% 18 16.82% No 44 57.89% 68 63.55% Not sure 14 18.42% 21 19.63% Total 76 100% 107 100% Table 4 4 . Summary statistics for the answer to purchase experience Real Choice Experiment NFC FCOJ OJD Yes 14.47% 77.63% 98.68% No 85.53% 22.37% 1.32% Not sure 0% 0% 0% Experimental Auction NFC FCOJ OJD Yes 19.63% 76.64% 99.07% No 74.77% 18.69% 0.93% Not sure 5.61% 4.67% 0% Table 4 5 . Summary statistics for how familiar with orange juice Real Choice Experiment NFC FCOJ OJD Very Unfamiliar 44.74% 3.95% 0% Unfamiliar 51.32% 28.95% 3.95% Familiar 2.63% 65.79% 67.11% Very familiar 1.32% 1.32% 28.95% Experimental Auction NFC% FCOJ OJD Very Unfamiliar 39.25% 4.67% 0.94% Unfamiliar 54.20% 32.71% 4.67% Familiar 6.54% 55.14% 56.08% Very familiar 0% 7.48% 38.32%
103 Table 4 6 . MNL estimates for choice experiments Mean St. Err. Price 0.25 *** 0.02 NFC 7.02 *** 2.50 FCOJ 6.67 *** 1.73 OJD 1.12 1.77 NFC : Age 0.04 ** 0.02 NFC : Income 1.43 ** 0.57 NFC : Income 2 0.11 *** 0.04 NFC : Gender 1.02 0.73 NFC : Number of Children 0.39 0.32 NFC : Education 0.47 0.38 NFC : Ranking 1.09 0. 63 NFC : Intention 1.11 ** 0.52 NFC : Middle 1.09 *** 0.21 NFC : High 0.27 0.56 FCOJ : Age 0.63 0.56 FCOJ : Income 0.04 *** 0.01 FCOJ : Income 2 0.64 0.40 FCOJ : Gender 0.07 ** 0.03 FCOJ : Number of Children 0.40 0.62 FCOJ : Education 0.11 0.28 FCOJ : Ranking 0.01 0.16 FCOJ : Intention 0.41 0.43 FCOJ : Middle 0.21 0.49 FCOJ : High 0.47 0.47 OJD : Age 0.02 0.01 OJD : Income 1.36 *** 0.42 OJD : Income 2 0.10 *** 0.03 OJD : Gender 0.82 0.66 OJD : Number of Children 0.22 0.29 OJD : Ranking 1.02 *** 0.14 OJD : Intention 0.17 0.44 OJD : Middle 0.40 0.51 OJD : High 0.23 0.48 Number of Obs. = 3040 LR Chi2 (33) = 907.88 Prob>Chi2 = 0.00 Log likelihood = 1108.47
104 Tab le 4 7 . Multivariate T obit model estimates for experimental auctions NFC FCOJ OJD Age 0.05* 0.01 0.04*** (0.03 ) a (0.01) (0.01) Gender 0.24 1.55 *** 0.03 (0.93 ) (0.47 ) (0.24 ) Income 0.24 0.12 0.07 (0.93 ) (0.46) (0.23 ) Income 2 0.03 0.02 0.01 (0.06) (0.03) (0.01 ) Education 0.96 0.94 * 0.58 *** (0.90) (0.44) (0.22) # Child 0.41 0.12 0.27* (0.63 ) (0.31) (0.15) Ranking 3.16 *** 0.03 0.45 *** (0.44 ) (0.22 ) (0.12) Intention 0.14 1.14* 0.76*** (1.09) (0.55) (0.26) Middle 0.81 0.80 0.67 ** * (1.05 ) (0.52) (0.25 ) High 1.98 ** 1.79 *** 1.02 *** (1.04 ) (0.51 ) (0.25 ) Constant 16.72 *** 8. 91 *** 6.63 *** Sigma1 1.95 *** Rho12 0.57 *** (0.04 ) (0.06 ) Sigma2 1.24 *** Rho13 0.10 (0. 41 ) (0.06) Sigma3 0.54 *** Rho23 0.12 * * (0.41 ) (0.06) Number of observation = 321 Log likelihood = 2435.32 Wald chi2(30 ) = 216.63 Prob. > chi2 = 0.00 Note a The numbers in the parentheses are standard errors calculated in the conventional manner.
105 Table 4 8 . Endowment e ffect 20 compensation 10 compensation Alternatives Values Values Mean NFC 11.85 12.88 FCOJ 8.09 7.83 OJD 3.56 2.90 Median NFC 10 10 FCOJ 8 7 OJD 3 3 Standard deviation NFC 7.40 9.27 FCOJ 3.70 5 OJD 1.83 1.76 Percentage of NFC 7.5% 7.7% zero bid FCOJ 1.9% 7.7% OJD 6.5% 15.4% Percentage of Bids above 10 NFC FCOJ OJD 42% 12% 0% 50% 23% 0% Sample size 107 26
106 Figure 4 1 . Purchase intention question .
107 CHAPTER 5 CONCLUSION make their production plan accordingly to maximize their profit, and food policy makers can predictions of WTP value are so critical. for different food labels overlooked the interactive effects between diff erent labels. In t he first essay , attempts have been made to understand the interactive effects betw een USDA organic label and COOL . The interactive effects have been determined and t he results suggest that respondents would prefer domestically produced USDA organic broccolis to imported USDA organic broccolis . When estimating the interactive effect between USDA organic label and COOL, information treatment about the Equal Organic Stan dard Rule was introduced to see whether respondents would have different attitudes towards imported food carrying the same USDA organic label. Results show that only on certain cases, respondents who have the information would willing to pay more than thos e who don't. The second essay of this dissertation tries to answer why there are huge gaps among three commonly used non hypothetical experiments, RCE, RCVM, and EA. This essay focuses on the process of each experiment, and argues that because the bargain ing environments are different and some experiment provide more bargaining power to respondents than others, people may behave differently in these experiments.
108 This essay uses that becaus e different experiment provides different bargainin g environment, respondents who are very aggressive in price bargaining would reveal their WTP significantly lower than those who are not when they have a lot of bargaining power; but this WTP gap can be ve ry essay has confirmed this hypothesis and suggests that WTP gaps among valuation methods can come from the differences of barg aining environments . This is, however, not the only r eason for the estimat ed WTP discrepancies. The third essay of this dissertation extends the method comparison study and suggests that purchase intention is ano ther possible explanation of the estimated WTP gaps. Purchase intention has been studied in EA but not in RCE in prev ious literature s . Because the process of the RCE and EA are differently, purchase intention may have differently impacts in these two methods, and the differently impact may result in estimated WTP gaps. Using the same data from the second essay, the third essay shows that purchase intention has significantly impact in EA for non novel products, but no effects on novel product; however, in the RCE, purchase intention only has significant impacts on novel product. Both Chapter 3 and Chapter 4 provide eviden ce that the results of non purchase intention. The finding of the significant different impacts of these elements emphasizes the importance of being cautious when conducting experiments to evaluate hypothetical experiments
109 can provide consistent orde estimations. More factors should be considered in the experiments, especially the personality of the respondents, the circumstance of their purchase plan that day, etc. Combining with the finding s in the first essay, this dissertation studies some important features that researchers should be consider in their analysis of preferences and WTP. It contributes to the literature by providing potential explanations on why different non hypot hetical experiments give us inconsistent estimations on
110 APPENDIX A INFORMATION PROVIDED TO RESPONDENTS Information about three types of orange juice or organic drink: 1) Not From Concentrate (NFC) orange juice: orange juice processed an d pasteurized by flash heating immediately after squeezing the fruit without removing the water content from the juice. No additional water or other ingredients are added in 100% NFC orange juice. There are only a few NFC orange juice in the Chinese market , such as Paisengbai NFC and some imported brands such as NFC orange juice from Florida and Australia. At the time of this study, the price of 250ml of 100% NFC orange juice ranges from 5 to 12. 2) Frozen Concentrated Orange Juice (FCOJ): orange juice ob tained from concentrated orange juice (COJ) that is reconstituted with water. FCOJ is orange juice made by removing, through evaporation, the water from the orange juice of fresh, ripe oranges that have been squeezed in extraction machines. No other ingred ients are added in 100% FCOJ except for the same amount of water that was evaporated. So far, FCOJ has the largest market share in China. For example, Huiyuan 100% FCOJ, Farmer's Orchard 100% FCOJ, and Great Lake 100% FCOJ are very common in the market. Th e price for 450ml 100% FCOJ ranges from 4 to 8. 3) Orange Juice drink (OJD): a sweetened beverage that is made of diluted fruit juice containing no less than 10% orange juice with other ingredients added (such as sweetener). OJD is also very popular in t he orange juice drink market. You can find OJD in the market very easily. Minute Maid, Uni President, and Master Kong are the
111 common brands that carry orange juice drinks in China. The price for 450ml OJD ranges from 1 to 5.
112 APPENDIX B INFORMATION IN EXPERIMENTAL DESIGN B 1. Attribute levels for RCE Product attribute Price attribute NFC, 500ml 17, 21, 25, and 29 FCOJ, 500ml 6, 8, 10, and 12 OJD, 500ml 2, 2.5, 3, and 3.5 B 2. Price levels for RCVM Product Price NFC Starting price: 23, low: 17, high: 29 FCOJ Starting price: 9, low: 6, high: 12 OJD Starting price 2.5, low: 2, high: 3.5 B 3. Examples of RCE, RCVM, and EA questions RCE A. A bottle of 500ml 100% NFC, 21 B. A bottle of 500ml 100% FCOJ, 8 C. A bottle of 500ml 10% OJD, 3 D. None of them RCVM A. Y es B. No 12 for it? A. Yes B. No 6 for it? A. Yes B. No EA For the following product, please indicate the most you are willing to pay. A bottle of 500ml 100% NFCOJ ________
113 LIST OF REFERENCES Agrawal, J., & Kamakura, W. A. (1999). Country of origin: A competitive advantage? International Journal of Research in Marketing, 16 (4), 255 267. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior. Aprile, M. C., Cap uto, V., & Nayga Jr, R. M. (2012). Consumers' valuation of food quality labels: The case of the European geographic indication and organic farming labels. International Journal of Consumer Studies, 36 (2), 158 165. Batte, M. T., Hooker, N. H., Haab, T. C., & Beaverson, J. (2007). Putting their money where their mouths are: Consumer willingness to pay for multi ingredient, processed organic food products. Food Policy, 32 (2), 145 159. Becker, G. M., DeGroot, M. H., & Marschak, J. (1964). Measuring utility by a single response sequential method. Behavioral Science, 9 (3), 226 232. Bettman, J.R., Luce, M. F. & Payne, J.W. (1998) . Construct ive consumer choice processes. The Journal of Consumer Research , 25, 187 217. Bilkey, W. J., & Nes, E. (1982). Country of origin effects on product evaluations. Journal of International Business Studies, 13 (1), 89 100. Bockstael, N. E., Hanemann, W. M., & Kling, C. L. (1987). Estimating the value of water quality improvements in a recreational demand framework. Water Resources Rese arch, 23 (5), 951 960. Briggeman, B.C., & Lusk, J.L. (2011). Preference for fairness and equality in the food system. European Review of Agricultural Economics 38 (1), 1 29. Brooks, K., & Lusk, J. L. (2010). Stated and revealed preferences for organic and c loned milk: Combining choice experiment and scanner data. American Journal of Agricultural Economics, 92 (4), 1229 1241. Browne, A. W., Harris, P. J., Hofny Collins, A. H., Pasiecznik, N., & Wallace, R. R. (2000). Organic production and ethical trade: Defin ition, practice and links. Food Policy, 25 (1), 69 89. Brucks, M., & Schurr, P. H. (1990). The effects of bargainable attributes and attribute range knowledge on consumer choice processes. Journal of Consumer Research, , 409 419. Buchan, N. R., Croson , R. T., & Johnson, E. J. (2004). When do fair beliefs influence bargaining behavior? E xperimental bargaining in japan and the united states. Journal of Consumer Research, 31 (1), 181 190.
114 Cameron, T. A., Poe, G. L., Ethier, R. G., & Schulze, W. D. (2002). Alternative non market value elicitation methods: Are the underlying preferences the same? Journal of Environmental Economics and Management, 44 (3), 391 425. Campbell, N. C., Graham, J. L., Jolibert, A., & Meissner, H. G. (19 88). Marketing negotiations in France, Germany, the United K ingdom, and th e United S tates. The Journal of Marketing, 52 , 49 62. Champ, P. A., & Bishop, R. C. (2001). Donation payment mechanisms and contingent valuation: An empirical study of hypothetical bias. Environmental and Resource Economics, 19 (4), 383 402. Cornick, J., Cox, T. L., & Gould, B. W. (1994). Fluid milk purchases: A multivariate tobit analysis. American Journal of Agricultural Economics, 76 (1), 74 82. Corrigan, J. R. & Rousu, M.C. (2008). Testing whether field auction experiments are demand revealing in practice. Journal of Agricultural and Resource Economics , 33, 290 301. Davies, A., Titterington, A. J., & Cochrane, C. (1995). Who buys organic food?: A profile of the purchase rs of organic food in northern I reland. British Food Journal, 97 (10), 17 23. Dentoni, D., Tonsor, G. T., Calantone, R. J., & Peterson, H. C. (2009). The direct and indirect effects of 'locally grown' on consumers' attitudes towards agri food products. Agricultural & Resource Economics Review, 3 8 (3), 384. Durham, C. A. (2007). The impact of environmental and health motivations on the organic share of produce purchases. Agricultural and Resource Economics Review, 36 (2), 304. Ehmke, M. D., Lusk, J. L., & Tyner, W. (2008). Measuring the relative imp ortance of prefer ences for country of origin in China, France, Niger, and the United S tates. Agricultural Economics, 38 (3), 277 285. Erickson, G. M., Johansson, J. K., & Chao, P. (1984). Image variables in multi attribute product evaluations: Country of or igin effects. Journal of Consumer Research, 11 , 694 699. Evans, K. R., & Beltramini, R. F. (1987). A theoretical model of consumer negotiated pricing: An orientation perspective. Journal of Marketing, 51 (2) Fiebig, D. G., Keane, M. P., Louviere, J., & Wasi, N. (2010). The generalized multinomial logit model: Accounting for scale and coefficient heterogeneity. Marketing Science, 29 (3), 393 421. Fishbein, M. (1979). A theory of reasoned action: Some applications and implications.
115 Fishbein, M., & Ajzen, I. (197 5). Belief, attitude, intention and behavior: An introduction to theory and research Friedland, M. T. (2005). You call that organic the USDA's misleading food regulations. NYU Envtl. LJ, 13 , 379. Gao, Z., & Schroeder, T. C. (2009). Effects of label information on consumer willingness to pay for food attributes. American Journal of Agricultural Economics, 91 (3), 795 809. Gracia, A., Loureiro, M. L., & Nayga, R. M. (2011). Are valuations from non hypothetical choice experiments different from those of experimental auctions? American Journal of Agricultural Economics, 93 (5), 1358 1373. Graham, J. L. (1985). Cross cultural marketing negotiations: A laboratory experiment. Marketing Science, 4 (2), 130 146. Grebitus, C., Lusk, J. L., & Nayga, R. M. (2013). E xplaining differences in real and hypothetical experimental auctions and choice experiments with personality. Journal of Economic Psychology, 36 (C), 11 26. Hanemann, M., Loomis, J., & Kanninen, B. (1991). Statistical efficiency of double bounded dichotomou s choice contingent valuation. American Journal of Agricultural Economics, 73 (4), 1255 1263. Harrison, G. W., & RutstrÃ¶m, E. E. (2008). Experimental evidence on the existence of hypothetical bias in value elicitation methods. Handbook of Experimental Econo mics Results, 1 , 752 767. Hausman, J., & McFadden, D. (1984). Specification tests for the multinomial logit model. Econometrica: Journal of the Econometric Society, 52 , 1219 1240. Hoefkens, C., Verbeke, W., Aertsens, J., Mondelaers, K., & Van Camp, J. (200 9). The nutritional and toxicological value of organic vegetables: Consumer perception versus scientific evidence. British Food Journal, 111 (10), 1062 1077. Horowitz, J.K. (2006). The Becker DeGroot Marschak mechanism is not necessarily incentive compatible, even for non random goods. Economic Letters , 93, 6 11. Hulland, J., TodiÃ±o Jr, H. S., & Lecraw, D. J. (1996). Country of origin effects on sellers' price premiums in competitive P hilippine markets. Journa l of International Marketing, 4 , 57 79. Hutchins, R. K., & Greenhalgh, L. (1997). Organic confusion: Sustaining competitive advantage. British Food Journal, 99 (9), 336 338.
116 Irwin, J. R., Slovic, P., Lichtenstein, S. and McClella nd, G. H. (1993). Preference reversals and the measu rement of environmental values. Journal of Risk and Uncertainty , 6, 5 18. Jones, P., Clarke Hill, C., Shears, P., & Hillier, D. (2001). Retailing organic foods. British Food Journal, 103 (5), 358 365. Krinsky, I., & Robb, A. L. (1986). On approximating the statistical properties of elasticities. The Review of Economics and Statistics, 68 (4), 715 719. Krystallis, A., Fotopoulos, C., & Zotos, Y. (2006). Organic consumers' profile and their willingness to pay for sel ect ed organic food products in G reece. Journal of International Consumer Marketing, 19 (1), 81 106. Kuhfeld, W. F. (2005). Experimental design, efficiency, coding, and choice designs. M arketing Research Methods in SAS : Experimental Design, Choice, Conjoint, and Graphical Techniques, 47 97. Lea, E., & Worsley, T. (2005). Australians' organic food beliefs, demographics and values. British Food Journal, 107 (11), 855 869. Lee, D. Y. (2000). Retail bargaining behavior of A merican and Chinese customers. European Jo urnal of Marketing, 34 (1/2), 190 206. Leggett, C. G. (2002). Environmental valuation with imperfect information the case of the random utility model. Environmental and Resource Economics, 23 (3), 343 355. Lim, K. H., Hu, W., Maynard, L. J., & Goddard, E. (2 and willingness to pay for Country of Origin Labeled beef steak and food safety enhancements. Canadian Journal of Agricultural Economics/Revue Canadienne d'Agroeconomie, 61 (1), 93 118. List, J. A. (2001). Do explicit warnings eliminate the hypothetical b ias in elicitation procedures? E vidence from field auctions for sportscards. The American Economic Review, 91 (5), 1498 1507. List, J. A., & Shogren, J. F. (1998). Calibration of the difference between actual and hypothetical va luations in a field experiment. Journal of Economic Behavior & Organization, 37 (2), 193 205. Loureiro, M. L., & Umberger, W. J. (2003). Estimating consumer willingness to pay for country of origin labeling. Journal of Agricultural & Resource Economics, 28 ( 2) Loureiro, M. L., & Umberger, W. J. (2003). Estimating consumer willingness to pay for country of origin labeling. Journal of Agricultural & Resource Economics, 28 (2)
117 (2005). Assessing consumer preferences for country of origin labeling. Journal of A gricultural and Applied Economics, 37 (1), 49 63. (2007). A choice experiment model for beef: What US consumer responses tell us about relative preferences for food safety, country of origin labeling and traceability. Food Policy, 32 (4), 496 514. Louvie re, J. J., Hensher, D. A., & Swait, J. D. (2000). Stated choice methods: Analysis and applications Cambridge University Press. Louviere, J. J., Meyer, R. J., Bunch, D. S., Carson, R., Dellaert, B., Hanemann, W. M., Hensher, D., & Irwin, J. (1999). Combining sources of preference data for modeling complex decision processes. Marketing Letters, 10 (3), 205 217. Louviere, J. J., Street, D., Burgess, L., Wasi, N., Islam, T., & Marley, A. A. (2008). Modeling the choices of individual de cision makers by combining efficient choice experiment designs with extra preference information. Journal of Choice Modelling, 1 (1), 128 164. Louviere, J., & Eagle, T. (2006). Confound it! that pesky little scale constant messes up our convenient assumptio ns. Paper presented at the Proceedings of the Sawtooth Software Conference, 211 228. Louviere, J., Street, D., Carson, R., Ainslie, A., DeShazo, J., Cameron, T ., Hensher, D ., Kohn, R ., & Marley, T. (2002). Dissecting the random compo nent of utility. Marketing Letters, 13 (3), 177 193. Lusk, J., & Fox, J.A. ( 2003 ). Economics Letters , 79, 27 34. Lusk, J. L., & Schroeder, T. C. (2004). Are choice experiments incentive compatible? A test with quality differentiated beef steaks. American Journal of Agricultural Economics, 86 (2), 467 482. (2006). Auction bids and shopping choices. Advances in Economic Analysis & Policy, 6 (1) MacQueen, J. (1967). Some methods for classification and analysis of multivariate observations. Paper presented at the Proceedings of th e Fifth Berkeley Symposium on Mathematic al Statistics and Probability, 1 , 281 297 . Magnusson, M. K., Arvola, A., Hursti, U. K., Ã…berg, L., & SjÃ¶dÃ©n, P. (2001). Attitud es towards organic foods among S wedish consumers. British Food Journal, 103 (3), 209 227. (2003). Choice of organic foods is related to perceived consequences for human health and to environmentally friendly behavior . Appetite, 40 (2), 109 117.
118 McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. Meyer, R. J. (2008). Formal choice models of informal choices: What choice modeling research can, and can't, learn from behavioral theory. Review of Marketing Research, 4 , 3 32. Michelsen, J., Hamm, U., Wynen, E., & Roth, E. (1999). The E uropean market for organic products: Growth and development. Morey, E. R., Rowe, R. D., & Watson, M. (1993). A repeated nested logit model of Atlantic salmon fishing. American Journal of Agricultural Economics, 75 (3), 578 592. Murphy, J. J., Allen, P. G., Stevens, T. H., & Weatherhe ad, D. (2005). A meta analysis of hypothetical bias in stated preference valuation. Environmental and Resource Economics, 30 (3), 313 325. Onozaka, Y., & Mcfadden, D. T. (2011). Does local labeling complement or compete with other sustainable labels? A conj oint analysis of direct and joint values for fresh produce claim. American Journal of Agricultural Economics, 93 (3), 693 706. Payne, J.W., Bettman, J. R. & Johnson, E.J. (1992). Behavioral decision research: A constructive processing perspective. Annual Review of Psychology , 43, 87 131. Peterson, H. H., Bernard, J. C., Fox, J. A. S., & Peterson, J. M. (2013). Japanese consumers' valuation of rice and pork from domestic, US, and other origins. Journal of Agricultural & Resource Economics, 38 (1) Roitner Sch obesberger, B., Darnhofer, I., Somsook, S., & Vogl, C. R. (2008). Consumer perceptions of organic foods in Bangkok , Thailand . Food Policy, 33 (2), 112 121. Rousu, M., Huffman, W. E., Shogren, J. F., & Tegene, A. (2007). Effects and value of verifiable infor mation in a controversial market: Evidence from lab auctions of genetically modified food. Economic Inquiry, 45 (3), 409 432. Sackett, H. M., Shupp, R., & Tonsor, G. T. (2012). Discrete choice modeling of consumer preferences for sustainably produced steak and apples. Paper presented at the 2012 AAEA/EAAE Food Environment Symposium, may 30 31, Boston, MA, (123517) Savage, S. J., & Waldman, D. M. (2008). Learning and fatigue during choice experiments: A comparison of online and mail survey modes. Journal of A pplied Econometrics, 23 (3), 351 371. Scarpa, R., Thiene, M., & Train, K. (2008). Utility in willingness to pay space: A tool to address confounding random scale effects in destination choice to the Alps . American Journal of Agricultural Economics, 90 (4), 9 94 1010.
119 Schifferstein, H. N., & Oude Ophuis, P. A. (1998). Health related determinants of organic food consumption in the Netherlands . Food Quality and Preference, 9 (3), 119 133. Schurr, P. H., & Ozanne, J. L. (1985). Influences on exchange processes: Buyers' preconceptions of a seller's trustworthiness and bargaining toughness. Journal of Consumer Research, 939 953. Shi L., House, L., & Gao, Z. (2013). Impact of purchase intentions on full and partial bids in BDM auctions: willingness to pay for organi c and local blueberries. Journal of Agricultural Economics , 64(3), 707 718. Sonnier, G., Ainslie, A., & Otter, T. (2007). Heterogeneity distributions of willingness to pay in choice models. Quantitative Marketing and Economics, 5 (3), 313 331. Steckel, J. H ., & Vanhonacker, W. R. (1988). A heterogeneous conditional logit model of choice. Journal of Business & Economic Statistics, 6 (3), 391 398. Steenkamp, J. E. (1990). Conceptual model of the quality perception process. Journal of Business Research, 21 (4), 3 09 333. Swait, J., & Louviere, J. (1993). The role of the scale parameter in the estimation and comparison of multinomial logit models. Journal of Marketing Research (JMR), 30 (3) Thompson, G. D. (1998). Consumer demand for organic foods: What we know and w hat we need to know. American Journal of Agricultural Economics, 80(5) , 1113 1118. Thompson, G. D., & Kidwell, J. (1998). Explaining the choice of organic produce: Cosmetic defects, prices, and consumer preferences. American Journal of Agricultural Economics, 80 (2), 277 287. Thurstone, L. L. (1927). A law of comparative judgment. Psychological Review, 34 (4), 273. Tobin, J. (1958). Liquidity preference as behavior towards risk. The Review of Economic Studies, 25 (2), 65 86. Tonsor, G. T., Schroeder, T. C., & Lusk, J. L. (2013). Consumer valuation of alternative meat origin labels. Journal of Agricultural Economics, 64 (3), 676 692. Train, K. (2009). Discrete choice methods with simulation Cambridge University press. Train, K. E. (1998). Recreation demand models with taste differences over people. Land Economics, 74 (2) Train, K., & Weeks, M. (2005). Discrete choice models in preference space and willingness to pay space Springer.
120 Tsakiridou, E., Boutsouki, C., Zotos, Y., & Mattas, K. (2008). Attitudes and behavior towards organic products: An exploratory study. International Journal of Retail & Distribution Management, 36 (2), 158 175. Tversky, A. & Simonson, I. (1993). Context dependent preference. Management Science , 39, 1 179 1 189. UreÃ±a, F., BernabÃ©u , R., & Olmeda, M. (2008). Women, men and organic food: Differences in their attitudes and willingness to pay. A Spanish case study. International Journal of Consumer Studies, 32 (1), 18 26. USDA/AMS. (2012). Agricultural Marketing Services . Country of Orig in Labeling. Accessed April 1, 2014, available at http://www.ams.usda.gov/AMSv1.0/cool USDA/AMS. (2012). National Organic Program. United States Department of Agriculture, Agricultural Mark eting Service (USDA/AMS, Washington, D.C. Accessed February 21, 2014, available at http://www.ams.usda.gov/AMSv1.0/ams.fetchTemplateData.do?template=Templ ateA&navID=NationalOrganicProgram&leftNav=NationalOrganicProgram&page= NOPNationalOrganicProgramHome&acct=AMSPW USDA/ERS. (1999). Broccoli: Super Food for All Se asons. United States Department of Agriculture, Economic Research Service (USDA/ERS), Washington, D.C. Accessed February 21, 2014, available at http://www.ers.usda.gov/public ations/agoutlook/apr1999/ao260b.pdf . (2011). US. Broccoli Statistics. United States Department of Agriculture, Economic Research Service (USDA/ERS), Washington, D.C. Accessed February 21, 2014, available at http://usda.mannlib.cornell.edu/MannUsda/viewDocumentInfo.do?documentID=1 816 Verlegh, P. W., & Steenkamp, J. E. (1999). A review and meta analysis of country of orig in research. Journal of Economic Psychology, 20 (5), 521 546. Verlegh, P. W., Steenkamp, J. E., & Meulenberg, M. T. (2005). Country of origin effects in consumer processing of advertising claims. International Journal of Research in Marketing, 22 (2), 127 13 9. Wandel, M., & Bugge, A. (1997). Environmental concern in consumer evaluation of food quality. Food Quality and Preference, 8 (1), 19 26. WFM blog, (2012). Is organic from China possible? Accessed February 21, 2014, available at http://blog.wholefoodsmarket.com/2008/06/is organic from china possible/
121 Wier, M., & Calverley, C. (2002). Market potential for organic foods in Europe . British Food Journal, 104 (1), 45 62. Wil kins, J. L., & Hillers, V. N. (1994). Influences of pesticide residue and environmental concerns on organic food preference among food cooperative members and non members in Washington State . Journal of Nutrition Education, 26 (1), 26 33. Windle, J., & Rolf e, J. (2011). Comparing responses from internet and paper based collection methods in more complex stated preference environmental valuation surveys. Economic Analysis & Policy, 41 (1) Worner, F., & Meier Ploeger, A. (1999). What the consumer says. Ecology and Farming, 20 (2), 14 15. Yiridoe, E. K., Bonti Ankomah, S., & Martin, R. C. (2005). Comparison of consumer perceptions and preference toward organic versus conventionally produced foods: A review and update of the literature. Renewable Agriculture and Fo od Systems, 20 (04), 193 205. Zanoli, R., & Naspetti, S. (2002). Consumer motivations in the purchase of organic food: A means end approach. British Food Journal, 104 (8), 643 653.
122 BIOGRAPHICAL SKETCH Jing Xie received her Ph.D. in food and resource e conomics at the University of Florida in 2014. She a inance from Xiamen Unive rsity research interests f ocus on market research, consumer behavior, survey and experimental design, and quantitative methodologies.