MEDIA ADVERTISING AND BALLOT INITIATIVES: THE CASE OF ANIMAL WELFARE REGULATION

MEDIA ADVERTISING AND BALLOT INITIATIVES: THE CASE OF ANIMALWELFARE REGULATION

TIMOTHY RICHARDS, WILLIAM ALLENDER and DI FANG

Spending on political advertising increases with every election cycle, not onlyfor congressional or presidential candidates, but also for state-level ballot initiatives.There is little research in marketing, however, on the effectiveness of political advertis-ing at this level. In this study, we conduct an experimental analysis of advertisementsused during the 2008 campaign to mandate new animal welfare standards in Califor-nia (Proposition 2). Using subjects’ willingness-to-pay for cage-free eggs as a proxyfor their likely voting behavior, we investigate whether advertising provides real infor-mation to likely voters, and thus sharpens their existing attitudes toward the issue, orwhether advertising can indeed change preferences. We find that advertising in supportof Proposition 2 was more effective in raising subjects’ willingness-to-pay for cage-freeeggs than ads in opposition were in reducing it, but we also find that ads in support ofthe measure reduce the dispersion of preferences and thus polarize attitudes toward theinitiative. More generally, political ads are found to contain considerably more “hype”than “real information” in the sense of Johnson and Myatt [Johnson, J. P., and D. P.Myatt. “On the Simple Economics of Advertising, Marketing and Product Design.”American Economic Review, 96, 2006, 756–84]. (JEL D12, D72, K32, L66, M37)

I. INTRODUCTION

In 2008, the citizens of California voted infavor of Proposition 2, which bans the use ofcages for housing egg-laying chickens.1 Simi-lar propositions in Arizona (Proposition 204 inNovember 2006) and other states mandate stall-free and crate-free housing for sows and calvesfor veal, respectively. Media advertising is usedheavily in state-level propositions because ofthe narrow focus of the issue, the geographic

Richards: Morrison School of Agribusiness and ResourceManagement, Arizona State University, 7171 E. SonoranArroyo Mall, Peralta 335U, Mesa, AZ 85212. Phone1-480-727-1488, Fax 1-480-323-2294, E-mail [email protected]

Allender: Morrison School of Agribusiness and ResourceManagement, Arizona State University, 7171 E. SonoranArroyo Mall, Peralta 335D, Mesa, AZ 85212. Phone1-480-727-1488, Fax 1-480-323-2294, E-mail [email protected]

Fang: Morrison School of Agribusiness and ResourceManagement, Arizona State University, 7171 E. SonoranArroyo Mall, Peralta 335E, Mesa, AZ 85212. Phone1-480-727-1488, Fax 1-480-323-2294, E-mail [email protected]

1. In July 2010, the governor signed follow-on legis-lation that banned the sale of caged eggs in California,preventing the importation of eggs from other states thatare produced in a way that is deemed inconsistent with theprinciples laid out in Proposition 2.

concentration of likely voters, and the (typically)highly polarized nature of the campaigns. Thereis evident support for this issue among someconsumers as cage-free eggs sell for a signifi-cant premium in retail stores (sometimes $1.75per dozen or more). What is less clear, how-ever, is whether voting in initiatives similar toProposition 2 is driven by advertising-inspiredmass-support of the issue at hand, or whether itis highly motivated support by a small segmentof the population that cannot be influenced byadvertising. In this study, we examine the role ofmedia advertising in the initiative process usingan experimental analysis of ads used by bothsupporting and opposing sides for Proposition 2in California in November 2008.

Polling voters’ intentions is fraught with dif-ficulties. Often, likely voters will tell the poll-ster what they think the pollster wants to hear.Second, more important for the narrow issuesthat are often the subject of public referenda,many voters simply do not care, or are unaware

ABBREVIATIONS

BDM: Becker–DeGroot–MarschakWTP: Willingness-to-Pay

1

Contemporary Economic Policy (ISSN 1465-7287)doi:10.1111/j.1465-7287.2011.00292.x© 2011 Western Economic Association International

2 CONTEMPORARY ECONOMIC POLICY

of the issue (Rothschild 1978). In the issue athand, only a certain percentage of the popula-tion are even egg consumers or animal loversso the issue seems abstract, at best. Third, vot-ers sometimes lack a sense of consequentialityif they are asked their intent and they are notactually in the voting booth so put little thoughtinto their response. In this study, we circumventthese problems by facing voters—consumersin this case—with real economic incentives inan experimental environment to determine theirwillingness-to-pay (WTP) for cage-free eggs.WTP, while important from a food market-ing perspective, is also a valuable proxy forlikely voting intentions and an accurate gaugeof the importance consumer-voters place on anissue. Ultimately, if a policy is a good oneand “deserves” a vote, it should be welfare-improving. Therefore, in this study we assess thevalue of political advertising not on how manyvotes it generates, but on how it impacts theway the vote should go, or how the ad affectseconomic welfare.

WTP is not without limitations as a measureof the expected improvement in welfare becauseof a policy change. Hamilton, Sunding, andZilberman (2003) argue that individuals vote inreferenda in ways that reflect their choices overpublic goods, and not necessarily private goods,so their choices regarding public policy alterna-tives may not reflect their private WTP. Further,Lusk and Norwood (2009a,b) and Norwood andLusk (2011) believe that conventional methodsof estimating WTP for public-good attributes aresubject to a significant “social desirability bias”whereby a consumer’s stated valuation includesan estimate of what he/she thinks the researcherwould regard as appropriate. These authors showthat social desirability bias with respect to ani-mal welfare issues can be mitigated througha method of “inferred valuation,” while Luskand Norwood (2010) and Olnyk, Tonsor, andWolf (2010) address social desirability bias inan animal welfare context through indirect ques-tioning, or asking survey respondents to expresshow they think the “average American” wouldrespond. In the current study, we are not asconcerned with social desirability bias as votingoutcomes only depend upon whether the WTPfor cage-free eggs is greater than for conven-tional eggs, not necessarily how much.

Animal welfare is an important issue inits own right, and has become increasinglyprominent as consumers become more consciousabout what they eat, and where it came from. In

addition to the studies cited above, others studiessuggest that consumers perceive products withanimal-friendly attributes to have a higher qual-ity due to ethical beliefs, taste, food safety, orhealth benefits (Harper and Makatouni 2002;Lusk and Norwood 2007; Ophuis and Oude1994).2 Regulating animal welfare, however,presumes a market failure, but how does themarket fail to adequately provide for the well-being of animals used to produce our food? Tothe extent that the method of production is a cre-dence attribute—an attribute that can impact theindividual agent’s health, well-being, or satisfac-tion with the product, but is not readily apparentin inspecting or consuming the product—thenthe usual asymmetric-information argumentsarise (Hartl and Herrmann 2009; Larue et al.2002; Lusk, Roosen, and Fox 2003). Alterna-tively, if we consider animals in general to bean appropriable resource similar to water, min-erals, or lumber, then many believe that pro-ducing food according to commercial methodsimposes a negative production externality onsociety. Producing cage-free eggs or stall-freepork are thus means of internalizing the negativeexternality. Further, Carlsson, Frykblom, andLagerkvist (2007a) argue that consuming meatproduced using conventional methods imposes anegative consumption externality on society. Ineither case, advertising in animal welfare bal-lot initiatives can therefore be interpreted asa means of convincing voters to voluntarilyself-impose a tax meant to address the exter-nality problem, much like a tax on fossil fuelis a means of reducing the costs imposed bygreenhouse gas accumulation.

Others do not attempt to address whether themarket fails, but estimate the premium for foodsproduced in ways that are consistent with pre-sumed animal welfare standards. For example,Tonsor, Olnyk, and Wolf (2009a) also use WTPas a proxy for consumers’ likely voting behav-ior and report a premium of $1.89 per poundfor pork raised without the use of gestationcrates, while these authors report a premium of$2.11 per pound in a separate study (Tonsor,Olnyk, and Wolf 2009b). Carlsson, Frykblom,and Lagerkvist (2007b), on the other hand, find

2. The issue has assumed a global dimension as ini-tiatives similar to Proposition 2 in Canada have begun totake shape on University campuses and among other animal-welfare interest groups (Potstra 2008). Uzea and Hobbs(2009), however, suggest that a subset of Canadian con-sumers has very high standards for animal welfare and tendto drive legislation that risks “over-regulating” industry.

RICHARDS, ALLENDER & FANG: MEDIA ADVERTISING 3

that other factors associated with meat pro-duction are more important than animal wel-fare attributes in a sample of European meatconsumers, while Gracia, Louriero, and Nayga(2009) find a statistically significant premiumof approximately 0.42 Euro for ham markedwith a label indicating that it was producedusing methods that respect the animal’s wel-fare. Among studies that estimate the demandimpact of Proposition 2 using secondary data,Lusk (2010) interprets the media advertisingduring the campaign for and against Proposi-tion 2 as an exogenous shock to demand. Com-paring the demand for regular and cage-freeeggs in California with the demand in a marketnot subject to Proposition 2 advertising, he findsthat the demand for cage-free eggs increased by180% over the period of the campaign. Sim-ilarly, Allender and Richards (2010) find thatconsumers in California are willing to pay asignificant premium for eggs labeled as cage-free using a large-scale household-panel data set.Meanwhile, Chang, Lusk, and Norwood (2010)find that consumers are willing to pay a 57%premium for cage-free eggs, but estimate that42% of this premium is due instead to the colorof cage-free eggs and not the cage-free attributeitself. Taken together, therefore, there is consid-erable evidence that consumers place a positivepremium on animal-welfare attributes, but verylittle evidence on the differential effects of adsfor or against the passage of animal welfare reg-ulations or the welfare effects that result fromshifting or rotating demand curves.

The welfare impacts of taxing egg-producersor imposing inefficient production techniquesare critically dependent on whether advertisingshifts the demand curve by changing prefer-ences (referred to as “hype” by Johnson andMyatt 2006 and typically associated with Dixitand Norman 1978) or rotates the demand curvethrough the provision of “real” informationabout the product and its alternatives (Nelson1970, 1974). While the current orthodoxy inempirical advertising studies is to either assumeadvertising only shifts the demand curve, or tointeract advertising and prices to obtain sometype of ad hoc measure of the rotational effect,we apply a new empirical approach to dis-tinguish between the two. If consumers areindeed heterogeneous in their preferences, thena demand curve essentially represents a proba-bility distribution of WTP across consumers thatare highly knowledgeable and involved with theproduct, and those with only casual knowledge

of, and preference for, the product. This is theobservation that led Johnson and Myatt (2006)to develop a formal model of how advertis-ing, and other media activities, really affectsdemand. We develop the model more completelybelow, but the intuition is straightforward. Ifa product appeals to a “niche” market, or ifthe issue is of intense interest to a concen-trated group of voters, then advertisements thatappeal to the core of this market will increasethe dispersion of consumers’ WTP and rotate thedemand curve clockwise, increasing the WTPof the marginal consumer whose WTP is abovethe mean and reducing it if the marginal con-sumer is below the mean. Reducing the disper-sion of demand means that the advertisementcontains real information as it allows consumersto sharpen their opinion of the product or, in thiscase, the issue. If a consumer strongly prefers aproduct or side of an issue, then real informa-tion regarding that issue will merely reinforcehis or her position and move them away fromthe mean of all voters. If, on the other hand,the issue is of interest to the mass market orall voters who have at least a passing interestin the topic, then a successful ad will reducethe dispersion of demand and rotate the demandcurve counter-clockwise (make it more elastic).In this case, the ad will increase the WTP of themarginal consumer who begins with a valuationbelow the mean.

In either case, if the preference-effect of thead dominates, then the pure shift effect domi-nates either rotation effect. If successful, expo-sure to the ad will change everyone’s attitudetoward the issue and move the entire distributionof preferences in the direction of either favoringor opposing the issue. Econometrically, the testis straightforward: if the ad reduces the varianceof WTP, then it appeals to the diffuse mass ofvoters with casual interest, but if it increasesthe dispersion of WTP then we conclude thatthe ad appeals to a small, yet focused groupof voters. If the dispersion does not change,but preferences do, then the preference-effect ismore important. Either way, we have a betterunderstanding of how political advertisementscan be better crafted depending on the nature ofthe issue at hand.

Our econometric model provides another wayof testing for asymmetric effects of differ-ent types of information. Positive and negativeinformation tend to have fundamentally differenteffects on consumer’s attitude toward a prod-uct or an issue. The same is likely true in


the case of political initiatives, where adver-tisement is inherently manipulative. Swartz andStrand (1981) and Smith, van Ravensway, andThompson (1988) conduct empirical studies ofpositive and negative information regarding con-taminated food scares and find that the initial,negative information has a far stronger effect inreducing demand than subsequent positive infor-mation has in restoring demand. Similarly, Fox,Hayes, and Shogren (2002) use an experimen-tal method similar to the one used here to findthat negative information regarding the potentialhealth effects of irradiated pork greatly outweighany positive information. Strongly negativeinformation effects can be explained by the loss-aversion theory of Viscusi, Magat, and Huber(1987) in which consumers are particularly sen-sitive to any negative variation in utility from thestatus quo. Losses are magnified while gains areminimized. The Viscusi–Magat–Huber theory,however, is observationally equivalent in thesesituations to the predictions of prospect theorydeveloped by Kahneman and Tversky (1979).Prospect theory maintains that individuals assesslosses and gains with respect to a referencepoint, which in this case is their expectationthat eggs are produced in uncontroversial ways.When confronted with information that realityis somewhat worse than their expectations, theymagnify the negative result, while ignoring anyinformation that may lead them to believe thathens are more happy than they expect. Ourexperimental data, and the econometric modelused to test our hypotheses regarding WTP arewell-suited to examining the issue of asymmetryand loss-aversion in the informative content ofpolitical ads.

Political advertising has received little recentattention in the marketing literature, despite itsevident, and growing, importance with eachelection cycle. Palda (1975) conducts a sem-inal econometric analysis of the effectivenessof advertising expenditure on determining theoutcome of Congressional races. Arguing that“revealed preference” data are sufficient toanalyze the effectiveness of campaign spend-ing, Palda (1975) shows a significantly pos-itive marginal return to campaign spendingin Quebec, even after explicitly recognizingthe endogeneity of campaign spending. For-malizing Palda’s (1975) insights, Rothschild(1978) develops a conceptual model of politicaladvertising in which he challenges the ortho-doxy at the time—the “limited effects model”—which maintains that media advertising can have

little impact on an otherwise stable politicalprocess. Rather, in a low involvement environ-ment such as voting, advertisement is likely tobe highly successful. He offers empirical datain support of his thesis, but only of an aggre-gate form. Similarly, Soley and Reid (1982)show that spending on political advertising hasas great an effect on voting outcomes as partyaffiliation and incumbency—two factors thatare widely regarded as the most important indetermining electoral results. Each of these stud-ies, however, investigate campaign spendingin general election cycles and not on specificinitiatives.

An explosion of ballot initiatives in recentyears, many supported by aggressive advertis-ing campaigns, has re-ignited interest in the roleof advertising in the exercise of direct democ-racy. Gerber (1999) finds that ad spending infavor of ballot propositions is not effective,while spending against is. The implication ofthis research is important: “economic groups,”that is groups that are small in membership, butwell-financed, cannot simply buy outcomes indirect democracy as is commonly believed. Thisis an important outcome as it suggests that directdemocracy through the initiative process is morelikely to result in policies that reflect major-ity interests (are closer to the median voter’spreferences) than is the more usual legisla-tive process. Specifically, Gerber (1999) findsthat “. . .initiatives that received majority supportfrom citizen interests passed at a significantlyhigher rate than those that received majority sup-port from economic interests. . . .” She also findsthat “. . .the set of successful initiatives receiveda greater share of supporting contributions fromcitizen interests than from economic inter-ests, whereas unsuccessful initiatives receiveda much larger share of supporting contribu-tions from economic interests” (pp. 119–120)which suggests that higher spending on cam-paigns, and hence more media advertising, isnot necessarily effective in increasing the like-lihood that the proposition will be successful.Stratmann (2006) is one of the few empiricalstudies to directly estimate the effectiveness ofad spending not on Congressional or Presiden-tial elections, but on ballot measures similar tothe one under study. Stratmann (2006) addressesthe received wisdom in the academic literaturethat interest-group campaigning is, in fact, noteffective either for or against a measure, butinstead the status quo or the side advocatingno change has a strong advantage. However,


Stratmann (2006) recognizes that previous sta-tistical analyses do not control for the endogene-ity of campaigning—that spending is inherentlystrategic—so any estimate that does not controlfor endogeneity will be biased and inconsistent.By developing a research design that accountsfor endogeneity (a two-way fixed effects model),he finds that special interest spending doesindeed exert a positive and significant effecton the outcome of the vote. Understanding theeconomic role of the initiative process is crit-ical to appreciating the importance of advertis-ing for or against ballot propositions. Matsusaka(2005) explains the rise of the referendum pro-cess in terms of the three economic functionsit provides: (1) resolving principal-agent prob-lems between voters and their representatives,(2) correcting fundamental asymmetric informa-tion problems on many issues, and (3) bundlingissues together in one focused decision pro-cess. For these reasons, political initiatives havebecome a central part of how democracy worksin the United States. But, in order for the ini-tiative process to be effective, each side hasto be able to inform (or persuade) the elec-torate, and advertising is the primary way thisis done.

This article makes a number of contributionsto the policy and marketing literature. First, wedevelop a new method of estimating the effectof marketing activities in an important, yet lit-tle studied category of advertising expenditure:voter behavior in public referenda. Second, oureconometric model provides a way of separat-ing the preference effect of advertising from itsinformation effect. In doing so, we also offera new explanation for the observed asymmetriceffects of positive and negative information onconsumers’ preferences. Third, we propose andapply a metric against which policies may bejudged as being either for or against the publicinterest. The role of marketing in the formationof public policy is too often overlooked, despiteits obvious importance in the exercise of directdemocracy.

Our research has three objectives. The firstand primary objective of this study is to deter-mine the WTP for food products raised in a“humane” way, or one that is fundamentally dif-ferent from current practice. The second objec-tive is to determine the relative effect on WTPof media advertising presented either in supportof an animal welfare initiative, or counter toit. The third objective is to determine whethermedia advertising shifts or rotates the demand

curve, and the welfare implications of whetherthe dominant effect is a shift or a rotation.Ultimately, we seek a better understanding ofthe role of campaign spending on matters thatmaterially impact the ways in which firms canconduct themselves.

II. RESEARCH METHOD AND EXPERIMENTALDESIGN

To determine the WTP for cage-free eggs,we use a nonhypothetical experiment in whichwe offer subjects the opportunity to purchaseeggs that are clearly labeled as cage-free. Statedpreference methods, such as surveys or choiceexperiments (conjoint analysis), are often usedto determine consumers’ WTP for new foodproducts (Lusk et al. 2001). If the productsdo not currently exist, however, it is difficultto elicit “homegrown” or personal values forthese products as the buyers have no basis foraccurate comparison. Moreover, participants instated choice data-gathering exercises have noreal incentive to reveal their true demand asthey have no economic stake in the outcomeof the survey. List and Gallet (2001) find thatrespondents state values 2–20% greater in hypo-thetical questions relative to nonhypotheticalvaluation questions. A nonhypothetical exper-iment, on the other hand, has the ability touncover consumers’ true WTP because partic-ipants in the experiment are provided real eco-nomic incentives to make decisions that providethe most benefit at the lowest possible cost. Con-sequently, we conduct a nonhypothetical exper-iment in which we use real economic incentivesto elicit consumers’ WTP for cage-free eggs.Because these attributes are not apparent byinspecting the goods themselves in the absenceof labels to that effect, we are essentially obtain-ing likely market prices for what are termed“credence” attributes, or attributes that the con-sumer must trust the producer to include in theproduct.

Recruitment services were contracted to athird-party marketing firm, Octagon Marketing,Inc. of Phoenix, AZ. We chose Arizona residentsfor our sample because they were not exposedto the television ads that ran in Californiaduring the Proposition 2 campaign. Subjectswere told that they had to be at least 18 yearsof age, and a consumer of eggs to participate.Each participant was guaranteed to earn noless than $40 plus a candy bar and case ofeggs for participating. Subjects who agreed to


participate were allocated to 1 of 10 potentialtime slots from 8:00 a.m. to 3:00 p.m. on eitherSaturday or Sunday of the test weekend on afirst-come-first-served basis. Our final sampleconsists of 178 subjects who have identifiedthemselves as egg consumers, selected from thegeneral population to be broadly representativeon demographic and socioeconomic grounds.

All auction procedures are carried out usingthe z-Tree (Fischbacher 2007) software system,which is an open-source software tool thatallows for a collaborative bidding process andthe automated assignment of winners and losersand, of course, market prices for the commodityunder bid.

On the two days of the auction, the exper-iment consisted of seven steps. Each of thesesteps was followed in identical fashion from onesession to the next. The procedure and designelements in each of the seven steps is describedbelow:

Step 1: In step 1, each participant wasassigned to a computer terminal when theyarrived at the lab. When all participants expectedfor that session were seated, they were instructedto ensure that their machine had a gray screenwith the z-Tree logo on it. All subjects werethen shown a Powerpoint presentation of theexperiment instructions, which were read by theexperimenter. We informed each participant thattheir information would be kept strictly confi-dential and their participation in the experimentwas completely voluntary. Next, we describedthe animal welfare issue and how it relates to theway in which eggs are produced on farms. Wethen confirmed that the research was not affili-ated with any organization other than ArizonaState University and took any questions thatparticipants might have had. Participants wereasked not to communicate with one another fromthat point forward as any exchange of informa-tion regarding eggs or animal welfare may biasthe results.

Step 2: In step 2, participants were intro-duced to the auction and the auction procedure.Subjects were provided an initial endowmentof $45.00, a regular-size chocolate bar and adozen regular (noncage-free) eggs. They werethen instructed that they could either use the$45.00 to upgrade from a regular-size chocolatebar to a king size in the practice round, upgradefrom regular to cage-free eggs in the auction, ortake the money and items without completing atransaction. All participants were told that oneround of the auction would be chosen at random

to be binding, and that they were to pick up theireggs, candy bar, and payment after the sessionwas complete and they had signed their paymentreceipt. We use a Becker–DeGroot–Marschak(1964, BDM) auction mechanism to determineWTP. Using a BDM mechanism is intendedto ensure truthful value elicitation while mini-mizing the possibility that we exclude marginalbidders (those who are likely to have a lowvalue for cage-free eggs) from the auction. Themechanics of the BDM procedure were carefullyexplained to the subjects, including the fact thatit is incentive compatible, or in their best inter-ests to report their true indifference amounts.3

The BDM auction works as follows. All subjectsare initially endowed with a regular-size candybar and a dozen regular (noncage-free) eggs.In each round (and in the practice round), sub-jects are asked to submit their WTP to upgradeto either a king-size candy bar (practice round)or cage-free eggs (actual auction). The exper-imenter then draws a price at random from auniform distribution between zero and the max-imum WTP submitted. All bidders willing topay equal to or more than this random pricereceive the upgrade, while those below the ran-dom price do not. The “market price” in theauction is the random price. The BDM mech-anism is demand revealing because the bidderdoes not know what the market price will be, butdoes understand that every price between zeroand the maximum bid has an equal probabilityof becoming the market price. Therefore, if heor she were to shade their bid in order to savemoney, they may lose an item they value and ifthey bid above their true value, they risk buyingsomething for more money than it is worth.

Step 3: Once the BDM mechanism had beenexplained carefully, we went through a simpleexample involving candy bars that demonstratedhow the BDM mechanism works. Once we wereconfident that all the participants understoodhow the game was going to be played andanswered any questions one might have had,

3. The incentive compatibility of the BDM mechanismhas been questioned by Horowitz (2006), among others. Weassume that the argument advanced there, that the agent’sWTP depends on the distribution of future values, is ofminor consequence. A popular alternative is the random nth-price auction (Shogren et al. 2001). These authors show thatthe nth-price auction is not only demand-revealing (incentivecompatible), but is better able to elicit WTP from marginalbidders, or those who may place only a small value on cage-free eggs than a more usual, second-price auction. However,a BDM mechanism is equally adept at eliciting marginalbidders and is not subject to the over-valuation problemsascribed to the nth-price auction (Lusk and Shogren 2007).


we conducted a practice auction involving candybars. A practice round was necessary to ensureparticipants understood how the BDM auctionworks and to familiarize them with the software(z-Tree). In the practice round, participants wereshown a regular size candy bar (Snickers orMilky Way) and were told that they werecurrently endowed with the regular size bar,meaning that they could leave at any time andtake the bar home with them. They were thengiven the opportunity of buying up to a king-size version of the same bar through the BDMauction. They were instructed to enter their bidsfor what a king-size bar is worth to them overa regular-size bar and then wait for the results.Once all bids were submitted, a random pricewas drawn between 0 and the highest bid, thewinners were tabulated, and all participants weretold whether they won automatically via z-Tree.They were also shown the random price that wasdrawn.

Step 4: After completing the practice round,the participants were then instructed that wewould begin the cage-free egg auction. The sam-ple was divided into ten groups of 15 partici-pants each: two groups bid on cage-free eggswith only prior information regarding the exis-tence of Proposition 2 (the control group); threegroups bid after being shown a short media clipfrom a popular television show in which we hadembedded ads developed in support of Propo-sition 2 (procage-free legislation); three groupsbid after being shown ads against Proposition 2(anticage-free legislation); and two groups bidafter being shown both sets of ads. The orderin which the groups bid was varied randomly,with subjects in each group submitting bids infive separate auction rounds. One round wasselected (randomly) as binding in order to deter-mine whether the subject successfully upgradedto cage-free eggs, or retained his or her endow-ment of regular eggs. All bids were recordedelectronically through the z-Tree software andsaved in a file for further processing.

The procage-free ads generally consisted ofemotional appeals from Hollywood celebritiessuch as Ed Asner, Alicia Silverstone, and othernotable animal-rights supporters to vote “Yes”on Proposition 2 in order to prevent cruelty tofarm animals. On the other hand, the anticage-free ads were less emotional and appealed toviewers’ affinity with the small farmers whoraise eggs in California, their concern withgrowers’ freedom of choice, and their owntestimonial that chicken-cages are not as harmful

as they are purported to be. Ads for boththe “pro” and “anti” Proposition 2 sides wereembedded in a short clip from The Simpsons inorder to mimic the actual viewing experience asnearly as possible.4

Step 5: After all rounds were completed, thesubjects were asked to fill out a survey designedto elicit a complete set of demographic, socioe-conomic and attitudinal information, includingage, income, education, employment, maritalstatus, frequency of egg consumption, attitudetoward animal welfare and knowledge of leg-islative propositions focusing on animal welfare.The complete survey instrument is availablefrom the authors. Determining each participant’sattitude toward animal welfare and the Califor-nia proposition were important to control forprior beliefs in forming opinions regarding aparticularly polarizing issue. Frequency of eggconsumption is also an important piece of dataas each participant is likely to have a differentlevel of involvement with eggs and, therefore, adifferent incentive to either gather more infor-mation, be receptive to the information contentof the ads, or to try alternative egg products. Wealso included a question regarding participants’awareness of the salmonella outbreak traced toeggs from Iowa in September 2010. Because thelargest egg recall in history occurred the weekprior to our experiment, we were concerned thatparticipants may associate cage-free eggs witheggs that are less likely to be contaminated by abacteria that is easily passed among animals inconfined spaces.5

Step 6: Once the surveys were all completed,each participant was instructed to leave thecomputer lab and enter the adjoining room toreceive their check, eggs, and candy bar, andto sign the payment receipt form. Of the 178participants, only a few expressed any degree ofuncertainty regarding either the validity of thechecks or their expectation to be paid, in cash,immediately.

Step 7: The final step involves downloadingand analyzing the bid and survey data. Oncethe data are collected, we estimate a regres-sion model of WTP values and media treatmentsusing a random-parameters multinomial logitestimation method (mixed logit) described inmore detail below. With the mixed logit model,

4. All video clips are available from the authors uponrequest.

5. Our empirical results found that their awareness ofthe salmonella outbreak did not have a significant impact onWTP for cage free eggs.


we are able to accurately estimate the marginalWTP for the specific attributes of interest—the“cage-free” attribute and whether pro- or anti-Proposition 2 messaging is more effective—forconsumers in different demographic market seg-ments. We then use these values to calculateaggregate welfare measures by market segmentbased on existing egg demands and the expectedmarket adjustments that will occur if cage-freeproduction methods are mandated.

In this way, the results of our research maybe used to inform policymakers as to the trueeconomic cost of the propositions, the distribu-tional effects of regulating production practices,and to inform marketing managers in egg mar-keting firms as to how they should price andpromote new food products to maximum advan-tage. More importantly, the empirical modeldescribed in the next section is able to determinewhether media advertising of either type “shifts”the demand curve for cage-free eggs or rotatesit. This distinction is important both in the calcu-lation of welfare outcomes, and in determiningthe nature of the advertising effect: whether itprovides information to consumers/voters, caus-ing the demand curve to become more elastic(Nelson 1970, 1974), or whether it is inherentlypersuasive, changing tastes and causing demandto become less elastic (Dixit and Norman 1978).

III. EMPIRICAL MODEL OF MEDIA ADVERTISING

We test for shift or rotation effects ofmedia advertising using an empirical modelthat captures the theoretical effects of adver-tising described by Johnson and Myatt (2006).Conceptually, the information content of adver-tising is difficult to separate from what mayotherwise be described as hype, or the pref-erence effect. While the historical debate cen-ters on whether advertising changes preferences(Dixit and Norman 1978), provides informa-tion (Norman 1974), or provides a comple-mentary good to the product being purchased(Becker and Murphy 1983), Johnson and Myatt(2006) develop a general theory of demand thatrelies on none of these behavioral assumptions.Rather, Johnson and Myatt (2006) develop afundamentally new perspective on how adver-tising works. Rather than simply shift demand,advertising in their model operates on the disper-sion of consumer valuations for the product. Ifadvertising provides “real information,” in theirterminology, then the dispersion of valuationsfor the product is likely to rise. Consumers who

value the product relatively highly before theadvertisement will like it even more after thead, and those who value it less highly will like iteven less. Demand rotates clockwise. For firmsthat sell a homogeneous product, designed tosell to the mass-market, this rotation in demandreduces the valuation of the marginal consumer,reducing profits. In this case, the marginal con-sumer is “below average” in terms of his WTPfor the product. Firms that sell highly differenti-ated products, designed to appeal to a niche mar-ket, however, prefer advertisements that rotatedemand in this way because their marginal con-sumer is “above average.” The WTP for thisconsumer rises with a clockwise rotation indemand. Firms that sell homogeneous goods,therefore, prefer to use advertising that resultsmainly in a preference effect as it shifts demandoutward at each price. These firms would ratherminimize the dispersion of valuations and thusrotate the demand curve counter-clockwise, rais-ing the valuation of their marginal consumerand, hence, profits.

The analogy to political ads is straightfor-ward. Polarizing issues that tend to have bothpassionate supporters and equally passionatedetractors are “highly differentiated products,”while more mundane issues that are not likelyto inspire as much controversy are more akinto “homogeneous products.” Same-sex marriageis an example of the former, while bond issuesfor local sports stadiums are good examples ofthe latter. The marginal voter in a polarizingcampaign is likely to have a valuation greaterthan the mean, so ads that provide real infor-mation are likely to increase the dispersion ofdemand, rotate the demand curve clockwise,increase the valuation of the marginal voter,and raise the “total take” on voting day. Onthe other hand, real information in a run-of-the-mill campaign is expected to reduce the disper-sion of demand, essentially moving voters to thecenter of the issue, rotating the demand curvecounter-clockwise, and increasing the WTP ofthe marginal voter who began with a valuationthat is below the mean.

In the current example, the debate surround-ing Proposition 2 suggests that animal welfareis indeed a polarizing issue. Proponents of theinitiative, or supporters of animal welfare leg-islation, are regarded by their detractors asextremely liberal, out of touch with reality, andantibusiness. Opponents, on the other hand, areregarded by animal-rights supporters as cruel,barbaric, and corporate apologists. To the extent


that advertisements on either side contain ele-ments of both real information and the pref-erence effect, we expect to find both shift androtation effects for both types of ads. In termsof the rotation effect, we expect the dispersionof WTP to rise for both proanimal welfare andantianimal welfare ads. The shift effect, on theother hand, is expected to be positive (increasingWTP) for proanimal welfare ads and negative(decreasing WTP) for antianimal welfare ads.Because of these opposing effects, and the miti-gating impact of rotation, the net welfare effectsare an empirical question.

Our model also provides another way ofthinking about the asymmetric information effectof positive and negative media coverage. Swartzand Strand (1981) document a stronger negativeeffect on the demand for oysters from contami-nated water, while Smith, van Ravensway, andThompson (1988) find a similar asymmetry incoverage of heptachlor in Hawaiian milk andBrown and Schrader (1990) for the purportednegative health effects of consuming shell eggs.In an effect akin to the framing observationsof Kahneman (2003), consumers are shown toreact more strongly to negative news regardinga product than positive. While the usual expla-nation for this effect is that consumers are inher-ently loss averse, Richards and Patterson (1999)argue that it is due rather to the convexity ofutility with respect to information. In the contextof the preference heterogeneity model of John-son and Myatt (2006), negative media is likelyregarded as real information as no agent has anincentive to reveal negative information aboutthe product (except, perhaps, for sellers of a sub-stitute good, but this seems implausibly strategicbehavior). Therefore, negative information willrotate the demand curve clockwise, and reducethe WTP of the average mass-market consumer,or voter as the case may be. Positive coverage,on the other hand, is not likely to be viewed ascredible so will be regarded as a pure preferenceeffect, shifting the demand curve to the extentthat it is effective at all. We investigate this pos-sibility in the empirical application below.

When both types of ads are consideredtogether, the real information conjecture des-cribed above will likely give way to confusion.If both sides are viewed as credible sources ofinformation, then the dominant effect will be theone that is regarded as the most convincing. Thereal information effect causes demand curves torotate clockwise, raising the WTP for consumerswhose valuations were initially above the mean,

but lowering them for above-average consumers.If neither is viewed as credible because theclaims tend to contradict each other, then theshift or preference effect will likely dominate.Demand curves will still rotate, but our hypoth-esis is that this effect will be less important,particularly in terms of net welfare, when vot-ers hear a cacophony of messages. Fortunately,this theory is imminently testable through exper-imental methods.

More formally, consider the definition ofa rotation in the demand curve described byJohnson and Myatt (2006). Assume there is aunit mass of consumers, each willing to payw for one unit of the item in question. Thedistribution of w is represented by Fs(w), whichis twice continuously differentiable in both s andw with density fs(w). The parameter s governsthe shape of the distribution of valuations suchthat an increase in s represents a spread in thedensity of w and, hence a clockwise rotation ofFs(w) about some point w (figure 1, Johnsonand Myatt 2006). We next derive the effect ofa spread in valuations on the distribution ofmarket demand. At any price, p, the proportionof consumers who purchase the good is givenby: q = 1 − Fs(p). Inverting this expressiongives an expression for the inverse demandcurve: Ps(q) = F−1

s (1 − q), so a change in srotates the inverse demand curve in a manneranalogous to the change in the distribution ofvaluations (figure 2, Johnson and Myatt 2006).Namely, if demand is below the pivot point,q, then an increase in the spread of valuationscauses a rise in the market price, and viceversa, or:

q < q => (∂Ps(q))/∂s > 0,(1)

q > q => (∂Ps(q))/∂s < 0.

Equation (1) implies that if we are below thepivot-point in demand, greater dispersion invaluations causes the valuation of the marginalconsumer, and hence the market price, to riseand if we are above the pivot-point in demand,an increase in the dispersion of demand causesthe price to fall. In the former case, the issuewith the product is more likely to be of interestto a concentrated special interest, or niche groupof voters, and in the latter the issue is likely tobe of interest to the mass of voters.

We use the theoretical framework developedin this model to derive a structural model ofpro- and antianimal welfare regulation advertis-ing. Our model is structural in the sense that it is


derived directly from the utility-theoretic modelof the shift or rotation effects associated withadvertising. We derive the WTP for cage-freeeggs in a random utility framework in which thedistribution of consumer heterogeneity reflectsthe distribution of marginal valuations in thetheoretical model above. In the random utilitymodel, consumer utility is the sum of a deter-ministic and stochastic part such that:

Uij = Vij + εij ,(2)

for product j by consumer i, where Vij is thedeterministic component of utility, and εij is aniid error term. Utility, in turn, is a function ofattributes of the chooser (xi) and of the choice(zj ), a vector of advertising exposures (ak) andincome (yi). The marginal value consumer iplaces on product j = 1 is defined as the amountof income that leaves his or her utility at leastas great with and without the purchase:

Vi0(z0, a0, xi , yi)(3)

+ εi0 ≤ Vi1(z1, a1, xi , yi − ci1) + εi1,

where ci1 is the marginal value of product 1 byconsumer i. We solve for the WTP by consumeri by invoking the random utility assumption andrecognizing that:

Pr(WTPi1 ≥ ci1) = Pr(Vi0 + εi0 ≤ Vi1 + εi1).(4)

Assuming the consumer decides to either buyproduct 1 or not, and that the error term isdouble-exponential distributed with mean 0 andvariance π2μ2/3, the WTP becomes:

Pr(WTPi1 ≥ ci1)(5)

= [exp(Vi1/μ)]/[1 + exp(Vi1/μ)],

where μ is the logit scale parameter. Solvingfor the WTP from this expression, we write theodds ratio of choosing product 1 relative to notchoosing it as:

(Pr(j = 1))/(1 − Pr(j = 1))(6)

= [exp(Vi1/μ)/(1 + exp(Vi1/μ))]/

[1/(1 + exp(Vi1/μ))] = exp(Vi1/μ),

where we normalize exp(Vi0/μ) to one andPr(j = 1) is the probability of purchasinggood 1. Taking logs of both sides of the oddsratio gives an expression for the WTP byconsumer i as a function of choice and chooserattributes, the level of advertising and the scale

parameter (which we normalize to 1 without lossof generality in the empirical application below):

ln ([Pr(j = 1)]/[1 − Pr(j = 1)])(7)

= WTPi1 = Vi1/μ.

With an appropriate specification for Vi1 it ispossible to test for both the direct effect ofpro- and antianimal welfare advertising on theWTP for cage-free eggs, and the indirect effectthrough the dispersion of valuations.

Utility in a random utility framework is typi-cally additive over attribute arguments. WritingVi1 in terms of an empirical, or estimable, modelof utility, we assume that:

Vij (zj , amj , xi , yi) = αj +K∑

k=1

βkzjk(8)

+L∑

l=1

γlxil +M∑

m=1

δmamj + ξj ,

where αj is a choice-specific constant, βk aremarginal values for each product attribute, γl

represents the influence of each demographicattribute on WTP, δm is the impact of advertisingof type m (pro- or antianimal welfare) on indi-rect utility, and ξj is the iid econometric errorterm. Advertising, however, is hypothesized tohave both a direct effect through changing pref-erences and shifting the demand curve, and anindirect effect through the dispersion of valua-tions. We model this latter effect by recognizingthat the advertising response term is a functionof unobserved consumer heterogeneity throughthe distribution of preferences, Fs above. Eachadvertising-impact parameter is randomly dis-tributed according to:

δm = δm0 + δm1σm + νm, νm ∼ N(0, 1),(9)

where δm0 is now interpreted as the direct effectof advertising of type m (shift effect), δm1 is theindirect or rotational effect caused by changes inthe dispersion of valuations, and σmj is the vari-ability in tastes associated with each type of ad.Substituting this utility model into the expres-sion for WTPi1 provides an estimable model ofthe WTP for eggs under each advertising type.Note that, although consumers are assumed tomake discrete choices among differentiated egg-brands in the store, the estimated bid functionimplied by the random utility decision processis continuous. This continuity reflects the under-lying indirect utility functions that drive con-sumers’ discrete decisions. Table 1 summarizesthe hypotheses we test with this model.


TABLE 1Hypotheses for Advertising Effect

Hypothesis Parameters Expectation

1. “Pro” advertising increases WTP for all voters δ10 > 0 “Pro” ads are pure hype and shift demandcurve out for all voters

2. “Pro” advertising increases the dispersion ofpreferences

δ11 > δ01 “Pro” ads have real information and rotatedemand clockwise

3. “Anti” advertising increases WTP for all voters δ20 > 0 “Anti” ads are pure hype and shift demandcurve out for all voters

4. “Anti” advertising increases the dispersion ofpreferences

δ21 > δ01 “Anti” ads have real information and rotatedemand clockwise

5. “Pro/Anti” advertising increases WTP for allvoters

δ30 > 0 “Pro/Anti” ads are pure hype and shiftdemand curve out for all voters

6. “Pro/Anti” advertising increases the dispersion ofpreferences

δ31 > δ01 “Pro/Anti” ads have real information androtate demand clockwise

Note: δ00 and δ01 are constant shift and rotation terms, respectively. Index m = 1 refers to “Anti” ads, index m = 2 refersto “Pro” ads, and m = 3 refers to “Pro/Anti” ads.

IV. RESULTS AND DISCUSSION

With the empirical model developed above,we test whether each type of ad, alone orin combination, is successful in changing thevoting behavior of the electorate as a whole (ashift of the demand curve) or if the ad servedrather to sharpen positions on each side (rotationof the demand curve). In this section, we reporttests of these hypotheses, and a number of otherinsights into the experimental data.

Prior to presenting the parameter estimatesfrom the formal econometric model of WTPdeveloped in the previous section, we begin byfirst presenting some summary statistics on theexperimental WTP data. Inspecting the exper-imental data provides some insight into howdifferent groups of consumers value animal wel-fare attributes. To that end, Table 2 provides across-tabulation of WTP values by income, race,gender, and household size. From the data in thistable, it appears as though higher income partic-ipants are willing to pay less for cage-free eggsthan those in lower income groups. This find-ing is somewhat surprising as cage-free eggsare well understood to sell for a premium so,to the extent that eggs are a “normal” good interms of income elasticity, we should see greaterdemand among higher income consumers. Fur-ther, although the standard deviation of the WTPestimate is high, Hispanic-Americans have thehighest mean WTP among all the racial groupsin our data.6 There appears to be a more signif-icant difference in WTP between genders, with

6. Note that many participants refused to provide theirracial background. For the econometric estimation, theserespondents were coded as the excluded dummy group.

females bidding a $0.923 premium over con-ventional eggs and males a $0.748 premium.Finally, the WTP for cage-free eggs is slightlylower among larger households. This finding islikely due to the fact that these households pur-chase in greater volumes, so are more sensitiveto any attribute that would cause them to facehigher prices.

Summarizing the data by treatment providesa preliminary indication of whether the type ofmedia exposure influences WTP (Table 3). Pool-ing bids from all rounds, we find that the “Pro”ads are associated with a premium relative tocontrol of nearly $0.14/dozen, while the “Anti”ads a discount of almost $0.28/dozen. When wecombine both “Pro” and “Anti” ads, there is stilla discount of $0.07/dozen so it appears as thoughthe “Anti” ads represent a dominant influenceon consumers’ choices. Considering only thefirst bids, we find a similar pattern, althoughthe absolute values of the bids are significantlylower in each case. Most importantly, the dif-ferences between each treatment and control areall significant, so media advertising clearly hasan effect on WTP.

The data in Table 3, however, do not take intoaccount the possible intercorrelation betweensome of the explanatory variables in the WTPmodel. If some of the variables explain the sameeffect, then the summary statistics presentedabove may be misleading. Therefore, even with-out the random parameter structure in the abovemodel, at least a multiple-regression approach ispreferred. A simple regression model, however,does not account for the unobserved heterogene-ity that is likely to be an important determinantof the WTP for cage-free eggs. We test whether


TABLE 2Summary of WTP by Demographic Strata

Variable Classification Mean Std. Dev Min Max N

Income <$10,000 0.776 0.783 0.000 3.950 384$10,001–20,000 0.927 0.792 0.000 3.990 336$20,001–30,000 0.899 0.798 0.000 5.000 245$30,001–40,000 0.872 0.629 0.010 2.670 66

>$40,000 0.569 0.664 0.000 2.070 36Race African-American 0.524 0.515 0.000 1.500 18

Asian-American 0.718 0.560 0.000 1.750 30Asian 0.365 0.727 0.000 3.000 30

Hispanic 0.604 0.403 0.000 1.220 18Hispanic-American 1.093 0.915 0.100 3.330 60

Mixed Race 0.521 0.393 0.000 1.000 18White 0.748 0.635 0.000 1.500 30

Marital status Married 0.851 0.780 0.000 5.000 1067Household size 1 0.873 0.814 0.000 3.330 264

2 0.903 0.768 0.000 3.990 4323 0.845 0.800 0.000 3.750 1324 0.781 0.722 0.000 2.900 1385 0.729 0.836 0.000 5.000 846 0.332 0.090 0.250 0.500 11

Note: Units are all $/dozen.

TABLE 3Summary of WTP by Treatment

Control Pro t-ratio Anti t-ratio Pro/Anti t-ratio

All bids 0.9672 1.1056 4.2138 0.6907 −8.0755 0.8941 −2.17310.7480 0.7698 0.8322 0.8056

First bids 0.7380 1.0879 10.3615 0.5758 −5.0165 0.6263 −3.63140.7102 0.8447 0.7824 0.7114

Note: All values are in $/dozen. Values below bids are standard deviations. t-ratios compare treatment value to controlvalue.

unobserved heterogeneity is indeed important byconducting a series of specification tests on theWTP model. These statistics, along with thefull set of parameter estimates, are given inTable 4. The first specification test involves ajoint test of the significance of the scale param-eters of the random-parameter model. If theseparameters are jointly equal to zero, we rejectthe random parameters model in favor of aconstant-parameters alternative. From the chi-square statistic reported in Table 4 (1306.183),it is clear that the scale parameters are jointlydifferent from zero so we conclude that the ran-dom parameters model is preferred by this met-ric. Second, we conduct t-tests of each randomparameter to determine whether, individually,a variable is better represented by a constant-parameter specification. Again, we reject thenull hypothesis of a constant parameter in each

case and conclude that the random parametermodel is preferred. Therefore, we use this spec-ification to test our hypotheses regarding theeffect of pro- and antianimal welfare ads on theWTP for cage-free eggs and to examine morecarefully how WTP varies among participants,and rounds of the experiment.

The results in Table 4 summarize our findingsboth with respect to the shift and rotation effects,and the demographic and experimental-designquestions. We find that bidding falls signifi-cantly from round 1 to round 2 and round 3,relative to the round 6 benchmark, and con-tinues to fall through rounds 4 and 5, albeitnot in a statistically significant way. While thisfinding is contrary to the experimental resultsof Fox, Hayes, and Shogren (2002), it is notuncommon in the experimental auction liter-ature (Lusk and Shogren 2007). Because our


TABLE 4Random Coefficient WTP Model

Nonrandom Parameters Random Parameters

Variable Estimate t-ratio Variable Estimate t-ratio

Bid order 1 −0.272∗ −5.948 Means of random parametersBid order 2 −0.180∗ −3.449 Constant 0.341∗ 3.086Bid order 3 −0.123∗ −2.139 Pro 0.253∗ 7.525Bid order 4 −0.076 −1.399 Anti −0.070∗ −2.158Bid order 5 −0.071 −1.125 Pro/Anti 0.211∗ 5.626Female 0.068∗ 2.923Household size −0.029∗ −2.678 Std. deviations of random parametersHousehold income −0.003∗ −2.724 Constant 0.533∗ 48.638Marital 0.279∗ 10.223 Pro 0.352∗ 16.805Vegetarian 0.080∗ 2.439 Anti 0.132∗ 7.901Pet owner 0.193∗ 7.679 Pro/Anti 0.416∗ 15.763African-America −0.075 −0.751Asian −0.032 −0.388 Variance of random effectsAsian-American −0.579∗ −3.479 Sigma 0.356∗ 80.011Hispanic −0.057 −0.548Hispanic-American 0.318∗ 4.899Mixed race 0.042 0.413White 0.275∗ 5.783Food safety 0.029 1.223Heard of Prop 2 −0.080∗ −6.562Animal welfare 0.690∗ 18.429Local produce 0.433∗ 13.247Organic 0.630∗ 16.838Omega 3 0.169∗ 3.449Number of eggs 0.006∗ 5.312Log-likelihood function −595.725Chi-square 1306.183

∗Significance at the 95% level.

participants are shown the “market price” aftereach round, and the market price is bound byzero and the highest bid, high bidders in earlyrounds tend to reduce their bids in subsequentrounds as they become more certain that theirbid has earned them the cage-free eggs. Thebid distribution thus becomes more concentratedas rounds progress and bidders learn how themarket works. Further, we find that women arewilling to pay more for cage-free eggs thanmen, while members of larger households tendto bid less. These results are consistent withthe summary statistics presented above, andare not particularly surprising. The estimatedincome-effect also confirms the summary resultsas the econometric model shows that higherincome participants are willing to pay less forcage-free eggs than lower-income participants.Among the statistically-strongest effects, we findthat single people are willing to pay significantlymore for cage-free eggs than married partici-pants. While striking, this outcome is likely not

suggestive of any deeper, generalizable resultthat should raise any concern for egg marketers.Of potentially more importance, however, arethe findings that pet owners and vegetarians arewilling to pay more for cage free eggs. Althoughthese findings are again not particularly surpris-ing, they do emphasize that concern over animalwelfare is real, and help identify, and quantify,market segments that are likely to be particularlysensitive to animal welfare concerns. Among theother effects of interest, we find that partici-pants who had heard of Proposition 2, while notnecessarily knowledgeable about the nuances ofthe issue, were willing to pay less for cage-free eggs than other respondents. Given thatour experiment was conducted in Arizona, thisresult likely reflects a more general sentimentin the state that voters in California tend to bemore concerned with “liberal” issues than theconservative electorate of Arizona. Finally, inresponse to our question “Which of the follow-ing attributes is most important. . .” in selecting


eggs, we find that respondents who regard ani-mal welfare, local produce, organics, and omega3 content as important are likely to bid morethan otherwise. Clearly, these consumers arehighly involved in their egg-purchase decisionand take issues other than price or color intoaccount.

Of greater interest, however, are the ele-ments of the random-coefficient advertising-effect function. The first set of results, the“Means of Random Parameters” refers to theshift, or preference effect associated with eachtype of advertising, or the absence thereof. John-son and Myatt (2006) interpret the shift effectas hype because real information would insteadrotate the demand curve. The constant term pro-vides a benchmark mean WTP established bythe control group. Relative to control, the “Pro”parameter implies that ads in favor of Propo-sition 2 were effective in causing participantsto be willing to pay more for cage free eggsthan control—some $0.25/dozen more. Next,the “Anti” parameter indicates that ads in favorof the status quo, which emphasized the impor-tance of the issue for small farmers in the state,led participants to bid $0.07/dozen less than con-trol. At least with respect to this direct effect, ourresults support the findings of Swartz and Strand(1981) and Fox, Hayes, and Shogren (2002)more recently. Combining both types of infor-mation, however, we find that participants arewilling to pay nearly as much as if they only sawthe “Pro” ads—$0.21/dozen more than control.In terms of loss-aversion theories of either Vis-cusi, Magat, and Huber (1987) or Kahneman andTversky (1979), if the “Pro” ads represent neg-ative information to the extent that they revealsomething bad about the egg production pro-cess that people were initially unaware of, ourrespondents clearly exhibit response asymme-tries consistent with either explanation.

The “Standard Deviations of Random Param-eters” are interpreted as the “rotation” effects ofthe different types of ad as they refer to theimpact on the dispersion of the distribution ofresponses in each case (Figures 1 and 2). John-son and Myatt (2006) show that such disper-sion effects are isomorphic to rotation effects.Insignificant rotation effects would suggest thatthe advertising contains only a preference effectand no real information: the stronger the rota-tion, the greater the real information content.Based on these estimates, we find that the“Pro” ads contain nearly three times the realinformation as “Anti” ads as measured by its

FIGURE 1Rotation of the Distribution Function

Willingness to pay wD

istr

ibut

ion

Fun

ctio

n F

s(w

)

w

FIGURE 2Rotation of the Inverse Demand Curve

w

q Pivot Point

Mass MarketNiche Market

Proportion of Consumer q

Ps(

q) =

Fs–1

(1–

q) ^

embodiment in a rotating demand curve. Rotat-ing the demand curve in a clockwise direc-tion (more inelastic) means that consumers whocare about animal welfare become more cer-tain of their preferences and are thus willing topay more than previously. On the other hand,participants who do not care how their eggsare produced will pay even less than beforefor cage-free eggs as they are more convincedof their position that cage-free eggs are onlyfor “animal lovers” or “environmentalists” andnot people like themselves. To the extent thatthe marginal consumer is to the left of thepivot point of the distribution of demand, “Pro”


ads cause the WTP to rise significantly. If themarginal consumer lies to the right of the pivotpoint, the “Anti” ads cause the WTP to fallonly marginally. The sharpness of the distinctionbetween these two parameters suggests that wehave found another explanation for the appar-ent asymmetry of good and bad news describedby Fox, Hayes, and Shogren (2002). Whereastheir theoretical explanation relies on behavioralinsights from prospect theory and risk aversion,our explanation lies merely in a more carefulexplication of the underlying changes in utility.If preference heterogeneity is properly regarded,such asymmetries are easily explained in termsof the effect of advertising (or information moregenerally) and not necessarily some violation ofthe underlying postulates of economics and mar-keting theory.

These insights are supported by the parameteron the “Pro/Anti” variable. If participants aregiven a balanced perspective and are shownboth types of ads, we find that the demandcurve rotates even more sharply clockwise. Asexpected, the real information content of the“Pro” ads is reinforced by the reference pointprovided by the “Anti” ads. If participants areallowed to assess the internal validity of bothsides of the argument on their own, they arebetter able to judge which is more plausible, andtheir evaluation of the issue more certain. Thisoutcome supports similar findings by previousstudies as the one that is closest to ours inspirit and method, Fox, Hayes, and Shogren(2002) find that providing simultaneous positiveand negative information reinforces the negativeeffect. If an ad is able to convince the consumeror voter that “things are not as good as theyseem,” then the presence of contrary informationmerely reinforces their worst fears.

In terms of voting tendencies, the “Pro” adsmay appeal to a core group of “citizen interests,”in Gerber’s (1999) terminology, increasing theintensity of support among voters who may havenot been aware of this issue before the cam-paign, but are sensitive to how their food isproduced in general. In this case, the ads inopposition to Proposition 2 made a very cleareconomic argument, transparently in support ofbusiness interests in the industry. The estimatesfrom our preference heterogeneity model sug-gest that this group was likely to be sufficientlysmall to have little impact on the overall out-come of the vote as neither the pure preferenceeffect of the “Anti” ads nor the real informationwere sufficient to overcome the “Pro” ads. While

the content of the “Anti” ads was economic, thecontent of the “Pro” ads was emotional, popular,and clearly representative of a position that con-sumers concerned about where their food comesfrom would want to associate with. At the endof the day, the outcome of the vote would notbe surprising had this analysis been conductedex ante.

Ultimately, however, good policy shouldincrease economic welfare, and the advertisingthat generates the greatest increment in welfareshould rule the day. Welfare estimates, and theproportion of the change because of either ashift or a rotation are shown in Table 5. Becausevoters in the real world see only a combina-tion of ads in support of and in opposition tothe proposition (the “Pro/Anti” scenario above),we calculate the combined welfare effects ofboth types of ads shown together. In the firstsection of Table 5, we show the shift and rota-tion effects relative to the base case in utility-measure. Beginning from a base utility of 0.799,the shift effect of supporting and opposing adsincreases utility to 0.885, while the rotationeffect reduces it back to 0.805. Although therotation effect causes the WTP of some con-sumers to rise sharply enough to cause the meanWTP to rise, the net effect is to reduce con-sumer welfare because the mass of consumerslower their WTP as a result of the ads. By divid-ing the utility effect by the marginal utility ofincome, we convert changes in utility to a dollar-metric, so the lower two sections of Table 4show the dollar values of political advertisingon a per-dozen basis, and then on an annual,per household basis. On a per-dozen basis, thenet effect of the animal welfare ads is a posi-tive $0.297/dozen, with $0.278/dozen from theshift effect and $0.019/dozen from the rotationeffect. On an annual basis, using the averageconsumption rate per household in the data set

TABLE 5Welfare Effects of Cage-Free Egg Advertising

Mean Std. Dev.

Base utility 0.799 0.659Shift effect 0.885 0.679Rotation effect 0.905 0.862Shift effect per dozen (cents) 27.781 46.464Rotation effect per dozen (cents) 1.881 87.240Total effect per dozen (cents) 29.663 99.616Annual shift effect ($) 83.822 14.019Annual rotation effect ($) 5.677 26.321Annual total effect ($) 89.498 30.056


to convert per-dozen values, the ads generate$83.82 in welfare from the shift effect and $5.67from the rotation effect for a total of $89.50 inadditional welfare. Because the “Pro” ads dom-inate the “Anti” ads in terms of their welfareeffect then, at least on purely economic grounds,Proposition 2 should have been passed—as itwas. The fact that the rotation effect accountsfor 6.3% of the total welfare created is striking.Real information in this case is apparently worthonly $5.67 on an annualized basis, while thepreference-effect value of advertising is worthsome 93.7% of the total, or $83.82 per house-hold per year. While somewhat depressing, thisresult is not surprising as the “Pro” ads madeheavy use of celebrities with little real infor-mation on the pros and cons of cage-free eggswhile the “Anti” ads were just as partisan butapparently less convincing.

As Johnson and Myatt (2006) point out, alladvertising consists of varying amounts of realinformation and hype. With the method usedhere, we sort out how much of each lies ineach ad. Our insights thus generalize beyond theanimal welfare case to any type of advertisingprogram in which advertising is neither purelyinformative nor purely persuasive. Staying in thepublic policy realm, “issue ads” are now theorder of the day given the unrestricted amountsinterested parties can contribute to political cam-paigns. Issue ads are often even more partisanthan state-level ballots so this method could beused to evaluate how voters perceive the mes-sage contained in the ads—do they understandthe true motivations of the money behind thead? From another perspective, tools like this canhelp ad agencies design political ads to max-imize the effectiveness of ad spending. If anissue is particularly misunderstood among theelectorate, then designing an ad with a high realinformation content may be able to skew the dis-tribution of preferences such that the “WTP” forthe issue at hand rises significantly. Ad agencymanagers can also use this approach to esti-mate the effectiveness of their ads, and per-haps to target specific ads to different marketsegments, depending on their demographic andsocioeconomic profile.

V. CONCLUSIONS AND IMPLICATIONS

In this study, we evaluate the effective-ness of advertisements on either side of the2008 California Proposition 2 campaign. Propo-sition 2, which was eventually passed and

further strengthened in the summer of 2010,sought to restrict agricultural production prac-tices to ensure that animals, specifically egg-laying hens, were treated humanely. We developan empirical approach based on the prefer-ence heterogeneity model of Johnson and Myatt(2006) to determine whether each type of ad, ora combination thereof, was able to change thepreferences of the entire electorate, or if theymerely sharpened positions on each side. Weapply this model to experimental data gatheredfrom a nonhypothetical auction for cage-freeeggs.

Our results show that the preference or shifteffect dominates the rotation effect for each typeof ad. Moreover, the ability of the ads sup-porting Proposition 2 were sufficiently effectivein changing preferences to outweigh the neg-ative effects associated with the opposing ads.Because the notion that animals may be mis-treated in the production of food comes as arevelation to many consumers (who are also vot-ers), this result can be explained by Kahnemanand Tversky’s (1979) prospect theory. Framedby the assumption that all farm animals arehappy, the ads supporting Proposition 2 revealedsometimes shocking images of animals beingmistreated and caused consumers to perceivea measure of risk to animal welfare that theydidn’t fully appreciate before the campaign. Thissmall amount of “bad news” regarding animalwelfare was enough to outweigh many strong,but conventional economic arguments, on theother side.

The real information content of the ads wasnot inconsequential. Perhaps because the sup-portive ads revealed information that was notpreviously known to many voters, and partic-ipants in our study, the rotation effect for the“Pro” ads was nearly three times the strengthof the rotation effect for “Anti” ads. Using mea-sures of consumer welfare to convert our econo-metric estimates to a dollar-measure, we foundthat over 6% of the change in consumer wel-fare associated with the ads came from the realinformation effect as opposed to the preferenceeffect. Therefore, the ads both changed pref-erences and managed to harden some voters’opinions on either side.

Based on campaign ad spending during the2010 election cycle, studying the behavioraleffects of political ads—for congressional, pres-idential, state, local, and ballot measures—is anarea ripe for further research in marketing. Pol-itics has become as much about marketing as


it is about political science or public admin-istration. Therefore, marketing researchers canhave a significant impact on discourse in thisarea by bringing the tools of consumer behav-ior and marketing research to study politicalissues. In terms of our specific study, thereare a number of issues in conducting nonhy-pothetical experiments that should be addressedin a political context: hypothetical bias (dif-ference between hypothetical and nonhypothet-ical responses), social isolation bias (differencebetween responses made in private and in pub-lic), and consequentiality (the perception that aparticipant’s responses will have a real effect)are only three important topics that remain tobe explored.

REFERENCES

Allender, W., and T. J. Richards. “Consumer Impact ofAnimal Welfare Regulation in the California PoultryIndustry.” Journal of Agricultural and Resource Eco-nomics, 35, 2010, 424–42.

Becker, G. S., M. H. DeGroot, and J. Marschak. “MeasuringUtility by a Single-Response Sequential Method.”Behavioral Science, 9, 1964, 226–32.

Becker, G. S., and K. M. Murphy. “A Simple Theory ofAdvertising as a Good or Bad.” Quarterly Journal ofEconomics, 108, 1983, 941–64.

Brown, D. J., and L. F. Schrader. “Cholesterol Informationand Shell Egg Consumption.” American Journal ofAgricultural Economics, 72, 1990, 548–55.

Carlsson, F., P. Frykblom, and C. J. Lagerkvist. “FarmAnimal Welfare-Testing for Market Failure.” Journalof Agricultural and Applied Economics, 39, 2007a,61–73.

. “Consumer Willingness to Pay for Farm Ani-mal Welfare: Mobile Abattoirs versus Transportationto Slaughter.” European Review of Agricultural Eco-nomics, 34, 2007b, 321–44.

Chang, J. B., J. L. Lusk, and F. B. Norwood. “ThePrice of Happy Hens: A Hedonic Analysis of RetailEgg Prices.” Journal of Agricultural and ResourceEconomics, 35, 2010, 406–23.

Dixit, A., and V. Norman. “Advertising and Welfare.” BellJournal of Economics, 9, 1978, 1–17.

Fischbacher, U. “z-Tree: Zurich Toolbox for Ready-MadeEconomic Experiments.” Experimental Economics, 10,2007, 171–78.

Fox, J. A., D. J. Hayes, and J. F. Shogren. “ConsumerPreferences for Food Irradiation: How Favorable andUnfavorable Descriptions Affect Preferences for Irra-diated Pork in Experimental Auctions.” Journal of Riskand Uncertainty, 24, 2002, 75–95.

Gerber, E. R. The Populist Paradox: Interest Group Influ-ence and the Promise of Direct Legislation, Princeton:Princeton University, 1999.

Gracia, A., M. L. Loureiro, and R. M. Nayga, Jr. “ValuingAnimal Welfare Labels with Experimental Auctions:What Do We Learn from Consumers?” Paper pre-sented at the International Association of AgriculturalEconomists Conference, Beijing, China, 2009.

Hamilton, S. F., D. L. Sunding, and D. Zilberman. “PublicGoods and the Value of Product Quality Regulations:The Case of Food Safety.” Journal of Public Eco-nomics, 87, 2003, 799–817.

Harper, G. C., and A. Makatouni. “Consumer Perception ofOrganic Food Production and Farm Animal Welfare.”British Food Journal, 104, 2002, 287–99.

Hartl, J., and R. Herrmann. “Do They Always Say No? Ger-man Consumers and Second-Generation GM Foods.”Agricultural Economics, 40, 2009, 551–60.

Johnson, J. P., and D. P. Myatt. “On the Simple Economicsof Advertising, Marketing and Product Design.” Amer-ican Economic Review, 96, 2006, 756–84.

Kahneman, D., and A. Tversky. “Prospect Theory: AnAnalysis of Decision under Risk.” Econometrica, 47,1979, 263–91.

Larue, B., G. E. West, C. Gendron, and R. Lambert.“Consumer Response to Functional Foods Producedby Conventional, Organic or Genetic Manipulation.”Agribusiness: An International Journal, 20, 2004,155–66.

List, J. A., and C. Gallett. “What Experimental Proto-col Influences Disparities between Actual and Hypo-thetical Stated Values?” Environmental and ResourceEconomics, 20, 2001, 241–54.

Lusk, J. L. “The Effect of Proposition 2 on the Demandfor Eggs in California.” Journal of Agricultural andFood Industrial Organization, 8, 2010. Accessed May5, 2011. http://www.bepress.com/jafio/vol8/iss1/art3/.

Lusk, J. L., and F. B. Norwood. “A Survey to Deter-mine Public Opinion about the Ethics and Gov-ernance of Farm Animal Welfare.” Journal of theAmerican Veterinary Medical Association, 233, 2007,1121–26.

. “An Inferred Valuation Method.” Land Economics,85, 2009a, 500–14.

. “Bridging the Gap between Laboratory Experimentsand Naturally Occurring Markets: An Inferred Valu-ation Method.” Journal of Environmental Economicsand Management, 58, 2009b, 236–50.

. “Direct vs. Indirect Questioning: An Application tothe Well-Being of Farm Animals.” Social IndicatorsResearch, 96, 2010, 551–65.

Lusk, J. L., and J. Shogren. Experimental Auctions: Meth-ods and Applications in Economic and MarketingResearch, New York: Cambridge University Press,2007.

Lusk, J. L., M. S. Daniel, D. R. Mark, and C. L.Lusk. “Alternative Calibration and Auction Institu-tions for Predicting Consumer Willingness to Payfor Non-Genetically Modified Corn Chips.” Journalof Agricultural and Resource Economics, 26, 2001,40–57.

Lusk, J. L., J. Roosen, and J. A. Fox. “Demand for Beef fromCattle Administered Growth Hormones or Fed Genet-ically Modified Corn: A Comparison of Consumers inFrance, Germany, the United Kingdom and the UnitedStates.” American Journal of Agricultural Economics,85, 2003, 16–29.

Matsusaka, J. G. “Direct Democracy Works.” Journal ofEconomic Perspectives, 19, 2005, 185–206.

Nelson, P. “Information and Consumer Behavior.” Journalof Political Economy, 78, 1970, 311–29.

. “Advertising as Information.” Journal of PoliticalEconomy, 82, 1974, 729–54.

Norwood, F. B., and J. L. Lusk. “Social Desirability Biasin Real, Hypothetical and Inferred Valuation Experi-ments.” American Journal of Agricultural Economics,93, 2011, 528–34.

Olnyk, N. J., G. T. Tonsor, and C. A. Wolf. “Consumer Will-ingness to Pay for Livestock Credence Attribute ClaimVerification.” Journal of Agricultural and ResourceEconomics, 35, 2010, 261–80.

Ophuis, P., and A. M. Oude. “Sensory Evaluation of‘Free Range’ and Regular Pork Meat under Different


Conditions of Experience and Awareness.” Food Qual-ity and Preference, 5, 1994, 173–78.

Palda, K. S. “The Effect of Expenditure on Political Suc-cess.” Journal of Law and Economics, 18, 1975,745–71.

Potstra, A. “How Green Is Brock?” Brock Press, October 7,2008, p. 2. Accessed October 11, 2011. http://butler.ac.brocku.ca/faculty/chys/rmitchell/HowGreenisBrock.pdf.

Richards, T. J., and P. M. Patterson. “The Economic Valueof Public Relations Expenditures: Food Safety andthe Strawberry Case.” Journal of Agricultural andResource Economics, 24, 1999, 440–62.

Rothschild, M. L. “Political Advertising: A Neglected PolicyIssue in Marketing.” Journal of Marketing Research,15, 1978, 58–71.

Shogren, J., M. Margolis, C. Koo, and J. A. List. “A Randomnth-Price Auction.” Journal of Economic Behavior andOrganization, 46, 2001, 409–21.

Smith, M. E., E. O. van Ravenswaay, and S. R. Thomp-son. “Sales Loss Determination in Food Contami-nation Incidents: An Application to Milk Bans inHawaii.” American Journal of Agricultural Economics,70, 1988, 513–20.

Soley, L. C., and L. N. Reid. “Promotional Expendituresin U.S. Congressional Elections.” Journal of PublicPolicy and Marketing, 1, 1982, 147–55.

Stratmann, T. “Is Spending More Potent For or Against aProposition? Evidence from Ballot Measures.” Ameri-can Journal of Political Science, 50, 2006, 788–801.

Swartz, D. G., and I. E. Strand. “Avoidance Costs Associatedwith Imperfect Information: The Case of Kepone.”Land Economics, 57, 1981, 139–50.

Tonsor, G. T., N. Olnyk, and C. Wolf. “Consumer Voting andDemand Behavior Regarding Swine Gestation Crates.”Food Policy, 34, 2009a, 492–98.

. “Consumer Preferences for Animal Welfare Attri-butes: The Case of Gestation Crates.” Journal of Agri-cultural and Applied Economics, 41, 2009b, 713–30.

Uzea, A., and J. Hobbs. “Farm Animal Welfare and QualityVerification.” Working Paper, Department of Biore-source Policy, Business and Economics, University ofSaskatchewan, Saskatoon, Saskatchewan, 2009.

Viscusi, W. K., W. A. Magat, and J. Huber. “An Investi-gation of the Rationality of Consumer Valuations ofMultiple Health Risks.” RAND Journal of Economics,18, 1987, 465–79.

MEDIA ADVERTISING AND BALLOT INITIATIVES: THE CASE OF ANIMAL WELFARE REGULATION

Documents

Transcript of MEDIA ADVERTISING AND BALLOT INITIATIVES: THE CASE OF ANIMAL WELFARE REGULATION