EXPERIMENTAL AUCTIONS VS REAL CHOICE EXPERIMENT: AN EMPIRICAL APPLICATION ON CONSUMER VALUATION FOR FOOD QUALITY ATTRIBUTES By Angelos Lagoudakis A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Agricultural, Food and Resource Economics Master of Science 2019 ABSTRACT EXPERIMENTAL AUCTIONS VS REAL CHOICE EXPERIMENT: AN EMPIRICAL APPLICATION ON CONSUMER VALUATION FOR FOOD QUALITY ATTRIBUTES By Angelos Lagoudakis Real choice experiments (RCEs) and experimental auctions (EAs) are two non - market valuation methods which have increasingly been willingness to pay (WTP) for food products. This paper aims to determine whether EAs and RCE s derive different welfare estimates or not and examine s their incentive compatibility using data from a U.S. consumer study. We compare the bidding behavior of consumers in three different incentive compatible EAs to the behavior of consumers who made non - hypothetical discrete choices for egg products. We find that the valuations elicited from EA s differ significantly from those obtained from RCEs . Nonetheless, for the goods evaluated, These finding s hold relevant implications for the design of real choice experiments an d experimental auctions. The practical implications for food retailers are also discussed. iii To my Grandmother Rena iv ACKNOWLEDGEMENTS Throughout the writing of this thesis, I have received a great deal of support and assistance. I would first like to thank my thesis advisor Dr. Caputo Vincenzina, Assistant Professor of the Department of Agricultural, Food, and Resource Economics at Michi gan State University. The about my research or writing. She consistently steered me in the right direction throughout the process of designing, implementing and wri ting this thesis. Furthermore, I would like to thank Dr. Caputo and Dr. Robert Shupp, Associate Professor of the Department of Agricultural, Food, and Resource Economics at Michigan State University for funding the data collection used for this research p roject. Dr. Shupp also assisted me throughout the design, the implementation, and the review of this research paper with his valuable comments, edits, and additions. Also, I would like to thank the participants in my survey, who have willingly shared their precious time during the whole process. I am also grateful to the other two members of my committee (in addition to Dr. Caputo and Dr. Shupp) who were involved in the review process of this paper: Dr. Robert Myers, Associate Chairperson, Graduate Program Director, and University Distinguished Professor of the Department of Agricultural, Food, and Resource Economics at Michigan State University and Dr. Eduardo Nakasone, Assistant Professor of the Department of Agricultural, Food, and Resource Economics and the Department of Media and Information at Michigan State been successfully completed. I also wish to t hank Dr. Andreas Drichoutis, Assistant Professor of the Department of Agricultural Economics and Rural Development at the Agricultural University of Athens , for v his advice during the initial steps of this research project. Moreover, Dr. Drichoutis has always been a role model for me and he was the one who ignit ed my interest in purs uing an academic career. Thanks should also go to Dr. Christos Papadas, Associate Professor of the Department of Agricultural Economics and Rural Development at the Agricultural University of Athens, who was instrumental in guiding m e during my application process for graduate programs in the United States. I would not have made it that far without him. I would also like to extend my deepest gratitude to my colleague Danielle Kaminski for her assistance and support throughout the design and the implementation of the data collection. Her unwavering support helped me through good and bad times in the last couple of years. I would like to express my sincere thanks to April Athnos and Christine Sauer, very good friends and Ph.D. candi dates of the Department of Agricultural, Food, and Resource Economics at Michigan State University, as the second readers of this thesis, and I am gratefully indebted to them for their valuable comments and friendship. The completion of my thesis would not have been possible without the support and nurturing of my colleagues and beloved friends within the department of Agricultural, Food, and Resource Economics. Brian Bartle, Dan Ochs, Sarah Chase - Walsh, Tanner Connors, Ari Kornelis, Sean Posey, Sa rah Kopper, Danielle Ufer, Kelsey Hopkins, Simone Faas, Ben DeMuth, Maria - Elena Marescotti, Petjon Balco and Carolina Vargas are only a few from the people who never wavered in their support. I cannot begin to express my thanks to my family, who has always been the most important pillar of my life. My grandparents, my parents, my siblings and my brothers in life Elias and Kostas, who have always been there for me and supported me in the hardest decision I have ever made; to leave Greece and follo w my dreams abroad. I would specifically like to vi thank my mother for keeping me harmonious and reminding me who I am when myself or other people tended to forget. I will be grateful forever for your love. Finally, I must express my very profound gratitude to my wife for providing me with unfailing support, love and continuous encouragement throughout the last year of my study and through the process of researching and writing this thesis. This accomplishment would not have been possible without her. Thank y ou. vii TABLE OF CONTENTS LIST OF TABLES ................................ ................................ ................................ .................. vi ii LIST OF FIGURES ................................ ................................ ................................ .................. i x 1. INTRODUCTION ................................ ................................ ................................ ................. 1 2. BACKGROUND AND PREVIOUS STUDIES ................................ ................................ .... 5 2.1 Experimental Auctions (EAs) ................................ ................................ ..................... 5 2.2 Real Choice Experiment (RCE) ................................ ................................ ................ 11 2.3 Experimental Auctions versus Real Choice Experiment s ................................ ........ 1 4 3. EXPERIMENTAL PROCEDURES A ND SURVEY DESIGN ................................ .......... 1 7 4. EMPIRICAL MODELS A ND SPECIFICATION ................................ .............................. 2 5 5. RESULTS ................................ ................................ ................................ ........................... 3 2 6 . DISCUSSION AND CONCLUSION S ................................ ................................ ............... 4 6 APPENDICES ................................ ................................ ................................ ......................... 4 9 Appendix A: Inverse Cumulative Density Functions (CDFs) of WTP for the eggs ...... 5 0 Appendix B: Empirical Estimates for the MNL specification in the RCE ..................... 53 Appendix C: Empirical Estimates for the RCE in preference space .............................. 54 Appendix D: Number of all zero bids and all no buy options ................................ ........ 55 Appendix E: Instructions for the elicitation methods ................................ ..................... 56 Appendix F : Consent form for participation in the study ................................ ............... 66 REFERENCES ................................ ................................ ................................ ........................ 6 7 viii LIST OF TABLES Table 1 : Experimental Auctions ................................ ................................ ................................ 6 Table 2 : Comparison of EAs using homegrown experiments ................................ ................... 8 Table 3 : Real choice experiment versus experimental auctions literature table ...................... 1 4 Table 4 : Prices in Discrete Choice Tasks ................................ ................................ ................ 2 3 Table 5 : Sample Characteristics (%, unless stated) ................................ ................................ . 3 2 Table 6: WTP Bids by Auction Mechanism and Egg Type ................................ ..................... 3 3 Table 7: Marginal WTP Bids by Auction Mechanism and Egg Type ................................ .... 3 4 Table 8: Variables used in auction data analysis (Tobit model for each auction type and pooled Tobit models ) ................................ ................................ ................................ ............... 3 5 Table 9 : Tobit results for marginal WTP per auction mechanism ................................ ........... 3 5 Table 1 0 : The effect of Auction Institution on marginal WTP: Random Effects Pooled Tobit Estimates ................................ ................................ ................................ ................................ .. 3 7 Table 1 1 : Estimates for the Mixed Logit Model in WTP space ................................ .............. 4 0 Table 1 2 : Mean marginal WTP per elicitation mechanism ................................ ..................... 4 2 Table 1 3 : The effect of institution on marginal WTP: Random Effects Pooled Tobit Estimates ................................ ................................ ................................ ................................ .................. 4 4 Table B 1: Empirical Estimates for the RCE ................................ ................................ ............ 5 3 Table C 1: Empirical Estimates for the RCE in preference space ................................ ............ 5 4 Table D 1: Participants will all zero bids/all no - buys ................................ ............................... 5 5 ix LIST OF FIGURES Figure 1: Labels used in our study ................................ ................................ ........................... 19 Figure 2: Example of Choice Experiment Question ................................ ................................ 2 2 Figure A1 : Distribution of Willingness - to - Pay for Conventional Eggs ................................ .. 50 Figure A 2: Distribution of Willingness - to - Pay for C age Free Eggs ................................ ........ 50 Figure A3 : Distribution of Willingness - to - Pay for USDA Organic Eggs ............................... 51 1 1. INTRODUCTION Real choice experiments (RCEs) and experimental auctions (EAs) are two of the most popular non - hypothetical non - willingness to pay (WTP) for food attributes and products. In RCEs respondents are faced with multiple choice scenarios and are asked to make decisions between different product alterna tives . The alternatives are defined by different attributes and attribute levels. Once a respondent completes the experiment one of these choice questions is randomly selected as binding; the respondent has to buy the chosen alternative in the binding choi ce question and pay the price indicated in the chosen alternative. In EAs consumers place their bids for a specific product characterized by different attributes and attribute levels. Once the experiment is completed, the highest bidder(s) buys the product , paying the market price which is defined from the auction mechanism. The vast majority of the research implementing th e se valuation methods (i.e.: EAs and RCEs) elicit homegrown values, defined as the subjective valuations of participants for a good (Mur phy et al., 2010). From a theoretical p er spective , the RCE and EA methods are both incentive compatible because every participant is incentivized to truly reveal their preferences. 1 Hence , i f participants truly reveal their preferences, then welfare estima tes such as the marginal and total WTP should not differ across the two methods. However, data from a number of studies suggest that WTP estimates for food attributes differ across RCE s and EAs. For instance, Gracia et al. (2011) , using a random n th price auction mechanism and a RCE to elicit WTP for cured - ham products , observed that the estimates derived from the two elicitation methods differ in 1 Not all EAs are considered incentive compatible. For example, the English auction is not considered incentive compatible while the second price and the random n th price auctions are. In our comparison we only use incentive compatible EA methods. However, previous studies have raised some concerns regard ing the incentive compatibility of the BDM auction (Horowitz, 2006) . One of the goals of this paper is to reach a conclusion on whether the BDM auction is incentive compatible or not. 2 magnitudes and are inconsistent across different product profiles. Similarly , Lusk and Schroeder (2006) find that auction bids for beef steaks, under a second price auction mechanism, are significantly lower than the valuations obtained from their RCE. The previously mentioned studie s compare RCEs with only one type of incentive compatible auction mechanism ( nth price auction or second price auction ) . Hence, it is still not known whether results between RCE s and EAs differ consistently across alternative auction mechanisms commonly used in food choice literature ( seco nd price, random n th price, and Becker De Groot Marschak (BDM ) ) . There are two primary goals of this study: First, we i nvestigate whether homegrown RCEs and EAs produce different marginal WTP for food attributes. Unlike previous studies, this study compares a RCE with three different popular auct ion mechanisms ; specifically, the seco nd price (Vickrey, 1961) , the random n th price (Shogren et al., 2001) , and the BDM (Becker, DeGroot, and Marschak, 1964) auction mechanisms . Second, we c ompare valuation estimates across the aforementioned EAs . Additionally, w hile research has been carried out on comparing welfare estimates derived from different EA mechanisms, only a few studies have compared the auction institutions as elicitation mechan isms for food attributes (Lusk et al. 2004 , , with some of the most prominent previous studies focusing on collectible trading cards ( Lucking - R e iley , 1999 ; List, 2003) . This is important because consumers routinely purchase food products in a typical posted - price marke t such as grocery stores. To achieve the research objectives, we used a between - subjects approach and conduct ed a RCE and three EAs : (i) second price, (ii) BDM , and (iii) random n th price . In total, 271 consumers participated in our study. In all experiments, the participants were asked to indicate their preferences for a dozen large, grade A, brown eggs produce d either with the conventional method , the cage free system or the USDA organic production requirements . O ur 3 results indicate that the WTP estimates differ across RCE and one of the EAs. Specifically, t he BDM auction yiel ds higher estimates than the other mechanisms. The second price auction, the random n th price auction and the RCE do not generate statistically different estimates from each other . This suggests that the implementation of a different EA mechanism affects s individuals to make choices which might not necessarily represent their true preferences. This study advances the experimental economics literature in three important ways. First, t o the best of our knowledge, this i s the first study to provide comprehensive results from a comparison of RCE and EAs. Comparisons of RCE and EA choice mechanisms are essential to establish whether the mechanisms are incentive compatible and whether they induce different estimates of WTP . Our study inform s the mechanism selection in future studies and h as understanding of consumer preferences . For e xamp le , due to their convenience, RCEs could potentially replace the use of BDM auctions in developing co untries . RCEs present similar benefits with the BDM auction (no need to form group s of people to run the experiments, individual decision - making) and they ha ve the advantage of replicat ing an actual market scenario, which is very familiar to consumers. Additionally , several concerns have been raised for the incentive compatibility of the BDM (Horowitz, 2006) . Second , while numerous studies comparing auctions focus on collectible trading cards ( Lucking - R e iley, 1999; List , 2003), only a few applications of EAs have been implemented in the food choice literature ( , 1998; Lusk et al., 2004 ) . Hence, this study adds to this research area by providing a deeper understanding o f whether valuations are equivalent across theoretically incentive compatible 4 Third, this study offers a detailed overview of prior research implementing RCEs i n food economics, EA studies comparing multiple auction mechanisms , and prior literature comparing RCE s to EA s . This article is structured as follows: first, we provide a brief description of the previous studies comparing EAs with one an other and with RCEs . This is followed by the experimental procedures and the econometric analyses that were implemented in the RCE and EAs . The final section discusses the results and presents the conclusions . 5 2. BACK G ROUND AND PREVIOUS STUDIES In contrast with most widely used available elicitation mechanisms , which ask consumers hypothetical questions, EAs and RCEs scenario where they pay real money in exchange for real goods. In the following paragraphs we describe the EAs and the RCE mechanisms in more detai l and review the existing literature. 2.1 Experimental Auctions (EAs) EA s constitute one of the most widely used elicitation methods in recent literature (Corrigan and Rousu, 2006 ; Lusk et al. , 2004) . The main advantage of EAs is that the bids constitute a Researchers increasingly use EAs because the environment of the experiment is real ( real products and real money are used ) and hence participa nts have an incentive to reveal their true preferences for the products being examined (Fox et al., 1996). When implementing EA s in a laboratory setting, external factors should be held constant . This protects participants by un observable , etc.) . Thus, researchers must have the ability to control for the effects of external factors ( location of study , instructions, etc.) in order to i solate the effect of changes in our variables of interest (Lusk and Shogren, 2007). There are several EA mechanisms that have been widely used in market research and decision making during the la st decade. Lusk and Shogren (2007) , in their seminal study, describe the most significant EA mechanisms, which are summarized in Table 1. 6 Table 1: Experimental Auctions Elicitation Mechanisms Participant Procedure Market Price Rule Number of Winners English Sequentially offer ascending bids Last offered bid Highest bidder pays market price 1 Second Price Simultaneously submit sealed bids Second Highest bid Highest bidder pays market price 1 Nth Price Simultaneously submit sealed bids Nth highest bid n - 1 highest bidders pay market price n - 1 BDM (Becker - De - Groot - Marschak ) Simultaneously submit sealed bids Randomly drawn price (random number generator) People pay market price if bid exceeds randomly drawn price Individually determined Random Nth price Simultaneously submit sealed bids Randomly drawn (nth) bid n - 1 highest bidders pay market price n - 1 Collective Auction Simultaneously submit sealed bids Mean bid Each individual pays market price (subject to unanimity rule) if sum of bids exceeds sum of costs None or all Source: Lusk and Shogren (2007) Past research has established tha t t he selection of the EA mechanism depends on a number of factors . First, s everal lines of evidence suggest that w hen selecting the appropriate mechanism, bidder affiliation should be avoided (Milgrom and Weber , 2006; Lusk et al., 2004) . 7 That is, f or s trategic equivalence across auctions , independent valuations from the rivals are required (i.e. nonaffiliated participants and/or non - collaborative bids ). Second, each auction mechanism implies a different experimental setting choice behavior (Noussair et al., 2004) . This is evident in second price auction where partici pants submit their bids together with other participant s . In contrast , in BDM auction , participants make decisions individually , as experiments are conducted one - to - one . As such , if one of the primary goals of the researcher is to observe how participants interact during an EA, then the second price auction would be preferred over the BDM auction. Moreover, every auction has a different training factor included which depends on h ow familiar participants are with the mechanism ( Noussair et al., 2004 ). For exampl e, the English auction is very well known to participants and hence, it requires only a short training . On the other hand, the BDM auction is unfamiliar to most participants and requires a longer training . Additionally , several studies suggest that convenience is an another important for selecti on of the auction mechanism. This is certainly true in the case of the BDM auction. Although the BDM mechanism is not necessarily incentive compatible (Horowitz, 2006), its use is widespread because it can be administered to one individual at a time. This is especially important in experimental design in international development settings (De Groote et al., 2001). According to Wertenbroch and Skiera, 2003 , the benefit of implementing BDM with one individual at a time is that it allows for the experimenter to do the experiment at real points of purchas e under the specific conditions he/she desires. 2 Another factor t o consider when selecting an auction mechanism is whether the researcher is interested in more accurately estimating the upper or lower end of the demand curve. High value bidders help researchers estimate the upper end of the demand curve , while 2 For example, the experimenter can implement the BDM auction next to an outdoors market with participants who just shopped from the market. 8 low value bidders aid in estimat ion of the lower end of the demand curve (Lusk and S hogren, 2007). To illustrate , Knetsch et al. ( 2005 ) examined bidding behavior in a second price auction and conclude d that the auction might not engage low - value bidders who have the perception that they will never win (i.e.: second price auction might fai l to engage off - margin bidders ) . A ccording to Lusk and Shogren (2007), it is also essential to consider how personal characteristics inform demand for the products up for auction . The selection of one type of auction instead of another has important implications in terms of welfare estimates , too . In this regard, there exists a plethora of studies comparing different incentive compatible EA mechanisms to elicit WTPs using homegrown experiments (Shogren et al. , 2004; , 1998; Lucking - R e iley , 1999 ; List, 2003; Lusk et al., 2004 ) . With a few exceptions (Shogren et al., 1994), r esults from most of th e se studies suggest that different EA mechanism s may lead to different welfare estimates Lusk et al., 2004). Table 2 summarizes studies which have compared EA s using homegrown experiments. Table 2 : Comparison of EAs using homegrown experiments Authors Country/Year Subject Experimental Auction s Compared Results Shogren et al. USA / 1994 I rradiation to control the food borne pathogen Trichinella (Food Economics) S econd price auction (2PR) , random n th price auction (RNP) , and combinatorial private - collective (PC) auction WTP 2PR = WTP RNP =WTP PC USA/ 1998 G ourmet chocolate truffles (Food Economics) English auction (EN) ; s econd price auction , and BDM auction WTP EN < WTP 2PR WTP BDM < WTP 2PR 9 Table 2 Lusk et al. USA / 20 04 B eef ribeye steaks (Food Economics) Second price auction, random nth price auction, BDM auction, and English auction WTP 2PR > WTP BDM WTP 2PR > WTP RNP WTP 2PR > WTP EN WTP RNP < WTP EN WTP RNP < WTP BDM Lucking - Riley USA/1999 (Internet) Collectible trading cards English auction , Dutch auction , first price auction (1 ST ) , and second price auction WTP EN =WTP 2PR WTP DUTCH >WTP 1ST List USA/2003 (field experiment ) Sports cards Second price auction and random nth price auction WTP RNP > WTP 2PR To illustrate, i n their longitudinal study, Shogren et al. (1994) compared a second - price auction, a random nth - price auction and a combinatorial private - collective auction. The combinatorial auction was created by combining a second price auction with a collective auction. P articipants were provided a sandwich and a $ 25 endowment. Then, a nother sandwich was auctioned with the difference being that it had been treated with irradiation to control the food borne pathogen Trichinella. Participants were told that a sandwich must be eaten to receiv e their take - home income. Results from this study indicate no statistically significant 10 d ifference between the second price and random nth price auctions; as well as between the second price and the combinatorial private - collective auction s . 3 (19 98) compared the bids from English, second price , and BDM auctions. The experiment design implemented conducted comparisons across different sub - samples of the participants varying the auction mechanism . The product was a box of gourmet chocolate truffles . Results from the study indicate that , on average, English and BDM auction bids are lower than the second price auction bids . Lusk et al. (2004) compared the results from the following incentive compatible auction mechanisms: second price, Englis h, BDM, and random nth price. To conduct such a comparison, they asked consumers to evaluate several different types of beef ribeye steaks. Results indicated that consumers bid higher in the second price auction compared to the English, BDM, and random n th price auctions. This result was magnified in the bidding rounds of the random nth price auctions . Furthermore, the authors found that random n th price auction s generate lower valuations than English and BDM auctions , on average . In contrast, Lucking - R e iley (1999) found that estimates of WTP generated from an English auction and a second price auction are not statistically different . Comparing results from intern et auctions (English, Dutch, first price sealed bid , and second price) using collectible trading cards as the product for sale , the author created two pairs of auctions and auctioned a copy of the same card in each institution. The two pairs were Dutch vs. first price auction and English vs. second price auction. 4 Lucking - Reiley concluded that Dutch auctions 3 Specifically, using the overall averages for the second - price auction and the random nth price auction, Shogren et al. were not able to reject the hypothesis that the average bids in t rial 1 and in trial 10 (the first and the last of 10 trials) were equal. 4 In th e Dutch auction, the price is continuously reduced until a buyer is found . Both in the Dutch and in the - e sealed bid auction, each participant submits a bid and the highest bidder wins the good, paying his/her bid. 11 earn approximately 30 percent mor e revenue than first price auctions and that English and second - price auction s exhibited revenue equivalence. Similar to Lucking - R e iley (1999) , List (2003) auctioned sports cards and compared the mean bids between a second price and a random n th price auction . Results from this study indicate that mean bids from th e two types of auction are significantly different with the second price auction yielding lower WTPs than the random n th price auction when actual auctions were implemented. Taken toget her , results from these studies suggest that different auction mechanisms generate different welfare estimates . Our study adds to this e xisting literature by exploring whether three widely used EA mechanism s ( i.e., second price auction, BDM auction, and random n th price auction ) produce equivalent welfare estimates using a generic commodity , eggs . A generic commodity is important because consumers routinely buy it, and eggs can be purchased in a traditional posted - p rice market such as grocery stores, convenience stores and on internet auction sites like E - bay. In addition, our study contributes to the literature by adding another piece of evidence against the equivalence of valuations derived under different EA mecha nisms ; showing that one auction mechanism (BDM) derives statistically different results from the other EAs. Our ultimate purpose is to investigate whether an EA ( second price/BDM/random n th price) method should be preferred over the other two methods ( BDM - random n th price/second price - random n th price/second price - BDM ) or not and explore the reasons for potential differences across EAs. 2.2 Real Choice Experiment (RCE) The choice experiment approach has its foundation in Lancastrian consumer theory (La ncaster, 1966) and random utility theory (see McFadden, 1974; Hanemann and Kanninen, 2001). Lancaster (1966) suggested that the total utility for a specific product can be broken down into sub - utilities for each attribute of the product. Random utility the ory is based on the classical 12 economic assumption that individual agents act rationally and always select the option that maximizes their derived utility. Therefore, the higher the utility each alternative provides (among the different alternatives), the h igher the probability for the consumer to select that alternative. McFadden (McFadden 1986; McFadden and Train 2000; McFadden 1974) has theory of paired comparisons . As for their practical applications, discrete choice experiments were originally used in the fields of transportation and marketing with the ultimate purpose of predicting demand for new products. The first discrete choice experiment (as we know it today ) is widely considered to have been created from Louviere and Hensher (1982) in the field of transportation. Since then , d iscrete c hoice e xperiments have become one of the most popular non - market valuation method s employed in different fields of applied economics including food economics ( Chang et al., 2009 ; Lusk and Schroeder , 2004, 2006 ; Alfnes et al. , 2006 ; van Loo et al. , 2011, 2014 ; Van Wezemael et al. , 2014 ; Lusk and Tonsor , 2016 ; Ortega et al. , 2011 ; Caputo et al. , 2013, 2018a, 2018b; Hu et al., 2006 ) , marketing (Ashok et al., 2003) , development (Ortega and Ward, 2016; Otieno, 2011) , transportation (Hess et al. , 2008; Rose and Bliemer, 2009) , and environmental economics (Ferrini and Scarpa, 2007; Scarpa et al. , 2012) . The popularity of discrete choice experiment s has increase d due to its advantages compared to other preference elicitation methods . Lusk and Schroeder ( 2004) summarize these advantages as follows . First, discrete choice experiments are more flexible than EAs because the evaluation of product alternatives or food attributes occurs simultaneously . To illustrate, in EAs, the number of attributes is usually minimized to facilitate operations and the products are usually evaluated one - by - one. On the other hand, in DCE s , product profiles are described by multiple attributes that can be simultaneously valued by the participants. Second, they are 13 consistent . 5 Third, choice tasks (i.e . , choice questions) are designed in a way that closely mirror s actual shopping situations (e.g., making a choice among multiple products offered at different p rices ) . Most of the choice experiments implemented in the field of agricultural and food economics are hypothetical (e.g., van Loo et al. (2011, 2014), Van Wezemael et al. (2014), Lusk & Tonsor (2016), Ortega et al. (2011), Caputo et al. (2013, 2018a, 2018b). However, due to the existence of hypothetical bias, a n increasing number of studies are now implementing RCEs ( Chang et al. , 2009 ; Lusk & Schroeder 2004, 2006 ; Alfnes et al. , 2006 ; Bazzani et al. , 2017 ) . 6 For instance, Chang et al. (2009) compared RCE and hypothetical choices concerning ground beef, wheat flour, and dishwashing liquid. Lu sk and Schroeder (2004, 2006) elicited willingness to pay from RCEs regarding beef products. Alfnes et al. (2006) derived c onsumers' w illingness to p ay for the c olor of s almon and Bazzani et al. (2017) examined valuation for local versus organic food . O ur study contribute s to th e growing body of RCE research by eliciting consumers WTP for a generic commodity (eggs) and most importantly, by comparing RCE s and three commonly used EAs. 7 5 According to the total utility for a specific product consists of the sub - utilities for each attribute of the product , whil e following the classical economic assumption on which random utility theory is based, individual agents act rationally and always select the option that maximizes their derived utility. 6 Hypothetical bias occurs when participants do not have to support t heir choices with real monetary commitment (i.e. , buy the binding product) (Lusk and Schroeder, 2004; De - Magistris et al., 2013). 7 WTP, which is elicited with the use of experimental methods (EAs, RCE, Contingent Valuation), is often used as an input or surrogate for demand mea surement in welfare analyses of food policies (Gao and Schroeder, 2009). In addition, WTP normally provides feedback for various food labeling programs (Lusk and Anderson, 2004; ccurately has far reaching implications as it influences the decision makers of the market (i.e. , policy makers, producers, intermediaries). The ability to determine which attributes are important to consumers has become more and more important over the la st decade for two reasons: first, the use of labeling has drastically increased in food markets over the past Identifying which attributes a re important to consumers can also lead to more directed marketing and could enhance the branding and labeling of food products. 14 2.3 Experimental Auctions versus Real Choice Experiment s Whether hypothetical or real, choice experiment s and EA s are now the most widely used experimental methods in valuing goods and attributes . RCEs and EAs are considered to have similar design features, to be incentive compatible and hence to yield equivalent outcomes (Lusk and Schroeder, 2006). Notably, only a few studies have compared RCEs and EAs with each other. Results from those studies have revealed that RCE and EA result in different welfare estimates (WTPs). Table 3 summarizes the studies which have compared a RCE with an EA . Table 3 : Real choice experiment versus experimental auctions literature table Authors Country/ Year Subject Experimental Auction Real Choice Experiment Results Mechanism (Sample Size) Rounds Attributes Choice Tasks (Sample Size) Gracia et al. Spain/2011 Cured - Ham Products Random n th p rice (N=62) 1 4 price levels; 4 animal welfare l evels 16 ( N =107) WTP RCE EA Lusk and Schroeder USA/2006 Beef Steaks Second price (N=35) 5 4 price levels; 5 beef steak types 17 (N=67) WTP RCE > WTP EA Shi et al. China/ 2018 Orange Juice Products BDM (N=107) 1 4 price levels; 3 types of orange juice 10 (N=76) WTP RCE > WTP BDM 15 Gracia et al. (2011) , using a between sample approach, compared RCE and EA by means of cured - ham products , differentiated by four different levels of an animal welfare label : (a) standard animal welfare; (b) improved animal housing; (c) improved transport conditions ; and (d) comprehensive animal welfare (comprising the last two improvements). In the RCE , respondents were presented with 16 choice questions, each represented by two cured - ham products and a no - purchase option. In the EA, respondents submitted bids for each product using a random n th price auction mechanism . Both experiments were conducted in Spain (Zaragoza) with actual consumers . Their results indicate that the W TP estimates under both elicitation procedures have the same sign (positive) but of different magnitudes. S tatistically significant differences were also found between the elicitation methods for most demographic profiles in the case of the comprehensive a nimal welfare label. 8 The authors concluded that W TP estimates vary across elicitation methods . This could be due to the more direct approach of RCEs compared to EAs, the familiarity consumers feel with the RCE, the existence of peer pressure in EAs, and the different price settings between RCE and EA (Gracia et al., 2011). In the same vein, Lusk and Schroeder (2006 ) investigated whether WTPs for steak attributes differ between a second price auction and a RCE. Using a between - sample approach, c onsu mers were asked to participate either in an auction market (second price auction) or in a choice task including five beef ribeye steaks: (a) a generic steak; (b) a guaranteed tender steak; A ngus beef steak. 9 In the RCE , respondents were faced with 17 choice questions, while in the EA people were asked to place bid s . Findings from this study can be summarized as follows: a) auction bids were significantly lower as compare d to WTPs from the RCE , b) own - price elasticities of demand for higher quality products were notably higher when derived from the auction data than when 8 Their hypothesis (WTP RCE = WTP EA ) was rejected in most cases. 9 In a between sample approach, every participant in the sampl e participates only in one experiment. 16 derived from the RCE data , and c ) the two elicitation methods. R ecent research by Shi et al. (2018) f ound that BDM auction bids for three different types of orange juice were significantly lower than WTP estimates derived from the RCE experiment. By controlling for - proneness , the authors also found that higher levels of deal - proneness lead to lower bid s in BDM auctions , while it did not affect WTP estimates from the real double - bounded dichotomous contingent valuation ( RCVM ) and the RCE. 10 In addition, lower level of deal - proneness led to smaller differences in WTP estimates across BDM auction and RCE . These results suggest that the - prone people may be influenced by the BDM auction mechanism and hence, the bids derived from this mechanism may be u nderstated . Although these studies provide evidence about bidding behavior across experiments, all of them only compared one type of EA with the RCE rather than comparing a RCE with multiple EA mechanisms . Hence, it is still unclear whether the documented differences in WTP estimates between RCE and EAs are a result of the type of auction mechanism selected for the EA. This study adds to the existing literature by comparing a RCE with multiple EAs (i.e. , second price auction, BDM auction, random n th price auction). Moreover, our study will provide more recent evidence to the pre - existing research in this area ; t he studies from Gracia et al. (2011) and Lusk and Schroeder (2006) are 8 and 13 years old respectively. As those elicitation methods have been increasingly u sed during the last decade, newest research might provide more relevant evidence given that consumers are on average more familiar with EAs and RCEs today than they used to be 10 years ago. Finally , this research is the second study to be implemented in the U.S. , after the one from Lusk and Schroeder (2006) . 10 Deal - 17 3 . EXPERIMENTAL PROCEDURES AND SURVEY DESIGN Data Collection In order to assess consumer s WTP across EAs and a RCE , participant s were recruited from the general population of the college town of East Lansing and the neighboring capital city of Lansing , Michigan , during Spring 2018. Participants were recruited through by the College of Communication and Arts and Science at Michigan State University. 11 The experiments were conducted at Michigan State University . Selected participants were older than 18 years, responsible for grocery shopping, had purchase d eggs during the last three months , were not lactose intolerant and did not follow a vegan diet . They all received $13 cash to participate in the study . Sessions were scheduled Monday through Sunday during morning, afternoon and evening hours to avoid timing eff ects (Lusk and Shogren, 2007) . The duration of each session was approximately 45 minutes. 12 The products used in the experiment were three different types of eggs : (i) Conventional, (ii) Cage free, and (iii) USDA organic . All eggs were provided by the dozen and were of similar size ( large ) , grade (A) , color (brown) and packaging. Eggs were selected in this study because they are a generic commodity; that is: widely available in various food outlets including grocery stores, convenience stores , and farmers markets all over the United States. 13 Although several brands exist for eggs, non - branded products were used to avoid branding - effects . 11 Participants were notified through email for the availability of the study. 12 The cost of each of the four treatment s we implemented was approximately $1,135 (not including the labor cost of the experimenters involved in the data collection) . 13 U nlike items used in previous studies such as tickets, mugs , and coupons, consumers frequently purchase eggs, and eggs can be found in every typical posted - price market. 18 Upon arriving at the lab for a session , p articipants were randomly assigned to one of the four following treatments: a second price auction, a random n th price auction, a BDM auction, or a RCE. That is, we conducted between - subject comparisons across elicitation methods. We used the between - subject approach to avoid bias introduced by the participation in multiple experiments (Lusk and Schroeder, 2006), fatigue effects (Charness et al., 201 2) and the potential demand reduction when consumers purchase more than one product (Lusk and Schroeder, 2006). The second price and the random nth price auctions were conducted in groups of five people, whereas the BDM auction and the choice experiment wer e conducted in a private interview setting (one - on - one). 14 After assigning participants to one of the treatments, they received the participation fee in cash and were asked to read and sign the consent form (see Appendix F ) . Subsequently, they were asked to complete a questionnaire , which included demographic s , prior knowledge, consumption habits , and other behavioral questions. Next, the experiment started, and participants had the chance to examine the di fferent products ( c onventional, USDA o rganic, c age f ree) featured on a display table. In addition, a captioned picture describing each of the egg types was read aloud and shown to participants (see Figure 1) . After that, bid or choices were made depending on the treatment participants were assigned to. 14 We randomly assigned participants to groups upon arrival in the research area. Small group size helps to avoid the disengagement of off - margin bidders from the auction procedure (Shogren et al., 2001). In addition, there is evidence from recent theoretical (Banerji and Gupta, 2014) and empirical studies (Rosato and Tymula, 2016) that the equilibrium bid is lower when the number of bidders is larger and that can potentially lead to a confo unding bidding effect. Moreover, the number of participant s in each group was kept constant to keep everyone equally engaged (Drichoutis et al., 2017). 19 Figure 1: Labels used in our study Sources: Priscilla ( 2013) , , Hurst ( 2016) After the completion of the experimental part of the study, participant s completed a second questionnaire including questions concerning animal welfare and environmental attitudes, risk preferences, involvement , and competitiveness. Both the questionnaires were implemented using tablets (iPad), while the auctions and the choice experiment were conducted o n paper . In what follows, we describe the experimental procedures for each of the elicitation methods. Expe rimental Auctions Participants that were randomly selected to participate in an EA, were first subjected to a hypothetical auction for four different candy bars to familiarize themselves with the procedure of each type of auction. Following the candy bar auction, consumers participated in an auction for each of the dozens of eggs. The following subparagraphs illustrate the experimental proc edures and steps followed to implement the three EAs: second price, r andom n th price, and BDM. 20 Second Price Auct ion The basic procedure for the implementation of the second price auction wa s as follows: - Step 1: A total of three rounds with 5 participants were conducted, one for each product: a ) the conventional eggs, b) the USDA organic eggs, and c) the cage free eggs . At the beginning of each round, the participant s received a bid sheet and were asked to simultaneously bid for the product up for auction in that round . T he bidding sheets for ea ch type of eggs were collected before the next ones were handed out. - Step 2: The experimenter rolled a four - sided die to determine which egg auction was binding. Importantly, all the auctions had an equally likely chance of being binding. - Step 3: The bids in the chosen auction were confidentially ranked from highest to low est. The person with the highest bid for the eggs purchased the eggs but, he/she paid the 2 nd highest bid for the eggs. - Step 4: For the chosen egg auction, the experimenter wrote the winning bidder(s) number and the price paid (second highest bid) on th e board for everyone to see. - Step 5: The winning bidder(s) came forward after the completion of the second questionnaire, paid the second highest bid and obtained the eggs. All other participants paid nothing and received nothing. Random n th price auction The basic procedure for the random nth price auction wa s as follows: - Step 1 and Step 2 for the random n th price auction were the same with the first two steps for the second price auction (mentioned above). - Step 3: The bids in the chosen auction we re then ranked from highest to lowest. Next, a random number (N) was drawn by rolling a die to determine how many participants will win the eggs. The random number (N) was somewhere between 2 and 5 (number of 21 participants). The N - 1 highest bidders in the b inding egg auction purchased the eggs and paid the nth highest bid. - Step 4: For the chosen egg auction, the experimenter wrote the winning bidder(s) number and the price paid (nth highest bid) on the board for everyone to see. - Step 5: The winning bidder(s ) came forward after the completion of the second questionnaire, paid the nth highest bid and obtained the eggs. All other participants paid nothing and received nothing . Becker - DeGroot - Marschak (BDM) auction The basic procedure for the BDM auction which was implemented on a 1 - 1 interview setting was as follows: - Step 1 and Step 2 for the BDM auction were exactly the same with the other two auctions, except from the fact that the BDM auction was implemented on a 1 - 1 private interview setting and not in groups of 5 people. - Step 3: The experimenter then rolled a 10 - sided die two times (one for the second decimal and one for the first decimal) and a 7 - sided die one time to determine a randomly drawn price between $0.00 and $6.00. If the bid for the binding eggs was greater than or equal to the randomly drawn price, the participant purchased the eggs and paid the randomly drawn price. If the bid for the binding eggs was less than the randomly drawn price, the participant paid nothing and received noth ing. - Step 4 : For the chosen eggs auction, the experimenter wrote the randomly drawn price (between $0.00 and $6.00) on the board. - Step 5 : The winning bidder came forward after the completion of the second questionnaire, paid the randomly drawn price and ob tained the eggs. 15 15 The bidder won the auction if his/her bid was higher than or equal to the randomly drawn price. In any other 22 Real Choice Experiment Participants who were randomly selected to participate in a RCE , were first subjected to a hypothetical choice experiment over a selection of four candy bars in order to familiarize themselves with the procedure. After that consumers participated in the egg RCE. The R CE closely follow ed protocols used in related studies (Lusk and Schroeder, 2006; Gracia et al., 2011 ; Bazzani et al. , 201 7) . During the experiment, our participants were faced with repeated choice question s , each represented by three alternatives: two types of eggs and a no - option, which was included to mirror what people experience in real shopping situations . 16 For each choice question, they were then asked to select their preferred one. Figure 2 provides a sample ( CE ) question. Figure 2 : Example of Choice Experiment Question The eggs ( dozen large, grade A, brown eggs ) were described by three attributes and the ir respective levels : price ($1.59, $2.59, $3.59, and $4.59) , USDA - organic label 16 Adamowicz et al. (1998) introduced the no - choice option to such frameworks given the fact that a no - buy option is a fundamental element of shopping/choice behavior. 23 (present/absent) , and cage free label (present/absent). Egg price levels were chosen to reflect the prices in local grocery stores and retail prices reported by the U.S. Department of Agriculture - at the time of the experiment . The number of the choice questions that were presen ted to the participants was determined by an optimal orthogonal in the differences (OOD) design developed by Street et al. (2001) . 17 Given the number of attributes and levels and using the generator (1, 1, 1), the design resulted in 8 choice questions , with a D - Optimality of 96.6% (Table 4 ) . T he 8 choice questions were then randomly divided in two blocks of 4 choice questions each. To alleviate any ordering effects, the order in which the choice tasks were presented was randomized. Table 4 : Prices in Discrete Choice Tasks Alternative 1 Alternative 2 Choice Set Price Cage Free USDA Organic Price Cage Free USDA Organic Block 1 1 $1.59 - - $2.59 2 $3.59 - $4.59 - 3 $2.59 - $3.59 - 4 $ 4 .59 $1.59 - - Block 2 5 $2.59 $3.59 6 $4.59 $1.59 7 $ 1 .59 $2.59 - - 8 $3.59 - $4.59 - The basic procedure for the R CE w as as follows: 17 The OOD is a special type of a sequential orthogonal design. The orthogonality of the design allows us to an optimal (D - optimal) design ensures that attributes common across alternatives never take the same level during the experiment (Burgers and Street, 2005; Street and Burgess, 2004; Street et al., 2001, 2005). In general, D - optimal designs are intended to maximize the attribute level differences. 24 - Step 1: The RCE was conducted with one participant at a time. Each participant received a choice sheet and he/she was faced with four choice questions, one at a time . F or each choice question, the participant was asked to select their preferred egg product at the listed price or the no - purchase option and record the choice o n the choice shee t. - Step 2: After the participant had finished responding to all four choice sets, the experimenter collected the choice sheet. - Step 3: T he experimenter rolled a four - sided die to determine which choice question was binding. That is, if a 1 is rolled, and the participant had chosen one of the two types of eggs in the first - choice question, he/she was given the product he/she selected and was a sked to - then he/she was not given any type of eggs and paid nothing. 25 4. EMPIRICAL MODELS AND SPECIFICATION Experimental a u ction model and s pecification In the EA s , participants simultaneously submitted bids for each one of the three types of eggs (i.e., conventional eggs; cage free eggs; and USDA organic eggs ). Using the bids collected from each auction mechanism (i.e., seco nd price auction; BDM auction; random n th price auction) we calculated both the marginal and total WTP estimates. The marginal WTP was for the other corresponding types of egg s . The total WTP coincides with the participant bids for each of the auctioned type s of eggs. Using the total and marginal WTPs, we then explore whether there exists a statistically significant difference in WTP elicited from the three different EA mechanisms: Failing to reject this hypothesis, we would conclude that there is statisti cal equality among the WTPs elicited from the three auction mechanisms . In order to test this hypothesis, we compared the total WTP and the marginal WTP for each egg type auctioned , from all auction mechanisms , by carrying out two traditional F - T est s (one - way Anova) , respectively . Our goal was to compare the means (i.e. , mean total and mean marginal WTPs) of all auction s . Subsequently, we estimated post - hoc pairwise comparisons which are described in detail later in this section. Consequently , we estimated three Tobit models (one for each type of auction). In EAs , bids are greater than or equal to zero for a ll of the auctioned goods (Gracia et al., 2011; Lusk and Schroeder, 2006) . As a result, the T obit model is one of the most widely used econometric models to analyze the WTPs elicited through EAs. Accordingly, to determine whether the 26 we estimated three random effects T obit models 18 . The T obit model, incorporati ng random effects can be expressed as follows : (1) where is the (auction) bid for the i th participant and the t th eggs type (conventional, cage free, USDA organic), which is detected only at non - negative values ; is a vector of independent variables including dummy variables identifying egg - type and sociodemographic characteristics ; is the (conformable) vector of coefficients ; is an individual specific disturbance for participant i ; and is the overall er ror term (Lusk et al. , 2004). 19 Finally, to check the robustness of the results from the three separate Tobit models, we estimate d three pooled T obit model s . In the first pooled Tobit model (Pooled Tobit 1) , w e includ ed the auction mechanisms and the products as in dependent (factor) variable s and the marginal WTP as the dependent variable. Th e second model (Pooled Tobit 2) adds to the first one by also including interaction terms between the auction mechanisms and the type of eggs. Finally, the third pooled Tobit model (Pooled Tobit 3) includes additional interaction terms between the demographic variables and the auction mechanisms. F ollowing the method used of the elicitation mechanisms ( the second price auction in our case ) and one of the types of the eggs ( the conventional in our case ) w ere treated as baseline . Real Choice Experiment (RCE) In RCEs, consumers make a discrete choice from a set of presented product alternatives , each represented by a number of att ributes with different levels, combined within choice sets. According to random utility theory, a given alternative within each choice set will be selected 18 We incorp orated random effects into the Tobit models in order to account for the panel nature of the data (i.e., each participant submitted multiple bids for different types of eggs). 19 In our analysis, we estimated two models based on equation (1) : a reduced one, which only includes the two labelled attributes indicator variables in the marginal WTP regression (model 1) and an extended one, which also includes the set of interaction terms between the type of the eggs and the selected sociodemographic variables (mod el 2). 27 if the perceived utility provided by such alternative is the highest among the alternative ones . ; he/ she can only observe the characteristics of the alternatives and the choice made by the individual. As previously mentioned, RCEs are consistent with the Random Utility Theory . T he utility that individual n derives from alternative j at choice situation t can be defined by a deterministic component and a stochastic component : (2) Different discrete choice models can be specified depending on the assumptions regarding the joint distribution of the vector of random error term s as well as the functional form of the deterministic portion of the utility function. As shown by Train (2009), a ssuming are distributed iid type I extreme value, the multinomial logit (MNL) specification results in: (3) where is the probability to choose alternative in the choice occasion by person . In addition to the independence from irrelevant alternatives (IIA) property (Ben - Akiva and Lerman, 1985; McFadden , 1986), the MNL imposes the restrictive assumption that the representative utility is deterministic, and the random terms are independently an d identically distributed ( iid, i.e., uncorrelated over alternatives). Importantly , it also assumes preference homogeneity in the sample, implying that all coefficients of the utility expression in equation ( 2 ) are the same across individuals (i.e., parame ters of are fixed) . Thus , if some heterogeneity is expected , the MNL specification is not sufficient for the purposes of our analysis. H ence , we estimated a mixed logit model (MXL) (Train 2009). Unlike the MNL, the MXL model allows for random state variation , unrestricted substitution patterns, and correlation in unobserved factors over time . In our RCE , participants 28 provide d a sequence of four choice responses. Hence, a panel data approach is used to allow for correlation among ind ividual preferences in a series of choice decisions ( four choice sets per participant in our case) . According to Train (2003), consider a sequence of observed choices i by individual n , one for each time period (i.e.: choice task) , conditional on , t he probability that individual n makes this sequence of choices is represented by the following joint probability : ( 4 ) The unconditional probability is the integral of this product over all values of in the space of the distribution: (5) Following Train (2003), the parameters of the model are estimated by simulated maximum likelihood estimation techniques. In this application, the utility in (2) was specified in WTP space (see Train and Weeks 2005; Scarpa, Thiene, and Train 2008) r ather than in preference - space to allow f or heterogeneity in the price coefficient ( Scarpa , Thiene, and Train 2008 ; Daly, Hess and Train 2012) . 20 In addition, utility in willingness to pay space provide s directly the WTP for each attribute in the design (i.e.: cage free and USDA organic) (Scarpa and Willis, 2010). Following the analysis from Train and Weeks (2005), Scarpa et al. (2008) and Bazzani et al. (2019), a ssuming the utility is linear in the parameters , the utility that individual n obtains from alternative j at choice situation t is specified as follows : ( 6 ) 20 To test the robustness of our results, we implemented utility specifications in preference space, too. The results from the estimation in preference space are presented in the Appendix C . 29 where is a random positive scalar representing the price/scale parameter; is the price (continuous) variable generated by the price levels in our experimental design; and are dummy variables for the cage free and the USD A organic attributes. They take a value of 1 when the label is present in the product, and 0 otherwise; and are the random coefficients of the estimated WTP values; and is the (random) error term which follows a Type I Extreme Value di stribution. The coefficients for the USDA organic and the cage free labels are assumed to be random following a normal distribution. 21 Comparison of Experimental Auction Data and Real Choice Experiment Data In an effort to compare the auction data with the RCE data, it must be highlighted that EA data are continuous in nature while the RCE data are discrete. In order to compare the two types of data, either auction data must be converted to simulated choice data or vice ver sa. F ollowing Lusk and Schroeder (2006), in this application we compared the WTP derived from the RPL model with our auction bids. Using the individual WTPs, we then explore whether there exist s a statistically significant difference in WTP elicited from the RCE and the three different EA mechanisms : Failing to reject this hypothesis, we would conclude that there is statistical equality among the WTP s elicited from th e se four methods . In order to test this hypothesis, we first need ed to estimate the marginal WTP for each label (i.e.: cage free and USDA organic) in the RCE and for each participant (i.e. . As previously discussed, while in the EAs the 21 We used NLOGIT 6.0 for the calculation of the WTP space model by setting the price coefficient to 1 and its standard deviation to 0. For our estimation, we used 500 Halton draws. We used Halton draws (Train, 2003) instead of random draws since the Halton draws provide a more efficient simulation procedure for the RPL (Train, 1999). 30 marginal WTPs can be simply derived by subtracting the bids for the conventional eggs from the other type of eggs, in the RCE , individual WTP is derived by implementing conditional inference . To illustrate, following the methods described by Train (2003), we have two ( probabilistic ) events: choice and taste . Let be the conditional distribution of taste for the participants who chose y , when faced with scenario x , and let denot e the population parameters. is drawn from the population distribution . T he distribution of h has a lower variance than the distribution of g, since the latter includes all the distributions for all possible choices. Hence, its use is preferabl e for statistical inference and comparisons like the one we are trying to implement. B : ( 7 ) Using this formula, we have the ability to derive various statistics conditional on the choices of the participants y (e.g . : mean , variance, and predicted choice probabilities) . In our case, we derived the i ndividual speci fi c WTP conditional on choice . For each individual i , we calculate d their WT P for label j . As we explained earlier, the derivation of WTP differs across preference space and WTP space. 22 After deriving the each attribute (USDA organic and Cage Free), we compared them with the respective auction marginal bids from all auction mechanisms . We did so by performing a traditional F - test (one - way A nova). 23 Subsequently, we estimated post - hoc pairwise comparisons. 24 W e implemented pairwise comparisons based upon the 22 23 We estimated the comparisons in Stata 14.0 statistical software . 24 Post - hoc pairwise c omparisons are commonly used when there are three or more levels of a factor (4 treatments in our case) (UCLA, 2019). The statistical software we used (Stata) has three built - ins pairwise methods (Sidak, Bonferroni and Scheffe) in the F - test command. Altho ugh these comparisons are easy to implement, these methods are considered to be too conservative for pairwise comparisons (UCLA, 2019; Hedayat and Kirk, 2006). Scheffe is the most conservative method of the three, followed, in order by Bonferroni and Sidak . 31 Studentized Range distribution. 25 Finally, to check the robustness of our results we estimated a pooled Tobit model applying the same procedures followed in the c omparison of the EAs. In the model, t he elicitation mechanisms and the products were enter ed as independent (factor) variables , while the marginal WTP w as the dependent variable . In the estimation, the second price auction was treated as baseline. 25 The IDRE Statistical Consulting Group (UCLA, 2019) has developed three different programs for the use of the three following methods: (1) the Tukey HSD, (2) the Tukey - Kramer and (3) the Fisher - Hayter respectively (Gleason, 2019). The three methods perf orm the same test statistic when the cell sizes (sample size of each treatment in our case) are equal but will differ when cell sizes are unequal. The Tukey - Kramer or the Fisher - Hayter tests are usually preferred when the cell sizes are unequal (UCLA, 2019 ; Hedayat and Kirk, 2006). 32 5. RESULTS Sample Characteristics For our experiments, we recruited consumers from the local community , rather than university students, in an effort to ensure that participant s were the primary shoppers of their households (Chang et al., 2009 , Gracia et al., 2011 ) . Table 5 shows the sociodemographic characteristics of the four sam ples used in the EA s and RCE. Most participants in our study were women ; in general , women have a higher participation rate in research studies and traditionally also make the food shopping decisions for the ir household. The participant age ranged from 1 8 to 7 8 years and the average household was comprised of two to three people. Around 34 % of the participants declared that they had a university degree and 26% indicated to have either completed a graduate degree or were a graduate student at the time of the research. Approximately 5 8 % of our participants had a n annual household net income lower than $50 , 0 00. The sociodemographic characteristics for all four samp les are similar . Specifically, the reciprocal tests ( chi - square test or AN O VA ) of equality between the sociodemographic variables in our four samples ( EAs and RCE ) were not rejected at the 5% significance level. Table 5 : Sample Characteristics (%, unless stated) Variable Definition Real Choice Experi ment Second Price Auction BDM Auction Rando m n th Price Auction p - Value Gender 0.89 9 0=male 29 .0 33.3 28.6 32.9 1=female 71 .0 66.7 71.4 67.1 Age 31. 8 33.4 30.6 28.4 0.1 71 Household size 2.7 2.3 2.4 2.3 0.20 1 Education 0. 257 High school 4 0.3 26.1 31.4 27.1 College 2 5.8 37.7 37.1 35.7 Postgraduate studies 2 2.6 30.4 22.9 28.6 Average household income 0. 803 Low income = less than $49,999 5 1.6 62.3 58.6 57.1 Medium income = between $49,999 and 99,999 2 5.8 17.4 20 14.3 33 High income = more than $100,000 1 4.5 11.6 12.9 15.7 Note: We conducted analysis of variance to test the equality between the sociodemographic variables within the 4 treatments; our null hypothesis was not rejected at the 5% significance level for the sociodemographic characteristics. In addition, to check the robustness of those tests, a non - parametric test such as the Kruskal W allis test by ranks was implemented. Results from the test indicate our null hypothesis was not rejected at the 5% significance level for all sociodemographic characteristics. Bids from Experimental Auction s Auction bids segregated by auction treatment and egg type are reported in Table 6 . Table 6: WTP Bids by Auction Mechanism and Egg Type Egg Type Second Price Auction BDM Auction Random n th Price Auction p - Value Conventional Eggs 0.363 Mean 0.867 1.064 0.926 Standard Deviation 0.831 0.718 0.945 Cage Free Eggs 0.004 Mean 1.148 1.726 1.304 Standard Deviation 1.031 1.032 1.081 USDA Organic Eggs 0.080 Mean 1.592 2.042 1.634 Standard Deviation 1.360 1.155 1.366 Note: We conducted a chi - square test or analysis of variance to test the equality between the auction bids within the three auction mechanisms; our null hypothesis was not rejected at the 5% significance level for the conventional and the USDA Organic Eggs. Our nul l hypothesis was rejected at the 10% significance level for the USDA organic eggs. Results indicate that p articipants were willing to pay on average $0.87 for a dozen of conventional eggs and $1.59 for a dozen USDA organic eggs in the second price auction. These values increased to $1.06 and $2.04 respectively, in the BDM auction. A s imple comparison of bids across the auction mechanisms show s that the second price auction bids are similar to random n th price auction bids, but lo wer than bids in the BDM auction. Implementing the Tukey - Kramer and the Fisher - Hayter post - hoc tests, we conclude that the BDM auction generates differences in means for the cage free eggs and the USDA organic eggs and that the second price and the random n th price auctions are not statistically different. 34 In an effort to implement a robust comparison of the auction bids from the three EA mechanisms, t he entire distribution of WTP s may be of interest (Lusk and Schroeder, 2006). Figures A1 through A3 in Appendix A display the inverse cumulative density functions of WTP for the conventional , Cage Free, and USDA organic eggs, respectively . The marginal WTP bids (presented in T able 7 ) are derived by subtracting the bids for the conventional e ggs from the bids for the cage free eggs and the USDA organic eggs for each participant , respectively . The results from the mean marginal WTP bids are consistent with the results from the mean total WTP bids . Table 7 : Marginal WTP Bids by Auction Mechanism and Egg Type Egg Type Second Price Auction BDM Auction Random n th Price Auction p - Value Cage Free Eggs 0.00 6 Mean 0 .281 0 .66 2 0 .378 Standard Deviation 0 .71 8 0 .79 5 0 .640 USDA Organic Eggs 0. 257 Mean 0 .725 0 .97 8 0 .70 8 Standard Deviation 1.194 1.04 8 0 .99 1 Note: We conducted a chi - square test or analysis of variance to test the equality between the auction bids within the t hree auction mechanism s ; our null hypothesis was not rejected at the 5% significance level for the USDA Organic Eggs. In addition, to che ck the robustness of those tests, a non - parametric test such as the Kruskal Wallis test by ranks was implemented. Results from the test indicate our null hypothesis was not rejected at the 5% significance level for all sociodemographic characteristics. T o determine whether the three auction mechanisms provide differences that are statistically significant , we estimated a random effects t obit model for each auction mechanism . Two models were specified : Model 1, in which the dependent variable is represented by the marginal WTPs from the different auctions and the independent variable is defined by the type of the eggs ( i.e ., cage free eggs , USDA o rganic eggs); and Model 2, which adds to Model 1 by also including the demographic variables as independent variables . Table 8 describes the 35 variables used in our auction data analysis , while Table 9 provides the estimates from models 1 and 2 . 26 Table 8 : Variables used in auction data analysis (Tobit model for each auction type and pooled Tobit models ) Variables Description D ependent Variable Marginal WTP The marginal auction bids for all type s of eggs Ind ependent Variables Product 2 =cage free eggs; 3=USDA Organic eggs Gender 1=female; 0=male Age Age in years Household size Total number of people leaving in the same house with the participant (including himself/herself) Education Years of education H ousehold income Total household a nnual income in dollars/10000 Table 9 : Tobit results for marginal WTP per auction mechanism Second Price Auction BDM Auction Random n th Price Auction Model 1 Model 2 Model 1 Model 2 Model 1 Model 2 Cage Free Eggs (CF) 0.281* * (0.11 5 ) 0.265** (0. 119 ) 0.662** (0.1 04 ) 0.667** (0 .113 ) 0.378** (0.10 2 ) 0.368** ( 0.114) USDA Organic Eggs (USDA) 0.725** (0.11 5 ) 0.687 ** (0. 119 ) 0.978** (0.1 04 ) 0.973** (0. 113 ) 0.708** (0.10 2 ) 0.721** ( 0.114) Female 0.15 2 (0. 150 ) 0. 237 (0. 157 ) 0.019 (0.125) Age 0.0 13 * * (0.00 5 ) 0.01 0* (0.00 6 ) - 0.01 0* (0.0 0 6 ) Education 0.0 67** (0.0 2 9 ) 0.0 23 (0.0 31 ) 0.0 49* (0.0 25 ) Household i ncome - 0.0 50 * * (0.0 18 ) 0.03 4** (0.0 19 ) 0.0 16 (0.0 14 ) Household Size 0.0 71 (0.0 59 ) 0.0 03 (0. 0 62 ) - 0.006 (0.0 53 ) Log likelihood - 239.29 - 211.77 - 228.29 - 210.84 - 211 .69 - 18 7.33 N 69 63 70 64 70 61 Note: Standard errors are in the parentheses. * denotes statistically significant variables at the 10% level and ** denotes statistically significant variables at the 5%. 26 We did not use any censoring for the models estimating the marginal WTP since we had negative values created when we calculated the marginal bids. 36 The marginal WTP estimates from Model 1 are reported in the second , fourth and sixth column s of Table 9 . Results generally in dicate that consumers are willing to pay a price premium for both egg types, although the marginal WTP for the USDA o rganic eggs is higher than the one for cage free eggs in all EAs. Most notably , differences across EAs were found. More specifically, consistent with previous studies ( Shogren et al. 1994 ) , bids for both products from the BDM auction are higher than the bids from the se cond price auction and random n th price auctions. In addition , differences are also found between the second price and random n th price auction . To illustrate, the ordering s of bids for the various egg products are not consistent ; the second price auction yields higher bids for the USDA o rganic eggs but lower bids for the c age free eggs as compared to the random n th price auction counterpart bids . This finding is consistent with Lusk et al. (2004) , who also found inconsistencies acr oss bidding rounds and products . Turning to the marginal WTPs from Model 2 ( third , fifth and seventh column of Table 9), it can be seen that adding socio - demographics characteristics to Model 1 result s in small changes in the magnitude of the coefficients of the Cage Free and the USDA organic eggs. In addition , a s it can be seen from the table (above), age was the only demographic characteristic found to have a statistically significant effect on the marginal WTPs for Cage Free and USDA organic eggs in all type of auctions . Furthermore, Table 9 illustrates that e ducation had a statistically significant effec t on the marginal WTPs for Cage Free and USDA organic eggs in the second price and the random n th price auction , while household income had a statistically significant effect on the marginal WTPs (for Cage Free and USDA organic eggs) in the second price and the BDM auction. To illustrate, for each additional year of age, consumers are willing to pay on average 1 more cent for eggs (independent ly of the attributed characteristics of the eggs). Moreover, for each additional year of education , participants are willing to pay 7 more cents in the second price auction and 5 more cents in the random n th price auction , ceteris 37 paribus . Regarding the ho usehold income, an additional annual revenue of $10,000 would make participants to pay, ceteris paribus, 3 more cents for eggs in the BDM auction and 5 less cents in the second price auction. Subsequently , we estimated three pooled T obit model s (Po ol ed Model 1, Pooled Model 2, and Pooled Model 3) to determin e the effect of auction institution on bids . Pooled Model 1 uses the Model 1 specification (from Table 9) to explore the effect of EA on bids , ceteris paribus. Pooled Model 2 adds to Pooled Model 1 by also including the set of interaction terms between the type of the eggs and the type of the EAs as independent variables. Finally, Pooled Model 3 adds to Pooled Model 2 by also includ ing interaction terms between the type of the auction and the demographic variables as independent variables. Table 10 summarizes the results . Table 1 0 : The e ffect of Auction Institution on marginal WTP: Random Effects Pooled Tobit Estimates Independent Variables Marginal WTP Pooled Model 1 Pooled Model 2 Pooled Model 3 Cage Free Eggs (CF) 0. 441 * * (0. 062 ) 0.281** (0.107) 0.265** (0.115) USDA Organic Eggs (USDA) 0. 804 ** (0 .062 ) 0.725** (0.107) 0.687** (0.115) BDM Auction (BDM) 0.211** (0.091) 0.000 (0.126) 0.430 (0.676) Random n th price auction (RNP) 0.026 (0.091) 0.000 (0.126) 0.938 (0.673) CF*BDM 0.381** (0.151) 0.401** (0.162) USDA*BDM 0.253* (0.151) 0.286* (0.162) CF*BDM 0.097 (0.151) 0.103 (0.164) USDA*RNP - 0.018 (0.151) 0.034 (0.164) Female 0.152 (0.143) BDM*Female 0.085 (0.206) RNP*Female - 0.134 (0.202) Age 0.013** (0.005) 38 BDM*Age - 0.003 (0.007) RNP*Age - 0.023** (0.008) Education 0.067** (0.028) BDM*Education - 0.044 (0.041) RNP*Education - 0.018 (0.040) Household income - 0.050** (0.018) BDM*Household income 0.088** (0.025) RNP*Household income 0.065** (0.024) Household size 0.071 (0.056) BDM*Household size - 0.068 (0.081) RNP*Household size - 0.077 (0.082) Log likelihood - 686.022 - 682.15 - 611.39 N 209 209 188 Note: Standard errors are in the parentheses. * denotes statistically significant variables at the 10 % level and ** denotes statistically significant variables at the 5 %. It is apparent from this table that w hen it comes to the type of eggs, in all models ( Pooled Model 1, Pooled Model 2, Pooled Model 3), the coefficients of the Cage Free and the USDA organic eggs are of positive magnitude and statistically significant. Further, the coefficient for the USDA organic eggs is higher than the coeffici ent of the Cage Free eggs in every case . This indicates that consumers are always willing to pay higher for the USDA organic eggs compared to the Cage Free eggs. However , differences were found across auction mechanisms. To illustrate, the results of Pooled Model 1 indicate that the coefficient of the BDM auction variable is positive and statistically significant ( 0.211 ), while the coefficient of the random n th price auction , although positive, is no t statistically significant. This evidence suggests that, on average, marginal WTP s from BDM are higher than the second price auction bids , while the second price and random 39 n th price auctions produce similar marginal WTP . This result is consistent with Shogren et al. (1994), who also found no statistically significant differences in mean (total) WTP estimates between second price and random n th price auctions . A potential explanation for th is equivalence could be the similarities in design those two elicitation methods have (see section 2.1, Table 1). In Pooled Model 2 we also includ ed the interaction terms between auction mechanisms and type of eggs . When we include an in teraction term, t he coefficients of the original variables can be tricky to interpret (Wooldridge, 2016). For example, in the Pooled Model 2 , the coefficients of the auction mechanisms are now interpreted as the effect of auction mechanism on marginal WTP when the product is zero. This effect is not of interest at any case. 27 Turning to the effects on consumer valuation for eggs across EAs, it can be seen that the coefficients of the interaction terms of the BDM auction with both types of eggs (USDA organic and Cage F ree) are positive and statistically significant suggesting that the marginal WTP for both type s of eggs is higher when the auction mechanism implemented is the BDM. On the other hand, the coefficients of the interaction terms of the random n th price auction with both types of eggs are not statistically significant. Hence, we can conclude that the random n th price auc tion does not have a differential effect on the marginal WTP compared to the second price auction ( for e ither type of eggs ) . Differences between BDM auction and the other two EA mechanisms (second price and random n th price) might be explained by the lack of peer pressure or competition during the BDM auction. 28 For instance, recent stud y by Rosato and Tymula (2019) suggests that in homegrown value auctions bids decline with increased competition. 29 27 We are interested in the value of the coefficients when the product=2 (Cage F ree eggs) and when the product=3 (USDA o rganic eggs). Hence, the fact that the coefficients of the auction mechanisms in Pooled Model 2 take value of zero is not of interest for our analysis. 28 To illustrate, while participants in the BDM auction bid indi vidually, in the second and the random nth price auction they submit bids in groups of five people. 29 The authors conducted an experiment where participants bid ed in multiple second price auctions for real objects and induced value items , with each auction having a different number of bidders . 40 The authors argued that this finding is consistent with loss aversion behavior ; w ith an increase in the number of bidders, participants perception is that there is a lower chance of winning and the y interpret that as a loss , and hence their willingness to pay gets reduced . 30 In Pooled Model 3 , when additional to the interaction terms included in Pooled Model 2, we included interaction terms between the type of auction mechanism s and the demographic characteristics , w e observed that age, education and household income have a statistically significant effect on marginal WTPs , ceteris paribus. 31 In addition, our results reveal that the random n th price auction has differential effect on age and household income while the BD M auction has a differential effect on income (i.e.: all three coefficients are positive and statistically significant). Real Choice Experiment The estimation results from the mixed logit model specified in WTP space are reported in Table 1 1 . 32 Table 1 1 : Estimates for the Mixed Logit Model in WTP space Parameters Estimates Cage Free Eggs (CF) 0. 270 (0.3 76 ) USDA Organic Eggs (USDA) 0.693 * * (0. 344 ) Price 1.0 No Buy - 2.45 3 ** (0.4 54 ) Standard deviations of parameter distributions 30 The study by Rosato and Tymula ( 2019) suggests that while in real - object auctions bids decrease with the increase of competition, in induced - value auctions, bids do not vary with the magnitude of competition. The authors conclude that participants may behave differently in homegrown auctions than in induced - value ones. 31 F or every additional year of age, participants were willing to pay on average 1.3 more cen ts for eggs; for every additional year of education, participants were willing to pay on average 6.7 more cents for eggs; and for every additional household income of $10,000, participants were willing to pay on average 5 less cents for eggs. 32 Other eco nometric models were estimated, all leading to a lower predictive performance. The results of the basic Multinomial specification are reported in Appendix B . The results of the Random Parameter Logit model with fixed price coefficient (RPL (FPC)) and the R andom Parameter Logit model with price following a triangular distribution (RPL (RPC)) are reported in Appendix C . 41 Table 11 Cage Free Eggs (CF) 0.891 * * (0. 356 ) USDA Organic Eggs (USDA) 1.485 ** (0. 396 ) Price 0.0 N 248 Log likelihood - 192.9 7 2 15 8.96 Pseudo - R 2 0.29 Note: Standard errors are in the parentheses. * denotes statistically significant variables at the 10 % level and ** denotes statistically significant variables at the 5 %. Results indicate that, on average, respondents are willing to pay a price premium of $0.693 for USDA organic eggs (dozen) , while the WTP premium for cage free ($0.270) is not statistically significant . The standard deviation of cage free is significant at the 5% significance level . This indicates that although participants , on average, ar e not willing to pay a price premium for cage free, a sub - group of consumers ha s a significant price premium for the product with the cage - free label . These results are consistent with the current status of the egg market in the US, where, according to (Lusk, 2018) choices impl y that half of consumers are willing to pay no more than a 30 cents/doz. premium for cage - free eggs examining the marginal effects of e gg a ttributes in c hanging m arket s hares ( for the control group), concluded that the existence of the USDA organic label would have higher impact than the addition of the cage free label. Our data also indicate that the USDA o rganic label produces a higher WTP estimate. 33 33 In addition, as noted by Train and Weeks (2005), the specification we used in WTP space provided lower and more realistic estimates for our two attributes compared with the specifications in preference space (see Appendix C ). 42 Comparing Real Choice and Experimental Auctions Willingness to Pay To compare the results from the RCE and EAs, we derived marginal WTP estimates from both the RCE and EAs institutions for all labelled characteristics analyzed. We derived the marginal WTP for each labelled characteristic in the EAs by subtracting the bids for the conventional eggs from the bids for the cage f ree and USDA organic eggs. Furthermore , in order to implement our comparison, we derived the individual WTP s from the RCE ( as described in S ection 4 ) . 34 Table 1 2 shows the results from the F - test we implemented for the comparison of the mean marginal WTPs from all four treatments. Table 1 2 : Mean m arginal WTP per e licitation m echanism Elicitation Mechanism Marginal Estimated WTP WTP Cage Free Real Choice Experiment 0.265 (0.503) Second Price Auction 0.281 (0.718) BDM Auction 0.662 (0.795) Random n th Price Auction 0.378 (0.640) F - Test 5.05 p - value 0.002** WTP USDA Organic Real Choice Experiment 0.702 (1.157) Second Price Auction 0.725 (1.194) BDM Auction 0.978 (1.048) Random n th Price Auction 0.708 (0.991) F - Test 1.02 p - value 0.385 Note: H 0 : WTP(RCE) = WTP(2PR) = WTP(BDM) = WTP(RNP), H 1 WTP(BDM) deviations of mean marginal WTP s are reported in the parentheses. * denotes statistically different WTP at the 10 % level and ** denotes statistically different WTP at the 5 %. In addition, to check the robustness of those tests, a non - parametric test such as the Kruskal Wallis test by ranks was implemented. Results from the test indicate our null hypothesis of equality of the means was rejected at the 5% significance level for t he Cage Free eggs. This finding confirms the results of the F - Tests implemented. 34 In Appendix D , we report the number of zero bids for all products per EA and the number of no - buys for all choice sets (per participant) in the RCE . 43 The WTP s (mean parameter) for the RCE are 0.265 for the cage free attribute, and 0.702 for the USDA organic attribute . As demonstrated in Table 1 2 , mean marginal WTP s under the four different elicitation procedures are positive but of different magnitudes. WTPs for the USDA organic eggs are almost the same for the second price auction, the random n th price auction and the RCE. WTPs for the cage free eggs are similar fo r the RCE and the second price auction and the WTP derived from the random n th price auction is slightly higher . In contrast, the BDM auction derives significantly higher WTP s than all other mechanisms for both attributes. Overall, t he mean marginal WTPs do not differ when equality F - Tests (ANOVA) are conducted for the USDA organic attribute but do differ for the c age free attribute. It is worth to note that w hile the marginal WTPs for USDA organic eggs are not statistically different across elicitation methods, results are useful for illustrative purposes and could provide insights when used alongside the overall statistical analysis we implemented (Lusk, 2003). Taken together these results, alongside with the results from the post - hoc tests , indicate that the BDM auction is the on ly mechanism that generates differences in means and, only for the c age free attribute . Furthermore, the second price auction, the random n th price auction and the RCE do not provide statistically different means at the 5% level for either of the two attributes ; USDA organic and cage free . Hence, our null hypothesis ( WTP RCE = WTP EAs ) is rejected for the c age free eggs but not for the USDA organic eggs. Subsequently, to check the robustness of the results from the F - Test and to determine the effect of elicitation mechanism on marginal WTP we estimated a pooled Tobit model . Similar to the comparison of the EAs, we used the products (i.e. cage free and USDA organic) and the treatments (i.e. all EAs and the RCE) as independent variables and the marginal WTP as the dependent variable. The results of the pooled Tobit model for all elicitation methods are summarized in Table 1 3 . 44 Table 1 3 : The e ffect of i nstituti on on marginal WTP: Random Effects Pooled Tobit Estimates Independent Variables Marginal WTP Cage Free Eggs (CF) 0. 4 01 * * (0. 0 55 ) USDA Organic Eggs (USDA) 0. 781 ** (0 .0 55 ) BDM Auction 0.211** (0.0 89 ) Random n th price auction 0.026 (0.0 89 ) Real Choice Experiment - 0.013 (0.092) Log likelihood - 887.643 N 2 71 Note: Standard errors are in the parentheses. * denotes statistically significant variables at the 10% level and ** denotes statistically significant variables at the 5%. Consistent with the results from the EAs, holding the type of eggs constant, the BDM coefficient is posit ive and statistically significant (0.211). On the other hand, the coefficient of the random n th price auction is no t statistically significant . Moreover , the results indicate that the marginal WTP from the RCE is lower than the second price auction bids on average , ceteris paribus . These findings are consistent with some of the findings of Gracia et al. (2011), where for some of the attributes the RCE yielded lower estimates than the EA. In general, we can conclude that BDM bids are higher than the s econd price auction bids on average. Furthermore, e stimates from the second price auction, the random n th price auction and the RCE were found to be statistically equivalent . Interestingly, t hese results contradict our earlier finding from the implementation of F - tests, where the mean marginal WTPs d id not differ across the four elicitation methods for the USDA organic attribute but d id differ for the cage free attribute . Th is inconsistency in results derived from the two different tests (i.e.: F - Test and Pooled Tobit Models) could be associated with the reduction in sample size that occurs when splitting the data by egg type. 45 Overall , c age f ree eggs are found to be valued $0.40 more than the conventional eggs , on average, while the USDA organic eggs have a $0.80 WTP premium when compared with conventional eggs, ceteris paribus. 46 6. DISCUSSION AND CONCLUSION S In recent years, there has been a growing interest in utilizing non - hypothetical elicitation methods such as EAs and RCEs for economic research . However, it is still unclear whether these experimental methods provide consistent welfare estimates such as WTP values. To the best of our knowledge, only three studies have compared valuati ons from EAs with those from RCE s ( Lusk and Schroeder, 2006 ; Shi et al., 2018; Gracia et al., 2011). However, the authors compare RCEs with only one type of incentive compatible auction mechanism ( second price auction, BDM auction, and random n th price auction respectively ) . Our study contributes to the existing literature by examining whether and how valuations from RCE s differ from three different EAs commonly used in food choice literature : seco n d price, Becker DeGroot Marschak (BDM) , and random n th price . In addition, in contrast with the previous studies where the RCE was consisted of a large number of choice tasks while the EA was conducted in a small number of rounds, in our study we implement ed 4 choice tasks for the RCE and 3 rounds (one for each type of eggs) for each EA (see Table 3) . This could make our comparison across elicitation methods more robust. Overall, our findings indicate that the USDA organic label is valued more highly tha n the C age F ree label and that preference rankings across these egg types remain consistent across elicitation methods. O ur results also suggest that the WTP values derived from the BDM auction are statistically different and higher than those derived from the other EAs (second price and random n th price) and the RCE when holding the type of eggs constant (Pooled Tobit Model s ) . When testing for eq uivalence of the elicitation methods within the egg types (through separate F - Tests), we found that the BDM marginal WTPs are statistically different from the other elicitation methods for the cage free eggs; in contrast, we found that all elicitation meth ods provide statistically equivalent marginal WTPs for USDA organic eggs. 47 Comparisons across the three auction mechanisms suggest that the BDM auction tend s to produce higher bids when comparing the three EAs holding the egg type constant, while the secon d price auction and the random n th price auction yield equivalent marginal WTPs. This finding could be attributed to the fact that the BDM auction has been found not to be incentive compatible (Horowitz, 2006). In addition, t he observed increase in WTPs could be attributed to the fact that participants individually revealed their preferences in the BDM auction, while they were in groups of five in the second and the random n th price auction . Hence , peer pressure or competition might have influenced their decisions in the second price and the random n th price auctions (Rosato and Tymula, 2019). Moreover, results from our models reveal that age, education, and household income have a statistically significant effect on marginal W TPs (in all three EAs) , ceteris paribus. When comparing WTPs across EAs and RCE, our results reveal that these methods derive equivalent WTPs for USDA organic eggs; while the BDM auction derive s higher ( and statistically significant ) WTPs for the C age F ree eggs. Focusing on the WTP estimates from the BDM auction and the RCE, our findings indicate that RCE yields lower WTPs, which are also consistent with the other two EAs ( second price and random n th price ). This evidence provide s further support for the hypothesis that the RCE better simulate s a market scenario with participants making purchasing decisions rather than submitting bids. It may be the case therefore that subjects might be less familiar with bidding and this might create a barrier in their e fforts to reveal their true preferences. Finding s from this study are of interest for researchers who utilize EAs and/or RCE methods. In this regard , our findings question the extended use of BDM auctions (especially in developing countries). RCE s c ould potentially provide estimates of WTP that are consistent with second price and random n th price auction mechanisms , while preserving the same logistic advantages of a BDM auction (no need to form groups of people to run the 48 experiments, individual decision - making) . Furthermore, RCEs have the advantage of simulating an actual market scenario (with posted prices), which is very familiar to consumers (in contrast with the BDM auction setting which is new to mos t participants) . However, with a small sample size, caution must be applied, as the findings might not be extrapolated to all studies implementing BDM auctions . O ur findings are also relevant for marketing teams in food production, retail companies , and po licy makers. Specifically, we found that consumers are willing to pay a higher premium for USDA organic eggs, which indicates that the use of the production methods underlined by the USDA is welcomed by consumers. The generalizability of our results is su bject to certain limitations. For instance, we had a relatively small sample size (similar limitation with the previous studies comparing RCEs with EAs) . To illustrate, all four elicitation methods had a sub - sample of equal to or less than 70 participants each. Having a larger sample size might provide more accurate mean values, easier identification of outliers that could skew the data and a smaller margin of error. Sample size limitations are especially evident when we make comparisons of the elicitation methods over the data that are split by egg type. Finally, f urther research should be undertaken to check the robustness of the results by using scanner data and/ or induced value experiments. This would be a necessary step to reach a concrete conclusion on whether welfare estimates from EAs are different from those elicited from RCE. 49 APPENDICES 50 Appendix A: Inverse Cumulative Density Functions (CDF s ) of WTP for the eggs Figure A1 : Distribution of Willingness - to - Pay for Conventional Eggs Figure A2 : Distribution of Willingness - to - Pay for Cage Free Eggs 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% $0.00 $1.00 $2.00 $3.00 $4.00 $5.00 Percent Willing to Pay Willingness - to - Pay for Conventional Eggs Second price auction BDM auction Random nth price auction 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% $0.00 $1.00 $2.00 $3.00 $4.00 $5.00 $6.00 Percent Willing to Pay Willingness - to - Pay for Cage Free Eggs Second price auction BDM auction Random nth price auction 51 Figure A3 : Distribution of Willingness - to - Pay for USDA Organic Eggs The figures above (CDF s ) can be interpreted as demand curves as long as we assum e that each participant only purchases ( consumes ) one dozen of eggs and, for each figure, no other egg alternative exists in the market (Lusk and Schroeder, 2006). For all three types of eggs , the dis tribution s of WTP implied from the BDM auction tends to lie above the WTP distributions implied from the second price and the random n th price auctions (see Figures A1 to A3). However, for the conventional eggs, and for price levels higher than $2.00 , the inverse CDF from the BDM auction lies exclusively to the left of the inverse CDF s from the second price and the random n th price auction (see Figure A1) . In addition, Figures A1 and A3 show that the inverse CDF s for the conventional and the USDA organic eggs from the second price and the random n th price auction s are virtually indistinguishable for all price levels. Nevertheless, as shown in Figure A2, the inverse CDF from the random n th price auction for Cage Free eg gs, lies on the right of the inverse CDF from the second price auction. Overall, these results indicate that the BDM auction derives on average, a higher WTP than the second price and the random n th price auctions for all types of eggs. In addition, we ca n conclude that the second price and the random n th price auctions derive similar WTPs, 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% $0.00 $2.00 $4.00 $6.00 $8.00 Percent Willing to Pay Willingness - to - Pay for USDA Organic Eggs Second price auction BDM auction Random nth price auction 52 except from the case of the Cage Free eggs, where the random n th price auction derives slightly higher WTPs than the second price auction. As noted by Lusk and Schroeder (200 6 ) , following the inverse CDF s, we could calculate optimal prices that would maximize profit given an assumed fixed marginal cost . In this case the profit is calculated by multiplying the number of participants with WTP greater tha n the price charged and the difference in the price and the ( assumed ) marginal cost. 53 Appendix B : Empirical Estimates for the MNL specification in the RCE Table B 1: Empirical Estimates for the RCE Parameters MNL Cage Free Eggs (CF) 0. 169 (0. 208 ) USDA Organic Eggs (USDA) 0.698** (0. 216 ) Price - 0 . 973 ** (0.11 9 ) No Buy - 1.587 ** (0.3 55 ) N 2 48 Log likelihood - 2 10.557 Note: Standard errors are in the parentheses. * denotes statistically significant variables at the 10 % level and ** denotes statistically significant variables at the 5 %. Table B 1 provides the multinomial logit results, which were rejected in favor of a random parameters logit (RPL) for the reasons we specified in chapter 4 . The RPL model s (Table C 1) uses the panel data structure of our data to take into account the fact that each individual made four choices (Gracia et al., 2011; Train 2003). 54 Appendi x C : Empirical Estimates for the RCE in preference space Table C 1 : Empirical Estimates for the RCE in preference space Parameters RPL (FPC) RPL (RPC) Cage Free Eggs (CF) 0.391 (0. 365 ) 0.407 (0. 321 ) USDA Organic Eggs (USDA) 1.001* * (0. 436 ) 1.245 * * (0. 412 ) Price - 1. 428 ** ( 0.204 ) - 2. 077 ** (0.3 55 ) No Buy - 2.454 ** (0. 487 ) - 3.657 ** (0. 681 ) Standard deviations of parameter distributions Cage Free Eggs (CF) 1.267* * ( 0.496 ) 0. 768 (0. 640 ) USDA Organic Eggs (USDA) 2.129 ** (0. 509 ) 1.629** (0. 561 ) Price 2.077** (0.355) N 2 48 2 48 Log likelihood - 192.95 - 183.39 2 1 59.02 178.13 Pseudo - R 2 0.2 9 0.3 3 Note: Standard errors are in the parentheses. * denotes statistically significant variables at the 10 % level and ** denotes statistically significant variables at the 5 %. In the second column of the results presented in table C 1 , we assumed that the coefficients for the two the price coefficient is fixed. In the third column, we assumed that the coefficients for the two labelled characteristic coefficient is random (RPC) following a triangular distribution. The two labelled characteristics were c age f ree and USDA o rganic . 55 Appendix D : Number of all zero bids and all no buy option s Table D 1: P articipants wi th all zero bids/all no - buys Experimental treatment Number of zero bids/all no - buys Total number of participants Percentage of zero bids/no buys Second Price Auction 8 69 12% BDM Auction 5 70 7% Random n th Price Auction 7 70 10% Real Choice Experiment 15 62 35 24% 35 Originally, 69 people participated in our RCE treatment. In the beginning of our analysis we dropped 7 observations due to the fact that they followed an irrational pattern during the course of their choices and hence they were considered to not engage wi th the experiment. 56 Appendix E : Instructions for the elicitation methods Instructions to Participant s: Treatment 1, Homegrown Value Second Price Auction Introduction Today you will be participating in an auction in which you will bid to potentially purchase a product (eggs). You will be shown four different types of eggs. These eggs will differ in terms of labelled characteristics. In order to purchase the eggs, you wi ll have to bid for them. How you bid will be explained soon. First, a quick overview of what follows. ( i) We will describe the auction and implement a practice round with four candy bars to help you better understand the mechanism (this round will not co be bidding for. (iii) real auction bids on the bidding sheets which will be given to you later. (i v ) The outcomes of the auction will be determined. As you may have noticed, there are five participants in this room. You will be competing with these participants when participating in the egg auctions. In the end, only one of the four egg auctions will be randomly selected to count or be binding. If you are the winning bidder in the binding auction you will purchase the eggs. You will bid for each type of eggs. Once you enter bids for all four egg auctions, one of them will be randomly selected as binding. The person with the highest bid in the binding eggs auction will purchase the eggs BUT, he/she will NOT pay what they bid, but will pay the 2 nd highest bid. Practice Round Step 1: First, each of you will receive a bid sheet for the practice round with the candy bars. The practice bid sheet includes places to enter your four bids, one for each candy bar. On the bid sheet, you will enter the most you are willing to pay for each of the candy bars you see in the pictures in the center of the room. Note: You will write four bids, one for each candy bar. Your bids are private information and should not be shared with anyone. [I will show them one by one the 4 candy bars; they will write their bids in the practice sheet.] Step 2: After you have finished writing your bids, I will go around the room and collect your bid sheets. Step 3: We will then roll a four - sided die to determine which candy bar auction is binding. For example, if a 1 is rolled, only the auction with candy bar No 1 will count and all other auctions and bids will be ignored. Importantly, all the auctions have an equally likely chance of being binding. Note: since this is the hypothetical example, the highest bidder will not pay and will not receive the candy bar. 57 Step 4: The bids in the chosen auction will then be ranked from highest to lowest. The person with the highest bid for the candy bar will purchase the candy bar BUT, he/she will pay the 2 nd highest bid for the candy bar. [I will do the ranking privately.] Step 5: For the chosen practice aucti on, we will write the winning bidder(s) number and the price paid (second highest bid) on the board for everyone to see. Step 6 : If this practice auction was real, the highest bidder would come forward and pay the 2 nd highest bid and obtain the candy bar. All other participants will pay nothing and receive nothing. Real Auctions Now that we are done with the practice, we will begin the real auctions. Here in the front of the room, we have four types of eggs: Convent ional, USDA Organic, Cage Free and USDA Organic & Cage Free. Other than differences in these characteristics, the eggs are of similar size, color, etc. We will now conduct an auction for each type of egg. The auction mechanism will be the same as in the p ractice round, except that you will write bids one at a time and we will collect the bid after each auction. Recall that for each bid, you should indicate the most you are willing to pay for one dozen of that type of egg. To refresh your memory as to how the auction works, I will go through the instructions again. Step 1: You will receive a bid sheet (for one of the types of eggs). On the bid sheet you will enter the most you are willing to pay for one dozen of that type of egg. You will bid for each of the following: a) the conventional eggs, b) the USDA Organic eggs, c) the Cage Free eggs and d) the USDA Organic & Cage Free eggs. You will get one bidding sheet at a time, they will be collected before the next one is handed out. Your bids are private in formation and should not be shared with anyone. Important Notes Because we randomly choose only one auction to be binding, you cannot purchase more than one type of eggs. That is, under no bidding scenario will you take home more than one dozen eggs. If there is a tie (more than one participant bids the same highest bid), then all of the highest bidders will pay the next (or 2 nd ) highest bid and purchase the eggs. The highest bidder(s) will actually pay money for the eggs. This set of auctions is not hypothetical, and you cannot make changes. In this type of auction, the best strategy is to bid exactly what each dozen of eggs is worth to you. Consider the following: if you bid more than the eggs are worth to yo u, you may end up having to buy eggs for more than you really want to pay. Conversely, if you bid less than the eggs are really worth to you, you may end up not winning the auction even though you could have bought a dozen of eggs at a price you were actua lly willing to pay. Thus, your best strategy is to bid exactly what the dozen eggs is worth to you. 58 It is acceptable to bid $0.00 for any type of eggs. Step 2: After you have finished writing your bids for the first type of eggs, I will go around the ro om and collect your bid sheet. This procedure will be repeated 4 times; once for each type of eggs. [Here I will give them one - by - one the bidding sheets and they will bid for each type of eggs.] Step 3: We will then roll a four - sided die to determine which egg auction is binding (either the Conventional, USDA Organic, Cage Free, USDA Organic and Cage Free). For example, if a 1 is rolled, only the first auction (conventional in our case) will count and all other auctions and bids will be ignored. Importantly, all the auctions have an equally likely chance of being binding. Step 4: The bids in the chosen auction will then be ranked from highest to lowest. The person with the highest bid for the eggs will p urchase the eggs BUT, he/she will pay the 2 nd highest bid for the eggs. [I will do the ranking privately.] Step 5: For the chosen egg auction, we will write the winning bidder(s) number and the price paid (second highest bid) on the board for everyone to see. Step 6: The winning bidder will come forward and pay the 2 nd highest bid and obtain the eggs. All other participants will pay nothing and receive nothing. Instructions to Participant s: Treatment 2, Homegrown Value BDM Auction [Partici pants will read the instructions and implement the experiment step - by - step with me] Introduction Today you will be participating in an auction in which you will bid to potentially purchase a product (eggs). You will be shown four different types of eggs. These eggs will differ in terms of labelled characteristics. In order to purchase the eggs, you will have to bid for them. How you bid will be explained soon. First, a quick overview of what follows. ( i) We will describe the auction and implement a prac tice round with four candy bars to help you better understand the mechanism (this round be bidding for. (iii) real auction bids on the bid ding sheets which will be given to you later. (i v ) The outcomes of the auction will be determined. For each of you, this is a decision - other people. You will bid against a randomly chosen price f rom a uniform distribution on the interval from $0.00 to $6.00. You will bid for each type of egg. In the end, only one of the four egg auctions will be randomly selected to count or be binding. If you win the binding auction you will purchase the eggs. 59 On ce you enter bids for all four egg auctions, one of them will be randomly selected as binding. I will then randomly draw a price (from the uniform distribution between $0.00 and $6.00) for the binding auction. If your bid is equal to or greater than the randomly drawn price you will purchase the eggs BUT, you will NOT pay what you bided, but will pay the randomly drawn price. (If your bid is less than the randomly drawn price you will not purchase the eggs and you will pay nothing.) Pra ctice Round Step 1: First, each of you will receive a bid sheet for the practice round with the candy bars. The practice bid sheet includes places to enter your four bids, one for each candy bar. On the bid sheet, you will enter the most you are willing to pay for each of the candy bars you see in the pictures in the center of the room. Note: You will write four bids, one for each candy bar. Your bids are private information and should not be shared with anyone. [I will show them one by one the 4 candy bars; they will write their bids in the practice sheet.] Step 2: After you have finished writing your bids, I will go around the room and collect your bid sheets. Step 3: We will then roll a four - sided die to determine which candy bar auction is binding. For example, if a 1 is rolled, only the auction with candy bar No 1 will count and all other auctions and bids will be ignored. Importantly, all the auctions have an equally likely chance of being binding. Note: since this is the hypothetical example, you will not pay anything and will not receive any candy bar. Step 4: Then, we will roll a 10 - sided die two tim es (one for the second decimal and one for the - $0.00 and $4.00. If your bid for the candy bar is greater than or equal to the randomly drawn price, you will win the c andy bar practice auction BUT, you will pay the randomly drawn price, not your bid (unless they are the same) for the candy bar. Step 5: For the chosen practice auction, we will write the randomly drawn price (between $0.00 and $4.00) on the board. Step 6 : If this practice auction was real, the winning bidder would come forward and pay the randomly drawn price and obtain the candy bar. All other participants will pay nothing and receive nothing. Note: since this is the hypothetical example, the highest bidder will not pay and will not purch ase the candy bar. Real Auctions Now that we are done with the practice, we will begin the real auctions. Here in the front of the room, we have four types of eggs: Conventional, USDA Organic, Cage Free and USDA Organic & Cage Free. Other than differe nces in these characteristics, the eggs are of similar size, color, etc. 60 We will now conduct an auction for each type of egg. The auction mechanism will be the same as in the practice round, except that you will write bids one at a time and we will collec t the bid after each auction. Recall that for each bid, you should indicate the most you are willing to pay for one dozen of that type of egg. To refresh your memory as to how the auction works, I will go through the instructions again. Step 1: You wil l receive a bid sheet (for one of the types of eggs). On the bid sheet you will enter and the most you are willing to pay for one dozen of that type of egg. You will bid for each of the following: a) the conventional eggs, b) the USDA Organic eggs, c) the Cage Free eggs and d) the USDA Organic & Cage Free eggs. You will get one bidding sheet at a time, they will be collected before the next one is handed out. Your bids are private information and should not be shared with anyone. Important Notes Because we randomly choose only one auction to be binding, you cannot purchase more than one type of eggs. That is, under no bidding scenario will you take home more than one dozen eggs. If your bid is greater or equal than the randomly drawn price will ac tually pay money for the eggs. This set of auctions is not hypothetical, and you cannot make changes. In this type of auction, the best strategy is to bid exactly what each dozen of eggs is worth to you. Consider the following: if you bid more than the egg s are worth to you, you may end up having to buy eggs for more than you really want to pay. Conversely, if you bid less than the eggs are really worth to you, you may end up not winning the auction even though you could have bought a dozen of eggs at a pri ce you were actually willing to pay. Thus, your best strategy is to bid exactly what the dozen eggs is worth to you. It is acceptable to bid $0.00 for any type of eggs. Step 2: After you have finished writing your bids for the first type of eggs, I will go around the room and collect your bid sheet. This procedure will be repeated 4 times; once for each type of eggs. [Here I will give them one - by - one the bidding sheets and they will bid for each type of eggs] Step 3: We will then roll a four - sided die to determine which type of eggs auction is binding (either the Conventional, USDA Organic, Cage Free, USDA Organic and Cage Free). For example, if a 1 is rolled, only the first au ction (conventional in our case) will count and all other auctions and bids will be ignored. Importantly, all the auctions have an equally likely chance of being binding. Step 4: We will then roll a 10 - sided die two times (one for the second decimal and one for the - $0.00 and $6.00. If your bid for the binding eggs is greater than or equal to the randomly drawn price, you will purchase the eggs BUT, you will pay the r andomly drawn price, not your bid (unless they are the same) for the eggs. 61 Step 5: For the chosen eggs auction, we will write the randomly drawn price (between $0.00 and $6.00) on the board. Step 6: The winning bidder will come forward and pay the rando mly drawn price and obtain the binding eggs. Instructions to Participant s: Treatment 3, Homegrown Value Random n th Price Auction [Participants will read the instructions and implement the experiment step - by - step with me] Introduction Today you will be participating in an auction in which you will bid to potentially purchase a product (eggs). You will be shown four different types of eggs. These eggs will differ in terms of labelled characteristics. In order to purchase the eggs, you wi ll have to bid for them. How you bid will be explained soon. First, a quick overview of what follows. ( i) We will describe the auction and implement a practice round with four candy bars to help you better understand the mechanism (this round will not co be bidding for. (iii) real auction bids on the bidding sheets which will be given to you later. (i v ) The outcomes of the auction will be determined. As you may have noticed, there are five participants in this room. You will be competing with these participants when participating in the egg auctions. In the end, only one of the four egg auctions will be randomly selected to count or be binding. If you are the winning bidder in the binding auction you will purchase the eggs. You will bid for each type of eggs. Once you enter bids for all four eggs auctions, one of them will be randomly selected as binding. The bids for the binding type of eggs will be ranked from highest to lowest. Next, a random number will be drawn to determine how many participants will win the binding egg auction. The random number will be between 2 and 5 (the number of participants). Call this random number N. The N - 1 highest bidd ers in the binding egg auction will purchase the eggs BUT, they will NOT pay what they bid, but will pay the Nth highest bid. For example, if the random number is a 3, then the 2 highest bidders would each purchase the eggs BUT pay the 3 rd highest bid for them. Practice Round Step 1: First, each of you will receive a bid sheet for the practice round with the candy bars. On the bid sheet, you will enter your ID number and the most you are willing to pay for each of the candy bars you see in the pictures in the center of the room. Note: You will write four bids, one for each candy bar. Your bids are private information and should not be shared with anyone. [I will show them one by one the 4 candy bars; they will write their bids in the practice sheet.] Step 2: After you have finished writing your bids, I will go around the room and collect your bid sheets. 62 Step 3: We will then roll a four - sided die to determine which candy bar auction is binding. For example, if a 1 is rolled, only the auction with cand y bar No 1 will count and all other auctions and bids will be ignored. Importantly, all the auctions have an equally likely chance of being binding. Note: since this is the hypothetical example, the highest bidder will not pay and will not receive the ca ndy bar. Step 4: The bids in the chosen auction will then be ranked from highest to lowest. Next, the random number will be drawn by rolling a 6 - sided die to determine how many participants will win the candy bar. As noted above, the random number (N) will be somewhere between 2 and 5 (number of participants), so if a one or six is rolled, we will roll again. The N - 1 highest bidders in the binding practice cand y bar auction will purchase the candy bar BUT, they will NOT pay what they bid (except in the unlikely event of a special tie as explained later) but will pay the Nth highest bid. [I will do the ranking privately. I will announce what is the N number (e. g.: 4 or 5).] Step 5: For the chosen practice auction, we will write the winning bidder(s) number and the price paid (Nth highest bid) on the board for everyone to see. Step 6 : If this practice auction was real, the N - 1 highest bidders would come forward and pay the Nth highest bid and obtain the candy bar. All other participants will pay nothing and receive nothing. Real Auctions Now that we are done with the practice, we will begin the real auctions. Here in the front of the room, we have four types of eggs: Conventional, USDA Organic, Cage Free and USDA Organic & Cage Free. Other than differences in these characteristics, the eggs are of similar size, color, etc. We will now conduct an auction for each type of egg. The auction mechanism will be the same as in the practice round, except that you will write bids one at a time and we will collect the bid after each auction. Recall that for each bid, you should indicate the most you are willing to pay for one doz en of that type of egg. To refresh your memory as to how the auction works, I will go through the instructions again. Step 1: You will receive a bid sheet (for one of the types of eggs). On the bid sheet you will enter your ID number and the most you a re willing to pay for one dozen of that type of egg. You will bid for each of the following: a) the conventional eggs, b) the USDA Organic eggs, c) the Cage Free eggs and d) the USDA Organic & Cage Free eggs. You will get one bidding sheet at a time, they will be collected before the next one is handed out. Your bids are private information and should not be shared with anyone. 63 Important Notes Because we randomly choose only one auction to be binding, you cannot purchase more than one type of eggs. That is, under no bidding scenario will you take home more than one dozen eggs. If there is a tie (more than one participant bids the same N - 1 th high est bid), then all the N - 1 th highest bidders will pay the N th highest bid and purchase the eggs. The only exception to this is if the tie is between the 4 th and 5 th highest bidders AND a 5 was rolled. In this case, everyone would buy and pay the tied bid, meaning that the 4 th and 5 th highest bidders would pay what they bid. The highest bidder(s) will actually pay money for the eggs. This set of auctions is not hypothetical, and you cannot make changes. In this type of auction, the best strategy is to bid e xactly what each dozen of eggs is worth to you. Consider the following: if you bid more than the eggs are worth to you, you may end up having to buy eggs for more than you really want to pay. Conversely, if you bid less than the eggs are really worth to yo u, you may end up not winning the auction even though you could have bought a dozen of eggs at a price you were actually willing to pay. Thus, your best strategy is to bid exactly what the dozen eggs is worth to you. It is acceptable to bid $0.00 for any type of eggs. Step 2: After you have finished writing your bids for the first type of eggs, I will go around the room and collect your bid sheet. This procedure will be repeated 4 times; once for each type of eggs. [Here I will give them one - by - one the bidding sheets and they will bid for each type of eggs] Step 3: We will then roll a four - sided dice to determine which type of eggs auction is binding (either the Conventional, USDA Organic, Cage Free, USDA Organic and Cage Free). For example, if a 1 is rolled, only the first a uction (conventional in our case) will count and all other auctions and bids will be ignored. Importantly, all the auctions have an equally likely chance of being binding. Step 4: The bids in the chosen auction will then be ranked from highest to lowest. Next, the random number will be drawn by rolling a 6 - sided die to determine how many participants will win the eggs. The random number (N) will be somewhere between 2 and 5 (number of participants), so if a one or six is rolled, we will roll again. The N - 1 highest bidders in the binding eggs auction will purchase the eggs BUT, they will NOT pay what they bid (except in the unlikely event of a special tie) but will pay the Nth highest bid. [I will do the ranking privately. I will announce what is the N nu mber (e.g.: 4 or 5).] Step 5: For the chosen egg auction, we will write the winning bidder(s) number and the price paid (Nth highest bid) on the board for everyone to see. Step 6: The winning bidder(s) will come forward and pay the Nth highest bid and ob tain the eggs. All other participants will pay nothing and receive nothing. 64 Instructions to Participant s: Treatment 4, Homegrown Value Real Choice Experiment [Participants will read the instructions and implement the experiment step - by - step with me] I ntroduction Today you will be participating in a choice experiment in which you will have a chance to purchase a product (eggs). You will be shown four different choice sets. Every choice set involves two different types of eggs and a no - purchase option. The eggs will differ in terms of labelled characteristics. In order to purchase the eggs, you will have to choose them. How you choose will be explained soon. First, a quick overview of what follows. ( i) We will describe the shopping scenario task and im plement a practice round with candy bars to help you better understand the mechanism (this round will not count for payment purposes). (ii) real choices on your choice sheet which will be given to you later. (i ii ) The outcomes of the task will be determined. You will make a choice for each of those four different shopping scenarios. In the end, only one of the shopping scenarios will be randomly selected, and this will be the one which will determine if you purchase eggs or not. Practice Round Step 1: First , you will receive a practice choice sheet. On the choice sheet, write your ID number. The practice choice sheet includes places to make four different shopping choices (Choice Question 1, etc.) based on four different sopping scenarios. Every choice scena rio involves two different candy bars and a no - purchase option. In each of those you will choose the candy bar you prefer to purchase given the listed prices. Alternatively, you may choose not to purchase either product. Please carefully examine each optio n before you make a decision and choose the product that you prefer most. [Note: your choices are private information and should not be shared with anyone. Given that the RCE practice round will be implemented with multiple participants.] Step 2: After y ou have finished responding to the four choice sets, I will collect your choice sheet. Step 3: After reviewing your choices, we will roll a four - sided die to determine which choice task is binding. For example, if a 1 is rolled, then the first - choice ques tion will be binding, etc. That is, if a 1 is rolled, and if you chose one of the two candy bars in that choice question, you will be given the product you selected and be asked to pay the price listed in the choice . If you - , then you will not be given any candy bar and you will pay nothing. It is important to understand that all 4 questions have the same chance of being selected as binding. Note: since this is the hypothetical example, you will not pay and will not purchase the candy bar. 65 Real Choice Tasks Now that we are done with the practice, we will begin the real choice tasks. Here in the front of the room, we have four types of eggs: Conventional, USDA Organic, Cage Free and USDA Organic & Cage Free. Ot her than differences in these characteristics, the eggs are of similar size, color, etc. For the real choices, you will be presented with 4 shopping scenarios . Each scenario involves two of the different types of eggs and a no - purchase option. The procedu res for making choices in this task are exactly the same as the candy bar practice round. To refresh your memory as to how the choice task works, I will go through the instructions again. Step 1: You will receive a choice sheet. On the choice sheet, wr ite your ID number. For each shopping scenario , please choose the type of eggs you would prefer to purchase given the listed prices. Alternatively, you may choose NOT to purchase any product. Please carefully examine each option before you make a decision and choose the product that you prefer most and indicate your choice on the choice sheet. It is important to understand that all 4 choice tasks have the same chance of being selected as binding in the end. Thus, you should consider each choice question as if it is the real chosen choice. Because of this, it is important that you answer the choi ce questions truthfully. If you do not, you might end up buying a product at a higher price than what you are willing to pay, or you might end up not being able to get the product when you would have actually been willing to buy it. Important Notes CHOO SE only one option for each scenario: one of the two types of eggs or not to purchase ASSUME that the options we will show you are the only ones available Once you have made your choice and moved to the next question you cannot go back The choices are all separate, so you do not and should not try to remember previous choices when making any particular new choice. In other words, we are asking you to treat each round of questions as separate from the previous one At the end of the experiment, we will choose a binding scenario and if you did not select not to purchase, you will be ASKED TO BUY one dozen of the eggs you picked in that scenario. Step 2: After you have finished responding to the four choice sets, I will collect your choice sheet. Step 3: After reviewing your choices to check they are completed correctly; I will roll a four - sided dice to determine which scenario will be binding. If a 1 is rolled, then the first - choice question will be binding, etc. That is, if a 1 is rolled, and if you chose one of the two types of eggs in that choice question, you will be given the product you selected and be asked to pay the price listed in the choice . If you - , then you will not be given any type of eggs and you will p ay nothing. 66 Appendix F : Consent f orm for participation in the study Consent Form This study is aimed at assessing your preferences for eggs and milk. You will be asked to complete a questionnaire about consumption habits, behavior and demographics, part icipate in an auction or a real choice experiment, and then complete an additional survey. This study will take about 30 - 45 minutes to complete. Risk and Benefits: There are no anticipated risks in participating in this study. Your participation will assi st in the advancement of knowledge of consumer choice behavior. In addition, you will be given $13 for your participation. During the experiment you will have the option to buy eggs. All the products offered are approved by the Food and Drug Administration . There are no additional risks to consumption of the eggs above those associated with similar purchases from traditional retailers. Voluntary Participation: Your participation in this research is completely voluntary. Confidentiality: All information will be kept confidential to the extent allowed by applicable experiment and records with identifiable, personal data will not be kept except to reco rd whether participants appeared for their assigned time slot and received their compensation. Right to Withdraw: You are free to withdraw from the study at any time or refuse to answer any questions. Questions, Concerns and Complaints: If you have any que stions or concerns about this study, you may contact Dr. Vincenzina Caputo (vcaputo@msu.edu). For questions or concerns about your rights as a research participant, please contact irb@msu.edu or +1 (517) 355 - 2180. Consent: I have read this consent form and my questions have been answered. I hereby give my voluntary consent to participate in this study. SIGNATURE DATE ____________________________________ _____________________ 67 REFERENCES 68 REFERENCES - based stated choice - National Oceanic Atmospheric Administration, Washington, 7 - 31 January. Alfnes, F., American Journal of Agricultural Economics 88(4):1050 - 1061. - 46. f loss aversion: 6(1):91 - 133. valuation for local versus organic food using a non - hypot hetical choice experiment: - 154. - 232. Ben - Akiva, M. tical Planning and Inference 134(1):288 - 301. Economic Behavior & Organization 1 45:335 - 351. - 57. Journal of Agricultural and Resource Economics 57(4):465 - 482. 69 Chang, J. Economics 91(2):518 - 534. ds: Between - subject and within - - 8. 8 - 457. to pay for food quality with experimental auctions: the case of yellow versus fortified - 16. De - Magistris, Economics 95(5):1136 - 1154. good things come in small packages? Bottle size effects on willingness to pay for pomegranate wine and grape - 104. with c and management 53(3):342 - 363. logit model: accounting for scale and coefficient heterogen 29(3):393 - 421. - 2016 - 57133):1 - 7. Gao, Z., and Schroeder, T. - to - - 809. Gleason, R. J. 2019. tukeyhsd, tkcomp, fhcomp: test statistics. UCLA: Statistical Consulting Group. Retri eved from https://stats.idre.ucla.edu/stata/ado/analysis/. 70 of Agricultural Econo mics 93(5):1358 - 1373. valuation method in the US, EU, and developing countries. Oxford University Press, pp. 302 441. willingness to pay e E: Logistics and Transportation Review 44(5):847 - 863. - DeGroot - Marschak mechanism is not necessarily incentive compatible, even for non - ics Letter 93(1):6 - 11. HOW CAN I DO POST - HOC PAIRWISE COMPARISONS USING STATA? 2019. Retrieved August 7, 2018, from https://stats.idre.ucla.edu/other/mult - pkg/faq/general/faq - how - do - i - cite - web - pages - and - programs - from - the - ucla - statistical - consulting - group/ Economics 88(4):1034 - 1049. Hurst, D. L. 2016. Let The Hens Out! Cage Free Eggs. Retrieved January 2, 2018, from https://www.dirt - to - dinner.com/let - the - hens - out - cage - free - eggs/ trials: Is the Vickrey auction demand - 269. 74(2):132 157. acceptance of controversial food technology: A cross - country investigation of - 19. 71 - Journal of Regulator y Economics 23(2):193 - 205. - 17. Lubben, B. D. 2005. A welfare analysis of country - of - ori gin labeling and alternative policy choices for beef. Kansas State University. Lucking - 1080. Lusk, J. L. 200 - to - American Journal of Agricultural Economics 85(4):840 - 856. Lusk, J. L . 2018. Market Potential for Cage Free Eggs. Retrieved February 21, 2019, from http://jaysonlusk.com/blog/2018/1/ 1/market - potential - for - cage - free - eggs 86(2):467 - 482. Lusk, J. L., and Schroede r, T. C Economic Analysis & Policy 6(1). - of - origin labeling on meat producers 185 - 205. Economics 86(2):389 - 405. Lusk, J. L., and Shogren, J. F. 2007. Experiment al Auctions. Cambridge J. Econ. - 711. 5(4):275 297. 72 McFadden, D. 1986 economics 3(4):303 - 328. applied Eco nometrics 15(5):447 - 470. Econometrica: Journal of the Econometric Society 1089 - 1122. d home - Environmental and Resource Economics 47(1):111 - 123. - to - pay: A comparison of the BDM mec psychology 25(6):725 - 741. Agricu ltural Economics 47(5):493 - 504. 36(2):318 - 324. veloping countries: implementation, challenges - 247. Priscilla. 2013. Our Cage Free Farms. Retrieved January 2, 2018, from http://cagefree.com.au/#!/a - new - kind - of - egg/ Rosato, A., and Tym - 208. 29(5):587 - 617. - International Journal of Game Theory 27(3):427 - 441. 73 address confou Journal of Agricultural Economics 90(4):994 - 1010. - to - pay for renewable energy: Primary and discretionary choice of British households' for micro - Economics 32(1):129 - 136. Research 4 8(1). - 362. sitivity and the 76(5):1089 - 1095. - Journal of economic behavior & organization 46(4):409 - 421. - Theory and Methods 30(10):2149 - 2171. - optimal pairs for th e estimation of effects in 2 - - 2):185 - 199. optimal and nearly optimal stated choice in marketing 22(4):459 - 470. Train, K. E. 2009. Discrete choice methods with simulation. Cambridge University Press. Tra in, K. E . 1999. Halton Sequences for Mixed Logit. Retrieved from http://elsa.berkeley.edu/~train. 74 - to - al and resource economics. Springer, Dordrecht, pp. 1 - 16. Policy 49:137 - 150. Van Loo, E. J., Caputo, V., Nayga Jr, R. M., Meullenet, J. F., and Ricke, S. C. 2011. - 613. Van Wezemael, L ., Caputo, V., Nayga Jr, R. M., Chryssochoidis, G., and Verbeke, W. 2014. - - 176. Vickrey, W. 1961. of Finance 16(1):8 - 37. - 241. Wooldr idge, J. M. 2016. Introductory econometrics: A modern approach. Nelson Education.