Journal of Agricultural and Applied Economics, 41,3(December 2009):713–730 2009 Southern Agricultural Economics Association

Consumer Preferences for Welfare Attributes: The Case of Gestation Crates

Glynn T. Tonsor, Nicole Olynk, and Christopher Wolf

Animal welfare concerns are having dramatic impacts on food and markets. Here we examine consumer preferences for products with a focus on use of gestation crates. We examine underlying consumer valuations of pork attributes while considering preference het- erogeneity as well as voluntary and legislative alternatives in producing -free pork. Our results suggest that prohibiting swine producers from using gestation crates fails to improve consumer welfare in the presence of a labeling scheme documenting voluntary dis- adoption of gestation crates. Consumers are found to implicitly associate at- tributes with smaller . Preference heterogeneity drives notably diverse consumer welfare impacts when pork produced with use of gestation crates is no longer available for consumption.

Key Words: animal welfare, consumer welfare, economics of legislation, gestation crates, pork, swine, voluntary labeling, willingness to pay

JEL Classifications: Q11, Q13, Q18

There is increasing consumer interest in practices impact the livelihood of . the production practices used in modern food Given this lack of concrete definitions and the production. Examples currently circulating inherent range of public perceptions and throughout the meat industry include consumer knowledge on farm animal livelihoods, it is interest to know whether and how antibiotics or hardly surprising that opinions vary regarding growth hormones were used, whether the prod- acceptability of current production practices. uct was produced ‘‘locally’’ or ‘‘on family A particular issue facing the U.S. swine in- farms,’’ and whether animals were handled in dustry is the possible elimination of production an ‘‘animal friendly manner.’’ Although we are practices deemed by some consumers to be an- unaware of current standardized definitions of imal unfriendly. In particular, consumer pres- ‘‘animal friendly,’’ ‘‘proper animal welfare,’’ or sure is mounting for the industry to no longer related terms, throughout this article such use gestation crates (also known as gestation phrases are used consistent with ongoing public stalls). Gestation crates are metal crates that discussions on the subject of how production house female breeding stock in individually confined areas during an animal’s four-month pregnancy. Pork producer organizations suggest Glynn T. Tonsor, assistant professor, Nicole Olynk, that use of these crates may facilitate more ef- graduate student, and Christopher Wolf, associate professor, Department of Agricultural, Food, and Re- ficient pork production resulting in lower prices source Economics, Michigan State University, East for consumers. The use of these crates is deemed Lansing, MI. as cruel to the animal by some consumer groups This research was partially funded by the Michi- as the crates limit animal mobility. This con- gan Animal Initiative. The authors thank Jayson Lusk for helpful comments on an earlier ver- sumer group perception has resulted in ballot sion of this manuscript. initiatives having been passed by residents of 714 Journal of Agricultural and Applied Economics, December 2009

Florida and Arizona that will ban the use of interested in the size or country of origin of the gestation crates in their state (Videras, 2006). In operation producing their food, then an evalua- November 2008, California residents passed a tion of preferences for ‘‘animal friendly’’products similar ballot initiative. Oregon was the first must take this into account. Furthermore, the state to ban gestation crates using legislature. In optimal response of both policy makers and the addition to these state-specific changes, food meat industry should reflect this implicit associ- retailers (i.e., McDonald’s and Burger King) ation if it exists. have responded by sourcing an expanding share The objectives of this study are to: (1) esti- of their food from animal wel fare friendly— mate consumer willingness-to-pay for alterna- meaning crate free—sources (Martin, 2007). tive pork production practice attributes including Not surprisingly, this growing consumer use of gestation crates; (2) examine whether interest in more knowledge of production these preferences are related to preferences for practices has led to an increase in research on farm size and country-of-origin attributes; (3) the underlying perceptions and preferences of evaluate whether or under what conditions ban- consumers, as well as the economic impact and ning use of gestation crates may be justified on viability of making corresponding adjustments grounds of economic welfare enhancement; and (Darby et al., 2008; Lusk, Norwood, and Pruitt, (4) identify the distribution of welfare impacts of 2006; Nilsson, Foster, and Lusk, 2006). How- gestation crate bans across consumers. Our ap- ever, as noted by Norwood, Lusk, and Prickett proach allows us to directly examine whether the (2007), the views of consumers in ‘the animal public good benefits of a ban on the use of welfare debate’ are basically absent. In partic- gestation crates outweigh the private loss stem- ular, a question yet to be addressed is whether ming from a reduction in selection of products. these legislative changes are welfare enhancing Specifically, we examine whether a gestation for the representative consumer. Moreover, the crate ban enhances consumer welfare given a distribution of consumer welfare effects is rel- labeling scheme was in place documenting the evant. Economic welfare evaluation is particu- use or absence of gestation crates in production. larly warranted as the desires of a population Mixed logit and latent class models are subset (e.g., ban supporters) may restrict the employed to investigate the extent of consumer food choice set of an entire population. For in- preference heterogeneity influencing conclu- stance, the November 2002 ballot initiative sions to these individual objectives. To the best banning gestation crates in Florida passed by a of our knowledge, no previous research has margin of 55 to 45% (Videras, 2006). Similarly, examined consumer preferences for alternative Proposition 2 passed (63–37%) in California pork production techniques, while controlling in November 2008 banning confinement not for farm size and country of origin preferences, allowing animals to turn around freely, lie down, in assessing valuations of gestation crate use as stand up, and fully extend their limbs (Humane well as voluntary and mandatory omission of Society of the United States, 2008). These ballot use. This study was designed to provide a better initiatives have implications for all consumers understanding of these issues, to improve fu- in Florida and California and these implications ture assessments and appropriateness of possi- likely are not equal across consumers differing ble adjustments in swine production practices, in pork and animal welfare preferences. and to identify consumer welfare impacts of Another unresolved issue relates to the ques- banning gestation crates. tion of underlying perceptions that consumers have in mind when stating a preference for a Prior Research change in animal welfare practices. In particular, when consumers reveal a preference for ‘‘more Several studies have investigated what consumers animal friendly practices,’’ do they implicitly as- are willing to pay to avoid or obtain various food sociate these products with smaller and/or do- attributes (Alfnes, 2004; Burton et al., 2001; mestic U.S. farms? This is an important question Grannis and Thilmany, 2002; Lusk, Roosen, and to address because if consumers are truly more Fox, 2003; McCluskey et al., 2003; Roosen, Tonsor, Olynk, and Wolf: Consumer Preferences for Animal Welfare Attributes 715

2003; Roosen, Lusk, and Fox, 2003; Tonsor et al., Research Design: Data Collection and 2005). A few studies have focused on consumer Choice Experiment valuations of ‘‘animal friendly’’ products (Carls- son, Frykblom, and Lagerkvist, 2007a, 2007b; This study uses a choice experiment to estimate Lijenstolpe, 2008; Lusk, Nilsson, and Foster, willingness-to-pay (WTP) for pork attributes. 2007; Nilsson, Foster, and Lusk, 2006). More- As in previous experimental research (i.e., Lusk, over, some have focused on how consumers value Norwood, and Pruitt, 2006), we collected in- use of antibiotics in pork production (Lusk, formation about consumer perceptions and Norwood, and Pruitt, 2006), factors impacting preferences via a survey of consumers from only brand premiums earned by meat products (Parcell one state, Michigan. The surveys were reviewed and Schroeder, 2007), determinants of poultry by pork industry representatives and animal prices (Parcell and Pierce, 2000), and the impacts science faculty, updated to reflect suggestions, of generic advertising on pork demand (Capps and then mailed to Michigan households iden- and Park, 2002). However, we are not aware of tified by SSI, a global market research company. any studies evaluating U.S. consumer preferences In November 2007, 1,000 surveys were mailed regarding the use of gestation crates. and followed by a postcard reminder 2 weeks Grethe (2007) notes that future costs of later. The final response rate was 26%, and after complying with animal welfare standards in the eliminating incomplete surveys, there were 205 may be substantial enough to surveys available for this analysis. spur a relocation of production to other coun- Given the controversial nature of animal tries. In the context of our analysis, this raises welfare issues and the use of gestation crates, we important questions for U.S. pork producers provided three different information statements and consumers alike. If the costs of complying in the survey discussing gestation crates to ex- with gestation crate legislation (coupled with amine if and how provision of information im- other associated regulatory pressures) lead to pacts consumer pork valuations (Fox, Hayes, and an increasing proportion of U.S. pork con- Shogren, 2002). Consumers randomly received sumption from imports, how would that impact one of three types of information: (1) Industry consumer perceptions and preferences for use Information,(2)Consumer Group Information, of gestation crates by U.S. pork producers? or (3) Base Information. Appendix A contains Carlsson, Frykblom, and Lagerkvist (2007c) copies of these three information treatments. present an appealing method for examining ex- In addition to socio-demographic informa- ternality effects of food production practices that tion about each respondent, meat consumption may supersede effects internalized by voluntary habits and a multitude of other factors were market adjustments and hence justify legislative collected. Each respondent also completed a bans. Product labeling enables consumers to in- choice experiment designed to determine the ternalize the private costs of production adjust- amount consumers were willing to pay for vari- ment expenditures. A legislative ban however ous pork attributes. Choice experiments simu- may be justified if public costs or other exter- late real-life purchasing situations and permit nalities exceed the loss in option values associated multiple attributes to be evaluated, thus allow- with restricting consumer choice sets. In their ing researchers to estimate tradeoffs among application to use GM fodder in Swedish meat different alternatives (Lusk, Roosen, and Fox, production, Carlsson, Frykblom, and Lagerkvist 2003). In this choice experiment, consumers (2007c) do not find support for the hypothesis that were presented with a set of eight simulated a ban on GM fodder would be welfare enhancing shopping scenarios, each of which involved in the presence of adequate labeling of meat choosing a preferred alternative from two pork produced voluntarily without using GM fodder. chops and a no purchase option. Our study uses a choice experiment designed to Boneless pork chops were offered at three directly examine whether a ban on gestation crate different price levels ($3.49/lb, $4.99/lb, $6.49/ use in the U.S. swine industry can be justified on lb) selected to be consistent with local retail grounds of consumer welfare enhancement. prices. The base price ($3.49/lb) reflected the 716 Journal of Agricultural and Applied Economics, December 2009 average price ($3.53/lb) for boneless pork country of origin (see Table 1). An orthogonal chops over the 1998–2007 period (United fractional design (Kuhfeld, Tobias, and Garratt, States Department of Labor, Bureau of Labor 1994) was used to select scenarios in which Statistics, 2008). Two price increases of $1.50/ prices are uncorrelated, and which lb were incorporated in the experimental design allowed for identification of own-price, cross- to reflect possible price premiums associated price, and alternative specific effects. This with the evaluated hypothetical products.1 In process also allowed the choice experiment to addition to price, the pork chop attributes be of reasonable size for survey participants. varied by farm size, production practice, and An example choice scenario is:

Pork Chop Attribute Option A Option B Option C Price ($/lb) $3.49 $6.49 Average farm size Large Small Production practice Labeled gestation crate-free Gestation crate ban Neither A nor B is preferred Country of origin United States Canada I choose ...

As in Lusk, Roosen, and Fox (2003), the choice in desirable attributes are not significantly dif- experiments were hypothetical in that they did ferent from nonhypothetical valuations. De- not include exchange of actual money or pork scriptions included in the choice experiments of products. However, our instructions specifi- the specific product attributes are included in cally stated ‘‘The experience from previous Appendix A. similar surveys is that people often state a We consider that pork products produced with higher willingness to pay than what one actu- and without gestation crates, by voluntary and ally is willing to pay for the good. It is impor- mandatory initiatives, is timely and appropriate. tant that you make your selections like you In particular, U.S. consumers currently live in an would if you were actually facing these choices environment characterized by partial banning of in your retail purchase decisions.’’ This state- gestation crates (e.g., Florida, Arizona, Oregon, ment was included as part of a ‘‘cheap-talk’’ California) and significant use of typical strategy at reducing hypothetical bias by Table 1. Pork Attributes and Attribute Levels informing survey participants of the concept Evaluated in Choice Experiments prior to conducting the choice experiment (Cummings and Taylor, 1999; Lusk, 2003). Product Attribute Attribute Label Furthermore, given that our principal interest Country of Origin United States is differences in marginal willingness-to-pay Canada amounts, we are less concerned with the hy- Brazil pothetical nature of our survey. This reassur- Production Practice Typical ance is based upon Lusk and Schroeder’s Labeled Gestation Crate- (2004) research, which suggests that hypo- Free thetical willingness-to-pay for marginal changes Gestation Crate Ban Size Small 1 The selection of prices can be challenging in Median designing choice experiments that often include hypo- Large thetical products without existing market prices (Lusk, Roosen, and Fox, 2003; Tonsor et al., 2005; Lusk, Price ($/lb) $3.49 Nilsson, and Foster, 2007). Although worthy of addi- $4.99 tional research, Hanley, Adamowicz, and Wright (2005) $6.49 and Ohler et al. (2000) found results from models based upon choice experiments not to be sensitive to the Note: See Appendix A for a description of the attributes selection of prices in the experimental design. provided to consumers. Tonsor, Olynk, and Wolf: Consumer Preferences for Animal Welfare Attributes 717

Table 2. Demographic Variables and Summary Statistics of Choice Experiment Participants Variable Definition Mean Gender 1 5 Female; 0 5 Male 0.35 Total Participants 205 Age Average age in years 55.6 Education (highest level completed) 1 5 Did not attend college 24.51% 2 5 Attended College, No Bachelor’s 32.81% (B.S. or B.A.) Degree 3 5 Bachelor’s (B.S. or B.A.) College Degree 18.18% 4 5 Graduate or Advanced Degree (M.S., Ph.D., 14.62% Law School) 5 5 Other 9.98% Household income 1 5 Less than $25,000 13.42% 2 5 $25,000 to $49,999 34.63% 3 5 $50,000 to $74,999 22.08% 4 5 $75,000 to $99,999 13.85% 5 5 $100,000 to $124,999 8.23% 6 5 $125,000 or more 7.79% Pork consumption frequency 1 5 4 or more times per week 4.76% 2 5 2–3 times per week 21.03% 3 5 Once per week 24.60% 4 5 2–3 times per month 29.76% 5 5 Once per month or less 15.87% 6 5 Never 3.97% production practices that may include gestation Research Methods: Random Parameters crate use. Moreover, recent research suggests Logit, Latent Class Models, and U.S. consumers understand that animal welfare is Willingness-To-Pay Analysis impacted by their shopping decisions (Norwood, 2007). As such, the selections required in this Choice experiments are based upon the assump- choice experiment are applicable and timely as tion that individual i receives utility (U)from the debate of whether to ban use of crates is yet to selecting option j in choice situation t. Utility is be settled nationally. represented by a deterministic [V(xijt)] and a sto- Summary statistics of selected demographic chastic component (eijt) and is specified here as: attributes of survey respondents are provided in (1) U 5 V x 1 , Table 2. Male respondents outnumbered female ijt ð ijtÞ eijt respondents and the average consumer was 56 where xijt is a vector of pork chop attributes and 2 years of age. The education and income dis- eijt is the stochastic error component iid over all tribution is roughly consistent with U.S. Census individuals, alternatives, and choice situations data (United States Census Bureau, 2006). (Revelt and Train, 1998). Alfnes (2004) points Nearly all respondents are at least occasional out that this describes a panel data model where pork consumers, with more than 50% con- the cross-sectional element is individual i and the suming pork at least once per week. time-series component is the t choice situations.3

2 While we expected a larger proportion of female respondents, our conclusions are not sensitive to esti- 3 Consequently, our model estimation procedures mating models with responses weighted by gender or are carried out in LIMDEP (Greene, 2002) utilizing the by incorporating a gender dummy variable. program’s panel data specification. 718 Journal of Agricultural and Applied Economics, December 2009

Our estimated models specify the system- factors over time, thus eliminating three limi- atic portion of the utility function as: tations of standard logit models (Train, 2003; Revelt and Train, 1998). In the context of our (2) V 5 a9P 1 b x 8 j 5 Option A, Option B, ijt ijt i jt study, the RPL is appealing as some of the pork (3) V 5 d 8 j 5 Option C, ijt chop alternatives presented in our choice ex- where Pijt is price and xjt is a 6 1 vector of periment are similar, possibly making the in- pork attributes (xjt 5 [Smalljt, Largejt, Crate dependence of irrelevant alternatives assump- Banjt, Labeled Crate Freejt, Canadajt, Braziljt]. tion overly restrictive. The RPL model also These pork attribute variables were effects facilitates correlation in random parameters coded relative to the omitted, base pork chop and hence a thorough evaluation of relation- originating from a Median sized, U.S. based ships in preferences across attributes. This facet operation using Typical production practices.4 is particularly valuable given our interest in the The remaining terms in Equations (2) and (3) are relationships between preferences for produc- a, bi,andd which are parameters to be estimated. tion practice attributes with other controlled The model described by Equations (1) to (3) attributes (i.e., farm size and country-of- may be estimated assuming homogeneous origin). preferences for the evaluated sample of con- Application of the general random utility of sumers or by allowing preference heterogene- Equation (1) in a random parameters logit ity. A growing amount of research suggests model can be presented as: consumers possess heterogeneous preferences, so employing a model that allows for and (4) Uijt 5 l9i xijt 1 eijt, evaluates preference heterogeneity is appro- priate (Alfnes, 2004; Alfnes and Rickertsen, where xijt is a vector of observed variables, li is 2003; Lusk, Roosen, and Fox, 2003; Tonsor unobserved for each individual and varies et al., 2005). Our analysis examines prefer- within the population with density f(liju*) ence heterogeneity by applying two alternative where u* are the true parameters of this dis- models, random parameters logit (also known tribution, and eijt is the stochastic error com- as mixed logit) and latent class logit models. ponent iid over all individuals, alternatives, and Random parameters logit (RPL) and latent choice situations (Revelt and Train, 1998). For class models (LCM) are both increasingly be- maximum likelihood estimation of the RPL ing used as they encompass logit models as- model we need to specify the probability of suming homogeneous preferences, in turn each individual’s sequence of selections. Let providing valuable insight into differential j(i,t) denote the alternative that individual i welfare effects on a sample of potentially dif- chose in period t. The unconditional probability ferentiated consumers. of subject i’s sequence of selections is given by We apply both models to examine sensi- (Revelt and Train, 1998): tivity of conclusions regarding consumer pork Z preferences and impacts of gestation crate Y l9i xijði;tÞt e (5) Piðu Þ 5 P f ðliju Þdli. bans to alterative model assumptions. The l9i xijt t e RPL model allows for random taste variation j within the surveyed population, is free of the independence of irrelevant alternatives as- In the RPL model we specify the price coeffi- sumption, and allows correlation in unobserved cient to be fixed and focus on heterogeneity in preferences for each of the six pork chop at-

tributes. We do this by allowing bi in Equation 4 That is, the six attributes in Equation (2) take on a (2) to vary within our consumer population. value of 1 when applicable, a value of 21 when the Prior to proceeding, it is important to note that base pork chop attribute applies, and zero otherwise. these random coefficients could be correlated Effects coding is used to avoid confoundment with the Opt Out coefficient (d) (Ouma, Abdulai, and Drucker, (Scarpa and DelGiudice, 2004; Train, 1998). 2007). For instance, consumers who are concerned Tonsor, Olynk, and Wolf: Consumer Preferences for Animal Welfare Attributes 719 with the use of gestation crates might also Estimated coefficients from random utility value pork from smaller operations. To inves- models themselves have little interpretive tigate these possibilities, we let b represent value. However, relative combinations of select the vector of attribute coefficients and specify coefficients provide economically meaningful b; Nðb,WÞ. The resulting coefficient vector is insights on consumer preferences. For exam- expressed as b 5 b 1 LMwhere L is a lower- ple, traditional calculations of WTP from RPL triangular Cholesky factor of W such that LL95 model coefficients are based on the mean of the

W, and M is a vector of independent standard normal distribution (e.g.,bSmall) and implicitly normal deviates (Revelt and Train, 1998). Upon ignore the distribution of preferences around estimation, evaluation of the individual ele- the mean (e.g., relevant elements of L). To relax ments in L allows for a better understanding of this strong assumption, as well as consider correlations in preferences across attributes statistical variability in parameter estimates, we evaluated.5 utilize simulation techniques consistent with While continuous heterogeneity is assumed those described by Rigby and Burton (2005) in RPL models, latent class models specify and Hensher and Greene (2003). preference heterogeneity to occur discretely To consider the entire preference distribu- (Train, 2003). More specifically, LCM models tion of WTP (rather than just the mean and assume that individuals can be intrinsically standard deviation) and consider statistical sorted into a number of latent classes where variability in parameter estimates, we use a each class is characterized by homogeneous two-step simulation approach. First, we let h be preferences, but preferences are heterogeneous the vector of model point estimates (e.g., indi- across classes (Boxall and Adamowicz, 2002). vidual elements of a, d, b, and L), s 5 var(h), LCM models simultaneously assign each indi- and T be the lower-triangular matrix of s such vidual into latent classes probabilistically while that TT95s. We then take 1,000 draws from a also identifying utility parameters of each la- standard normal distribution for each element tent class. Within a given class, individual of h and place them in a vector m. For each of choices from one choice situation to another these 1,000 draws, we identify estimates of the are assumed to be independent and choice model parameters as h 1 Tm. Secondly, for probabilities are assumed to be generated by each of the simulated 1,000 parameter values, the logit model (Greene, 2006). The probability 1,000 preference drawings from a standard that individual iselects option j in choice sit- normal distribution are made to generate a uation t, given that he belongs to latent class s, distribution of WTP estimates. This provides a is: series of 1,000 estimates for any desired sta-

YT tistic, facilitating identification of confidence expðB x Þ (6) P ðijtjsÞ 5 s ijt , intervals for each statistic (e.g., 95% confi- i PJ t51 expðBsxijtÞ dence intervals for mean WTP). This simula- j51 tion process makes more complete use of valuable information provided by estimated where x is a vector of observed attributes as- ijt random parameters logit models, and results in sociated with alternative j and B is a class- s a much more complete mapping of consumer specific utility parameter vector (Ouma, Abdu- preferences. lai, and Drucker, 2007). Willingness-to-pay estimates from the LCM model were derived specific to each class, ac- counting for different preference structures. 5 Furthermore, in our situation of multiple, corre- While simulated WTP estimates stemming lated random parameters, standard deviations of b are from the RPL model require examination of not independent (Hensher and Greene, 2003). For both statistical variation and variation in pref- proper assessment we utilize Cholesky decomposition erences, corresponding examinations from the to identify attribute-specific standard deviations (e.g., Crate Ban) and attribute-interaction standard devia- LCM model incorporate variation in class tions (e.g., Crate Ban Small). membership probability as well as statistical 720 Journal of Agricultural and Applied Economics, December 2009 variation in class-specific utility parameters. A gestation crate ban enhances consumer wel- distribution of 1,000 values of each WTP esti- fare given a labeling scheme was in place mate was generated using a bootstrapping documenting the use or absence of gestation procedure proposed by Krinsky and Robb crates in production. (1986). More specifically, 1,000 observations were drawn from a multivariate normal distri- Results bution parameterized by using the coefficients and variance terms estimated by the LCM An array of alternative model specifications model. were considered prior to selecting the random The simulated WTP statistics from each utility model described above with log likeli- model were utilized to empirically test for hood tests rejecting the hypothesis that prefer- differences in WTP preferences. First, mean ences are jointly homogeneous (e.g.,b 5 b in WTP estimates and 95% confidence intervals the RPL and bs 5 bt, "s 6¼ t in the LCM) were identified incorporating both statistical and the hypothesis that the random parameters and preference (class membership) variability of the RPL model were uncorrelated (e.g., the in the RPL (LCM) model. Second, a combi- off-diagonal elements of W are jointly zero). national technique suggested by Poe, Giraud, We estimated separate models for each of the and Loomis (2005) was used to provide a three information treatments (see Appendix A) simple nonparametric evaluation of differences and compared the sum of the log-likelihood in WTP distributions. The difference between functional values to values from a pooled two simulated WTP series was evaluated with model constraining coefficient equality across this difference being calculated for all possi- information treatments (but allowing relative ble combinations of the two series. In other scale variation). Consumer choice experiment words, 1,000,000 differences (e.g., WTPa 2 responses were found to be insensitive to the WTPb "a,b; where a 5 1,...,1000 and b 5 information treatment they received as we 1,...,1000) were calculated for each test. failed to reject the hypothesis that we can pool Our methodological approach allows us to observations across consumers receiving the directly examine if a state-wide ban prohibit- three alternative information statements.6 The ing the use of gestation crates can be eco- finding that information differences had no nomically justified. In particular, our choice effect on pork chop selections may stem from experiment contains three different attribute the similarity in the underlying point of all levels for gestation housing: Typical, Labeled three, intentionally brief information treat- Gestation Crate-Free,andGestation Crate ments, or in strong prior beliefs held by con- Ban. Instructions preceding the choice exper- sumers. Our finding of pork preferences to be iment inform survey participants that the La- insensitive to differences in information pre- beled Gestation Crate-Free attribute guaran- sentations is similar to that of Lusk, Norwood, tees pork to have been raised by a producer and Pruitt (2006). As an outcome of these who voluntarily chose not to use gestation findings, the remainder of this analysis reports crates while the Gestation Crate Ban attribute results from pooled models with identical pa- guarantees the pork to have originated from an rameters and scales across the three informa- animal raised in a region (state or country) tion treatments. where the use of gestation crates is legally Estimates to RPL and LCM models are banned for all swine producers. This is con- provided in Table 3. In the RPL model, the sistent with the approach of Carlsson, Fryk- majority of the estimated means for the ran- blom, and Lagerkvist (2007c) and allows us to dom pork chop attribute parameters were directly test if the public good benefits of a ban outweigh the private loss of option values 6 (reduction in selection of products if pork These tests were conducted allowing the scale parameter to vary across the pooled data sets when raised using gestation crates is completely estimating the pooled model. See Louviere, Hensher, banned). Specifically, we examine whether a and Swait (2000) for additional tests details. Tonsor, Olynk, and Wolf: Consumer Preferences for Animal Welfare Attributes 721

statistically significant. To further evaluate ’’ preference heterogeneity we examine esti- mated Cholesky matrices (Appendix B). The diagonal values of each Cholesky matrix rep-

Class 4 resent the true level of variance for each ran- 0.1351 (0.1257)

Ban Preferring dom parameter once the cross-correlated pa- 2 0.5173** (0.0656) 1.3740** (0.1446) 1.8749** (0.3392) ‘‘ 2 2 2 rameters terms have been unconfounded (Hensher, Rose, and Greene, 2006). This for-

’’ mulation is an important distinction in our RPL model application. For instance, five of the six random parameters were estimated to have statistically significant standard deviation pa-

Class 3 rameter estimates.7 However, only the diagonal 0.5281 (0.9929) 0.0881 (0.3507)0.7734 (0.4506)0.3335 (0.4276) 0.8110** (0.1245) 0.1663 (0.1170) 0.1199 (0.3769) 2 2 2 2 2 Small Gestation

0.9408** (0.2548) Cholesky elements for and Price Conscious ‘‘ 2 Crate Ban in our final model were statistically significant. This implies that the statistically ’’ significant standard deviation parameters for the Labeled Gestation Crate-Free, Canada, and Brazil variables were attributable to cross- correlations with other random parameters and

Class 2 not heterogeneity around the mean of each 0.3138 (0.3357) 0.1572 (0.1121) 0.4426 (0.3034) 0.0396 (0.1246) 2 2 random parameter (Hensher, Rose, and Greene, 0.3662** (0.0632) 0.6001** (0.1092)0.4105** (0.1293) 2.3232** (0.2208) 0.3756 (0.3566) 1.4453** (0.1276) Attribute Conscious 2 2 2 2 2006). ‘‘ The statistical significance of diagonal Cho- lesky elements for Small and Gestation Crate ’’ Ban is evidence of preference heterogeneity persisting, even after allowing cross-correlations to exist across attribute parameters. Examina-

Class 1 tion of the off-diagonal elements of the 0.4429* (0.1817) Pork Enjoyers Cholesky matrix reveals several statistically 9.0966** (0.8546) 1.3306** (0.1495) 0.6630** (0.1956) 1.9029** (0.2690) ‘‘ 2 2 2 2 2 significant estimates, primarily stemming from the Small coefficient. This suggests significant cross-correlations among the random parame-

1,052, respectively) were estimated using NLOGIT 4.0, with Halton draws, and 500 replications forter simulated probability. estimates would have been inappropriately 2 confused within standard deviation estimates

a of each random parameter without Cholesky matrix decomposition and evaluation. Evalua- 1,215 and Elements of the random parameter model’s Cholesky matrix, along with corresponding correlation statistics, are presented in Appendix B. Mean 2

a tion of the correlation terms reveals the Small variable to be positively correlated with the 1.9123** (0.3079) 0.7317** (0.0622) 3.4415** (0.3786) 0.4666** (0.1227) 2 2 2 2 Gestation Crate Ban and Labeled Gestation

Random Parameters Model LCM - 4 Class Model Crate-Free variables. This suggests that farm size attributes are closer substitutes for pro- duction practices than suggested by the

7 These standard deviations, while provided by

Parameters (standard errors) for Two Choice Models NLOGIT, are not presented. In the context of corre- lated random parameters, these standard deviation parameters are not independent and Cholesky decom- position should be used to identify proper standard Opt Out Price LabelCanadaBrazil 0.7695** (0.2319) 0.5115* (0.2037) 0.5499** (0.1886) 0.1954 (0.1937) 0.3286** (0.1178) Large Standard errors are presented in parentheses. Ban 0.1264 (0.2131) Membership ProbabilitySmall 0.2033 (0.1234) 0.3221** (0.0268) 0.3326** 0.2956 (0.0261) (0.1836) 0.1412** (0.0343) 0.1705 (0.1157) 0.2040** (0.0331) Table 3. Variable Notes: Presented models (log likelihoods of *, ** Indicates statistical significance at the 0.05deviation and 0.01 level, respectively. terms (Hensher, Rose, Greene, 2006). 722 Journal of Agricultural and Applied Economics, December 2009 nonstochastic portion of our RPL model ’’ 3.93] (Alfnes, 2004). 3.06] 2 The latent class model estimates are also 2 presented in Table 3. To identify the number of 7.19, 0.30, 1.57] 0.81, 1.11] 1.42, 0.46] latent classes to be used in the analysis, we 4.14, 2 2 2 2 2 employed the Bayesian Information Criterion Class 4 Ban Preferring

as discussed by Boxall and Adamowicz (2002). ‘‘

This criterion is minimized in a four-class ud, and Loomis (2005). All model, leading to the estimates presented 8 in Table 4. Incorporating class member- ’’ ship covariates (i.e., demographics, attitudi- nal information, and information treatment 3.51, 1.11] ($5.35) [ 4.03, 0.20] $0.64 [ 0.97, 2.30]1.68, 1.75] $5.62 [4.18, 7.41] $3.13 [2.08, 4.39] 0.50, 2.73] $0.17 [ 1.70, 1.69] ($0.52) [ 1.73, 3.43] ($3.62) [ 2 2 2 2 2 2 dummies) failed to improve the model’s sta- 2 tistical performance. This result is not neces- Class 3

sarily surprising and is consistent with several Price Conscious other applications of latent class models to ‘‘ consumer food preferences that have found

observable consumer characteristics to be poor ’’ 9.00] ($0.89) [ 0.83] ($1.68) [ indicators of food preferences (Nilsson, Foster, 1.99] $0.73 [ 2 2 and Lusk, 2006). 2

The LCM results reveal significant hetero- LCM - 4 Class Model 20.00, 4.05, 5.44, 2.32, 0.40] $0.98 [ 0.23, 2.60] ($0.21) [ 2.06, 1.49] ($0.23) [ 2 2 2 2 2 geneity in consumer preferences across latent 2 Class 2 classes with associated class membership probabilities of 32%, 33%, 14%, and 20%, re- Attribute Conscious spectively. That is, there is a 32%, 33%, 14%, ‘‘ and 20% probability that a randomly chosen respondent belongs to the first, second, third, ’’ 2.19] ($13.13) [ 0.45] ($3.39) [ 0.15] ($0.89) [ and fourth class, respectively (Nilsson, Foster, 6.35] ($0.72) [ 2 2 2 and Lusk, 2006). The first and fourth classes 2 have significant, negative coefficients on the 3.72, 0.28, 0.99] ($2.29) [ 1.58, 1.36, 0.07, 1.09] $0.99 [ 7.54, 2 2 2 2 2 ‘‘opt out’’ parameter indicating a preference to 2 Class 1 retain pork in their choice set. Utility coeffi- Pork Enjoyers cients for the first class (32% of population) ‘‘ indicate a preference for pork Labeled Gesta- tion Crate-Free and dislike of Large, Gestation Crate Ban, and Brazil attributes. These pref- 7.00] ($2.90) [ 2.10] ($6.88) [ 0.64] ($0.70) [

erences, however, appear to be dominated by a 2 2 significantly negative ‘‘opt out’’ parameter. As 2 such, we refer to this class as the ‘‘Pork En- 3.01, 12.55, 0.79, 1.58] ($1.00) [ 1.97, 0.15, 1.26] $0.48 [ Model 2 2 2 2 joyers’’ group. The second class (33% of pop- 2 ulation) is characterized by a preference for

pork Labeled Gestation Crate-Free and dislike Random Parameters $1.44 [0.27, 2.64] $0.33 [ of nonU.S. pork and pork produced under a ban ($9.49) [ on gestation crate use (Gestation Crate Ban).

8 Furthermore, marginal reductions in the AIC Consumer Willingness-to-Pay [95% confidence intervals] for Pork Attributes (Akaike Information Criterion) reduce significantly as additional latent classes are added and inclusion States) of more than four latent classes results in classes less States) Opt Out ($2.60) [ Brazil (vs. United Label (vs. typical)Canada (vs. United $2.11 [0.71, 3.54] $0.84 [0.30, 1.39] $1.86 [0.50, 3.49] ($0.08) [ Ban (vs. typical) $0.34 [ Large (vs. median) ($1.27) [ Small (vs. median) $0.56 [ Table 4. Notes: Simulated confidence 95% confidence intervals are presented in brackets and were derived by the complete combinational approach of Poe, Gira than 10% in size. presented values are in $/lb units. Tonsor, Olynk, and Wolf: Consumer Preferences for Animal Welfare Attributes 723

Collectively, this leads us to refer to this group $2.29/lb for pork from the US over pork from as the ‘‘Attribute Conscious’’ class. Canada. Class 3 (‘‘Price Conscious’’)isthe The third class is the smallest group (14% of only class indifferent between pork from the population) and appears to be most concerned United States and from Brazil. Discounts for with the price of pork. The insignificance of Brazilian pork range from $2.90/lb for class individual pork attribute coefficients, as well 1(‘‘Pork Enjoyers’’) to $13.13/lb for class 2 being rather indifferent on maintaining pork in (‘‘Attribute Conscious’’). their choice set, compels us to refer to this class The four classes are also very diverse in as the ‘‘Price Conscious’’ group. The fourth their valuations of gestation crate use. More class (20% of population) is the only class with specifically, classes 1 and 2 (a combined ap- utility estimates suggesting a preference for proximately two-thirds of the population) place pork produced without use of gestation crates a significant premium on pork from producers originating under a ban over that originating voluntarily selecting not to use gestation crates from voluntary choices of farmers. Collec- (mean WTP of $0.84/lb and $1.86/lb, respec- tively, this leads us to refer to this group as the tively). However, these same consumers place a ‘‘ Ban Preferring’’ group. significant discount on pork known to have originated from regions operating under a Willingness-to-Pay gestation crate ban (mean WTP of 2$1.00/lb and 2$3.39/lb, respectively). Consumer willingness-to-pay estimates are of This is in contrast to class 3 (14% of pop- particular interest. Results (point estimates and ulation) that is unwilling to pay a premium for indication of statistical significance) of our either gestation crate use attribute relative to simulations are presented in Table 4. By ex- Typical production practices. Class 4 (20% of amining both point estimates and overlapping population) is the only class possessing a sig- of confidence intervals, the RPL model indi- nificant preference for pork produced without cates a significant preference for pork from use of gestation crates regardless of the vol- Canada over the United States (mean WTP of untary or mandatory nature of this production $1.44/lb) and a larger discount for Brazilian practice (mean WTP of $5.62/lb and $3.13/lb pork (mean WTP of 2$9.49/lb). The RPL for ban and voluntary label, respectively).9 model indicates indifference between small and median sized farms of origin, indifference be- tween pork from operations using typical pro- duction practices or operating under a gestation crate ban, and positive preference for pork 9 It is prudent to note the conclusions that can, and voluntarily produced without use of gestation cannot, be drawn from these results in the context of crates (mean WTP of $2.11/lb). Significance of current discussions regarding mandatory country of origin labeling (MCOOL) (Meyer, 2008). The RPL the Opt Out coefficient reveals our sample model suggests the representative consumer prefers population has a preference for having pork pork from Canada rather than the United States while chops in their food choice set. the LCM model suggests only one-third of consumers Table 4 also reveals notable diversity in (class 2, ‘‘Attribute Conscious’’) significantly differ- entiate between pork from Canada and the United consumer WTP values across the four latent States. Differences in valuation estimates reflect alter- classes suggested by the LCM model. Class native assumptions about preference heterogeneity and 1(‘‘Pork Enjoyers’’) is the only class (32% of the functional form of these underlying models. How- the population) willing to pay a significant ever, these valuation estimates are not sufficient to draw final conclusions regarding MCOOL. In partic- amount for farm size preferences, with a mean ular, this analysis intentionally (primarily as our focus WTP of $0.70/lb for pork from median, rather was on gestation crates and we limit experimental than large farms. Similarly, class 2 (‘‘Attribute complexity) did not include products carrying Product Conscious’’) is the only class significantly dif- of the U.S., Canada or Product of Canada, U.S. labels that in reality exist under MCOOL. Moreover, no ferentiating between Canadian and U.S. pork, evaluation of cost increases imposed by MCOOL is with consumers indicating a mean WTP of evaluated in this analysis. 724 Journal of Agricultural and Applied Economics, December 2009

Table 5. Comparison of Crate Ban and Labeled Crate-Free Willingness-to-Pay Gestation Crate Ban Labeled Gestation Model/Segment vs. Typicala Crate-Free vs. Typicala p-Valueb Random Parameters Model $0.34 $2.11 0.972 LCM-Class 1 ‘‘Pork Enjoyers’’ 2$1.00 $0.84 0.999 LCM-Class 2 ‘‘Attribute Conscious’’ 2$3.39 $1.86 0.999 LCM-Class 3 ‘‘Price Conscious’’ $0.73 2$0.08 0.228 LCM-Class 4 ‘‘Ban Preferring’’ $5.62 $3.13 0.005 a WTP values are derived from models presented in Table 4 and are in $/lb increments. b p-Values report results of the one-sided test that the Gestation Crate Ban distribution exceeds the Labeled Gestation Crate-Free distribution. These values were determined by applying the nonparametric combinational method of Poe, Giraud, and Loomis (2005).

Consumer Welfare Evaluation Collectively these results suggest that if a consumer is provided with adequate labeling of Table 5 presents results of nonparametric tests pork produced on farms certified to voluntarily comparing WTP series to evaluate consumer not use gestation crates, we find no economic welfare impacts of banning gestation crates. A support justifying a ban on the use of gestation ban is welfare enhancing, in the presence of crates on the grounds of improving general transparent labeling, if and only if consumer consumer welfare. Using the RPL model we willingness-to-pay for Gestation Crate Ban firmly reject the notion of gestation crate bans pork exceeds that of Labeled Gestation Crate- improving consumer welfare in the presence of Free (Carlsson, Frykblom, and Lagerkvist, voluntary labeling. This implies that the private 2007c). As shown in Table 5, consumers (20% loss of option values (reduction in selection of of the population) possessing the utility func- products if pork raised using gestation crates tion represented by class 4 (‘‘Ban Preferring’’ ) are completely banned) is offsetting any public of the LCM model are the only consumers good benefits of a ban that would be necessary identified as having a significantly higher WTP for a ban to enhance overall consumer welfare for Gestation Crate Ban pork than Labeled (Carlsson, Frykblom, and Lagerkvist, 2007c; Gestation Crate-Free pork. Estimated utility Hamilton, Sunding, and Zilberman, 2003). functions for the other three consumer classes However, use of LCM model estimates reveals in the LCM model, and for representative that conclusions are segment specific. For consumers modeled by the RPL model, indi- approximately 65% of the population we can cate either indifference between a gestation reject the notion of gestation crate bans im- crate ban and voluntary disadoption (class 3 of proving consumer welfare, but for the remain- LCM model) or actually discount pork pro- ing 35% we cannot. Identification of markedly duced under a gestation crate ban relative to different consumer welfare effects is consistent pork labeled to have been voluntarily produced with other applications of LCM models, most without use of gestation crates. Combined, notably that of Boxall and Adamowicz (2002). these findings suggest that only a subset (20% The remaining issue addressed in this paper belonging to class 4 of the LCM model) of the is identification of actual consumer welfare evaluated consumer population have pork effects our estimated models imply would oc- preferences consistent with justifying a ban cur in the event of a gestation crate ban. As on gestation crates. Stated differently, using explained by Lusk, Norwood, and Pruitt estimates from an RPL model we reject the (2006), the WTP valuations of Table 5 are only hypothesis that a ban on gestation crates one welfare measure of relevance to our study. would improve consumer welfare. We also re- These typical WTP estimates are not appro- ject this hypothesis using LCM estimates for priate welfare measures in situations where consumers in latent classes 1 and 2 (approxi- consumers may not make a purchase (i.e., ‘‘Opt mately 65%). Out’’) or in situations involving choice Tonsor, Olynk, and Wolf: Consumer Preferences for Animal Welfare Attributes 725 uncertainty. Following Morey (1999) and Lusk, estimates in $/choice occasion and aggregated Norwood, and Pruitt (2006) we note that ex- values for the population are presented as- pected maximum utility (EMU) from each suming 26,975 million choice occasions per consumer’s choice set selection is given by: year (Lusk, Norwood, and Pruitt, 2006). X Table 6 reveals that each model implies (7) EMU 5 ln eVj 1 C, statistically significant welfare losses. As an- ticipated, the welfare losses are larger for consumers who lose the ability to purchase where C is Euler’s constant and Vj is defined as in Equations (2) and (3). As such, the general two pork products (typical production and welfare change of moving from situation Y to voluntary gestation crate-free production) than situation Z is given by: for consumers who only lose the ability to purchase one product (typical production). 1 (8) EMUZ EMUY , Estimates for the four classes identified by the MUI LCM model reveal differential welfare im- where MUI is the marginal utility of income.10 pacts. Consumers belonging to class 2 (‘‘Attri- Note that consumers currently have choice sets bute Conscious’’) are found to experience containing pork produced under three condi- significantly larger welfare declines than con- tions: (1) under gestation crate bans, (2) using sumers in the other three segments. Consumers typical production, and (3) with voluntary dis- of class 4 (‘‘Ban Preferring’’), the only class adoption of gestation crates. However, when a with statistically significant preferences for ban is imposed, the consumer choice set is re- pork produced under a gestation crate ban duced and the latter two options are no longer (Table 4), also experience a welfare decline available for purchase. The welfare change that from a gestation crate ban. This potentially would result from choosing between three pork counter-intuitive finding corresponds with products and none to a situation of choosing consumers in this segment also possessing between one pork product and none can hence positive valuations of pork produced by farmers be identified by using Equations (7) and (8). voluntarily not using gestation crates. Further- Evaluation of Equation (8) provides a value that more, the general overall finding of negative may be interpreted as the amount consumers welfare impacts corresponds to the loss of would pay to maintain pork originating from purchasing options experienced by consumers typical production and voluntary disadoption of following a ban, including the ‘‘Ban Prefer- gestation crates in their choice sets. ring’’ class implied by the LCM model. This Table 6 presents estimates of the welfare important distinction underlies the necessity of impacts our utility models imply consumers not only evaluating traditional WTP measures would experience following a gestation crate in assessing welfare impacts (Lusk, Norwood, ban. Two sets of estimates are provided. The and Pruitt, 2006). first corresponds with the assumption that It is critical to note that these consumer consumers currently have access to pork pro- welfare measures are based upon the assump- duced by farmers voluntarily not using gesta- tion of no production cost adjustment and tion crates. Given the possibility that some hence no overall pork price adjustment. In re- consumers may not currently have access to ality there would be some nonzero produc- these products, we also present the welfare tion cost adjustment, resulting in an increase in impacts of consumers losing the ability to pork prices, further exacerbating the con- purchase Typical pork (but not pork labeled to sumer welfare estimates presented here (Lusk, have been produced by producers voluntarily Norwood, and Pruitt, 2006). That is, the pre- choosing not to use gestation crates). Welfare sented consumer welfare estimates reflect only changes in available pork choice sets and marginal valuations of alternative uses and 10 We use 21 times our estimated price coefficients regulation of gestation crates. For instance, as marginal utility of income estimates. Lusk, Norwood, and Pruitt (2006) incorporated 726

Table 6. Welfare Measures [95% confidence intervals] for Banning Gestation Crates Labeled Gestation Crate-Free Pork Available Labeled Gestation Crate-Free Pork NOT Available Welfare Change from Ban with Loss of Option to Buy Welfare Change from Ban with Loss of Typical and Labeled Gestation Crate-Free Pork Option to Buy Typical Pork

$/Choice Occasion Millions of Dollars/Year $/Choice Occasion Millions of Dollars/Year 2009 December Economics, Applied and Agricultural of Journal Random parameters model 2$0.17 2$4,626.21 2$0.07 2$1,971.87 [2$0.26, 2$0.11] [2$7,062.06, 2$2,872.84] [2$0.12, 2$0.04] [2$3,183.05, 2$1,173.41] LCM-Class 1 ‘‘Pork Enjoyers’’ 2$0.02 2$147.71 2$0.01 2$97.31 [2$0.04, 2$0.01] [2$369.27, 2$44.31] [2$0.03, 2$0.00] [2$255.45, 2$26.07] LCM-Class 2 ‘‘Attribute Conscious’’ 2$0.80 2$7,194.63 2$0.43 2$3,853.95 [2$1.40,$0.082$0.44] [2$12,605.41,2$308.622$3,951.75] [2$0.77,2$0.042$0.23] [2$6,872.50,2$154.312$2,072.79] LCM-Class 3 ‘‘Price Conscious’’ [2$0.40,2 2$0.01] [2$1,527.88, 2$31.62] [2$0.19, 2$0.00] [2$706.40, 2$13.72] LCM-Class 4 ‘‘Ban Preferring’’ 2$0.36 2$1,996.50 2$0.13 2$731.90 [2$0.58, 2$0.21] [2$3,218.73, 2$1,180.95] [2$0.23, 2$0.08] [2$1,285.51, 2$427.04] Notes: Values in brackets are simulated confidence 95% confidence intervals derived via the Krinsky-Robb bootstrapping method. These estimates are based upon the assumption of 26,975 million choice occasions per year (Lusk, Norwood, and Pruitt, 2006). Tonsor, Olynk, and Wolf: Consumer Preferences for Animal Welfare Attributes 727 an expected retail price increase of 0.81% in welfare. Considering preference heterogeneity their evaluation of banning antibiotic use, re- differently, estimates from the LCM model flective of farm-level production cost increases suggest that only a subset (approximately 20%) of 2.02% suggested by Brorsen et al. (2002). of the evaluated consumer population have Although certainly needed for a broader wel- pork preferences consistent with those that fare analysis of the issue (i.e., improved con- could justify a ban on gestation crates. sumer, producer, and society impacts), we are Given the close voting margin of some re- unaware of similar published estimates of pro- lated ballot initiatives (e.g., November 2002 duction cost impacts stemming from banning initiative in Florida), this work highlights the gestation crates. Moreover, such estimates are implications of ‘‘ban preferring’’ consumers dependent upon the production practices used disproportionally showing up to vote. Further- in lieu of gestation crates, an issue notably less more, this work supports many of the ‘‘politics certain than in discussions such as use of anti- by other means’’ conclusions made by biotics, growth hormones, or genetically mod- Schweikhardt and Browne (2001) as alternative ified feeds that likely do not require substantial methods, including consumer purchasing be- capital investments (Tonsor, Wolf, and Olynk, havior, voting on ballot initiatives, and exerting 2008). indirect pressure on food producers and dis- tributors increasingly being used by select Conclusions consumer groups to initiate changes in food production practices. The results of this anal- Increasing consumer interest in the production ysis imply that the desires (and corresponding practices employed in modern food production voting behavior) of these consumers have have led to growing analysis of consumer substantial impacts on the consumer welfare of preferences for production methods. This all consumers whose food product choice set is analysis focused on the growing consumer impacted. pressure for the U.S. swine industry to no These findings imply that the swine industry longer use gestation crates. In employing both may benefit by encouraging additional labeling random parameters logit and latent class of products originating from producers volun- models, we find strong consumer preference tarily choosing not to utilize gestation crates. If heterogeneity for pork chop attributes. RPL these products are currently not widely avail- model estimates revealed preferences for pork able to consumers, results of this study suggest from Small farms to be positively correlated that additional labeling may, in addition to with preferences for pork produced under a seizing market opportunities, potentially help gestation crate ban or produced by farmers alleviate some of the increasing pressure for voluntarily not using gestation crates. This sug- production practice changes associated with gests our evaluated sample of consumers hold gestation crates. farm size attributes as partial substitutes for use Given these findings, future work should of gestation crates. Inferences from the LCM further examine consumer perceptions and model further document preference heteroge- valuation of alternative methods of certifying neity and provide insights on differential con- voluntary disadoption of gestation crates. Fu- sumer welfare impacts of restrictions on ges- ture research could compare the findings based tation crate use. here on a sample of Michigan consumers with In our analysis, if a consumer is provided consumers from other U.S. states or regions. with adequate labeling of pork produced on Additional work could also examine if opera- farms certified to voluntarily not use gestation tion size is truly coupled with other credence crates, we find no economic support justifying attributes of current interest including ‘‘locally a ban on the use of gestation crates that impacts grown,’’ ‘‘natural,’’ ‘‘organic,’’ ‘‘food safety,’’ all consumers. Using estimates from the RPL and ‘‘free-range.’’ Finally, this analysis high- model we reject the hypothesis that a ban on lights the need for additional research esti- gestation crates would improve consumer mating the cost differentials of gestation crate 728 Journal of Agricultural and Applied Economics, December 2009 use (i.e., cost of production with and without ———. ‘‘Farm Animal Welfare – Testing for gestation crates) and alternative replacement Market Failure.’’ Journal of Agricultural and practices (incorporating implicit changes in Applied Economics 39(2007b):61–73. mortality, animal health, capital investments, ———. ‘‘Consumer Benefits of Labels and Bans on GM Foods – Choice Experiments with etc.) including indoor group housing and pas- Swedish Consumers.’’ American Journal of ture based systems. This future analysis should Agricultural Economics 89(2007c):152–61. examine impacts on producers varying in spe- Cummings, R.G., and L.O. Taylor. ‘‘Unbiased cialization (e.g., farrow-finish versus farrow- Value Estimates for Environmental Goods: A weanling) and current operation characteristics Cheap Talk Design for the Contingent Valua- (e.g., facility age, availability for physical tion Method.’’ The American Economic Review expansion of facilities). Combined, these esti- 89(1999):649–65. mates would lead to an improved understand- Darby, K., M.T. Batte, S. Ernst, and B. Roe. ing of the net consumer, producer, and societal ‘‘Decomposing Local: A Conjoint Analysis of impacts of alternative forms of gestation crate Locally Produced Foods.’’ American Journal of Agricultural Economics use and regulation. 90(2008):476–86. Fox, J.A., D.J. Hayes, and J.F. Shogren. ‘‘Con- sumer Preferences for Food Irradiation: How [Received November 2008; Accepted May 2009.] Favorable and Unfavorable Descriptions Affect Preferences for Irradiated Pork in Experimental References Auctions.’’ Journal of Risk and Uncertainty 24(2002):75–95. Alfnes, F. ‘‘Stated Preferences for Imported and Grannis, J., and D. Thilmany. ‘‘Marketing Natural Hormone-Treated Beef: Application of a Pork: An Empirical Analysis of Consumers in Mixed Logit Model.’’ European Review of the Mountain Region.’’ Interna- Agriculture Economics 31(2004):19–37. tional Journal (Toronto, Ont.) 18(2002):475–89. Alfnes, F., and K. Rickertsen. ‘‘European Con- Greene, W. NLOGIT Version 3.0 Reference sumers’ Willingness to Pay for U.S. Beef in Guide. Econometric Software, Inc., 2002. Experimental Auction Markets.’’ American Greene, W.H. Econometric Analysis, 5th Ed. Journal of Agricultural Economics 85(2003): Upper Saddle River, NJ: Prentice Hall, 2006. 396–405. Grethe, H. ‘‘High Animal Welfare Standards in Boxall, P.C., and W.L. Adamowicz. ‘‘Understand- the EU and International Trade – How to Pre- ing Heterogeneous Preferences in Random vent Potential ‘Low Animal Welfare Havens’?’’ Utility Models: A Latent Class Approach.’’ En- Food Policy 32(2007):315–33. vironmental and Resource Economics 23(2002): Hamilton, S.F., D.L. Sunding, and D. Zilberman. 421–46. ‘‘Public Goods and the Value of Product Quality Brorsen, B.W., T. Lehenbauer, D. Ji, and J. Connor. Regulations: The Case of Food Safety.’’ Journal ‘‘Economic Impacts of Banning Subtherapeutic of Public Economics 87(2003):799–817. Use of Antibiotics in Swine Production.’’ Jour- Hanley, N., W. Adamowicz, and R.E. Wright. nal of Agricultural and Applied Economics ‘‘Price Vector Effects in Choice Experiments: 34(2002):489–500. An Empirical Test.’’ Resource and Energy Burton, M., D. Rigby, T. Young, and S. James. Economics 27(2005):227–34. ‘‘Consumer Attitudes to Genetically Modified Hensher, D.A., and W.H. Greene. ‘‘The Mixed OrganismsinFoodintheU.K.’’European Review Logit Model: The State of Practice.’’ Trans- of Agriculture Economics 28(2001):479–98. portation 30(2003):133–76. Capps, O., and J. Park. ‘‘Impacts of Advertising, Hensher, D.A., J.M. Rose, and W.H. Greene. Attitudes, Lifestyles, and Health on Demand Applied Choice Analysis. Cambridge, UK: for U.S. Pork: A Micro-level Analysis.’’ Jour- Cambridge University Press, 2006. nal of Agricultural and Applied Economics of the United States. ‘‘Think 34(2002):1–15. Outside the Crate Campaign.’’ Internet site: Carlsson, F., P. Frykblom, and C.J. Lagerkvist. http://www.hsus.org/ (Accessed October 27, ‘‘Consumer Willingness to Pay for Farm Ani- 2008). mal Welfare: Mobile Abattoirs versus Trans- Krinsky, I., and A. Robb. ‘‘On Approximating the portation to Slaughter.’’ European Review of Statistical Properties of Elasticities.’’ The Review Agriculture Economics 34(2007a):321–44. of Economics and Statistics 64(1986):715–9. Tonsor, Olynk, and Wolf: Consumer Preferences for Animal Welfare Attributes 729

Kuhfeld, W.F., R.D. Tobias, and M. Garratt. Norwood, F.B. (2007). ‘‘Lessons Abound on Ani- ‘‘Efficient Experimental Design with Market- mal Welfare Issue.’’ The Voice of Agriculture- ing Research Applications.’’ JMR, Journal of American Farm Bureau. November 26, 2007. Marketing Research 31(1994):545–57. Norwood, F.B., J.L. Lusk, and R.W. Prickett. Lijenstolpe, C. ‘‘Evaluating Animal Welfare with (2007). ‘‘Consumers Express Views on Farm Choice Experiments: An Application to Swedish Animal Welfare.’’ Feedstuffs. October 8, 2007. Pig Production.’’ Agribusiness 21(2008):67–84. Ohler, T., A. Le, J. Louviere, and J. Swait. ‘‘At- Louviere, J.J., D.A. Hensher, and J.D. Swait. Stated tribute Range Effects in Binary Response Choice Methods: Analysis and Application. Task.’’ Marketing Letters 11(2000):249–60. Cambridge: Cambridge University Press, 2000. Ouma, E., A. Abdulai, and A. Drucker. ‘‘Mea- Lusk, J.L. ‘‘Effects of Cheap Talk on Consumer suring Heterogeneous Preferences for Willingness-to-Pay for Golden Rice.’’ American Traits among Cattle-Keeping Households in JournalofAgriculturalEconomics85(2003): East Africa.’’ American Journal of Agricultural 840–56. Economics 89(2007):1005–19. Lusk, J.L., T. Nilsson, and K. Foster. ‘‘Public Parcell, J.L., and V. Pierce. ‘‘Factors Affecting Preferences and Private Choices: Effect of Al- Wholesale Poultry Prices.’’ Journal of Agri- truism and Free Riding on Demand for Envi- cultural and Applied Economics 32(2000): ronmentally Certified Pork.’’ Environmental 471–8. and Resource Economics 36,4(2007):499–521. Parcell, J.L., and T.C. Schroeder. ‘‘Hedonic Retail Lusk, J.L., F.B. Norwood, and J.R. Pruitt. ‘‘Con- Beef and Pork Product Prices.’’ Journal of sumer Demand for a Ban on Antibiotic Drug Agricultural and Applied Economics 39(2007): Use in Pork Production.’’ American Journal of 29–46. Agricultural Economics 88(2006):1015–33. Poe, G.L., K.L. Giraud, and J.B. Loomis. Lusk, J.L., J. Roosen, and J. Fox. ‘‘Demand for ‘‘Computational Methods for Measuring the Beef from Cattle Administered Growth Hor- Difference of Empirical Distributions.’’ Amer- mones or Fed Genetically Modified Corn: A ican Journal of Agricultural Economics Comparison of Consumers in France, Ger- 87(2005):353–65. many, the United Kingdom, and the United Revelt, D., and K. Train. ‘‘Mixed Logit with States.’’ American Journal of Agricultural Repeated Choices: Households’ Choices of Economics 85(2003):16–29. Appliance Efficiency Level.’’ The Review of Lusk, J.L., and T.C. Schroeder. ‘‘Are Choice Exper- Economics and Statistics 80(1998):647–57. iments Incentive Compatible? ATest with Quality Rigby, D., and M. Burton. ‘‘Preference Hetero- Differentiated Beef Steaks.’’ American Journal of geneity and GM Food in the UK.’’ European Agricultural Economics 86(2004):467–82. Review of Agriculture Economics 32(2005): Martin, A. ‘‘Burger King Shifts Policy on Ani- 269–88. mals.’’ New York Times. March 28, 2007. Roosen, J. ‘‘Marketing of Safe Food through La- McCluskey, J.J., K.M. Grimsrud, H. Ouchi, and beling.’’ Journal of Food Distribution Research T.I. Wahl. ‘‘Consumer Response to Genetically 34(2003):77–82. Modified Food Products in Japan.’’ Agricultural Roosen, J., J.L. Lusk, and J.A. Fox. ‘‘Consumer and Resource Economics Review 32(2003): Demand for and Attitudes Toward Alternative 222–31. Beef Labeling Strategies in France, Germany, Meyer, S.R. ‘‘Implementation of Mandatory and the UK.’’ Agribusiness International Country of Origin Labeling (MCOOL) in the Journal (Toronto, Ont.) 19(2003):77–90. Pork Industry.’’ Choices (New York, N.Y. 1994) Scarpa, R., and T. DelGiudice. ‘‘Market Seg- 23(2008):39–42. mentation via Mixed Logit: Extra-Virgin Olive Morey, E.R. ‘‘TWO RUMS unCLOAKED: Oil in Urban Italy.’’ Journal of Agricultural and Nested-logit Models of Site Choice and Nested- Food Industrial Organization 2(2004):1–18. logit Models of Participation and Site Choice.’’ Schweikhardt, D.B., and W.P. Browne. ‘‘Politics Valuing Recreation and the Environment. J.A. by Other Means: The Emergence of a New Herriges and C.L. Kling, eds. Northampton, Politics of Food in the United States.’’ Review MA: Edward Elgar, 1999. of Agricultural Economics 23(2001):302–18. Nilsson, T., K. Foster, and J.L. Lusk. ‘‘Marketing Tonsor, G.T., T.C. Schroeder, J.A. Fox, and A. Opportunities for Certified Pork Chops.’’ Ca- Biere. ‘‘European Preferences for Beef Steak nadian Journal of Agricultural Economics Attributes.’’ Journal of Agricultural and Re- 54(2006):567–83. source Economics 30(2005):367–80. 730 Journal of Agricultural and Applied Economics, December 2009

Tonsor, G.T., C. Wolf, and N. Olynk. ‘‘Consumer d Median means the animal was raised on a farm Voting and Demand Behavior Regarding Swine that is smaller than about 50% and larger than Gestation Crates.’’ Working paper, 2008. about 50% of the firms in the industry, and Train, K.E. ‘‘Recreation Demand Models with d Large means the animal was raised on a farm Taste Differences over People.’’ Land Eco- that is larger than about 75% of the firms in nomics 74(1998):230–39. ———. Discrete Choice Methods with Simula- the industry. tion. Cambridge: Cambridge University Press, Production Practice is the method used in raising the 2003. animal where: United States Census Bureau (2006). Interna- tional Database Summary Demographic Data. d Typical means the animal was raised using Internet site: http://www.census.gov/ipc/www/ production practices typical for the industry, idb/ (Accessed October 27, 2008). d Labeled Gestation Crate-Free is the same as United States Department of Labor, Bureau of La- Typical except the animal is guaranteed to bor Statistics (2008). Databases, Tables & Cal- have been raised by a producer who volun- culators by Subject. Series Id: APU0000704212. tarily chose not to use gestation crates, and Videras, J. ‘‘Religion and Animal Welfare: Evi- d Gestation Crate Ban is the same as Typical dence from Voting Data.’’ Journal of Socio- except the animal is guaranteed to have been Economics 35,4(2006):652–59. raised in a region (state or country) where the Appendix A. Information Treatments and use of gestation crates is legally banned for all Choice Experiment Definitions swine producers. d Country of Origin refers to the country in Respondents randomly received one of the following which the animal was raised in and includes three information treatments: the United States, Canada, and Brazil. 1. Industry Information Treatment: Use of Gestation Crates in Pork Production Gestation crates refer to metal crates that house fe- male breeding stock in individually confined areas Appendix B. Cholesky Matrix and during an animal’s four-month pregnancy. Some Correlation Statistics from the Random pork producer organizations (such as National Pork Producers) suggest that using gestation crates may Parameters Logit Model facilitate more efficient pork production, leading to lower pork prices for consumers. 2. Consumer Group Information Treatment: Use of Gestation Crates in Pork Production Cholesky Matrix Gestation crates refer to metal crates that house fe- Small Large Ban Label Canada Brazil male breeding stock in individually confined areas during an animal’s four-month pregnancy. Some Small 0.36* consumer groups (including the Humane Society of Large 20.08 0.14 the United States and Sierra Club) suggest gestation Ban 0.99* 20.07 1.51** crates are inhumane devices. Label 0.93* 0.57 0.91* 0.86 3. Base Information Treatment: Canada 21.17* 20.21 0.20 20.20 0.95 Use of Gestation Crates in Pork Production Brazil 2.98* 20.68 20.27 0.42 20.78 0.88 Gestation crates refer to metal crates that house fe- Correlation Matrix male breeding stock in individually confined areas Small Large Ban Label Canada Brazil during an animal’s four-month pregnancy. Attribute descriptions included in the choice exper- Small 1.00 20.52 0.55 0.56 20.76 0.90 iments were: Large 20.52 1.00 20.32 0.00 0.28 20.64 Farm Size refers to the size of operation the animal Ban 0.55 20.32 1.00 0.75 20.30 0.43 was raised on where: Label 0.56 0.00 0.75 1.00 20.47 0.45 Canada 20.76 0.28 20.30 20.47 1.00 20.82 d Small means the animal was raised on a farm Brazil 0.90 20.64 0.43 0.45 20.82 1.00 that is smaller than about 75% of the firms in *, ** Indicates statistical significance at the 0.05, 0.01 level, the industry, respectively.