Author's personal copy - École Polytechnique

This article appeared in a journal published by Elsevier. The attached

copy is furnished to the author for internal non-commercial research

and education use, including for instruction at the authors institution

and sharing with colleagues.

Other uses, including reproduction and distribution, or selling or

licensing copies, or posting to personal, institutional or third party

websites are prohibited.

In most cases authors are permitted to post their version of the

article (e.g. in Word or Tex form) to their personal website or

institutional repository. Authors requiring further information

regarding Elsevier’s archiving and manuscript policies are

encouraged to visit:

http://www.elsevier.com/copyright

http://www.elsevier.com/copyright

Author's personal copy

A calibrated auction-conjoint valuation method: Valuing pork andeggs produced under differing animal welfare conditions

F. Bailey Norwood, Jayson L. Lusk n

Department of Agricultural Economics, Oklahoma State University, 411 Ag Hall, Stillwater, OK 74078, USA

a r t i c l e i n f o

Article history:Accepted 16 October 2010Available online 1 May 2011

Keywords:Animal welfareAuctionConjointDiscovered preference hypothesisWillingness-to-pay

a b s t r a c t

This paper develops a valuation method which generates consistent and systematicestimates of people’s preferences for complex multi-attribute goods by inextricablylinking auction bids with conjoint ratings. The advantage of the valuation approach isthat it permits the estimation of people’s values for many potential goods, allows one todecompose people’s values for a good into its sub-components, and permits the study ofpreference heterogeneity without distributional assumptions. We apply the method toan important and increasingly controversial topic: animal welfare. The method is usedto determine people’s preferences for eggs and pork produced from different productionsystems. Data from experiments conducted in three diverse U.S. locations (Chicago, IL;Dallas, TX; and Wilmington, NC) indicates that people are, on average, willing to pay$0.95 more for a dozen eggs raised in an aviary, pasture system vs. a cage system, andare willing to pay $2.02 more for two-pounds of pork chops raised in a pasture systemas opposed to a crate system.

& 2011 Elsevier Inc. All rights reserved.

1. Introduction

Although survey and experimental valuation methods have become standard tools in environmental valuation, thereremains significant debate on the malleability, contextuality, and even irrationality of preferences. When a behavioralanomaly is observed that violates a normative model of consumer behavior, two general options exist: (i) add parametersto the utility function to create a behavioral theory that is consistent with the observed choices, or (ii) utilize aninstitutional or elicitation setting that promotes more systematic and rational behavior. The explosive growth in the fieldof behavioral economics has tended to follow the first approach [16,24]. In this article, we pursue the second path bycombining the strengths of conjoint and auction value elicitation mechanisms in a hybrid approach we refer to as thecalibrated auction-conjoint method (CACM).1

The emergence of behavioral economics has resulted in new motivations for paternalistic policies seeking to rescuepeople from their own infallibilities.2 Behavioral economics has also led to increased skepticism of the validity of usingstated preference values in cost–benefit analysis. Elicited values are most useful in cost–benefit analysis if they arise from

Contents lists available at ScienceDirect

journal homepage: www.elsevier.com/locate/jeem

Journal ofEnvironmental Economics and Management

0095-0696/$ - see front matter & 2011 Elsevier Inc. All rights reserved.doi:10.1016/j.jeem.2011.04.001

n Corresponding author. Fax: þ1 405 744 8210.E-mail address: [email protected] (J.L. Lusk).1 Others have sought to obtain better estimates of people’s values by combining data from different elicitation methods ex post (e.g., [38]). Here we

combine different methods rather than sources of data.2 Note, however, that inconsistent preferences need not justify paternalism even on conceptual grounds. For example, Sugden [31–33] outlines

conditions under which one can respect individuals’ choices in a coherent manner even if people themselves behave incoherently. His work suggests thatmarkets can efficiently allocate resources even if some people have inconsistent preferences.

Journal of Environmental Economics and Management 62 (2011) 80–94


some systematic and stable process consistent with normative principles. Unfortunately, people do not always act in sucha manner.3

If one accepts the premise that analysts need to elicit people’s values to inform cost–benefit analyses, a natural questionthat arises is whether elicitation mechanisms can be created that that generate more systematic willingness-to-pay values.The purpose of this paper is to introduce just such a mechanism, the CACM. Thus, rather than seeking to identifypaternalistic policies in light of the findings from behavioral economics, we seek to develop a more paternalistic elicitationmechanism, but one that is libertarian in the sense that people are free to express their own preferences within thestructure of the approach.

The CACM can be described generally as follows. Like typical experimental auction methods [21], people place bids onseveral products. Like typical conjoint methods [20], the products are defined by a bundle of attributes. But unlike typicalauction and conjoint methods, the CACM involves an iterative and interactive decision making task where auction bids areinextricably linked to an underlying attribute-based utility function. The method is ‘‘self-calibrating’’ because the only wayfor people to alter their bids is by changing their utility coefficients via responses to rating scales. In essence, people mustcalibrate their attribute-based utility function to produce the auction bids they desire. The CACM was designed specificallyfor the application studied in this paper, animal welfare, and is particularly useful in other such applications where thereare a large number of complex attributes related to product outcomes; situations which commonly arise when dealingwith environmental issues.

The CACM possesses several advantages over existing valuation approaches. First, as previously indicated, the CACMgenerates consistent and systematic responses by linking auction bids with conjoint ratings. This consistency or rationalityworks in three ways. The first is that the method, by definition, imposes a mechanical or algebraic relationship betweenvaluations and utility. Secondly, and more importantly, people are able to directly see the consequences of their conjoint-ratings and the trade-offs implied in their auction bids. That is, the CACM puts the conjoint-ratings in an economic contextwhere choices have consequences that would otherwise be less transparent. Likewise, the CACM puts the bids in a contextwhere it is clear that the valuations correspond in a systematic way to underlying product attributes.

Finally, despite what is often assumed, people may not have well formed preferences for the goods being valued, whichcould lead to a number of behavioral anomalies. Plott [25], however, argued that although people may not have wellformed preferences, rational choice arises from a series of stages in which people gain experience with and receivefeedback from a particular market environment; something he referred to as the discovered preference hypothesis.Empirical support for the theory can be found in studies on preference reversals [7–9], the willingness-to-pay/willingness-to-accept gap [28], and violations of expected utility theory [17] (see [4] for a more complete discussion). An advantage ofthe CACM is that it entails an iterative valuation process that promotes learning and provides feedback, which according tothe discovered preference hypothesis, promotes the formation of rational preferences.

A second key advantage of the CACM is that it allows for a distribution-free characterization of heterogeneity inpreferences. For example, in typical contingent valuation or conjoint methods, people choose (or rank) which product oroutcome they most prefer. Assumptions must be made about the form of a representative utility function and thedistribution of errors in the random utility model for such responses to be meaningfully used. Although preferenceheterogeneity can be incorporated in discrete choice models by using advances in econometric techniques, such as mixedlogit models or hierarchical Bayes models [36], such approaches require assumptions about a functional form for theutility function and assumptions about the joint distribution of preferences. The CACM allows one to obtain utilitycoefficients for each individual without using econometric methods or assuming anything about the distribution ofpreferences in the population.

A final key advantage of the CACM is that it allows for the evaluation of a large number of attributes and attribute-levels, and permits the estimation of people’s values for a very large number of goods or products—much larger than thenumber of products people bid on in an experiment. For example, in the application we consider there are numerousattributes or characteristics that can be used to describe the well-being of farm animals. In one of our applications, forexample, we consider 10 non-price attributes, most of which were varied at 4 levels or more. This means there are at least410¼1,048,576 possible products that could be created based on the variation in attributes. To circumvent the problem ofdimensionality, the first stage of the CACM relies instead on the so-called self-explicated approach to preferencemeasurement, where respondents simply rate the relative desirability of attribute-levels and the relative importance ofattributes. The self-explicated approach has been shown to perform as well or better than traditional conjoint analysis[29,30].

2. Background on farm animal welfare

Animal welfare is rapidly becoming one of the most contentious issues in animal agriculture [13]. For example,it is estimated that about 50–60 pieces of legislation regarding animal welfare are introduced in the U.S. Congress each

3 For example, there is now an extensive literature on the preference reversal phenomenon, which began with [18,15]. This finding led to thedevelopment of competing behavioral theories, virtually all of which abandon classical preference theory [14,37]. An alternative is to ask whether theremight be environments in which people will act in a more systematic and consistent manner, and several studies have found that certain elicitationenvironments can significantly reduce if not eliminate preference reversals and other behavioral anomalies [7–9].

F.B. Norwood, J.L. Lusk / Journal of Environmental Economics and Management 62 (2011) 80–94 81


year [27]. Citizens in Arizona and Florida, and most recently California, have voted to pass constitutional amendments toban the use of gestation crates in hog production or cages in egg production, and similar measures are on the horizon inseveral other states.

There are several important public policy issues at play in the animal welfare debate, and in this paper we focus onaspects related to the private-good components of the problem which is important to food retailers and also to policymakers. For example, on the marketing front, food retailers such as McDonald’s and Burger King have adopted improvedanimal welfare standards as a result of pressures from activist groups and a perception that such changes were demandedby their customers. Further, there are many specialty grocers such as Whole Foods that are proactively appealing togrowing concern about animal welfare by selling ‘‘animal compassionate’’ meat. There are also certification agencieslabeling products ‘‘certified humane.’’ These developments suggest the need for a better understanding of people’spreferences for food produced under different conditions of animal well-being. Moreover, understanding the effects ofpublic policies, such as a ban on gestation creates or cages, requires an understanding of how consumer demand for meatand eggs will change once a ban is enacted. There is also a public-good dimension associated with animal welfare, and thisis an issue we take up elsewhere (see [23]).

One of the key challenges in studying preferences for the farm animal well-being is that it is a complex and multi-dimensional issue; a feature that is common to many environmental problems. Most previous studies on this topic havetended to focus on eliciting consumer preferences for a limited number of production practices, such as the use of cages,gestation crates, or mobile abattoirs (e.g., [2,5,35]). However, such results are necessarily limited to the specific productionpractice studied and do not generalize well when attempting to forecast what consumers think about entire productionsystems.

For example, consider the recent bans on gestation stalls in the U.S. The stalls confine sows to a metal crate slightlylarger than the sow herself, and the sows are kept in the stalls for two-thirds of their lives. Many contend this practice isinhumane, but what are the consequences of a ban on stalls? If stalls were banned, many producers would likely replacegestation stalls with gestation pens. The pens hold several sows, but the space per sow is roughly the same as the stall.Unfortunately, multiple sows in a cramped space frequently injure one another. In fact, scientific studies have concludedanimal well-being is roughly equivalent in gestation stalls and gestation pens [22,34]. Surely, consumers require suchinformation to properly value a gestation stall ban. Nevertheless, previous studies on consumer willingness-to-pay forpork produced without gestation crates have tended to utilized choice experiments, and the need for parsimony prohibitedthe studies from considering the full consequences of crate-free pork, which includes the positive value of increased sowspace and the negative value of increased sow injuries. Understanding how a change in production practices alters animalwell-being requires a holistic consideration of many attributes.

It would be a significant challenge to determine consumer preferences for the large number of attributes thatcharacterize the well-being of farm animals using conventional approaches like experimental auctions or conjoint analysis.For example, in trying to devise a system to characterize the well-being of sows housed in seven different productionsystems, Bracke et al. [3] utilized 37 different attributes (such as space per sow, space per pen, etc.) related to 12 differentwell-being outcomes (such as pain, illness, and aggression), but previous choice experiments have only focused on a verylimited number of these attributes (e.g., see [19]).

It should be clear from the preceding discussion that there is a need for a method that is able to consider a numerousand complex set of attributes, and is capable of eliciting preferences for these numerous attributes without producingsubject fatigue or irrational responses.

3. Methods and data

To study people’s preferences for animal well-being, three marketing research companies were hired to randomlyrecruit 100 people from the general population in Chicago, IL; Dallas, TX; and Wilmington, NC. Participants were recruitedby phone to participate in a ‘‘food preference study’’ and were paid between $60 and $85 (depending on the location)to participate in a 90 min session. These three locations were chosen because they are geographically disparate, andthus provide diversity in several demographic variables; they are also regions associated with high levels of porkconsumption [10].

Collecting data in person, rather than through phone, mail, or Internet survey was important in this application giventhe complex issues involved and the need to convey a substantial amount of information to people about animalproduction practices and answer questions for subjects who know little about livestock and poultry production. Withineach location, one half of the subjects (50) were randomly assigned to the egg layer treatment and the other 50 subjectswere assigned to the pork treatment. In each location, four sessions were held with approximately 25 people each, and werotated which treatment was assigned to a session (i.e., the first session held in Dallas was with 25 people evaluating eggs,the second session in Dallas was with 25 people evaluating pork, the third session in Dallas was with 25 people evaluatingeggs, and the last session in Dallas was with 25 people evaluating pork).

Once people arrived at a session they were seated in front of a lap-top while the moderator made several introductorycomments. The introductory comments stressed the points that: (i) the researchers were not affiliated with either animalproduction or animal rights organizations, (ii) all responses were completely confidential, (iii) there would be several real

F.B. Norwood, J.L. Lusk / Journal of Environmental Economics and Management 62 (2011) 80–9482


money, real product decision tasks, and (iv) the respondents should not try to answer in a ‘‘socially acceptable’’ way or in away to ‘‘please’’ the researchers, but rather to honestly provide their own preferences.

After the introduction, participants were given a lengthy presentation describing the attributes characterizing hog oregg laying production systems. The key attributes and attribute-levels were chosen by selecting the key factors affectinghog and layer welfare identified in previous animal science research [3,11]. The attributes included issues like space,optional surgeries, flooring materials, etc. The authors of these animal science studies were contacted, and they providedinformation on the range over which each of the attributes was varied and described how each production system relatedto each of the underlying attributes. Table 1 lists each of the attributes and attribute levels studied in the egg and porktreatments.4

Collection of data proceeded in three steps as shown in Fig. 1. The first two steps were designed to loosely follow themethods used by Srinivasan and Park [29]. In step 1, participants were shown numerous tables corresponding to eachof the attributes studied, and in each table, people were asked to rate the desirability of each attribute-level on a scale of1–10, where 1 was very undesirable and 10 was very desirable. Thus, unlike typical conjoint approaches where peoplerate, rank, or choose between different product options that differ by attributes, our approach simply asks people toevaluate the desirability of each attribute-level directly. In addition to the information presented by the moderators in thepreliminary instructions, on the computer screen additional information was provided about each attribute to aid thesubjects’ understanding of each attribute’s role in animal well-being.

After rating the relative desirability of each level within an attribute, in step 2 participants were asked to indicate therelative importance of each attribute. We accomplished this task in two stages to ease respondent burden. First, subjectswere simply asked to rate the relative importance of each of the attributes to them when purchasing eggs or pork chops ona scale of 1–7, where 1 was very unimportant and 7 was very important (step 2). Respondents were encouraged not to rateeverything as very important, but rather to think about the relative importance. In step 3, these ratings were used toprovide a starting point for ‘‘importance weights’’ assigned to each of the attributes, where the weights summed to 100across each of the attributes. In step 3, respondents adjusted the weights to match their own preferences.

Once step 2 had been completed, sufficient data were available to calculate each person’s attribute-based utility for apork chop or carton of eggs. In particular, a person’s utility for a product is calculated by multiplying the relativeimportance of each attribute (normalized to sum to one over all attributes), by each attribute’s rating. Let Ik represent the

Table 1Attributes and attribute levels used in preference elicitation experiment.

Attribute Levels

Eggs1. Price (dozen eggs) $0.50, $1.50, $2.50, $3.50, $4.502. Barn space per hen (sq inches) 48, 69, 100, 171, 252, Z3533. Barn floor space per hen (sq inches) r97, 111, 129, 155, Z1944. Beak trimming Beaks are not trimmed, beaks are trimmedo10 days old, beaks are trimmed when older

than 10 days5. Room for scratching, foraging, and dustbathing (sq feet per hen)

0, 1, 2

6. Nest availability No nests, group nests—no bedding, group nests—with bedding, individual nests—nobedding, individual nests—with bedding

7. Free range No free range, free range without predator protection or shelter, free range with predatorprotection, free range with shelter, free range with predator protection and shelter

8. Group size 43000 hens, 43000 hens with perches, 2000 hens, 2000 hens with perches,o7 hens,o7hens with perches

9. Type of feed Non-organic, non-organic with flaxseed to add omega 3 fatty acids, organic

Pork1. Price (2 lb package) $2, $4, $6, $8, $102. Space per gestating sow (square feet) 14, 30, 60, 90, 120, Z1503. Space per nursing sow (square feet) 14, 30, 60, 90, 120, Z1504. Space per growing pig (square feet) 8, 16, 24, 32, 40, Z485. Nesting provisions No straw/no privacy, with straw/no privacy, no straw/with privacy, with straw/with privacy6. Survival rate of farrows 50%, 70%, 80%, 90%, 99%7. Minor surgeries None, performed wheno7 days old, performed when older than 7 days8. Free range No free range, free range without shelter or pasture, free range with no shelter and with

pasture, free range with shelter and no pasture, free range with shelter and pasture9. Group size (number of sows) 1, 5, 10, 20, 3010. Provision of dry straw (inches) 0, 3, 6, 1211. Type of feed Non-organic, non-organic without hormones or antibiotics, organic

4 A copy of the presentations is available online at http://asp.okstate.edu/baileynorwood/Survey4/Default.aspx.



stated importance of the kth attribute, where SIk¼1. Further, let Rkl represent the rating of the lth level of the kth attribute,which is normalized so that the lowest rated level of each attribute has a scaled rating of 0 and the highest rated level ofeach attribute has a scaled rating of 1. This normalization ensures that the preference rankings of levels within an attributeare between zero and one and are not confused with the relative importance of attributes as a whole.

Individual i’s utility for pork chop or egg carton option j is

Vij ¼ Zij#IiPPj, where Zij ¼XK

k ¼ 1

XLk

l ¼ 1

WklðIkRklÞ ð1Þ

1 2 3 4 5 6 7 8 9 10

1 2 3 4 5 6 7 8 9 10

1 2 3 4 5 6 7 8 9 10

1 2 3 4 5 6 7 8 9 10

1 2 3 4 5 6 7 8 9 10

1 2 3 4 5 6 7 8 9 10

4

4

4

4

4

4

4

4

4

Increase 1% Decrease 1%


11.11%



11.11%



11.11%



11.11%



11.11%



11.11%



11.11%



11.11%



11.11%

100%

$0.00

$0.00

$0.00

$0.00

Step 1

Step 3

Step 2

Fig. 1. Steps in the calibrated auction-conjoint (CAC) valuation method. Step 1: rate the desirability of attribute levels. Step 2: indicate the relativeimportance of each attribute (used to provide starting weights in Step 3). Step 3: indicate bid for cage system eggs, and then alter the importance weighton each attribute to change bids for other egg products (user may also go back and change answers to previous questions.



where IiP (Ik) is the normalized ‘‘importance weight’’ placed on the price attribute (attribute k), Pj is the price of option j, Lkis the number of levels over which the kth attribute is varied, K is the number of non-price attributes, and Wkl is a dummyvariable that equals 1 if the product possesses the lth level of the kth attribute, and 0 otherwise.5 The term IkRkl can beinterpreted as a utility ‘‘part-worth,’’ which is the utility provided from the lth level of the kth attribute. The part-worth isanalogous to the coefficients in a random utility model estimated from a conjoint analysis, with Wkl being the explanatoryvariable for the presence or absence of an attribute in the conjoint analysis.

Eq. (1) is the basis from which a typical conjoint study would estimate consumer preferences and willingness-to-payfor product attributes. That is, a typical conjoint-type study would have stopped at this point. The problem with stoppingat this point and utilizing the utility function calculated in (1) is that people are unable to directly see the consequences oftheir rating decisions and the trade-offs implied. To put the ratings in an economic context which have real consequences,the CACM goes a step further. In particular, Eq. (1) was used to calculate, for each person, their willingness-to-pay for fivedifferent pork or egg products that differed according to the underlying attributes. Willingness-to-pay to have one productversus another is determined by calculating the non-price utility differences between the products (using Eq. (1)) anddividing by the ‘‘part-worth’’ on price, which represents the marginal utility of income. For example, individual i’swillingness-to-pay for product j rather than product t is

WTPij-WTPjt ¼ ðZij-ZitÞ=IiP : ð2Þ

Prior to revealing the willingness-to-pay calculation to respondents, however, two additional training/informationphases were necessary. First, people were introduced to the five products that systematically differed by the underlyingattributes. People were given another presentation describing each of the products which detailed exactly how theydiffered from one another on the underlying spectrum of attributes.6 Table 2 lists each of the five products/productionsystems and shows how they differ along the continuum of attributes studied in the hog and egg treatments. The productswere completely described by the underlying attributes.

Following the presentation, subjects were told that they would have the opportunity to participate in auction topurchase the five packages of pork chops (each were two-pound packages of fresh pork chops that were identical exceptfor the attributes in question) or five egg products (each were one dozen brown eggs that were identical except for theattributes in question). Then, people were trained on the bidding procedures used. In particular, people first bid to buy acandy bar using a Becker–DeGroot–Marshack (BDM) mechanism.

The participants were shown numerous examples and were provided several justifications for why it was in their bestinterest to bid truthfully—i.e., submit bids exactly equal to their maximum willingness-to-pay. Once participants weretrained using the single candy bar auction, another auction was held for five different types of candy bars to train peoplefor subsequent bidding for five pork/egg products.

At the completion of the training phase, participants were shown a screen on their laptop that looked like step 3 inFig. 1. Because the ratings only provide an indication of the relative desirability of products, Eq. (1) cannot identify an‘‘overall’’ or ‘‘total’’ willingness-to-pay amount—only differences. Thus, people were first asked to use the drop-down boxto place a bid for eggs from cage system (or pork from the crate system). Then, using each person’s previous responses andEq. (1), a bid was forecasted for each of the other four products. Respondents were told that the computed bids were‘‘intelligent guesses’’ of how they would value each of the products based on their previous answers, but that they wouldlikely want to alter their bids. To change the bids, respondents had to change the ‘‘importance weights’’ shown on the left-hand side of the screen assigned to each attribute (see step 3, Fig. 1) and they could also go back and change any of theirprevious ratings if they so desired. For example, if someone wanted to bid a higher amount for eggs from a free-rangesystem relative to the cage system, then they would need to increase the relative weight given to the attribute of freerange. If someone wanted to increase (or decrease) all bids proportionally, then they simply had to decrease (or increase)the ‘‘importance weight’’ assigned to price.

Respondents were given numerous examples of how to change the importance weights to generate bids that matchedtheir preferences and were given ample time to complete the task. Three experimental monitors assisted any participantsthat had difficulty making the bids fit their true preference. Once a participant was comfortable with their bids for theproducts from the five production systems, they hit a ‘‘submit bids’’ button and waited for other participants to finish. Byrequesting that people change the importance weights to alter their bids, each person was effectively calibrating Eq. (1) tomatch what they were willing to bid on each of the products; the result is a utility function that is systematically matchedwith people’s values in a context that has real economic consequences.

Once all bids were submitted, then one of the five pork/egg products was randomly selected as binding, and one of thebids for the product was randomly selected as binding. No one’s bid was made public information, and people wererepeatedly reminded that all bids were private information and would not be revealed. At the conclusion of theexperiment, the binding bidder was taken to a different room, and their bid was compared with the randomly drawn

5 Eq. (1) posits a linear attribute-based utility function. This seemingly strong assumption is not as problematic as it might first appear (see thediscussion section for more on this issue).

6 A copy of the presentations is available online at http://asp.okstate.edu/baileynorwood/Survey4/Default.aspx. The entire program used to elicitpreferences is also available at the web site.



‘‘secret price.’’ If the bid was greater than the secret price, the purchase ensued; otherwise, the participant paid nothingand received nothing.7

4. Results

In total, 291 people participated in the study. Across all treatments, 54% of subjects were male, while 46% were female.People of all age groups were well represented with 11%, 19%, 23%, 22%, 19%, and 7% of the sample falling in the followingin the respective age categories, 18–24, 25–34, 35–44, 45–54, 55–64, and 65 or older. Almost half the sample had aBachelor’s’s degree or higher. About 36% had an annual household income less than $40,000, about 37% had annualhousehold income between $40,000 and $80,000, and the remaining 27% had household income greater than $80,000. Interms of race, about 55% of the sample was white, 24% was black or African American, 12% was Hispanic, and 6% of therespondents were of Asian descent. Seven observations were removed from the data set for reasons such as: theparticipant fell asleep during the session, the participant entered answers on the lap-top before the moderators presentedinstructions on how to answer the questions, and the participant failed to answer one of the questions needed to completethe analysis.

Table 3 reports participant’s bids for a dozen eggs from each of five production systems. Across all locations, the meanbid for a dozen eggs from the cage system was $0.92, and this value increased by $0.61–$1.53 for eggs from the barnsystem. Interestingly, bids for eggs from the aviary system were only $1.41, which is less than that for eggs from the barnsystem. An aviary system is like a barn system except that it has vertical layers or floors such that more birds can be

Table 2Egg and pork production systems.

Attribute Production system

Cage Barn Aviary Aviary with freerange

Organic

Eggs2. Barn space per hen (sq inches) 69 155 186 186 1863. Barn floor space per hen (sq inches) 69 155 97 97 974. Beak trimming Trimmed o10

daysTrimmed o10days

Trimmed o10days

Trimmed o10days

Trimmed o10days

5. Room for scratching, foraging, and dustbathing (sq. feet per hen)

0 1.35 1.35 1.35 1.35

6. Nest availability No nests Individual nestswith bedding

Individual nestswith bedding

Individual nests withbedding

Individual nestswith bedding

7. Free range No free range No free range No free range Free range withshelter and predatorprotection

Free range withshelter andpredator protection

8. Group size o7 43000 43000 withperches

43000 with perches 43000 withperches

9. Type of feed Non-organic Non-organic Non-organic Non-organic Organic

Crate Pen Open barn Pasture Organic

Pork2. Space per gestating Sow (square feet) 14 24 90 90 903. Space per nursing sow (square feet) 14 14 90 90 904. Space per growing pig (square feet) 8 8 32 32 325. Nesting provisions With privacy/no

strawWith privacy/nostraw

With privacy/withstraw



6. Survival rate of farrows (%) 90 90 80 70 707. Minor surgeries Performedo7

daysPerformedo7days

None None None

8. Free range No free range No free range Free range withshelter and nopasture

Free range withshelter and pasture

Free range withshelter and pasture

9. Group size (number of sows) 1 5 20 20 2010. Provision of dry straw (inches) 0 0 12 12 1211. Type of feed Non-organic Non-organic Non-organic Non-organic Organic

7 If there is a perfect field substitute, then bids might be censored at the field price. However, most people do not know what type of eggs or pork theybuy or how the animal was raised. Thus, there are not perfect field substitutes for the products we auctioned in the sense that respondents cannotdetermine exactly how the animals were raised in the products they are buying from the supermarket. This makes our auctioned products relatively uniqueto the experimental setting. It is possible there is some downward pressure on cage and crate pork bids as a result of the availability of field substitutes;however, the fact that our egg and pork bids were almost universally less than the price of field substitutes suggests that censoring is not a major issue.



housed on a unit of land. Table 2 shows that the aviary system differs from a barn system in that it has more barn space perhen, but less barn floor space per hen. Apparently, the gain in total barn space was not sufficient to offset the disutility ofthe decrease in barn floor space, resulting in eggs from the aviary system being valued less, on average, than that from abarn system. One might question whether such a preference ranking is ‘‘rational.’’ The answer is yes; the objectivescientific rankings of animal well-being provided by De Mol et al. [13] indicate that animal well being is indeed slightlylower in an aviary system as compared to a barn system. The results in Table 3 indicate that the aviary with free range andorganic systems are most valued by consumers at $1.87 and $2.23 per dozen, respectively.

Table 4 reports descriptive statistics on participant’s bids for a two pound package of pork chops from five differentpork production systems. Across all locations, the mean bid for pork from the crate system was $1.47. Bids for pork from apen system were about $0.12 higher or $1.59. Although bids for pork from the pen system are higher on average than thecrate system (a difference which is statistically significant), the magnitude is not particularly large. Bans on gestationcrates have been proposed in several states (and have already happened in Arizona, California, Colorado, and Florida), andit is likely that many producers, if they stay in business, would transition from a crate-type system to one like a pen systemafter a ban. Our results suggest that while such a change is positively valued, it is not markedly so.

It is instructive to compare the average premium for gestation-crate free pork obtained in our study, $0.12 for twopounds of pork, to the $2.11 (per choice between one pound pork chops) estimated by Tonsor, Olynk, and Wolf (2008).8

Why the stark difference? There are many differences in the two studies, and as such there is unlikely to be a definitiveanswer to this question; however, it is important to note that the choice experiment used by Tonsor et al. [35] onlyincluded a single attribute related to animal welfare: the use or non-use of gestation crates. We hypothesize that thesubjects in our experiments significantly discounted space per sow when they recognized that space per sow is only one ofmany factors affecting animal welfare. Although our study only found a relatively small value associated moving from acrate to pen system, by contrast, bids were almost twice as high for the open barn, pasture, and organic systems ascompared to the crate system at $3.30, $3.49, and $3.80, respectively. As can be seen by looking at the attributes in Table 2,much of this increase results from of a move to a free range system.

In the introduction, we argued that the CACM can promote the rational formation of preferences. An indirect means ofasking whether the CACM caused people to behave differently than they might in an alternative mechanism that does notinvolve feedback and inter-activity is to ask whether the CACM yields answers that would differ from a simpler self-explicated approach. We can investigate this issue by investigating how the weights assigned to the product attributes

Table 3Distribution of bids for one dozen eggs from five egg production systems.

Cage Barn Aviary Aviary withfree range

Organic

Chicago, IL (N¼48)Median $0.80 $1.31 $1.22 $1.84 $2.28Mean $0.79 $1.49 $1.38 $1.94 $2.35Standard deviation $0.73 $1.42 $1.14 $1.40 $1.67Minimum $0.00 #$1.86 #$0.35 $0.00 $0.00Max $3.00 $6.52 $4.65 $5.64 $8.23Percent zero bids 27.08 12.50 12.50 12.50 12.50

Dallas, TX (N¼42)Median $0.75 $1.38 $1.29 $1.67 $2.04Mean $0.83 $1.57 $1.37 $1.88 $2.32Standard deviation $0.65 $1.48 $1.17 $1.68 $2.01Minimum $0.00 $0.00 $0.00 $0.00 $0.00Max $2.00 $7.72 $5.53 $8.81 $9.17Percent zero bids 23.81 19.05 21.43 19.05 16.67

Wilmington, NC (N¼49)Median $0.62 $0.96 $0.92 $1.26 $1.49Mean $1.11 $1.52 $1.47 $1.78 $2.02Standard deviation $1.76 $2.29 $2.13 $2.30 $2.58Minimum $0.00 #$0.91 $0.00 $0.00 $0.00Max $8.47 $10.15 $9.85 $10.14 $14.14Percent zero bids 10.20 4.08 4.08 2.04 2.04

Pooled (N¼139)Median $0.70 $1.26 $1.15 $1.55 $1.95Mean $0.92 $1.53 $1.41 $1.87 $2.23Standard deviation $1.19 $1.78 $1.56 $1.83 $2.12

8 Tonsor et al. [35] find a lower estimate from the simpler multinomial logit model ($1.13), but one that remains substantially higher than ourestimate of $0.12. They also report that willingness-to-pay for legislation to ban gestation crates is effectively zero.



changed as subjects moved from step 2 to step 3 (see Fig. 1). Our data indicates that 99% of participants changed theimportance weights when moving from step 2 (where there was no feedback or interaction) to step 3 (where there wasfeedback). Most people changed their weights to make the price attribute more important. For example, in the porktreatment, the importance weight on price increased, on average, by 33.3 percentage points from 11.8% to 45.1%.Nevertheless, there was significant heterogeneity in that almost 18% of participants decreased the weight placed on price.On average, the attribute experiencing the largest decrease in attribute weight was the survival rate of baby pigs (meandecline of 4.1%), but here again, there was significant heterogeneity with about 20% increasing the weight on this attributeand 80% decreasing the weight on this attribute. These results provide evidence that the CACM caused people to behavedifferently than they would if participating in a simpler non-interactive preference elicitation method.

Although one could have simply utilized traditional experimental auction procedures to generate the bids shown inTables 3 and 4, the CACM method yields more information than could be extracted from a conventional auction. Indeed,bids for each of the systems can be decomposed into people’s values for each of the underlying attributes comprising theproduction system. Given the calibrated attribute-based utility function, we can calculate willingness-to-pay for eggs orpork from any production system comprised of any combination of attributes shown in Table 1. To illustrate, Table 5reports mean and median willingness-to-pay for selected marginal changes in attribute level for pork.9

Table 5 also shows that respondents were willing to pay $0.46 to increase the survival rate of baby pigs from 60% to90%. The findings in Table 5 illustrate how complicated an issue like a gestation crate ban can be, and that the overall valuepeople might derive from such a ban is comprised of many factors—all of which may not be apparent to a person unlessthey have participated in an experiment like the one discussed in this article. For example, banning gestation andfarrowing crates would increase space per sow—something people value; however, it could results in group sizes largerthan 10 and could also result in lower survival rates. It is easy to see how adding up all these ‘‘partial effects’’ could resultin a negative value for a gestation and farrowing crate ban. Again, we emphasize that Table 5 only provides a partialpicture of the many values that the CACM provides.

Another purported advantage of the CACM is that it permits a parametric-free characterization of heterogeneity inpreferences. Table 5 showed that, on average, people prefer that baby pigs be administered the optional surgeries of taildocking and teeth trimming before seven days of age rather than no surgeries at all. Although such a finding might seem

Table 4Distribution of bids for two pounds of pork chops from five pork production systems.

Crate Pen Open barn Pasture Organic

Chicago, IL (N¼48)Median $1.50 $1.92 $3.23 $3.42 $3.57Mean $1.84 $2.01 $3.83 $3.96 $4.27Standard deviation $1.97 $2.08 $3.90 $4.04 $4.35Minimum $0.00 #$0.32 #$0.01 #$0.03 #$0.02Max $7.00 $6.85 $21.59 $22.44 $24.66Percent zero Bids 37.50 16.67 12.50 10.42 10.42

Dallas, TX (N¼48)Median $0.00 $0.00 $1.12 $1.36 $1.70Mean $0.74 $0.66 $2.58 $2.85 $3.37Standard deviation $1.12 $1.52 $5.16 $5.93 $8.32Minimum $0.00 #$5.78 #$0.07 #$0.07 #$0.02Max $5.00 $5.11 $34.79 $40.11 $57.32Percent zero bids 56.25 37.50 31.25 31.25 33.33

Wilmington, NC (N¼48)Median $1.18 $1.33 $3.41 $3.31 $3.40Mean $1.83 $2.10 $3.49 $3.64 $3.74Standard deviation $1.75 $2.14 $3.08 $3.24 $3.33Minimum $0.00 $0.00 #$0.68 #$0.67 #$0.19Max $8.80 $9.97 $16.53 $17.56 $17.56Percent zero bids 8.33 8.33 6.25 6.25 4.17

Pooled data (N¼144)Median $1.00 $1.09 $2.89 $2.92 $3.11Mean $1.47 $1.59 $3.30 $3.49 $3.80Standard deviation $1.72 $2.03 $4.14 $4.55 $5.73

9 The willingness-to-pay values shown in Table 5 and Fig. 3 exclude the responses of any individual that gave the attribute of price an importanceweight of 90% or higher. Such a high importance weight, in this context, implies that people did not want to buy any of the products at all, and thus, thevalues in Table 5 should be interpreted as willingness-to-pay conditional on ‘‘being in the market’’ or bidding at least some positive amount for one of thefive products. Our personal experience was that respondents who placed such a high weight on price made little effort to adjust the importance weightsof non-price attributes to reflect their personal preferences.



counterintuitive, one should consider the consequences of not performing such surgeries—injury and fighting later in life.Although the ‘‘average’’ person would prefer to see such surgeries performed, as long as they occur prior to seven days oflife, when the pain is temporary, one might expect a great deal of diversity in opinion on the issue. Fig. 2 plots thedistribution of willingness-to-pay for no surgeries vs. surgeries performed prior to seven days. As can be seen in Fig. 2,there is wide disagreement on the merits of surgeries, with the largest fraction of consumers either having slightly positiveor slightly negative willingness-to-pay.

Table 5Marginal willingness-to-pay values for selected changes in pork production practices.

Change Mean Median Lower 95% CIa Upper 95% CIa

Space60 vs. 14 ft2 per gestating sow $0.35nb $0.24 $0.27 $0.4360 vs. 14 ft 2 per nursing sow $0.26n $0.17 $0.19 $0.3332 vs. 8 ft 2 per growing pig $0.37n $0.24 $0.26 $0.48

Group size5 sows vs. 1 sow $0.06 $0.00 #$0.08 $0.2010 sows vs. 1 sow #$0.19n #$0.11 #$0.33 #$0.0530 sows vs. 1 sow #$0.45n #$0.29 #$0.59 #$0.31

SurgeriesPerformedo7 days vs. none $0.14 $0.01 $0.03 $0.24Performed47 days vs. none #$0.15n #$0.06 #$0.25 #$0.05

Survival rate of farrows90% vs. 60% $0.47n $0.39 $0.35 $0.5980% vs. 60% $0.28n $0.17 $0.19 $0.3770% vs. 60% $0.13n $0.07 $0.07 $0.19

OtherOrganic vs. non-organic feed $0.41n $0.13 $0.08 $0.74Free range with no pasture or shelter vs. no free range $0.08n $0.00 $0.03 $0.13Free range with shelter but no pasture vs. no free range $0.43n $0.13 $0.08 $0.77Free range with pasture and shelter vs. no free range $0.82n $0.42 $0.38 $1.27

Note: results are the average value across all three locations for people who placed less than 90% importance on price; N¼109.a Reported statistics are the upper and lower 95% confidence intervals for the mean WTP.b One asteristic (n) indicates the mean is significantly different from zero at the p¼0.05 level of lower according to a two-tailed t-test.

0%

5%

10%

15%

20%

25%

(-2.

4, -2

.2]

(-2.

2, -2

]

(-2,

-1.8

]

(-1.

8, -1

.6]

(-1.

6, -1

.4]

(-1.

4, -1

.2]

(-1.

2, -1

]

(-1,

-0.8

]

(-0.

8, -0

.6]

(-0.

6, -0

.4]

(-0.

4, -0

.2]

(-0.

2, 0

)

[0]

(0, 0

.2]

(0.2

, 0.

4]

(0.4

, 0.

6]

(0.6

, 0.

8]

(0.8

, 1]

(1 ,

1.2]

(1.2

, 1.

4]

(1.4

, 1.

6]

(1.6

, 1.

8]

(1.8

, 2]

Perc

enta

ge o

f Res

pond

ents

Willingness-to-Pay Category ($)

Fig. 2. Distribution of willingness-to-pay for no surgeries vs. surgeries performed within 7 days in pork production.



The recent interest in preference heterogeneity has generated refinements in statistical methods but relatively littleimprovement in the construction of valuation instruments. Latent class and mixed logit models rely on statistical modelingto infer preference heterogeneity, but require heterogeneity to be viewed through the lens of the assumed model. Theheterogeneity in Fig. 2 is not an inferred distribution; it is a direct observation of individual differences in values. As such,we can be more confident that the heterogeneity observed represents true differences in preferences.

4.1. Validity of the CACM

As with any new method it is important to ask whether the results generated are valid. Validity is a multifacetedconcept, but generally speaking, it can be thought of as the extent to which a particular empirical measurement accuratelycorresponds to what one wishes to measure. It is a difficult to matter to determine whether the WTP values resulting fromthe CACM are valid because the ultimate object of interest—people’s ‘‘true’’ preferences—are latent and unknown.Nevertheless, it is useful to examine several indicators of validity—construct and convergent validity.

According to Carson et al. [6], construct validity ‘‘refers to how well the measurement is predicted by factors that onewould expect to be predictive a priori.’’ Thus, construct validity relates to how well measures correspond with predictionsof economic theory (e.g., are demand curves downward sloping?, do responses pass the scope test?) in addition to whetherthe WTP measures relate intuitively to individual-specific characteristics in ways that would be expected (e.g., are heavierconsumers of pork WTP more for pork? Are higher income consumers WTP more than lower income consumers?). Contentvalidity can also be ascertained by investigating whether WTP for egg or pork product attributes change in ways thatwould be expected given what is known about the relationship between animal welfare and the product characteristics.10

The results clearly suggest that demand curves are downward sloping. At the aggregate level, the un-normalized ratingof desirability (ranging from 1 to 10) assigned to the highest price option for pork (eggs) was 2.56 (1.89), whereas themean desirability rating given to the lowest-price option was 7.63 (7.64). Thus, lower prices were preferred to higherprices (demand curves are downward sloping). These results also hold-up at the individual-level. For example, not a singleindividual in the pork studies indicated that a price of $4 was more desirable than a price of $2; and 138 out of 139 (or99.3%) said that than $1.50 eggs were more desirable than $1.00 eggs.

The results also pass the scope test. For example, in the hog studies, as the piglet survival rate changed from 50% to 70%to 80% to 90% to 99%, the average un-normalized desirability rating went from 3.72 to 4.83 to 6.30 to 7.72 to 8.46,respectively. Table 5 shows these results in terms of WTP estimates. At the individual level, over 90% of participantsindicated a preference for 99% survival rate over 50% survival rate. Similar results hold for other characteristics. Consider,for example, barn space per hen. As barn space per hen increased from 48 to 69 to 100 to 171 to 252 to Z353 squareinches, the mean un-normalized desirability rating increased from 1.76 to 2.67 to 3.56 to 4.61 to 6.74 to 8.70, respectively.At the individual level, over 95% of respondents indicated that Z353 square inches was more desirable than 48 squareinches.

Another indicator of construct validity is the extent to which WTP measures correspond to individual-specificcharacteristics in intuitive ways. One would expect that those individuals who consumed more pork and eggs would bewilling to pay more for pork and eggs. The results are consistent with this a priori expectation. For example, the mean bidfor cage eggs is about $1.42 for those who purchase more than four dozen eggs in a month (n¼11), whereas the mean bidfor cage eggs is only $0.87 for those who purchase fewer than four dozen eggs a month (n¼128). As with eggs, pork bidsare positively correlated with consumption of pork. For example, the mean bid for crate pork is about $2.29 for those whopurchase eight pounds of pork or more in a month (n¼27), whereas the mean bid for crate pork is only $1.15 for those whopurchase fewer than eight pounds of pork a month (n¼117).

One would also expect that participants with higher incomes would be less price sensitive, which in our context impliesa lower weight assigned to the price attribute. The results are suggestive of this hypothesis. For example, the correlationcoefficient between the importance score assigned to price and a participants’ income was #0.18 for pork and #0.10 foreggs. These correlations are significantly different from zero at the p¼0.03 and 0.22 levels of statistical significanceaccording to a two-tailed t-test.

A final aspect of content validity relates to whether WTP for various egg or pork product characteristics changes in waysthat would be expected given the literature on animal well-being. Table 5 shows that people are willing to pay $0.35, onaverage, for a gestating sow to have 60 rather than 14 square feet of space. Interestingly, the value for the same spacechange is less for nursing sows, $0.26 on average. The mean WTP for an increase in 46 square feet of space is significantlyhigher for gestating sows as compared to nursing sows according to a paired, two-tailed, t-test (p-value¼0.01). Thisfinding is completely logical because sows spend a much longer period of their life in gestation as compared to nursing,and this result suggests a strong systematic component to people’s behavior. Table 5 shows that while people value anincrease in group size from 1 to 5 sows, they actually dislike increasing group size from 1 to 10 sows. Again, this finding isa completely logical reaction to animal well being, with people valuing socialization as hogs are moved from isolation tosome level of companionship, but with people recognizing the decline in welfare that can result when group size becomes

10 By construction, the CACM passes at least one measure of content validity in that the auction bids must vary with utility function parameters inthe way described by Eqs. (1) and (2).



large and inter-fighting arises. The model developed by Bracke et al. [3] indicates that sow welfare increases when the sowgroup size rises from 1 to 5 sows, but begins to decline when the group size exceeds 10 sows.

It is also useful to consider issues related to convergent validity, which refers to whether results from the CACM arehighly correlated with other measurements or indicators of animal welfare. Support for convergent validity can be foundby comparing people’s bids with expert opinion on animal welfare. Bracke et al. [3] and De Mol et al. [11] constructedmodels based on surveys of the animal science literature to generate predictions of animal well being in differing pork andegg production systems, respectively. Fig. 3 plots the median bid (for the entire sample) for eggs/pork from the four non-organic production systems (we exclude organic because providing organic feed does not change animal well being)against the animal welfare scores from the models developed by Bracke et al. [3] and De Mol et al. [11]. Results suggest astrong positive relationship between median bids and objective scientific ratings of animal well being. Indeed, thecorrelation coefficients between bids and animal welfare scores are in excess of 0.90 for both eggs and pork, suggestingthat the results pass this particular test of convergent validity.

5. Discussion of the CACM

No preference elicitation method is a panacea, and most popular approaches, such as dichotomous choice referendaquestions or choice experiments, have advantages and disadvantages. The CACM is no different. As such, it is useful toconsider in more detail some of the advantages and disadvantages of the CACM relative to existing elicitation approaches.

One of the seemingly strong assumptions of the CACM method is the linear utility function used to translate the self-explicated conjoint ratings into valuations (see Eq. (1)). We argue that this assumption is not as restrictive as it might firstappear. One can easily modify the CACM to incorporate non-linearities by bundling attributes and describing levels thatcombine the attributes. In fact, for the free range attribute used in this application, the utility for free range, outsideshelter, pasture, and protection for predators is not additive given the way we constructed the attribute levels (see theresults in Table 5). That is, Eq. (1) makes the utility function appear linear, but only because each of the ‘‘free range’’interaction effects is described as a different ‘‘level’’ of an attribute. Thus, one can include interactions in CACM by carefullywording the different attribute levels respondents are asked to rate.11

Moreover, because we asked respondents to indicate their preferences for numerous discrete levels of each explanatoryvariable, preferences can vary across individuals in a multiplicity of ways that are non-linear in spite of the fact Eq. (1)appears to posit a linear form. For example, WTP for increased space per hen is not assumed to increase in space in a linear,quadratic, or logarithmic fashion. Each person’s WTP varies with space per hen according to their own choices. Toillustrate, consider Fig. 4, which shows five randomly selected participants’ values to increase gestating sow space from 14sq inches. One participant’s WTP is flat for all levels of space, another’s WTP is increasing at an increasing rate, and theother participants’ WTPs are increasing at a decreasing rate, but with different maxima. As Fig. 4 clearly shows, there is nouniform functional form imposed across individuals; subjects’ responses dictate the slope of the valuation function. In fact,

$0.50

$1.00

$1.50

$2.00

$2.50

$3.00

0

Med

ian

Bid

for P

rodu

ctio

n Sy

stem

Animal Welfare Score fromBracke et al. (2002) and DeMol et al. (2006) Models

Eggs

Pork

1 2 3 4 5 6 7 8

Fig. 3. Relationship between participant’s bids and animal welfare for eggs and pork from different production systems.

11 For example, if it is hypothesized that Attribute A can become more valuable in the presence of Attribute B, the CACM can handle this by askingsubjects to rate a scenario where both Attributes A and B are present, in addition to Attributes A and B appearing alone.



it is possible to interpret the CACM as having no functional form at all, save for the restriction that the total value for agood must equal to the sum of the value its attributes.

Random utility models (RUMs) of the sort used in analyzing data from choice experiments are not necessarily bound toa particular functional form for the utility function. One must assume a function form with a RUM to obtain empiricalestimates, but the data can also be analyzed using other functional forms to determine which form best fits the data.Nevertheless, the vast majority of RUM applications utilize a linear utility function without any specification tests. Forchoice experiments and conjoint analyses, this choice is often made because the experimental design used to generate thechoice or ranking data can only support the identification of a linear utility specification. Thus, most choice experimentsand conjoint analyses, like the CACM, impose assumptions about the utility function during data collection.

To analyze data from a RUM requires assumptions about: (1) the functional form of the utility function, (2) thedistribution of the overall random-utility stochastic error term (often assumed to be distributed type I extreme value), and(3) in the case of random parameter models, the distributions of the individual utility parameters in the population. TheCACM is, in some senses, much more general than RUMs because it only requires the first assumption—the functional formfor the utility function.

One advantage of RUMs is that they specify an explicit stochastic structure for analyzing choices. By contrast, the theoryof valuation by auction (see [21]) is deterministic as are most other models of consumer behavior. With the CACM, it isunclear whether or how a stochastic element might need to be added to account for inconsistent choice patterns orbehavioral deviations from the imposed functional form. Of course, different people provide different WTP estimates inthe CACM, and it is the variability across people that allows one to determine the degree of sampling error that underliesthe estimated valuation distribution (as, for example, with the 95% confidence internals reported in Table 5); however, theCACM does not provide an explicit means of analyzing different types of error. RUMs, by contrast, at least provide somemechanism for conceptualizing stochastic choice. Nevertheless, RUMs, as traditionally conceived, include an error term notbecause it is believed that an individuals’ utility is inherently stochastic, but because of an assumption that the analystcannot perfectly observe the utility function with data on only a few discrete choices. With the CACM, however, the analystobserves much more than a few discrete choices, and given the interactive and repetitive process embedded in the CACM,participants are given ample opportunity to fully describe their preferences in the context of the imposed preferencestructure.

There have been other methods proposed to improve the rationality of survey responses. Bateman et al. [1], forexample, proposed a ‘‘learning design’’ where people answered several repeated valuation questions. They showed, in thecontext of animal welfare, that repetition and learning were needed for preferences to converge to standard expectations.In particular, they showed that the well-known starting point bias that exists in double-bounded dichotomous choicequestions dissipated as respondents gained experience with the topic and answered repeated valuation questions.Bateman et al. [1] argued that their results were consistent with a model in which (p. 127),‘‘preferences converge towardsstandard expectations through a process of repetition and learning.’’

Our approach is similar to that used by Bateman et al. [1] in that it relies on learning and repetition; however, the CACMis more paternalistic because it imposes a consistency between utility and valuation. Whether this ‘‘forced rationality’’ is anegative or positive aspect of the CACM is debatable, and people of different ideologies are likely to come to differentconclusions. It should be noted that while the CACM paternalistically forces valuations to be consistent with utility, it does

$0.00

$0.20

$0.40

$0.60

$0.80

$1.00

$1.20

$1.40

30

Will

igne

ss-to

-Pay

Pre

miu

m O

ver 1

4 ft2

Space per Gestating Sow (ft2)60 90 120 150

Fig. 4. Effect of changes in space per gestating sow on willingness-to-pay premium over pork from sows given 14 ft2 space for five randomly selectedparticipants.



not place any constraint on the level or magnitude of valuations; if someone wants to pay more for pork from a systemwhere only 50% of piglets survive than one in which 99% survive, the CACM will allow them to do so. The CACM simplyforces the respondent to be consistent: if a respondent indicates with simple rating scales that 99% survival rates are moredesirable than 50% survival rates, then the CACM requires that WTP for 99% survival rate to be higher than that for 50%survival rate. It is difficult to imagine many people who would view such a constraint as an egregious violation ofconsumer sovereignty—especially when one realizes that the CACM allows respondents to freely change their answers torating scales if they do not like the resulting WTP values.

Another potential criticism of the ‘‘forced rationality’’ imposed by the CACM relates to the method’s applicability tofirms for use in marketing decisions. It may not be desirable to force people to be consistent if they fail to act in such amanner in the marketplace. Consumers often act in capricious ways, and retailers often make money as result.Nevertheless, such capriciousness may be a short-run phenomenon. For a marketing scheme to be effective over thelong-run, it should relate in some fundamental way to people’s ‘‘true’’ preferences, and withstand the whims of consumersand the provision of information. But, estimates of people’s private-good values are also needed for use in public-policyanalysis, and here it makes sense to put people in an environment where they are given their ‘‘best shot’’ at discoveringand revealing their preferences.

Finally, it is instructive to note that the irrationality documented in previous surveys and experiments may well be dueto the particular preference elicitation methods used. For example, choice experiments often require subjects to peruselarge choice sets containing numerous attributes, some of which are new to the subject, and include certain designfeatures that decrease choice consistence (e.g., see [12]). Such elicitation approaches pose a higher cognitive burden thanmost market settings, and encourages a type of irrationality that the CACM seeks to mitigate by asking simple ratingquestions.

6. Conclusions

A conflict exists when projecting the market impacts of food policies or the success of new products. Decision makersneed information on people’s preferences to make such projections; however, research suggests preference elicitationstudies often suffer from biases that raise questions about the validity of such information. In this paper, we introduced ahybrid valuation method that has the potential to assist people in providing more systematic responses than is the casewith many other existing methodologies. With the calibrated, auction-conjoint method (CACM), we link people’s bids forproducts, obtained in an incentive compatible auction, to their preference ratings of product attributes. Because theapproach forces people to change their preference ratings to alter bids, the method imposes an internal consistency onbehavior.

Our results show that people’s values for egg and pork products are affected by animal living conditions, and that theexpressed willingness-to-pay values are highly correlated with scientific models of animal well-being. The advantage ofthe CACM is that the auction bids can be decomposed to determine why, for example, people willing to pay more for porkfrom a pasture system as compared to a crate system. Results indicate that moving to a free-range system is highly valued,but only when accompanied with shelter and pasture. Although pasture systems can result in lower survival rates for babypigs, our results indicate that the disutility of a decline in survival rate is more than offset by the extra utility of a move topasture.

Although one can decompose the value of a product into its attributes using standard auction or conjoint methods, theCACM achieves the same outcome for many more attributes and attributes levels and without imposing a high level ofcognitive burden. The first two steps of the CACM involve the relatively simple rating tasks, requiring no more effort thanis needed to answer simple Likert questions. Step 3 of the CACM requires more thought that the previous two steps, butthe use of laptops greatly reduces the cognitive burden, and is an integral part of the ‘‘self-calibration’’ that is at the heartof the CACM. Not only is the CACM relatively easy for participants to complete, analyzing data obtained from the CACM isstraightforward especially considering the large amount of information extracted. The CACM delivers individual-level,point-estimates of peoples’ willingness-to-pay for each attribute level requiring no statistical model and no econometricestimation.

Of course, much remains to be learned. Although we argue that the CACM is conceptually superior to auctions orconjoint used alone, future research might further explore empirical issues related to internal and external validityassociated with the CACM. Some of the most interesting areas of inquiry relate to whether the CACM can alleviatebehavioral anomalies witnessed in previous research. Although our particular application related to people’s preferencesfor private goods, the general idea behind the CACM is perfectly applicable to valuing environmental or other public goods.It is of perhaps less useful in mail surveys (because of the need to ask questions that depend on previous answers), but theCACM is easily adaptable to phone surveys, internet surveys, or as our application demonstrates, in-person interviews. Aswith virtually all survey-based approaches to value public goods, it is a challenge to make the decision task incentivecompatible and this is true of the CACM as it is with conjoint rankings, choice experiments, and traditional contingentvaluation dichotomous choice questions. However, one could use still the interactive and iterative process imbedded in theCACM but simply refrain from actually holding a binding auction. Although the bids or willingness-to-pay amounts wouldbe hypothetical with such an approach, they would be linked to an underlying utility function and could be calibrated inthe same way as our approach. Realism and consequentiality might be increased by using the people’s answers to predict



how they might vote in a referendum on a public good; if respondents disagreed with the calculated vote, they would needto go back and re-calibrate the utility function.

Future research might focus on more general questions related to the appropriateness of paternalistic policies vs.paternalistic value elicitation. Previous research shows that people can make rational and systematic choices if given theproper institutional setting and feedback to do so [7,25,26]. The question is whether economists should use approaches toencourage people to behave more rationally. Our answer to this question is to ask: when would you feel comfortable withresearchers using your valuation for an unfamiliar good to inform public policy and cost–benefit analysis? When we placeourselves in the role of the participant rather than the researcher, an approach like the CACM has much appeal.

References

[1] I.J. Bateman, D. Burgess, W.G. Hutchinson, D.I. Matthews, Learning design contingent valuation (LDCV): NOAA guidelines, preference learning andcoherent arbitrariness, J. Environ. Econ. Manage. 55 (2008) 127–141.

[2] R.M. Bennett, Farm animal welfare and food policy, Food Policy 22 (1997) 281–288.[3] M.B.M. Bracke, B.M. Spruijt, J.H.M. Metz, W.G.P. Schouten, Decision support system for overall welfare assessment in pregnant sows a: model

structure and weighting procedure, J. Anim. Sci. 80 (2002) 1819–1834.[4] J. Braga, C. Starmer, Preference anomalies, preference elicitation and the discovered preference hypothesis, Environ. Resource Econ. 32 (2005) 55–89.[5] F. Carlsson, P. Frykblom, C.J. Lagerkvist, Consumer willingness-to-pay for farm animal welfare: mobile abattoirs versus transportation to slaughter,

Eur. Rev. Agr. Econ. 34 (2007) 321–344.[6] R. Carson, N.E. Flores, N.F. Meade, Contingent valuation: controversies and evidence, Environ. Resource Econ. 19 (2001) 173–210.[7] T.L. Cherry, T. Crocker, J.F. Shogren, Rationality spillovers, J. Environ. Econ. Manage. 45 (2003) 63–84.[8] T. Cherry, J.F. Shogren, Rationality crossovers, J. Econ. Psychol. 28 (2007) 261–277.[9] Y.P. Chu, R.L. Chu, The subsidence of preference reversals in simplified and marketlike experimental settings, Amer. Econ. Rev. 80 (1990) 902–911.

[10] C.G. Davis, B.H. Lin, Factors affecting U.S. pork consumption, U.S. Department of Agriculture, Economic Research Service, Report no. LDP-M-130-01,2005.

[11] R.M. De Mol, W.G.P. Schouten, E. Evers, H. Drost, H.W.J. Houwers, A.C. Smits, A computer model for welfare assessment of poultry productionsystems for laying hens, Ned. J. Agr. Sci. 54 (2006) 157–168.

[12] J.R. DeShazo, G. Fermo, Designing choice sets for stated preference methods: the effects of complexity on choice consistency, J. Environ. Econ.Manage. 44 (2002) 123–143.

[13] Future of Animal Agriculture in North America, Farm Foundation, Oak Brook, IL, 2006.[14] W.M. Goldstein, H.J. Einhorn, Expression theory and the preference reversal phenomenon, Psych. Rev. 94 (1987) 236–254.[15] D. Grether, C. Plott, Economic theory of choice and the preference reversal phenomenon, Amer. Econ. Rev. 69 (1979) 623–638.[16] D. Kahneman, A. Tversky, Prospect theory: an analysis of decision under risk, Econometrica. 47 (1979) 263–291.[17] G. Keren, W. Wagennar, Violations of utility theory in unique and repeated games, J. Exp. Psych. 13 (1987) 387–391.[18] S. Lichtenstein, P. Slovic, Reversals of preference between bids and choices in gambling decisions, J. Exp. Psych. 89 (1971) 46–55.[19] C. Liljenstolpe, Evaluating animal welfare with choice experiments: an application to swedish pig production, Agribusiness 24 (2008) 67–84.[20] J.J. Louviere, D.A. Hensher, J.D. Swait, Stated Choice Methods: Analysis and Application, Cambridge University Press, Cambridge, UK, 2000.[21] J.L. Lusk, J. Shogren, Experimental Auctions: Methods and Applications in Economic and Marketing Research, Cambridge University Press,

Cambridge, UK, 2007.[22] J.J. McGlone, E.H. von Borell, J. Deen, A.K. Johnson, D.G. Levis, M. Meunier-Salaun, J. Morrow Reeves, J.L. Salak-Johnson, P.L. Sundberg, Review:

compilation of the scientific literature comparing housing systems for gestating sows and gilts using measures of physiology, behavior,performance, and health, Prof. Anim. Sci. 20 (2004) 105–117.

[23] F.B. Norwood, J.L. Lusk, Compassion by the Pound: The Economics of Farm Animal Welfare, Oxford University Press, Oxford, UK, 2011.[24] T. O’Donoghue, M. Rabin, Doing it now or later, Amer. Econ. Rev. 89 (1999) 103–124.[25] C. Plott, Rational individual behavior in markets and social choice processes: the discovered preference hypothesis, in: K. Arrow, E. Colombatto,

M. Perlemann, C. Schmidt (Eds.), The Rational Foundations of Economic Behavior, Macmillan and St. Martin Press, London 1996, pp. 225–250.[26] C.R. Plott, K. Zeiler, The willingness to pay—willingness to accept gap, the ‘endowment effect,’ subject misconceptions and experimental procedures

for eliciting valuations, Amer. Econ. Rev. 95 (2005) 530–545.[27] B.E. Rollin, Animal agriculture and emerging social ethics for animals, J. Anim. Sci. 82 (2004) 955–964.[28] J.F. Shogren, S.Y. Shin, D.J. Hayes, J.B. Kliebenstein, Resolving differences in willingness to pay and willingness to accept, Amer. Econ. Rev. 84 (1994)

255–270.[29] V. Srinivasan, C.S. Park, Surprising robustness of the self-explicated approach to customer preference measurement, J. Marketing Res. 34 (1997)

286–291.[30] V. Srinivasan, P. deMaCarty, Predictive validation of multiattribute choice models, Marketing Res. 11 (1999) 29–34.[31] R. Sugden, The opportunity criterion: consumer sovereignty without the assumption of coherent preferences, Amer. Econ. Rev. 94 (2004) 1014–1033.[32] R. Sugden, Why incoherent preferences do not justify paternalism, Constit. Polit. Econ. 19 (2008) 226–248.[33] R. Sugden, Market simulation and the provision of public goods: a non-paternalistic response to anomalies in environmental evaluation, J. Environ.

Econ. Manage. 57 (2009) 87–109.[34] Task Force Report, A comprehensive review of housing for pregnant sows, J. Amer. Vet. Med. Assoc. 227 (2005) 1580–1590.[35] G.T. Tonsor, N. Olynk, C. Wolf., Consumer preferences for animal welfare attributes: case of gestation crates, J. Agr. Appl. Econ. 41 (2009) 713–730.[36] K.E. Train, Discrete Choice Methods with Simulation, Cambridge University Press, Cambridge, MA, 2003.[37] A. Tversky, P. Slovic, D. Kahneman, The causes of the preference reversal phenomenon, Amer. Econ. Rev. 80 (1990) 204–217.[38] R.H. von Haefen, D.J. Phaneuf, Identifying demand parameters in the presence of unobservables: a combined revealed and stated preference

approach, J. Environ. Econ. Manage. 56 (2008) 19–32.


Date post:	17-Jan-2022
Category:	Documents
Upload:	others
View:	4 times
Download:	0 times

Author's personal copy - École Polytechnique

Documents