Evaluative Conditioning Procedures and the …warrington.ufl.edu/departments/mkt/docs/janiszewski/Evaluative...Evaluative Conditioning Procedures and the ... evaluative conditioning

000

� 2010 by JOURNAL OF CONSUMER RESEARCH, Inc. ● Vol. 37 ● October 2010All rights reserved. 0093-5301/2010/3703-0012$10.00. DOI: 10.1086/653656

Evaluative Conditioning Procedures and theResilience of Conditioned Brand Attitudes

STEVEN SWELDENSSTIJN M. J. VAN OSSELAERCHRIS JANISZEWSKI

Changing brand attitudes by pairing a brand with affectively laden stimuli such ascelebrity endorsers or pleasant pictures is called evaluative conditioning. We showthat this attitude change can occur in two ways, depending on how brands andaffective stimuli are presented. Attitude change can result from establishing amemory link between brand and affective stimulus (indirect attitude change) orfrom direct “affect transfer” from affective stimulus to brand (direct attitude change).Direct attitude change is significantly more robust than indirect attitude change,for example, to changes in the valence of affective stimuli (unconditioned stimulusrevaluation: e.g., endorsers falling from grace), to interference by subsequent in-formation (e.g., advertising clutter), and to persuasion knowledge activation (e.g.,consumer suspicion about being influenced). Indirect evaluative conditioning re-quires repeated presentations of a brand with the same affective stimulus. Directevaluative conditioning requires simultaneous presentation of a brand with differentaffective stimuli.

Acommon approach to building brand equity is to createfavorable brand attitudes (Keller 1993, 2003). One

popular technique for building favorable brand attitudes isto use advertising, event sponsorship, product placement, orother forms of promotion to pair a brand with positive af-fective stimuli (Gibson 2008; Gorn 1982; Kim, Allen, andKardes 1996; Stuart, Shimp, and Engle 1987). These pair-ings can be performed in several different ways, collectivelyknown as evaluative conditioning procedures (for reviews,see De Houwer 2008, forthcoming; De Houwer, Baeyens,and Field 2005; De Houwer, Thomas, and Baeyens 2001;Walther, Nagengast, and Trasselli 2005). For example, abrand and a positive stimulus can be presented simulta-

Steven Sweldens is assistant professor of marketing at INSEAD, EuropeCampus, Boulevard de Constance, 77300 Fontainebleau, France ([email protected]). Stijn van Osselaer is professor of marketing at theRotterdam School of Management, Erasmus University, P.O. Box 1738,3000 DR Rotterdam, Netherlands ([email protected]). Chris Janiszewskiis the JCPenney professor of marketing, Warrington College of BusinessAdministration, University of Florida, P.O. Box 117155, Gainesville, FL32611-7155 ([email protected]). Support from the de-partment of marketing management at the Rotterdam School of Manage-ment, the marketing department at the Warrington College of BusinessAdministration, the Erasmus Research Institute of Management, and theTrust Fund Erasmus University Rotterdam is gratefully acknowledged. Thisarticle is based on the first author’s dissertation.

John Deighton served as editor and Susan Broniarczyk served as associateeditor for this article.

Electronically published May 7, 2010

neously or sequentially (Lascelles and Davey 2006). In ad-dition, a brand can be presented with the same or with adifferent positive stimulus on each occasion (Stahl and Un-kelbach 2009).

One might argue that exploring the differences in eval-uative conditioning procedures is not of substantive impor-tance. After all, evaluative conditioning has been docu-mented with both sequential and simultaneous pairings aswell as with pairings involving the same versus differentpositive stimuli (De Houwer 2008). It is possible, however,that the resilience of conditioned attitudes depends on theconditioning procedure. Different conditioning proceduresmight encourage different learning processes. First, someevaluative conditioning procedures may encourage the for-mation of associations between a brand name and affect-inducing concepts (De Houwer et al. 2001; Gawronski andBodenhausen 2006). Subsequent affective responses to thebrand name are indirect in that they rely on the activationof the affective concept upon perception of the brand. Sec-ond, other evaluative conditioning procedures may result inthe brand name generating the affective response directly(i.e., not mediated by a brand-concept association; Jones,Fazio, and Olson 2009). We propose that indirect affectiveresponses should be susceptible to any activity (e.g., retro-active interference, counterarguing, memory decay) thatweakens the brand-concept association.

Three experiments explore these ideas. Experiment 1 usesan affective stimulus revaluation procedure (i.e., changingthe valence of associated affective stimuli postconditioning)

000 JOURNAL OF CONSUMER RESEARCH

to identify conditioning procedures that promote indirectversus direct affective responses. Experiment 2 explores re-silience to retroactive interference, and experiment 3 ex-plores the influence of a persuasion knowledge prime (i.e.,warning of the persuasion attempt) on the evaluative con-ditioning of affective responses. Together, the three studieshighlight the importance of exploring how conditioning pro-cedures might influence learning and subsequent feelingstoward brands.

EVALUATIVE CONDITIONING

Evaluative conditioning is evidenced by an affective re-sponse to an initially neutral conditioned stimulus (CS) sub-sequent to its pairing with a valenced unconditioned stim-ulus (US; Levey and Martin 1975). Evaluative conditioninghas been shown to influence affect for words (Staats andStaats 1957), political slogans (Razran 1954), art (Leveyand Martin 1975), brand names (Stuart et al. 1987), bev-erages (Zellner et al. 1983), products (Gorn 1982), and ev-eryday objects (Hammerl and Grabitz 2000). In consumerresearch, evaluative conditioning typically consists of pair-ing brand names (CSs) with stimuli that evoke favorableaffective responses (USs: e.g., celebrity endorsers, pleasantimages, popular music; Allen and Janiszewski 1989; Bierley,McSweeney, and Vannieuwkerk 1985; Gibson 2008; Gorn1982; Janiszewski and Warlop 1993; Kim et al. 1996;Shimp, Stuart, and Engle 1991; Stuart et al. 1987).

Evaluative conditioning effects have been attributed tothe operation of different mental processes (De Houwer,forthcoming; De Houwer et al. 2001). Early behaviorist the-ories assumed that direct mental connections could be forgedbetween the CS and the physiological response (e.g., anaffective reaction) triggered by the US. The establishmentof such stimulus-response associations was supposed to bepossible without establishing a CS-US relation (Thorndike1911; Watson 1913). As such, the affective response to theCS (e.g., brand) is assumed to be direct (i.e., independentof an intermediating CS-US association).

As the subject population was expanded to include hu-mans, in addition to animals, a more cognitive view of eval-uative conditioning emerged (Bower 2000; Brewer 1974).Cognitive psychologists attributed conditioning effects to amemory association between the CS and the US. The re-peated pairings of a CS (e.g., brand) and a US (e.g., pleasantevent) create a memory association between the two stimuli.Subsequently, when the CS is presented on its own, it evokesconscious or unconscious thoughts of the US, which thengenerate the affective response. As such, the affective re-sponse to the brand is assumed to be indirect (i.e., dependenton the activation of the US in memory). In the evaluativeconditioning literature, this type of learning has been re-ferred to as stimulus-stimulus learning (e.g., Wickens 1959),referential learning (e.g., Baeyens et al. 1992), or associativelearning (Gawronski and Bodenhausen 2006).

Most recently, the pendulum seems to be swinging backin the direction of the early behaviorists. For example, Jones

et al. (2009) provide initial evidence for the existence ofdirect affective responses in evaluative conditioning. How-ever, as pointed out by De Houwer (forthcoming), the ma-jority of the findings in Jones et al.’s article are also con-sistent with indirect affective responses. Support for directaffective responses comes from other literatures. For ex-ample, Bower (1981) invokes stimulus-response learning toexplain a large body of findings in the literature on moodand memory.

Differentiating Direct and Indirect AffectiveResponses to Brands

An important reason why there is so much uncertaintyregarding the learning mechanisms underlying evaluativeconditioning effects is that the learning process cannot bedirectly observed. The learning process can only be inferredfrom outcomes by using appropriate manipulations or mea-sures. We propose a number of manipulations and measuresthat may help to distinguish direct versus indirect affectiveresponses to brands. All of these differentiators rely on themediating role of the unconditioned stimulus in the affectiveresponse to the brand (CS). First, and most important, in-direct affective responses should be sensitive to US reval-uation (Baeyens et al. 1992; Walther et al. 2009). US re-valuation occurs when there is a postconditioning changein the valence of the US (e.g., famous golf player TigerWoods admitting to marital infidelity). If affective responsesto the US change, indirect affective responses to the CS(e.g., brand) should track this change. Direct affective re-sponses to the CS should be unaffected by US revaluation,because the affective response to the CS is independent ofthe US at the time of revaluation.

Second, because direct affective responses do not rely onspecific CS-US associations, direct evaluative conditioningshould not necessitate conscious awareness of the CS-USrelation (Thorndike 1911; Watson 1913). In contrast, indi-rect affective responses should be sensitive to contingencyawareness (Shanks and St. John 1994). Awareness of thecontingency between the CS and US signals elaboration ofthe association. More elaboration leads to a stronger asso-ciation between the CS and US and a stronger indirect af-fective response. Third, indirect but not direct affective re-sponses should be sensitive to memory decay of the specificCS-US associations involved (Wixted 2004). Weakeningthe strength of the association between the CS and the USshould reduce the indirect affective response. Fourth, in-direct affective responses should be more susceptible to per-suasion knowledge activation than direct affective responses(Friestad and Wright 1994). Indirect affective responses relyon brand-affective stimulus associations, whereas direct af-fective responses do not. Hence, critical assessments of thebrand-affective stimulus association’s relevance for brandquality should affect indirect but not direct affective re-sponses (Gawronski and Bodenhausen 2006).

EVALUATIVE CONDITIONING 2.0 000

TABLE 1

DIRECT OR INDIRECT AFFECTIVE RESPONSES TO BRANDSAS A FUNCTION OF US PAIRINGS AND THE TIMING OF THE

UNCONDITIONED STIMULUS

Same US pairings Different US pairings

Sequential Simultaneous Sequential Simultaneous

Indirect OK OKDirect OK

Conditioning Procedures Promoting Direct versusIndirect Affective Responses to Brands

We hypothesize that the establishment of direct versus in-direct affective responses to brands depends on critical pro-cedural boundary conditions that can be traced to the learningprocedure (see table 1). Indirect affective responses criticallydepend on the strength of the association between a brand(CS) and a specific affective stimulus (US). Repeatedly pair-ing a CS with the same affective stimulus should strengthenthe association between the two and, thus, promote indirectaffective responses. The more often a brand (CS) occurswith a certain affective concept (US), the more likely itbecomes that future encounters with the brand will activatethe representation of the US in memory. In contrast, re-peatedly pairing a CS with different affective stimuli shouldnot promote indirect affective responses. Varying the affec-tive concepts (USs) that are paired with the CS should makeit unlikely that the CS will activate the representation of aparticular US. Thus, an affective stimulus should remain thesame across conditioning trials in order to establish indirectaffective responses to the brand.

The creation of direct affective responses should dependon the opportunity for affective confusion: positive affectmust be experienced in a way that makes it difficult to“determine” whether the US or CS was responsible. Thisprocess has most recently been referred to as an implicitmisattribution process of affect (Jones et al. 2009). Thisconfusion or misattribution can only occur when the US andCS appear simultaneously. Simultaneous presentation al-lows the CS to be present during the initial fast, nonspecificaffective response to the US (Ferguson 2007; Murphy, Mon-ahan, and Zajonc 1995; Murphy and Zajonc 1993). Thisinitial affective response, called “diffuse affect,” has beenhypothesized to “disperse, scatter, permeate, combine, fuse,blend, spill over, and become attached to totally unrelatedstimuli” (Murphy et al. 1995, 590). Yet, after less than 100milliseconds, an affective response loses its potential to spillover to other stimuli (Ruys and Stapel 2008; Stapel, Koo-men, and Ruys 2002). Thus, only a simultaneous CS-USpresentation has the potential to directly transfer affect tothe brand and lead to a direct affective response to the brand.

Whereas the effect of pairing a brand with the same ordifferent USs on indirect evaluative conditioning should beindependent of simultaneous versus sequential presentation,direct evaluative conditioning may be influenced by thenumber of unconditioned stimuli a brand is paired with.Pairing a brand with different unconditioned stimuli usingalways different CS-US presentations may further promotethe establishment of direct affective responses to the brand(CS), because the different pairings should lead to moreaffective confusion. Note that this should still only happenwhen CS and US are perceived simultaneously.

The combination of predictions derived from simultaneousversus sequential CS-US presentations, and from same versusdifferent CS-US pairings, is presented in table 1.

Hypotheses

The discussion of factors that differentiate indirect fromdirect affective responses, and the procedural constraints thatbound each type of learning, lead to the following predic-tions:

H1: Indirect affective responses to brands will occurwhen an evaluative conditioning procedure entailsthe repeated pairing of a brand with the sameaffective stimuli (independent of whether they aresequentially or simultaneously presented). Indi-rect affective responses are indicated by:

a) An evaluative conditioning effect.b) Sensitivity to unconditioned stimulus reval-

uation.c) Sensitivity to contingency awareness.d) Sensitivity to retroactive interference.e) Sensitivity to persuasion knowledge activa-

tion.

H2: Direct affective responses to brands will occurwhen an evaluative conditioning procedure en-tails the repeated simultaneous pairing of a brandwith different affective stimuli. Direct affectiveresponses are indicated by:

a) An evaluative conditioning effect.b) Insensitivity to unconditioned stimulus reval-

uation.c) Insensitivity to contingency awareness.d) Insensitivity to retroactive interference.e) Insensitivity to persuasion knowledge activa-

tion.

These predictions are structured to reflect the order in whichthey will be explored empirically. US revaluation is the mostcommonly accepted indicator of indirect versus direct affec-tive responses (Baeyens et al. 1992; De Houwer et al. 2001;Walther et al. 2009); thus, experiment 1 will focus on un-conditioned stimulus revaluation and contingency learning(hypotheses 1a–1c and 2a–2c). Experiment 2 will focus onmemory decay induced by retroactive interference. Experi-ment 3 will assess the effects of persuasion knowledge ac-tivation on explicit versus implicit brand attitudes.

Our reasoning above also implies that a sequential dif-ferent pairings conditioning procedure should result in little


or no evaluative conditioning because it should not be con-ducive to direct or indirect affective responses. This hy-pothesis will also be explored in experiment 1.

EXPERIMENT 1: AFFECTIVE STIMULUSREVALUATION

The most common method of distinguishing direct fromindirect affective responses to brands involves changing thevalence of the affective stimulus (US) after evaluative con-ditioning (the so-called US-revaluation effect; Baeyens et al.1992; Rescorla 1974, 1988; Walther et al. 2009). On the onehand, if an evaluative conditioning procedure establishes anindirect affective response to the brand, then any post-conditioning change in the affective response to the US shouldyield corresponding changes in the affective response to thebrand (i.e., CS). On the other hand, direct affective responsesto the brand are independent of memory traces to the US.When evaluative conditioning results in a direct affective re-sponse, postconditioning changes in affective responses to theUS should not affect the attitude toward the brand.

The influence of US revaluation was tested in the contextof the four evaluative conditioning procedures illustrated intable 1. We anticipated that basic evaluative conditioningeffects would be obtained for all procedures except the se-quential-different US pairings procedure. The sequential-different US pairings procedure should generate neither di-rect nor indirect affective responses to the brand. We alsoanticipated that US revaluation effects would be obtainedin the same pairing procedures because these proceduresshould generate indirect affective responses, but not in thesimultaneous-different US pairings condition that shouldsupport direct affective responses. Finally, we expected theevaluative conditioning effect to depend on contingencyawareness in the same pairings procedure, but not in thesimultaneous-different US pairings procedure. The predic-tions are consistent with hypotheses 1a–1c and 2a–2c.

Method

Participants and Design. Two hundred and ninety-fourundergraduate business students (166 female, 128 male) atthe University of Florida participated in the experiment inexchange for extra course credit. The experiment employeda 2 (US heterogeneity: same pairings, different pairings) #2 (presentation format: sequential presentation, simulta-neous presentation) # 3 (US valence: neutral, standard pos-itive, positive to-be-revaluated) mixed design with the firsttwo factors manipulated between subjects and the latter fac-tor manipulated within subject. The experiment was admin-istered in two fall semesters, separated by 1 year. The ex-perimental sessions were procedurally identical, except thatthe second run included contingency awareness measures atthe end of the experiment. The sessions were run on com-puters in a behavioral lab.

Conditioned Brands. Belgian beers unknown to theparticipant population were chosen to serve as to-be-con-

ditioned brands (CS). To avoid contamination by existingattitudes toward brand names, labels, bottles, and so on, weran a pretest using 74 participants from the same participantpool. The pretest measured participant attitudes toward 43Belgian beers. The nine beers with the most neutral (i.e.,closest to the midpoint of the scale) and most normallydistributed attitudes were selected to serve as conditionedbrands.

Affective Stimuli. Ten positively valenced and five neu-trally valenced images from the International Affective Pic-ture System (IAPS: Lang, Bradley, and Cuthbert 2005) wereselected to serve as US. Positive images depicted adultshaving fun in various ways, such as cuddling, waterskiing,or sailing. Neutral images depicted adults with neutral ex-pressions and engaged in such everyday activities as readinga newspaper on a bench or napping on the subway. On theIAPS’s 9-point affective rating scale, all neutral imagesscored between 4.5 and 5.5 for both male and female raters,whereas the positive images all scored above 7.0 for bothgenders. For every participant, the computer program ran-domly selected the five positive pictures that would remainpositive throughout the experiment (the standard positiveUSs) and the five positive pictures that would be made morenegative over the course of the experiment (the positive to-be-revaluated USs).

Procedure. The cover story explained that the studywas about assessing college students’ spontaneous attitudestoward Belgian beers. Participants were told that it was un-likely that they had ever seen these brands before, so a slideshow would be used to familiarize them with the brands.Participants were also told that the slide show included pic-tures of people engaged in various activities in an effort tomake it more interesting. Next, participants went throughdifferent phases in the experiment in the following order: aconditioning phase, a US revaluation phase, attitude as-sessment toward the conditioned brands, demand awarenessmeasures, a manipulation check of the US revaluation, acontingency awareness measure (only for the 197 partici-pants in the second fall semester), and some demographicquestions.

Conditioning Phase. In the first part of the experiment,three of the nine beer brands (i.e., the three control brands)were randomly selected to be paired with the neutral USs.Three other beer brands were randomly selected to be pairedwith the standard positive USs (that would remain positivethroughout the experiment). The remaining three beers werepaired with the positive to-be-revaluated USs (toward whichthe attitude would be changed in a later phase of the ex-periment).

The conditioning procedure consisted of five rounds ofpairings. Every brand appeared once in every round andwas paired with a US. Therefore, the total procedure con-sisted of 45 CS-US pairings: nine brands (CSs), each ofwhich would be paired with USs five times. For participantsin the same pairings conditions, each brand appeared five


FIGURE 1

TWO SAMPLE CONDITIONING ROUNDS IN EACH OF THE EXPERIMENT 1 PROCEDURES

times with the same US. For participants in the differentpairings condition, each brand appeared once with fiveunique USs. In the sequential conditioning procedure, thebrand was presented in the center of the screen for 1.5seconds, followed by an interstimulus interval of 0.5 seconds(blank screen), followed by the US presented for 1.5 sec-onds. The intertrial interval was 1.5 seconds. In the simul-taneous conditioning procedure, the picture of the US (whichcovered the entire screen) was shown with the brand su-perimposed on the bottom center of the screen for 3 seconds.Next, there was a 2 second intertrial interval. Note that thisprocedure guarantees equal slide show duration as well asequal total presentation duration of brand plus USs acrossconditions. Two illustrative rounds for each of the four con-ditions are presented in figure 1.

Affective Stimulus (US) Revaluation Phase. Partici-pants saw the USs from the conditioning phase a secondtime after a short introduction that stated that inconspicuous-looking people can commit serious crimes. In this secondphase, all USs were shown three more times. Felonies (e.g.,murderer, rapist, arsonist) were consistently displayed at thebottom of the to-be-revaluated positive USs. The neutraland standard positive USs were shown equally often (thrice),

but without accompanying felony information. The condi-tioned beer brands did not appear in any of the pictures.

Brand Attitude Assessment. After the affective stim-ulus revaluation phase was complete, participants’ attitudestoward all nine conditioned brands were assessed. For everybrand, participants were asked to provide their global atti-tude toward the beer on a 9-point scale (scale endpoints“extremely negative,” “extremely positive”). Next, partici-pants indicated how appealing they found the beer (scaleendpoints “totally unappealing,” “very appealing”) and theirlikelihood of buying the beer if it were available at a rea-sonable price (scale endpoints “extremely unlikely,” “ex-tremely likely”) using 7-point scales.

Demand Assessment. First, we asked participants toindicate all the factors that had contributed significantly totheir ratings of the beers. They could place check marksnext to eight answer categories, one of which was “Exper-imental demand: You rated the beers in the way you thoughtthe experimenter wanted you to rate them.” Second, we useda funneled debriefing technique in which participants re-sponded to four progressively revealing questions regardingthe nature of the experiment. The questions were “Sum-


FIGURE 2

BRAND (CS) ATTITUDES AS A FUNCTION OF AFFECTIVE STIMULUS (US) TYPE AND THE CONDITIONING PROCEDURE

marize below what you believe was the purpose of thisstudy,” “What do you think are the hypotheses being testedin this study?” “During the study, did you ever have athought that its purpose might be something other than youwere told? If so, when did this occur to you, and what wereyou thinking?” and, finally, “Did you notice any regularitiesin the presentation of the beers and the human pictures? Ifso, please describe what struck you as remarkable.”

Affective Stimulus (US) Attitude Assessment (Manip-ulation Check). We assessed participants’ attitudes towardall the USs that were presented over the course of the ex-periment on a 9-point scale with endpoints labeled “ex-tremely negative” and “extremely positive.”

Contingency Awareness. This part of the procedurewas only administered to the 197 participants in the secondwave of data collection. In order to assess participants’awareness of the CS-US relationship, we asked them toindicate for every beer brand a picture that it had occurredwith at least once. On nine consecutive screens, each beerwas presented on the left side of the screen, and the 15 USswere presented on the right side of the screen. Finally, afterresponding to demographic questions, the participants werefully debriefed.

Results

The attitudes toward the beer brands in this and all sub-sequent experiments were computed by transforming theglobal attitude ratings from a 9-point to a 7-point scale and

averaging them with the appealingness ratings (Cronbach’salphas were higher when likelihood to buy measures werenot included).

Affective Stimulus (US) Revaluation ManipulationCheck. As intended, after the US revaluation phase, theparticipants liked the standard positive USs better (M p6.84) than the positive to-be-revaluated USs (M p 5.93),(F(1, 290) p 83.65, p ! .001). This was not moderated byany of the between-subjects factors (all p 1 .20).

Brand (CS) Attitudes. The mean attitudes and standarderrors toward all the classes of conditioned brands are shownin figure 2. The three-way interaction between presentationformat, US heterogeneity, and US valence was significant(F(2, 580) p 3.75, p ! .03). To assess support for hypotheses1a and 1b, three tests were performed. First, consistent withhypotheses 1a and 1b, there was a significant main effectof US valence in the same US pairing conditions (F(2, 580)p 39.07, p ! .001). Second, consistent with hypothesis 1a,pairing the beer brands with the standard positive USs (Mp 4.67) resulted in higher brand evaluations than pairingthe beer brands with the neutral USs (M p 4.02; F(1, 290)p 54.82, p ! .001). This relationship held in the sequentialsame pairings (Mneut p 4.12, Mpos p 4.70; F(1, 290) p22.38, p ! .001) and simultaneous same pairings (Mneut p3.92, Mpos p 4.65; F(1, 290) p 33.21, p ! .001) conditions.Third, consistent with hypothesis 1b, the beer brands thatwere paired with positive to-be-revaluated USs resulted inlower brand evaluations (M p 3.94) than the beer brands


paired with the standard positive USs (M p 4.67; F(1, 290)p 70.66, p ! .001). This relationship held in the sequentialsame pairings (Mstd pos p 4.70, Mreval pos p 3.94; F(1, 290) p37.17, p ! .001) and simultaneous same pairings (Mstd pos p4.65, Mreval pos p 3.93; F(1, 290) p 31.03, p ! .001) con-ditions. We further note that the beer brands paired withpositive to-be-revaluated USs were not liked more than thecontrol brands paired with neutral stimuli in the same pair-ings conditions (both p 1 .20). These results show that USrevaluation reduced and effectively removed the evaluativeconditioning effect in the same US pairings conditions.

To assess support for hypotheses 2a and 2b, three testswere performed. First, consistent with hypotheses 2a and2b, there was a significant interaction between presentationformat and US valence in the different US pairing conditions(F(2, 580) p 12.12, p ! .001). Second, consistent withhypothesis 2a, pairing the beer brands with the standardpositive USs (M p 4.31) resulted in higher brand evalua-tions than pairing the beer brands with the neutral USs (Mp 3.54) in the simultaneous condition (F(1, 290) p 35.20,p ! .001), but not in the sequential condition (Mneut p 3.99,Mpos p 4.02; F(1, 290) p 0.04, p 1 .80). Third, consistentwith hypothesis 2b, pairing the beer brands with positiveto-be-revaluated USs (M p 4.28) resulted in higher brandevaluations than pairing the beer brands with the neutralstimuli (M p 3.54) in the simultaneous condition (F(1, 290)p 27.18, p ! .001), but not in the sequential condition (Mneut

p 3.99, Mreval pos p 3.89; F(1, 290) p 0.60, p 1 .40). Thislast set of tests shows that US revaluation did not removethe evaluative conditioning effect in the simultaneous dif-ferent US pairings condition. Furthermore, US revaluationdid not even reduce the evaluative conditioning effect inthis condition, as brands paired with positive to-be-revalu-ated USs were not liked less than brands paired with standardpositive USs (F(1, 290) p 0.05, p 1 .80).

Contingency Awareness. First, we analyzed partici-pants’ levels of awareness by comparing the percentage ofcorrect identifications in the awareness test with the per-formance expected from random guessing. Consistent withhypothesis 1c, we find relatively high levels of awarenessin the same pairings conditions. Random responses in theseconditions would yield 6.66% correct identifications, as foreach brand, only one in 15 pictures in the awareness testwas correct. In the sequential same pairings condition, par-ticipants indicated the correct US for 46% of the brands,significantly higher than the chance level (t(77) p 8.95,p ! .001). In the simultaneous same pairings condition, againparticipants also picked the correct US for 46% of brands,again much higher than chance level (t(71) p 8.49, p !

.001). In the different pairings conditions, there was a fivetimes higher probability for correct guesses (33% chancelevel), as five of the 15 pictures were correct responses inthe awareness test. Corroborating the claim that direct af-fective responses can be established without contingencyawareness (hypothesis 2c), there were only 38% correct re-sponses in the simultaneous different pairings condition, notsignificantly different from chance level (t(69) p 1.47, p 1

.14). In the sequential different pairings condition, therewere only 30% correct responses, also not significantly dif-ferent from chance level (t(73) p �0.86, p 1 .35).

Most informative is an item-based contingency awarenessanalysis as recommended by Pleyers et al. (2007). Pleyerset al. recommend looking separately at brands of which thecorrect US was indicated (we call these “aware brands”)versus those thought associated with an incorrect US (“un-aware brands”). We first explore the attitudes toward theaware and unaware standard positively conditioned brandsand compare them with the control brands. Means and stan-dard errors are represented in figure 3. Both direct and in-direct evaluative conditioning should lead to “aware posi-tively conditioned brands” being liked better than the controlbrands. This is indeed the case (in the sequential same pair-ings condition: Maware p 4.94, Mcontrol p 4.18; F(1, 510) p13.79, p ! .001; in the simultaneous same pairings condition:Maware p 4.89, Mcontrol p 3.94; F(1, 510) p 18.69, p ! .001;in the simultaneous different pairings condition: Maware p4.46, Mcontrol p 3.56; F(1, 510) p 16.57, p ! .001; but notin the sequential different pairings condition where therewas no conditioning to start with: Maware p 4.03, Mcontrol p3.94; F(1, 510) p 0.17, p 1 .65). However, only directevaluative conditioning should lead to “unaware positivelyconditioned brands” being liked better than neutral controlbrands. That is, an evaluative conditioning effect shouldeven be found for brands that participants indicated had beenpaired with a neutral or revaluated picture but that had inreality been paired with a positive picture. This should bethe case because the direct evaluative conditioning effectshould not depend on CS-US associations stored in memory(in the simultaneous different US pairings condition: Munaware

p 3.98, Mcontrol p 3.56; F(1, 510) p 3.72, p p .05). In theother conditions, “unaware positively conditioned brands”were not liked more than neutral control brands (all p 1 .50).In the sequential-different USs condition, there was no eval-uative conditioning regardless of awareness. For the same USconditions, “unaware positively conditioned brands” shouldnot be liked better than neutral control brands because (1)under indirect evaluative conditioning, evaluations of a brandshould depend on the evaluation of the USs with which abrand is believed to be associated, and (2) those USs for the“unaware positively conditioned brands” were most oftenneutral or revaluated pictures. Thus, these data provide an-other confirmation of hypotheses 1c and 2c.

Next, we performed the same analysis on the brands thatwere paired with positive to-be-revaluated USs. Means andstandard errors for the attitudes toward aware and unawarebrands paired with positive to-be-revaluated USs are pre-sented in figure 4. Only in the condition generating directaffective responses (simultaneous different US pairings)were the brands paired with positive to-be-revaluated USsliked better than control brands. This was true irrespectiveof whether the correct US was indicated, as the differenceis significant for “unaware brands” (Munaware p 4.25, Mcontrol

p 3.56; F(1, 497) p 8.29, p ! .01) and even for “awarebrands” (Maware p 4.16, Mcontrol p 3.56; F(1, 497) p 6.58,


FIGURE 3

ATTITUDES FOR BRANDS (CSs) PAIRED WITH STANDARD POSITIVE STIMULI (USs) (COMPARED WITH CONTROL BRANDS)AS A FUNCTION OF THE CONDITIONING PROCEDURE AND CONTINGENCY AWARENESS

p p .01). (In the simultaneous different US pairings condi-tion, participants were generally unaware of the CS-US re-lationship, as their total awareness scores did not exceedchance level. Nevertheless, for some brands, a correct US wasidentified. These brands are therefore classified as “awarebrands.”) The latter finding indicates that even when con-sumers indicate the correct (now devaluated) US as beingpaired with the brand, the devaluation did not affect theirpositively conditioned brand attitudes. Hence, the affectiveresponse is direct even when the correct US is indicated. Inno other condition were revaluated CSs (aware or unaware)liked better than control brands (all p 1 .10).

Mediation by Objectively versus Subjectively PairedUS Evaluations. We define the basic evaluative condi-tioning effect as the difference in attitudes toward brandspaired with standard positive stimuli and brands paired withneutral stimuli. This basic evaluative conditioning effect de-pended, of course, on the valence of the unconditioned stim-uli, as pairing a brand with standard positive stimuli resultsin more positive brand attitudes than pairing it with neutralstimuli. Thus, it would not be surprising if attitudes towardthe objectively paired unconditioned stimuli mediated the ba-sic evaluative conditioning effect. What is less obvious is theextent to which indirect and direct affective responses shouldbe sensitive to the participants’ perceptions of whether or nota US was paired with a CS. To appreciate this distinction,recall that indirect affective responses rely on the valence ofthe US that is activated in memory. Thus, indirect evaluative

conditioning effects should be fully mediated by the valenceof the US that participants subjectively remember to havebeen paired with a brand. In contrast, direct evaluative con-ditioning is not dependent on memories of USs, but on thevalence of the USs the brand was actually, objectively,paired with at the time of the learning phase. Thus, whereasthe evaluative conditioning effect in the same US conditionsshould be mediated more fully by the rated valence of thesubjectively paired USs than by the rated valence of theobjectively paired USs, the opposite pattern should obtainfor the simultaneous different pairings condition. Analysesof mediation of the basic evaluative conditioning effect bythe attitudes toward the USs that were actually paired withthe CS (objectively paired US), and the attitudes toward theUS that the participant thought was paired with the US(subjectively paired US; derived from the contingencyawareness instrument), confirmed these predictions. In thesame pairings conditions, we found that the effect of CScondition (as paired with standard positive versus neutralUS) on CS attitudes was completely mediated by the sub-jectively paired US (and only partially mediated by the ob-jectively paired US). In the simultaneous different pairingscondition, however, the effect of CS condition on CS atti-tudes was fully mediated by the objectively paired US (andonly partially mediated by the subjectively paired US). Fur-ther details of this analysis are available from the first author.

Demand Awareness. Two analyses were performed toassess whether demand awareness was responsible for the


FIGURE 4

ATTITUDES FOR BRANDS (CSs) PAIRED WITH TO-BE-REVALUATED POSITIVE STIMULI (USs) (COMPARED WITHCONTROL BRANDS) AS A FUNCTION OF THE CONDITIONING PROCEDURE AND CONTINGENCY AWARENESS

evaluative conditioning results. For the “indicate factors thatcontributed to your ratings” check mark question, only 18 outof 294 participants indicated that they had tried to provideevaluative ratings that matched experimenter expectations.Omitting these participants from the analysis did not influencethe statistical significance of any reported tests. For the open-ended questions, two independent coders, who were blind tothe experimental conditions, sorted the participants into fourdifferent categories of hypothesis awareness (84% agreement,conflicting assignments resolved through discussion). Onehundred and forty-eight participants were classified as “hy-pothesis unaware,” as they expressed no hypothesis-threat-ening beliefs. Twenty-one participants claimed that somebrands were systematically paired with more positive imagesthan others and were classified in a second category. Eighty-nine participants expressed the belief that affective stimulipaired with felons might reduce liking of associated brandsand were classified in a third category. Finally, 36 partici-pants expressed both of the previous beliefs and were clas-sified in the fourth and highest category of hypothesis aware-ness. Incorporating the level of hypothesis awareness as acovariate in the attitude analyses left the significance for thethree-way interaction unchanged (F(2, 578) p 3.76, p !

.03). Furthermore, the same overall pattern of results wasobserved at every level of hypothesis awareness. In otherwords, expressed demand awareness was uncorrelated withparticipants’ responses.

Discussion

Experiment 1 provides some preliminary evidence thatboth indirect and direct affective responses can result fromevaluative conditioning procedures. The results show that aprocedure consisting of the repeated pairing of a brand nameand the same valenced stimulus was sensitive to postcon-ditioning affective stimulus revaluation. Brands paired withdevaluated positive stimuli were liked less than brandspaired with normal positive stimuli and not liked better thancontrol brands that had never been paired with positive stim-uli. The inference is that this procedure promoted indirectaffective responses to the brand name. The results also showthat a procedure consisting of the repeated simultaneouspresentation of a brand name and different valenced stimuliwas insensitive to postconditioning affective stimulus de-valuation. Brands paired with devaluated positive stimuliwere liked as much as brands paired with normal positivestimuli and continued to be liked more than control brands.The inference is that the procedure promoted direct affectiveresponses to the brand name.

The results of a contingency awareness analysis supportedour inferences about the conditions under which indirect ordirect affective responses occur. First, the same affectivestimulus pairings conditions, which should be conducive toindirect affective responses, showed higher levels of con-tingency awareness, and this contingency awareness wasassociated with the strength of the evaluative conditioning


effect. Second, in the simultaneous different pairings con-dition, the awareness of CS-US associations was not abovechance level, consistent with the idea that direct affectiveresponses are not dependent on contingency awareness. Thisfinding, along with the insensitivity of the evaluative con-ditioning effect to affective stimulus revaluation, providesfurther evidence for the existence of direct affective re-sponses in human evaluative conditioning as well as for theirfundamental distinctness from indirect affective responses.The finding is also consistent with early behaviorists’ claimsthat stimulus-response learning can occur without awarenessof CS-US relationships (Thorndike 1911; Watson 1913).Particularly interesting is that in the simultaneous differentpairings condition, brands paired with positive to-be-deval-uated stimuli remain positively liked even when participantsknew they had been paired with devaluated USs. Hence, inthis condition, the brand attitudes are not derived from mem-ory associations to the affective stimuli, a hallmark propertyof direct affective responses.

Finally, our interpretation in terms of indirect versus directaffective responses was further supported by mediation anal-yses of the basic evaluative conditioning effect (i.e., thediverging attitudes toward standard positively versus neu-trally conditioned brands). We found that when the affectiveresponse was hypothesized to be indirect, conditioned brandattitudes were fully mediated by the attitude toward the af-fective stimulus that is subjectively associated with the brand(i.e., which comes to mind in the awareness test when thebrand is presented). However, when the affective responsewas hypothesized to be direct, the brand attitudes were fullymediated by the attitudes toward the affective stimuli thatwere objectively paired with the brand.

EXPERIMENT 2: RETROACTIVEINTERFERENCE

A second way in which direct affective responses maybe distinguished from indirect affective responses is by theproperty of retroactive interference (hypotheses 1d and 2d).Retroactive interference is a major source of forgetting andoccurs when the learning of new information makes pre-viously learned associations less accessible (Wixted 2004,2005). Indirect affective responses should be particularlysusceptible to retroactive interference because they dependon the brand-affective stimulus association. In contrast, di-rect affective responses should be insensitive to retroactiveinterference because the brand name directly evokes theaffective response, independent of any brand associationsto specific affective stimuli. These predictions were testedwith two evaluative conditioning procedures that are hypoth-esized to uniquely generate indirect affective responses (se-quential same pairings) versus direct affective responses (si-multaneous different pairings; the simultaneous same pairingsprocedure is somewhat more ambiguous because the simul-taneous presentations could allow for some direct affect trans-fer in addition to the indirect effect in this condition).

Method

Participants and Design. Forty undergraduate businessstudents (13 female, 27 male) at Erasmus University Rotter-dam participated in the experiment in exchange for coursecredit. The experiment employed a 2 (conditioning procedure:sequential same pairings, simultaneous different pairings) #2 (interference: no, yes) # 2 (US valence: neutral, positive)mixed design with the first two factors manipulated betweensubjects and the latter factor manipulated within subject.

Conditioning Phase. The conditioning procedure wasanalogous to sequential same US pairings and simultaneousdifferent US pairings conditions of experiment 1, with theexception that there were no brands paired with positive to-be-revaluated USs. Four Belgian beer brands (CSs) werepaired with positive USs, and four Belgian beer brands werepaired with neutral USs.

Interference by Subsequent Learning. After the con-ditioning phase, participants in the interference conditionentered a second learning phase in which they were askedto learn the brewers of a new set of eight Belgian beerbrands. Thus, we created interference by learning infor-mation about new brands rather than about the brands fromthe conditioning phase. The former is much more commonin daily life (Wixted 2005) and perhaps also more repre-sentative of interference in advertising. In the interferencephase, there were four Belgian breweries (e.g., “Van Hon-sebrouck”) that each brewed two of the new beers. Thislearning phase consisted of three presentation rounds, duringwhich a picture of every beer was presented with a breweryname for 3 seconds, with 1 second between the pairs. Afterthree presentation rounds, participants were asked to indicatethe brewery that produced each beer.

Dependent Measures. Only attitudes toward the beersthat appeared in the conditioning phase were assessed, anal-ogous to experiment 1. Familiarity with the eight condi-tioning phase beers and demographic measures completedthe measurement instrument.

Results

An initial analysis confirmed that the participants’ fa-miliarity with the beer brands was a low 6.25%. Means andstandard errors for the attitudinal index are presented infigure 5. The three-way US valence by conditioning pro-cedure by interference condition interaction was significant(F(1, 36) p 7.40, p p .01). Consistent with hypotheses 1aand 1d, the sequential same pairings condition revealed aUS valence by interference interaction (F(1, 36) p 14.88,p ! .001). Analyses of the simple effects revealed successfulevaluative conditioning without interference (Mneut p 2.99,Mpos p 4.64; F(1, 36) p 28.09, p ! .001), but not withinterference (Mneut p 4.03, Mpos p 3.94; F(1, 36) p 0.08,p 1 .75). Consistent with hypotheses 2a and 2d, the simul-taneous different US pairings condition revealed a signifi-cant main effect of US valence (F(1, 36) p 16.46, p ! .001)


FIGURE 5

BRAND (CS) ATTITUDES AS A FUNCTION OF AFFECTIVE STIMULUS (US) TYPE, THE CONDITIONING PROCEDURE,AND RETROACTIVE INTERFERENCE

that was not moderated by interference (F(1, 36) p 0.01,p 1 .90). The evaluative conditioning effect was significantwithout (Mneut p 3.34, Mpos p 4.29; F(1, 36) p 7.49, p !

.01) and with (Mneut p 3.28, Mpos p 4.26; F(1, 36) p 9.06,p ! .01) interference.

Discussion

This experiment explores the consequences of retroactiveinterference for evaluative conditioning. The results show thatpresumably indirect affective responses, in the sequentialsame pairings procedure, are critically susceptible to retro-active interference. Presumably direct affective responses, inthe simultaneous different pairings procedure, are insensitiveto retroactive interference. These results suggest that the di-rect affective responses in the simultaneous different USpairings condition are indeed fundamentally different fromthe indirect affective responses in the sequential same USpairings condition. These results also rule out the alternativeexplanation that affective responses in the simultaneous dif-ferent pairings procedure resulted from multiple, weakbrand-affective stimulus associations. If this were so, theopposite pattern of results should have occurred: multiple,weak associations should be much more susceptible to in-terference than the single, solid association in the sequentialsame pairings condition.

EXPERIMENT 3: PERSUASIONCORRECTION

Experiment 3 primed persuasion knowledge (Friestad andWright 1994), the knowledge consumers use to resist adver-tisers’ attempts to persuade them, to provide two insights.First, we hypothesized that direct affective responses (whichare independent of awareness) will be less susceptible to con-sumers’ counterarguing strategies than indirect affective re-sponses that depend on associations between brands and pos-itive stimuli (of which consumers are often aware; see theresults from experiment 1). This different susceptibility tocorrection would be consistent with hypotheses 1e and 2e.

Second, persuasion knowledge activation can provide in-sight in the associative versus propositional nature of theindirect affective response. Recently, Gawronski and Bod-enhausen (2006) claimed that there are two indirect, asso-ciation-based processes active in evaluations (they did notconsider direct affective responses). The first process is thesimple association between two stimuli (CS and US). Thesecond process is the first process plus propositional reasoningabout the validity of the association for one’s evaluative judg-ment. The question is whether the indirect affective responseis reflective of purely propositional reasoning or whether thereis also a more basic associative process that contributes tothe indirect affective response or a combination of both. Using


a persuasion knowledge prime, we provided participantswith a warning that the images contain no valuable infor-mation about the taste or quality of the beer brand and thusshould not influence their evaluative judgments. This ma-nipulation should correct for evaluative conditioning drivenby a propositional process and allow us to assess if there is(also) a more basic associative process at work.

This experiment will, in addition, assess implicit attitudestoward the conditioned brands. This measure should be muchless sensitive to propositional correction processes and morereflective of the simple association between the stimuli (case3 in Gawronski and Bodenhausen 2006).

Method

Participants and Design. The participants in this ex-periment were 214 undergraduate students (149 male, 65female) at Erasmus University Rotterdam who were re-warded with partial course credit. The design was a 2 (USvalence: positive, neutral) # 2 (conditioning procedure: se-quential same pairings, simultaneous different pairings) #2 (persuasion knowledge activation: control, active) design,with the first factor manipulated within subject.

Stimuli and Procedure. The evaluative conditioningpart of the experimental procedure was analogous to ex-periment 2. After the evaluative conditioning procedure, par-ticipants in the persuasion knowledge activation conditionread the following: “warning: You just saw these beers incombination with pictures. It’s important to realize that thesepictures contain no valuable information about the taste orquality of the beer. Therefore, it might be advisable not torely on the pictures the beers were shown with when youdetermine your attitudes toward the beers.” Participants inthe control condition received no such warning. The ma-nipulation followed rather than preceded the persuasion at-tempt (as is more common in persuasion knowledge re-search; Campbell and Kirmani 2000) because we did notwant participants to ignore the CS-US presentations. Thiswould likely interfere with the formation of both direct andindirect affective responses and thus compromise the ex-periment’s ability to distinguish direct from indirect affec-tive responses.

Immediately afterward, we assessed participants’ explicitand implicit attitudes toward the beers. The order of explicitversus implicit attitude measures was counterbalanced acrossparticipants. The explicit attitude assessment was identical toexperiments 1 and 2. The implicit attitude assessment reliedon an affective priming task (Fazio et al. 1986; Wittenbrink2007). In an affective priming procedure, primes (here theconditioned brands) are followed by target words of polar-ized valence. The participants’ task is to categorize the targetwords according to their evaluative connotation (positiveversus negative). Six positive words and six negative wordswere used as target words in the implicit test. The beerbrands from the conditioning phase were used as primes.To the extent the conditioning procedure created an implic-itly positive brand attitude, the brand prime should facilitate

correct classification of positive words and inhibit the correctclassification of negative words (Klinger, Burton, and Pitts2000). A trial in the affective priming task consisted of (1)a presentation of the prime (i.e., a conditioned beer brand)for 0.5 seconds, (2) a short time interval of blank screen(50 milliseconds), and (3) presentation of the target worduntil response (“P” for positive and “N” for negative).Twelve practice trials were followed by 96 test trials: eachof the eight conditioned brands was presented with everypositive and negative word.

The procedure ended with a demand awareness assess-ment, familiarity ratings of the eight brands, and demo-graphic measures. Demand awareness analyses paralleledthe results of experiment 1.

Results

Explicit Attitudes. The counterbalancing factor of at-titude assessment (explicit versus implicit attitudes first as-sessed) did not affect the results, so the data were collapsedover its levels. An initial analysis confirmed that the par-ticipants’ familiarity with the beer brands was a low 3.68%.Means and standard errors for the explicit attitude measureare presented in figure 6. As predicted by hypothesis 1e, inthe sequential same US condition, the main effect of USvalence (F(1, 210) p 35.44, p ! .001) was qualified by asignificant interaction with persuasion knowledge activation(F(1, 210) p 5.94, p ! .02). This interaction means thatthe explicit evaluative conditioning effect was larger in thecontrol condition (Mneut p 3.71, Mpos p 4.63, Mdiff p 0.92)than in the persuasion knowledge active condition (Mneut p3.96, Mpos p 4.34, Mdiff p 0.38). Persuasion knowledgereduced, but did not eliminate, the evaluative conditioningeffect as positively conditioned brands remained more wellliked than control brands (F(1, 210) p 6.12, p p .01). Thissuggests that propositional and “simple” associative learningboth contributed to the evaluative conditioning effect in thesequential same US pairings condition. Thus, we found evi-dence for both of Gawronski and Bodenhausen’s (2006)association-based processes.

As predicted by hypothesis 2e, in the simultaneous dif-ferent US pairings procedure, there was only a main effectof US valence (F(1, 210) p 37.24, p ! .001), which wasnot moderated by the persuasion knowledge activation (F(1,210) p 1.42, p 1 .20). These results imply that the eval-uative conditioning effect did not differ significantly be-tween the control (Mneut p 3.54, Mpos p 4.35, Mdiff p 0.81)and the persuasion knowledge active (Mneut p 3.85, Mpos p4.40, Mdiff p 0.55) conditions. This is again consistent withthe hypothesis that simultaneous different US pairings gen-erate direct affective responses to brands. Although not sta-tistically significant, there is a directional correction processin the simultaneous different US pairings condition as well.This is not completely unexpected because, in the extremecase, anyone can correct for evaluative conditioning’s in-fluence on explicit brand evaluations by simply rating allbrands neutrally. This has, however, the unfortunate con-


FIGURE 6

EXPLICIT BRAND (CS) ATTITUDES AS A FUNCTION OF AFFECTIVE STIMULUS (US) TYPE,THE CONDITIONING PROCEDURE, AND PERSUASION CORRECTION

sequence that the overall three-way interaction between USvalence, conditioning procedure, and persuasion knowledgeactivation was not statistically significant ( p ! .40). An aux-iliary, more powerful analysis was conducted in which the5% fastest responders on the explicit attitude questions wereremoved, and the counterbalancing factor (implicit versusexplicit attitudes first assessed) and the level of hypothesisawareness were included as full factors to reduce error var-iance. The 5% fastest respondents were most likely not tohave taken the experiment seriously or to have used a strat-egy that disregarded their attitudes toward the brands (e.g.,rating brands randomly), and this left us with 203 respon-dents for the analysis. In this analysis, the three-way inter-action between affective stimulus valence, the conditioningprocedure, and the persuasion knowledge factor was sig-nificant (F(1, 187) p 4.17, p p .04). This three-way in-teraction was itself not further qualified by the hypothesisawareness or counterbalancing factors (all ).p ≥ .10

Implicit Attitudes. Implicit attitudes were analyzed fol-lowing the recommendations by Fazio (1990). First, incor-rect responses were eliminated (2.8%). Second, outlier la-tencies were truncated at 200 and 1,500 milliseconds (1.8%;Gawronski, Bodenhausen, and Becker 2007). Third, responselatencies were log transformed to achieve a normal distri-bution. Fourth, a positivity index was computed for eachbrand by subtracting the average response latency to positivewords from the average response latency to negative words.Finally, the appropriate indices were averaged to obtain anindex for the positively conditioned brands and an index forthe neutrally conditioned brands.

As with the explicit attitudes, the counterbalancing factorhad no significant effects and was dropped. The implicitattitude index was analyzed as a function of US valence,the conditioning procedure, and the persuasion knowledgefactor as a mixed model with the first factor within subjectand the latter two between subjects. In the sequential sameUS pairings condition, there was a main effect of US valenceonly (F(1, 210) p 4.00, p ! .05), indicating that positivelyconditioned brands had a higher implicit positivity index(M p 0.04) than neutrally conditioned brands (M p 0.02).Contrary to the explicit attitude analysis, this main effect wasnot moderated by persuasion knowledge activation ( p 1 .40).In the simultaneous different US pairings condition, there alsowas a main effect of US valence only (F(1, 210) p 4.07,p ! .05), indicating that positively conditioned brands againhad a higher implicit positivity index (M p 0.06) than neu-trally conditioned brands (M p 0.04). Analogous to theexplicit attitudes in this procedure, there was no moderationby persuasion knowledge activation ( p 1 .90).

Discussion

In this experiment, participants were asked to correct forevaluative conditioning’s effect on their brand attitudes. Theresults suggest an important difference between indirect anddirect affective responses. Participants who receive a warn-ing about the irrelevance of the images for their brand at-titudes were able to correct their explicit attitudes in thesequential same US pairings condition. This suggests thatthe indirect affective response involved propositional rea-


soning, consistent with our findings in experiment 1. How-ever, the results also revealed that persuasion knowledgeactivation could not eliminate the evaluative conditioningeffect on the explicit attitude measure. In the sequential sameUS pairings condition, an evaluative conditioning effect re-mained even after correction. This suggests that evaluativeconditioning procedures can persuade because of simple butstill indirect associative learning that is devoid of proposi-tional reasoning. The results also show that persuasionknowledge activation did not significantly reduce the eval-uative conditioning effect in the simultaneous different pair-ings condition. This provides further support for direct af-fective responses to brands in this condition. Finally, theexistence of an evaluative conditioning effect for both eval-uative learning procedures when an implicit attitude measureis used provides further evidence that the attitudinal changescannot simply be dismissed as demand artifacts. The implicitattitudes also show that—at a more unconscious level—theconditioning effects on brand attitudes are resistant to per-suasion correction, irrespective of whether the affective re-sponse to the brand is direct or indirect.

GENERAL DISCUSSIONThree experiments suggest that different ways of pairing

brands and affective stimuli lead to brand attitudes that differin their strength and resilience. Reliable evaluative condi-tioning effects were found when a brand was consistentlypaired with a single positive stimulus or when a brand wassimultaneously paired with different positive stimuli. In ex-periment 1, we also found that the latter conditioning pro-cedure yielded brand attitudes that were more resistant to apostconditioning decline in the valence of the affective stim-uli. In contrast, the evaluative conditioning procedures thatconsistently paired a brand with the same affective stimuluswere seriously affected by such “US revaluation.” Finally,experiment 1 showed that evaluative conditioning in the si-multaneous procedure with multiple different affective stimuliwas less dependent on consumers’ awareness of the pairingsthan evaluative conditioning using the consistent, same pos-itive stimulus, procedures.

In experiment 2, we assessed the resilience of conditionedbrand attitudes to retroactive interference. We found thatexposing participants to new but related information afterthe initial evaluative conditioning led to a decline in brandevaluation when the brand had been consistently followed,after a brief delay, by the same positive stimulus duringinitial learning. However, such retroactive interference wasnot found when we used the simultaneous procedure withmultiple affective stimuli.

In experiment 3, we explored resilience to participants’attempt to correct for evaluative conditioning’s effect ontheir brand attitudes. Results indicated again that the si-multaneous procedure pairing a brand with multiple affec-tive stimuli yielded explicit brand evaluations that weremore resilient than the sequential procedure consistentlypairing a brand with the same affective stimulus. Resultsalso indicated that the evaluative conditioning effect was

largely unaffected by our manipulation of persuasion knowl-edge activation when using an implicit brand attitude mea-sure. This result obtained for both pairing procedures.

In sum, our results suggest that simultaneous pairing ofa brand with a different affective stimulus on each occasionyields much more resilient effects on brand evaluations thanconsistently pairing the brand with a single affective stim-ulus. Although our findings provide only preliminary evi-dence, these findings are consistent with our hypothesis thatthe conditioning procedures promote different learning pro-cesses. We hypothesized that pairing brands with positiveaffective concepts can make brands more well liked by es-tablishing direct or indirect affective responses. Indirect af-fective responses to brands depend on memory associationsbetween the brands and the valenced stimuli with whichthey have been paired. In the case of direct affective re-sponses, positive affect becomes directly associated with thebrand, independent of memory associations to the positiveevent that initially generated the affect.

Our findings suggest that the type of learning dependsstrongly on the way marketing stimuli are presented. Indirectaffective responses should be highly sensitive to the consis-tency of the affective stimuli with which a brand is paired.Because indirect affective responses require that specific as-sociations be retrieved (consciously or unconsciously), theywill be weak or absent unless a brand is consistently pairedwith the same affective stimulus on multiple occasions. Directaffective responses, in contrast, do not require consistentpairing because what becomes attached to the brand is theaffect itself, not a link to a specific affective stimulus. How-ever, because establishing direct affective responses relieson an affect misattribution process (Jones et al. 2009), itdoes require that the brand and affective stimulus are pre-sented simultaneously—which in turn is not required forestablishing indirect affective responses. Failing to considerthese necessary conditions limits the effectiveness of eval-uative conditioning. For example, advertising consisting ofthe sequential presentation of a brand with many differentaffective concepts will result in little to no evaluative con-ditioning because the sequential presentation prevents theestablishment of a direct affective response, and the diversebrand-concept pairings do not support the establishment ofindirect affective responses (experiment 1).

Branding Implications

Brand leveraging is used to refer to the effects of linkinga brand to another person, place, thing, or brand. As pointedout by Keller (2003, 595), “marketers often attempt to in-crease their brand equity by, in effect, borrowing equity fromothers.” From a brand-leveraging perspective, our findingspoint to a distinction between borrowing equity from othersand actually acquiring equity directly. Indirect affective re-sponses correspond to borrowed equity, as conceptualizedby Keller. They are obtained, for example, in advertising,by pairing a visual of a brand or product with the same orsimilar visual of a positive affective stimulus across multipleoccasions, employing either sequential or simultaneous pair-


ing. Direct affective responses imply that the brand can ac-tually acquire the equity. They are obtained by pairing a visualof a brand or product with visuals of a different positiveaffective stimulus on each occasion, employing simultaneouspresentation of brand and positive stimulus visuals. Our stud-ies demonstrated in multiple ways why acquiring equity ismuch more robust and thus, from the marketer’s perspective,more desirable than borrowing it.

Companies spend hundreds of millions of dollars annuallyon celebrity endorsers, sponsorships, product placements,and so on. This strategy to promote positive affective re-sponses to the brand can increase the value of the brand.The danger is when these once positive concepts turn neg-ative. For example, celebrity endorsers who have fallen fromgrace include Kobe Bryant, Floyd Landis, Michael Jackson,Michael Phelps, O. J. Simpson, Britney Spears, MichaelVick, and Tiger Woods. If a brand derives a large part ofits value by establishing an association with a celebrity en-dorser (e.g., Accenture and Tiger Woods), devaluation ofthe celebrity can reduce the value of the brand. An obviousstrategy to minimize this risk is using many endorsers, thusreducing the negative impact of any one celebrity’s down-fall. In addition to this basic risk-spreading benefit, our find-ings evoke the possibility of another benefit of using manyendorsers. If many endorsers are used who are each pairedwith the brand infrequently, and if brand and endorser arepresented simultaneously, direct evaluative conditioning mayobtain. That is, the brand directly acquires the equity and isinsulated from negative revaluations of the affective stimuli(i.e., the endorsers). Of course, several caveats are in orderhere. We did not use celebrity endorsers in our experiments,and using many endorsers may be expensive and impractical,although examples of brands using many endorsers do exist(e.g., Nike, l’Oreal).

A second important difference between borrowed equity(indirect affective responses) and acquired equity (direct af-fective responses) is borrowed equity’s dependence on thepreservation of a memory link between the brand and itsendorser. When brands acquire equity, for example, by pair-ing them with many different positive visual stimuli usingsimultaneous presentation of brand and positive stimulus,links to the source become unimportant. Memory associa-tions are most vulnerable immediately after they have beenlearned (Wixted 2004, 2005). This implies that advertisingstrategies leading to borrowed equity may be highly sus-ceptible to interference by advertising clutter. Our secondexperiment suggests that commercials that lead to acquiredequity may yield brand equity that is uniquely robust tocompetitors’ advertising clutter.

A final important difference between borrowed versusacquired equity applies to consumers’ ability to use theirpersuasion knowledge to withstand brand-leveraging tech-niques. Overall, our findings may justify a warning from aconsumer protection point of view. When the brand lever-aging process results in borrowed equity, consumers are atleast partially (but not completely) able to correct for ad-vertising effects on their brand attitudes, but only when they

are explicitly asked about their product opinions. Consum-ers’ implicit product attitudes, however, showed no evidenceof correction at all. This is noteworthy because implicitproduct attitudes are predictive of product choice when con-sumers are mentally preoccupied (Gibson 2008). When thebrand-leveraging process results in acquired equity, the po-tential implications from a consumer protection point ofview are even more worrisome. We found no significanteffect of persuasion knowledge activation on either implicitor explicit brand attitudes when the conditioning procedureattached affect directly to the brand.

Limitations

There are some anomalies in the data that deserve dis-cussion. In experiment 1 (see fig. 2), brands paired withneutral and positive stimuli were evaluated lower in thesimultaneous different US condition than in the same USconditions. The lower evaluation of brands in the simulta-neous different US condition is at odds with the claimedresilience advantage for this conditioning procedure. Twopieces of data address this concern. First, the simultaneousdifferent condition does not consistently produce lower rat-ings. In experiment 2, brands paired with neutral US wereevaluated somewhat more positively in the simultaneousdifferent condition than in the sequential same condition.Second, an unreported experiment (made redundant by ex-periment 1) showed no substantial differences between thepositively conditioned brand attitudes in the simultaneousdifferent and sequential same or simultaneous same con-ditions. Nevertheless, it is possible that some participantsmay have been frustrated by the more complex presentationsetup in the simultaneous different US condition. This sug-gests that simultaneous different conditioning proceduresmay yield more resilient but smaller or slower-to-build brandattitude effects than the same US procedures.

Another potential concern is that ratings close to the mid-point of our scale may not reflect neutral attitudes, but moreof a “don’t know” response. To some extent, this is con-sistent with the hypothesized process. For example, re-sponses near the middle of the scale for both positive andneutral US types in interference conditions using the se-quential same US procedure (experiment 2) should reflecta “don’t know” response. However, “don’t know” responsescannot be responsible for all responses close to the middleof the scale. A case in point might be the positive andpositive-to-be-revaluated stimuli in the simultaneous differ-ent US condition of experiment 1. If responses in these con-ditions represented “don’t know,” then the lower ratings inthe neutral condition could not have been a “more extremedon’t know.” Thus, the whole pattern of data seems moreconsistent with an interpretation in terms of somewhat de-pressed brand evaluations for the simultaneous different UScondition regardless of type of US (neutral, positive, positive-to-be-revaluated). Nevertheless, further research involving,for example, negative instead of neutral US or more stronglypositive US would provide insight into this issue.


Summary

Evaluative conditioning is a crucial determinant of con-sumers’ likes and dislikes (Gawronski and Bodenhausen2006). Despite its importance, research only recently startedto unravel the different processes involved (Jones et al.2009). In their review of the available evidence, De Houweret al. (2001) concluded that attitude formation through eval-uative conditioning is indirect in nature, thus always me-diated by establishing brand-affective stimulus associationsin memory. Gawronski and Bodenhausen (2006) argued theevaluative conditioning effect on brand attitudes can occurthrough establishing new positive associations with a brand,through alteration of the set of momentarily active associ-ations (in the case of brands with rich networks of preex-isting associations, see also Gibson 2008), and through prop-ositional reasoning about the positive stimuli linked to abrand. These processes have in common that the evaluativeresponse to the brand depends on intermediating brand-af-fective concept associations and thus were grouped in thisarticle as indirect affective responses. We hope to contributeto the nascent literature on the processes underlying eval-uative conditioning by providing initial evidence for theempirical ability of different conditioning methods to forgedirect as well as indirect connections between brands andaffective responses.

REFERENCESAllen, Chris T. and Chris A. Janiszewski (1989), “Assessing the

Role of Contingency Awareness in Attitudinal Conditioningwith Implications for Advertising Research,” Journal of Mar-keting Research, 26 (February), 30–43.

Baeyens, Frank, Paul Eelen, Omer Vandenbergh, and Geert Crom-bez (1992), “The Content of Learning in Human EvaluativeConditioning: Acquired Valence Is Sensitive to US-Revalu-ation,” Learning and Motivation, 23 (May), 200–224.

Bierley, Calvin, Frances K. McSweeney, and Renee Vannieuwkerk(1985), “Classical Conditioning of Preferences for Stimuli,”Journal of Consumer Research, 12 (December), 316–23.

Bower, Gordon H. (1981), “Mood and Memory,” American Psy-chologist, 36 (2), 129–48.

——— (2000), “A Brief History of Memory Research,” in TheOxford Handbook of Memory, ed. Endel Tulving and FergusI. M. Craik, New York: Oxford University Press, 3–32.

Brewer, William P. (1974), “There Is No Convincing Evidence forOperant or Classical Conditioning in Adult Humans,” in Cog-nition and the Symbolic Processes, ed. Walter B. Weimer andDavid S. Palermo, Hillsdale, NJ: Erlbaum, 1–42.

Campbell, Margaret C. and Amna Kirmani (2000), “Consumers’Use of Persuasion Knowledge: The Effects of Accessibilityand Cognitive Capacity on Perceptions of an Influence Agent,”Journal of Consumer Research, 27 (June), 69–83.

De Houwer, Jan (2008), “Conditioning as a Source of Liking: ThereIs Nothing Simple about It,” in The Social Psychology ofConsumer Behavior, ed. Michaela Wanke, New York: Psy-chology Press.

——— (forthcoming), “Evaluative Conditioning: A Review ofProcedure Knowledge and Mental Process Theories,” in As-sociative Learning and Conditioning Theory: Human and

Non-human Applications, ed. T. R. Schachtman and S. Reilly,Oxford: Oxford University Press.

De Houwer, Jan, Frank Baeyens, and Andy P. Field (2005), “As-sociative Learning of Likes and Dislikes: Some Current Con-troversies and Possible Ways Forward,” Cognition and Emo-tion, 19 (February), 161–74.

De Houwer, Jan, Sarah Thomas, and Frank Baeyens (2001), “As-sociative Learning of Likes and Dislikes: A Review of 25Years of Research on Human Evaluative Conditioning,” Psy-chological Bulletin, 127 (November), 853–69.

Fazio, Russell H. (1990), “A Practical Guide to the Use of Re-sponse Latency in Social Psychological Research,” in Reviewof Personality and Social Psychology: Research Methods inPersonality and Social Psychology, Vol. 11, ed. Clyde Hen-drick and Margaret Clark, Thousand Oaks, CA: Sage, 74–97.

Fazio, Russell H., David M. Sanbonmatsu, Martha C. Powell, andFrank R. Kardes (1986), “On the Automatic Activation ofAttitudes,” Journal of Personality and Social Psychology, 50(February), 229–38.

Ferguson, Melissa J. (2007), “The Automaticity of Evaluation,” inSocial Psychology and the Unconscious: The Automaticityof Higher Mental Processes, ed. John A. Bargh, New York:Psychology Press, 219–64.

Friestad, Marian and Peter Wright (1994), “The Persuasion Knowl-edge Model: How People Cope with Persuasion Attempts,”Journal of Consumer Research, 21 (June), 1–31.

Gawronski, Bertram and Galen V. Bodenhausen (2006), “Asso-ciative and Propositional Processes in Evaluation: An Inte-grative Review of Implicit and Explicit Attitude Change,”Psychological Bulletin, 132 (September), 692–731.

Gawronski, Bertram, Galen V. Bodenhausen, and Andrew P. Becker(2007), “I Like It, Because I Like Myself: Associative Self-Anchoring and Post-decisional Change of Implicit Evalua-tions,” Journal of Experimental Social Psychology, 43 (March),221–32.

Gibson, Bryan (2008), “Can Evaluative Conditioning Change At-titudes toward Mature Brands? New Evidence from the Im-plicit Association Test,” Journal of Consumer Research, 35(June), 178–88.

Gorn, Gerald J. (1982), “The Effects of Music in Advertising onChoice Behavior: A Classical Conditioning Approach,” Jour-nal of Marketing, 46 (Winter), 94–101.

Hammerl, Marianne and Hans-Joachim Grabitz (2000), “Affective-Evaluative Learning in Humans: A Form of AssociativeLearning or Only an Artifact?” Learning and Motivation, 31(November), 345–63.

Janiszewski, Chris and Luk Warlop (1993), “The Influence of Clas-sical Conditioning Procedures on Subsequent Attention to theConditioned Brand,” Journal of Consumer Research, 20 (Sep-tember), 171–89.

Jones, Christopher R., Russell H. Fazio, and Michael A. Olson(2009), “Implicit Misattribution as a Mechanism UnderlyingEvaluative Conditioning,” Journal of Personality and SocialPsychology, 96 (May), 933–48.

Keller, Kevin L. (1993), “Conceptualizing, Measuring, and Man-aging Customer-Based Brand Equity,” Journal of Marketing,57 (January), 1–22.

——— (2003), “Brand Synthesis: The Multidimensionality of BrandKnowledge,” Journal of Consumer Research, 29 (March),595–600.

Kim, John, Chris T. Allen, and Frank R. Kardes (1996), “An In-vestigation of the Mediational Mechanisms underlying Atti-


tudinal Conditioning,” Journal of Marketing Research, 33(August), 318–28.

Klinger, Mark R., Philip C. Burton, and G. Shane Pitts (2000),“Mechanisms of Unconscious Priming: I. Response Compe-tition, Not Spreading Activation,” Journal of ExperimentalPsychology: Learning, Memory, and Cognition, 26 (March),441–55.

Lang, Peter J., Margaret M. Bradley, and Bruce N. Cuthbert (2005),International Affective Picture System (IAPS): Affective Rat-ings of Pictures and Instruction Manual, Technical Report A-6, Gainesville, FL: University of Florida.

Lascelles, Kristy R. R. and Graham C. L. Davey (2006), “SuccessfulDifferential Evaluative Conditioning Using Simultaneous andTrace Conditioning Procedures in the Picture-Picture Paradigm,”Quarterly Journal of Experimental Psychology, 59 (March),482–92.

Levey, Archie B. and Irene Martin (1975), “Classical Conditioningof Human ‘Evaluative’ Responses,” Behaviour Research andTherapy, 13 (October), 221–26.

Murphy, Sheila T., Jennifer L. Monahan, and Robert B. Zajonc(1995), “Additivity of Nonconscious Affect: Combined Ef-fects of Priming and Exposure,” Journal of Personality andSocial Psychology, 69 (October), 589–602.

Murphy, Sheila T. and Robert B. Zajonc (1993), “Affect, Cognition,and Awareness: Affective Priming with Optimal and Sub-optimal Stimulus Exposures,” Journal of Personality and So-cial Psychology, 64 (May), 723–39.

Pleyers, Gordy, Olivier Corneille, Olivier Luminet, and VincentYzerbyt (2007), “Aware and (Dis)Liking: Item-Based AnalysesReveal That Valence Acquisition via Evaluative ConditioningEmerges Only When There Is Contingency Awareness,” Jour-nal of Experimental Psychology: Learning, Memory, and Cog-nition, 33 (January), 130–44.

Razran, Gregory (1954), “The Conditioned Evocation of Attitudes(Cognitive Conditioning?),” Journal of Experimental Psy-chology, 48 (October), 278–82.

Rescorla, Robert A. (1974), “Effect of Inflation of the Uncondi-tioned Stimulus Value Following Conditioning,” Journal ofComparative and Physiological Psychology, 86 (January),101–6.

——— (1988), “Pavlovian Conditioning: It’s Not What You ThinkIt Is,” American Psychologist, 43 (March), 151–60.

Ruys, Kirsten I. and Diederik A. Stapel (2008), “The Secret Lifeof Emotions,” Psychological Science, 19 (April), 385–91.

Shanks, David R. and Mark F. St. John (1994), “Characteristics

of Dissociable Human Learning Systems,” Behavioral andBrain Sciences, 17 (September), 367–95.

Shimp, Terence A., Elnora W. Stuart, and Randall W. Engle (1991),“A Program of Classical-Conditioning Experiments TestingVariations in the Conditioned-Stimulus and Context,” Journalof Consumer Research, 18 (June), 1–12.

Staats, Carolyn K. and Arthur W. Staats (1957), “Meaning Estab-lished by Classical Conditioning,” Journal of ExperimentalPsychology, 54 (July), 74–80.

Stahl, Christoph and Christian Unkelbach (2009), “Evaluative Learn-ing with Single versus Multiple Unconditioned Stimuli: TheRole of Contingency Awareness,” Journal of Experimental Psy-chology: Animal Behavior Processes, 35 (2), 286–91.

Stapel, Diederik A., Willem Koomen, and Kirsten I. Ruys (2002),“The Effects of Diffuse and Distinct Affect,” Journal of Per-sonality and Social Psychology, 83 (July), 60–74.

Stuart, Elnora W., Terence A. Shimp, and Randall W. Engle (1987),“Classical Conditioning of Consumer Attitudes: Four Exper-iments in an Advertising Context,” Journal of Consumer Re-search, 14 (December), 334–49.

Thorndike, Edward T. (1911), Animal Intelligence: ExperimentalStudies, New York: Macmillan.

Walther, Eva, Bertram Gawronski, Hartmut Blank, and Tina Langer(2009), “Changing Likes and Dislikes through the Back Door:The US-Revaluation Effect,” Cognition and Emotion, 23 (Au-gust), 889–917.

Walther, Eva, Benjamin Nagengast, and Claudia Trasselli (2005),“Evaluative Conditioning in Social Psychology: Facts and Spec-ulations,” Cognition and Emotion, 19 (February), 175–96.

Watson, John B. (1913), “Psychology as the Behaviourist ViewsIt,” Psychological Review, 20 (2), 158–77.

Wickens, Delos D. (1959), “Conditioning to Complex Stimuli,”American Psychologist, 14 (April), 180–88.

Wittenbrink, Bernd (2007), “Measuring Attitudes through Prim-ing,” in Implicit Measures of Attitudes, ed. Bernd Wittenbrinkand Norbert Schwarz, New York: Guilford, 17–58.

Wixted, John T. (2004), “The Psychology and Neuroscience ofForgetting,” Annual Review of Psychology, 55, 235–69.

——— (2005), “A Theory about Why We Forget What We OnceKnew,” Current Directions in Psychological Science, 14 (Feb-ruary), 6–9.

Zellner, Debra A., Paul Rozin, Michael Aron, and Carol Kulish(1983), “Conditioned Enhancement of Humans’ Liking forFlavor by Pairing with Sweetness,” Learning and Motivation,14 (August), 338–50.

Documents

Evaluative Conditioning Procedures and the …warrington.ufl.edu/departments/mkt/docs/janiszewski/Evaluative...Evaluative Conditioning Procedures and the ... evaluative conditioning