The Implicit–Explicit Construct Distinction
Many constructs of interest in cognitive and social psychology (and by extension in consumer psychology) are presumed to involve stable mental representations (e.g., relatively consistent, valenced evaluations in the case of attitudes) that are stored in memory and activated by contextual cues, leading to immediate changes in behavior.1 For example, if consumers have positive attitudes toward global brands, being exposed to globality information associated with a well-known brand like McDonald's may improve attitudes toward the brand and purchase likelihood (Dimofte, Johansson, & Ronkainen, 2008). However, an alternative, constructionist view of attitudes argues that consumers may in fact create attitudes toward global brands “on the spot” in a particular context (in other words, consumers may infer their attitudes from observing the ambient stimuli in a salient context and recalling their past behavior in such contexts; see Wilson & Hodges, 1992).
To accommodate these two accounts, Wilson, Lindsey, and Schooler (2000) proposed that when an individual is exposed to an object, the person's initial attitude toward it is automatically retrieved, although salient aspects of the context are also brought to bear in producing a response. Whether the initial attitude or the novel information is given more weight can vary, as described by several classic models of attitude formation and change, such as Chaiken's heuristic-systematic model (Chaiken, 1980) and Petty and Cacioppo's Elaboration Likelihood Model (Petty & Cacioppo, 1986). At the end of the process, a novel attitude is created and the old one is generally overwritten. However, when both the initial attitude and a newly formed one toward the same object are stored in memory, a dual attitude can result (Wilson, Lindsey, & Schooler, 2000; also see Cohen & Reed, 2006; Petty, 2006). The classic case of vice behaviors illustrates a situation in which the implicit–explicit discrepancy comes into play. In this example, explicit attitudes may involve an individual's conscious acknowledgment that engaging in a vice behavior is bad, whereas implicit attitudes point to a more positive underlying valence (see Fitzsimons, Nunes, & Williams, 2007). For a McDonald's customer, recently acquired knowledge of the negative aspects of fast food consumption (e.g., increased obesity levels or risk of heart disease) may lead to a downward adjustment of explicit attitudes toward the brand and perhaps (but not necessarily, as we will see below) reduced patronage of the chain. However, it is likely that the individual will continue to show a positive predisposition toward the brand and perhaps exhibit a smile when passing by the restaurant and absorbing the enticing smell of the “golden” french fries. In short, whereas a consumer's explicit, conscious attitudes toward a brand may become more negative, implicit or nonconscious attitudes may yet retain their highly positive automatic brand associations.
Petty, Briñol, and DeMarree's (2007) Meta-Cognitive Model proposes that attitudes consist of stored evaluative associations (positive and/or negative) along with accompanying true/false validity tags. Unlike the dual attitudes approach of Wilson, Lindsey, and Schooler (2000), this model argues for one integrated attitude representation and accommodates the potential discrepancy between implicit and explicit attitudes via the conscious consideration of the validity tag in the latter case (Petty, Briñol, & DeMarree, 2007). In our example, McDonald's may well elicit positive automatic thoughts, but they are largely tempered by a negative cognitive tag that our consumer retrieves when creating an explicit attitude toward the brand.
Whether speaking about attitudes, goals, or even self-esteem, it is possible that a certain level of dissociation exists between constructs at the conscious, effortful processing level and their nonconscious, implicit variants. Wilson, Lindsey, and Schooler (2000) reviewed such disparities between implicit and explicit constructs as varied as memory, attachment, dependency, and explanatory style. Importantly, these differences have direct relevance for the specific type of behavior that follows, and therefore an accurate understanding of the implicit–explicit construct distinction is conceptually critical. Once this dichotomy is acknowledged, the next step will necessarily involve a similar dichotomy in terms of the appropriate measurement instruments.
The Implicit–Explicit Measure Distinction
Explicit measures rely on individuals' self-reported assessments of the specific attributes or their intentions regarding potential behaviors and choices they face. Responses are often registered on Likert scales, by means of which individuals select numerical values to express the degree to which they possess an attribute or plan to engage in a particular behavior. This approach naturally assumes that individuals have conscious access to the relevant constructs in memory and that responses are not determined on the spot, as the constructionist model of attitudes suggests (cf. Wilson & Hodges, 1992). If either of these assumptions is not satisfied, the validity of the respective item or scale suffers significantly.
Other problems plaguing explicit measures have been widely acknowledged. For example, they may induce poor comprehension (due to complex or unclear wording), social desirability (due to perceived pressure to provide socially acceptable answers), acquiescence (due to a misplaced propensity to indiscriminately agree to items regardless of content), or extremity of response (for a more comprehensive review, see Oskamp & Schultz, 2005). On the other hand, implicit measures are arguably free of such methodological shortcomings and hold the advantage that individuals may not realize what is being measured or be able to consciously correct their answers within the allotted time constraints.
According to De Houwer and Moors (2010), a measure's implicit character is determined by whether the processes involved in measuring the attribute are automatic. For example, automatic processing occurs in the absence of particular processing goals on the part of the individual or operates even when the person is unaware of the object prompting the process. Different implicit measures can thus be implicit (i.e., automatic) in different ways, and one should specify the automaticity feature that characterizes the respective measure (De Houwer & Moors, 2010).
Implicit measures of attitudes are often structured to assess whether information processing is facilitated (i.e., shorter latencies) or hindered (i.e., longer latencies) by the presentation of an attitude object (Gawronski & Bodenhausen, 2007). Facilitation or impairment reflects the (lack of) compatibility between the process engaged by the activation of the attitude and some other processing demand. De Houwer (2003) dichotomized this difference into two processes, response compatibility (driven by the match between the tendencies associated with two different tasks, as in the Stroop paradigm) and stimulus compatibility (driven by semantic similarity, as in lexical decision tasks). Both of these processes make responses to implicit measures, unlike those to explicit measures, difficult to control.
Despite the general enthusiasm associated with the emergence and use of these novel methodological tools, several researchers have argued that more rigorous study is needed to better understand the value of implicit measures. For example, the fact that a particular construct is assessed via an implicit measure does not necessarily imply that the construct is an implicit or nonconscious one, but instead may simply suggest that motivational influences that occur downstream from attitude elicitation play a key role (as suggested by the MODE dual process model of Fazio & Towles-Schwen, 1999). At the same time, the finding that different implicit measures of the same construct often do not correlate very highly is not encouraging and begs for more inquiry into this problem (Fazio & Olson, 2003; Payne, Burkley, & Stokes, 2008).
In general, implicit and explicit constructs in a consumption context are well aligned and correlate highly. However, this is not always the case. In fact, it is in the very instance when this alignment is lacking that research findings have shown extremely interesting results. Here we turn to consumer research involving the most popular of the measures of implicit attitudes, the Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998).
IMPLICIT MEASURES IN CONSUMER RESEARCH
The psychology literature has been prolific of late in introducing a variety of implicit measures of cognition (see De Houwer & Moors, 2010, for a list of as many as 17 examples). The most commonly employed (and debated) measure has been the IAT (Greenwald, McGhee, & Schwartz, 1998); as its popularity has expanded, specific applications have emerged in the consumer research literature. Other such measures have also found their place in consumer applications, as discussed next.
The IAT assesses automatic associations between a bipolar target (e.g., competing brands such as Nike vs. Reebok in a marketing context) and a bipolar attribute concept (e.g., fast vs. slow) through a series of categorization tasks that require quick responses (see Schnabel, Asendorpf, & Greenwald, 2008). Shorter response latencies are expected to emerge when strongly associated concept pairings are elicited (e.g., Nike and fast, based on perception that Nike shoes provide better athletic performance) and share a common response key as compared to when they do not. In a typical IAT, the first block instructs participants to press the “D” key when a Nike logo appears on the screen and the “K” key when a Reebok logo pops up. In the following block, participants are trained to press “D” for fast words (e.g., speedy, quick) and “K” for slow words (e.g., sluggish, lethargic). The next, critical block combines the two discrimination tasks, such that participants are instructed to press “D” for Nike or fast and “K” for Reebok or slow. Naturally, there are also single and combined discrimination blocks that reverse the key assignment (i.e., such that Reebok logos are responded to with a press of the “D” key and that Nike and slow share a response key). The order of the combined blocks is usually counterbalanced across participants in order to control for the fact that IAT scores show stronger associations for categories that are paired (and learned) first. Recorded latencies in the combined tasks are then used for calculation of IAT scores, which are generally computed as the difference between mean response latencies to the second combined task and to the first combined task (Greenwald, McGhee, & Schwartz, 1998). If response management is attempted (say by a Reebok employee who is a closet fan of Nike or vice versa), response latencies and error rates increase notably. Detailing the specific scoring algorithms that can be used to measure the IAT effect is beyond the scope of this review, but analysis of reliability and validity suggests that the measure has good psychometric properties (Greenwald et al., 2009a; Maison, Greenwald, & Bruin, 2004).
According to Schnabel, Aspendorpf, and Greenwald (2008), much of the strength of the IAT comes from the fact that many social objects seem to have natural counterparts (e.g., males vs. females, whites vs. blacks, or even McDonald's vs. Burger King and Microsoft vs. Apple). Yet that is perhaps one of its weaknesses as well, since as a relative measure the IAT effect always involves a dual explanation. Thus, it is not necessarily clear whether the IAT effect described in the example above stems from an automatic association of Nike and fast or, alternatively, one between Reebok and slow.2 Although this may not be a major problem in a marketing context (i.e., the relative implicit preference for the two brands is clearly established), in cases where associations with unipolar concepts are of interest, other implicit measures may be more appropriate. Along these lines, the Single Category IAT (SC-IAT; Karpinski & Steinman, 2006; Steinman & Karpinski, 2008) employs a single unipolar element (e.g., Exxon) and one bipolar concept (e.g., good vs. bad) but otherwise features a procedure similar to the IAT.
The Go/No-Go Association Task (GNAT; Nosek & Banaji, 2001) is another example of an implicit measure. Like the IAT, it works by presenting various stimuli for brief periods of time that require prompt responses. However, unlike the IAT, the GNAT requires the same response (i.e., go—press the space bar) to items that belong to a category (e.g., insects) and an evaluative attribute (e.g., good). No response (i.e., no-go—do not press any key) is expected when items do not belong to the target category and attribute (i.e., are distractors; see Nosek & Banaji, 2001). In a consumer context, Bassett and Dabbs (2005) employed the GNAT to show that smokers had less negative implicit attitudes toward smoking than nonsmokers, although for both groups the GNAT correlated positively with explicit self-reports.
Finally, the Breadth-based Adjective Rating Task (BART; Karpinski et al., 2007) is an indirect paper-and-pencil measure of consumer attitudes that is based on the premise that individuals tend to describe expectancy-consistent information with more abstract and generic traits, whereas expectancy-inconsistent information is captured via more concrete and specific traits. This abstraction bias is captured in the BART by having participants rate how well trait adjectives of varying breadth and valence describe an attitude object (Karpinski et al., 2007). Initially developed in the context of information describing the self and interpreted as an indirect expression of a person's level of self-esteem, the measure has found use in consumer contexts as well (e.g., Steinman & Karpinski, 2009).
There are two general situations that warrant researcher recourse to implicit measures in order to reliably and validly capture consumer processing of, and response to, marketing stimuli. The first is the case of self-presentation biases that often plague marketing research data. To the extent that survey or experimental response items create consumer discomfort or entail the risk of the respondent coming across as less sophisticated, open-minded, or knowledgeable than is socially acceptable or expected (Kihlstrom, 2004), conscious adjustments may be undertaken that alter or even hide objectively true responses.
The second instance that may produce biased feedback is one of consumers lacking conscious access to their own cognitive processes or information stored in memory. Explicit measures may simply be inadequate to capture these types of data. In these situations, a theoretically interesting dissociation of explicit and implicit responses may occur, and the immediate question of whether explicit or implicit measures of cognition are more predictive of actual behavior becomes directly relevant. The self-presentation bias and the lack of awareness cases are each explored next.
Consumer Conscious Adjustment of Explicit Responses
In a consumption environment that features ever-changing social trends and norms, deviant attitudes and behaviors are often not readily admitted. Consider the case of consumers queried about their recycling behaviors and attitudes toward recycling. It is likely that because of the enhanced pressure to think and act in an environmentally conscious manner in today's society, many respondents may be reluctant to express unfavorable attitudes toward recycling or admit that they routinely fail to recycle. Accordingly, they may engage in response management strategies to conceal their true attitudes and instead provide socially desirable answers (see Meneses, 2010), which can lead to invalid inferences regarding their attitudes and behavior. A recent illustration of this effect was provided during the 2008 presidential electoral season. The American voting public was polled by a variety of national media outlets, but a significant number of individuals also took IAT tests that measured their implicit preferences for the two main candidates on an Internet Web site sponsored by an academic institution. Notably, whereas the electoral polls varied widely in their predictions and many were unable to clearly predict a winner given their margins of error, the IAT proved highly reliable in predicting the winning candidate (Greenwald et al., 2009b). It appears that revealing preferences to a pollster (or a marketer, for that matter) is less honesty-inducing than responding to an implicit measure.
In the same vein, Brunel, Tietje, and Greenwald (2004) assessed consumers' behavioral and attitudinal responses to spokesperson race in print advertising. Social psychology research had uncovered relatively low correlations between explicit and implicit measures of racial attitudes, which is in line with the expectation that respondents consciously adjust their answers due to a self-presentation bias (Greenwald, McGhee, & Schwartz, 1998). Brunel, Tietje, and Greenwald (2004) exposed subjects to advertisements featuring celebrity spokespeople of either Caucasian or African American ethnicity. An interesting interaction occurred between viewer ethnicity and type of measurement of respondents' attitudes toward the advertisement. Explicit self-report measures administered to Caucasian consumers were unable to detect any preference for a same-ethnicity spokesperson advertisement, whereas the IAT identified a significant pro-Caucasian preference among the same respondents. Alternatively, African American consumers self-reported an explicit preference for ads featuring same-ethnicity endorsers, but this effect was absent in the IAT. These results suggest that response management strategies may have been employed by both ethnic groups, driven by the perceived pressure to provide socially desirable or group-consistent responses.
Individuals are generally adept at engaging in a variety of cognitive defensive mechanisms in order to detect and protect the self from threatening incoming information. For example, in the domain of romantic relationships, researchers find evidence for a specific risk regulation system designed to deal with risky relationship situations (e.g., Murray, Holmes, & Collins, 2006). In this context of romantic relationships, Dimofte and Yalch (2010a) exposed subjects to information according to which a recent survey in the study's geographic location had uncovered a 3:1 ratio of females to males in the population. The immediate implication that it is much easier to date someone as a male was immediately salient to all respondents, including females for whom this novel information (factually true and therefore highly credible) was directly threatening. However, in line with classic findings on self-enhancement, these female respondents claimed that future dating in the respective location would not be a more difficult task when asked about it on explicit measures. This conscious, self-protective adjustment in self-reports was not possible on the IAT, which revealed stronger automatic associations between dating and difficult (than between single and difficult) for the threatened female respondents, including those in committed romantic relationships at the time of the study (Dimofte & Yalch, 2010a).
Finally, recent research by Dimofte, Brumbaugh, and Goodstein (2010) on the topic of consumer response to target marketing is germane to the point that individuals may choose to conceal their true responses in explicit self-reports. Over time, many product categories develop associations with particular user prototypes in the consumer's mind regarding who it is that tends to buy the particular products (e.g., environmentally conscious, progressive urban dwellers are hybrid car buyers). Yet some of these prototypes may not always fit with those brand managers had considered when developing their targeting strategies. In these cases, it could be argued that the marketplace has in effect created an undesirable product association (i.e., because the prototype is overly narrow or perhaps even completely off the mark relative to the firm's initial positioning). When firms create target advertising for such products, the user prototype knowledge is automatically activated and the response to advertising is often driven by the way the consumer compares to the prototype (and not to the ad-suggested target customer). In this case, if the user prototype is a member of an out-group relative to the consumer's in-group, a social comparison process is engaged. If this group comparison is unfavorable to the consumer (say for a Caucasian male exposed to advertising for basketball shoes associated with an athletic African American product user prototype), a decline in collective self-esteem may ensue, leading to unfavorable advertisement and product attitudes (Dimofte, Brumbaugh, & Goodstein, 2010). These declines are not readily observed in explicit self-reports, as subjects may guess the reason why ethnic collective self-esteem questions are being explicitly asked after exposure to the ad and may choose to artificially inflate their estimates. However, these self-esteem–enhancing adjustments are not as likely to occur with implicit measures, as these authors observed. In fact, results suggested that individuals' implicit self-esteem (captured via their IAT effect size) fully mediates the response of consumers to target advertising that elicits threatening user prototype knowledge (Dimofte, Brumbaugh, & Goodstein, 2010).
Consumer Lack of Awareness of Implicit Responses
It has been argued that the nonconscious nature of some consumer cognition resides in individuals' lack of awareness for a variety of processing-related elements. For example, Chartrand (2005) suggested that consumers may be unaware of the external cues that prompt the engagement of an automatic cognitive process, of this process itself, or of its outcomes. In other words, a TV viewer exposed to a commercial for Jackson Hewitt's tax preparation service may be unaware that the ad's slogan (“Get more in return”) has triggered an automatic processing of its multiple meanings, that the secondary meaning has been accessed and comprehended, or that an ad hoc perception has emerged according to which this tax service is simultaneously perceived as more affordable and better at getting deductions. In fact, that is precisely what Dimofte and Yalch (2007a) found in their research on polysemous (i.e., multiple-meaning) brand slogans. For a cellular phone provider that employed the slogan “Raising the Bar” to effectively convey two brand-favorable information cues (i.e., superior service relative to competitors and more signal bars when using the company's network), many consumers unexpectedly had more negative attitudinal responses to the brand than did those in a control group exposed to the slogan “Redefining the Best.” Yet the reason for the negative attitudes was not apparent when exploring participants' elicited thoughts, suggesting that self-report measures may be inadequate when it comes to fully capturing the language processing effects involved in consumer response to polysemous slogans. However, an IAT juxtaposing the respective cellular provider with a direct competitor using the evaluative categories affordable and expensive uncovered novel automatic associations between the advertised brand and perceptions of expensiveness. While certainly inadvertent and unintended on the part of the marketer, consumers apparently implicitly accessed a negative secondary meaning of the brand slogan, according to which they perceived that the firm raised the bar in terms of jacking up the prices it charged for its high-end service (Dimofte & Yalch, 2007a).
In a similar fashion, the expression “going down fast in Aspen,” employed to suggest the quality of the mountain resort's ski slopes, was instead implicitly (but not explicitly) construed by study respondents to imply the deteriorating quality of the resort's services over time, an effect certainly opposite to that intended by the advertiser (Dimofte & Yalch, 2007b). Importantly, in both cases consumers failed to mention the negative slogan aspects in self-reports, but demonstrated the implied negative associations via the IAT.
Forehand and Perkins (2005) used self-report and the IAT to assess consumer response to advertising using celebrity voices. They found that liking a celebrity produced a positive response to advertising featuring the celebrity's voice, but only for consumers who were unable to recognize the celebrity. However, consumers who recognized the celebrity, were motivated to eliminate irrelevant influences on their advertising response, and were able to consciously adjust their explicit response did not exhibit the same effect. The authors argued that the explicit measure adjustment involved a correction of the perceived influence of the celebrity (i.e., resetting) because of its actual irrelevance. This resetting implied a conscious evaluation that the IAT did not allow for, leading to the emergence of dissociation between the explicit and implicit results. This work (as well as that of Dimofte & Yalch reviewed above) shows the value of the IAT as a methodological tool for capturing cognitive processes that underlie effects observable on explicit measures of attitudes but not easily explainable from consumer self-reports (see Perkins et al., 2008 for a similar argument).
In evaluating consumer attitudes toward genetically modified foods, Spence and Townsend (2006) employed the GNAT to show that context-free implicit attitudes were in fact relatively positive, although corresponding explicit self-reports were neutral. At the same time, a downshift was observed when GNAT measurement occurred in the context of organic foods, as implicit attitudes toward genetically modified products were found to be neutral but not negative (Spence & Townsend, 2006). These results suggest that consumers have automatic approach tendencies toward these foods despite indifferent explicit attitudes. Since self-report measures did not show a reported preference for non-modified products (as self-presentation bias might have suggested), it appears the positive implicit effects are driven by consumers' lack of awareness of the actual favorable attitudes they show toward these products at nonconscious levels.
The largely positive behavioral response that American consumers have toward global brands was the focus of research by Dimofte, Johansson, and Ronkainen (2008). Unlike respondents in developing nations, who display an explicit preference for these brands due to their aspirational nature, U.S. consumers reported seeing no particular benefit or value associated with global brands (be they American or foreign). In a study employing a nationally representative panel of respondents recruited via the Internet, who presumably had no self-presentation motives, the only explanation for the favorable behavioral effect that global brands engendered (i.e., higher purchase levels than attitude–behavior consistency models would predict) was that U.S. consumers harbor positive implicit attitudes toward global brands. Indeed, an indirect test showed that an individual described along several attributes was liked better if presented as a global (vs. local) beer drinker, whereas an IAT uncovered implicit associations favoring global over local brands (Dimofte, Johansson, & Ronkainen, 2008).
Psychological research introduced a shifting standards model of evaluations (Biernat, Manis, & Nelson, 1991) according to which individuals routinely adjust their subjective (but not objective) judgment standards as they evaluate members of stereotyped social groups. For example, women are stereotypically expected to earn less than men if judged in annualized dollar amounts, but they are at the same time not expected to be less financially successful than men. (In fact, Biernat, Manis, & Nelson, 1991, found that the very same women earning less than men were perceived to be more financially successful than those men; thus, for a woman, such financial performance was quite impressive). Along these lines, Dimofte and Johansson (2009) uncovered the existence of a similar shifting standards effect in marketing with respect to consumer brand evaluations. They found that for inferior brands that engender strong expectations, consumers unconsciously lower their evaluative standards and “cut them slack” when responding to word-based, subjective scales (but not to number-based, objective scales). In other words, a Hyundai engine that puts out 150 hp is objectively unimpressive, but in subjective terms (i.e., on a scale anchored by “not powerful at all” and “extremely powerful”), a horsepower rating at that level sounds pretty good for a Hyundai. The automatic nature of this adjustment was captured via the IAT and further demonstrated by consumers' lack of acknowledgment that they had engaged in the evaluative shift when informed about it on a post hoc basis (Dimofte & Johansson, 2009).
Priluck and Till (2009) examined consumer brand perception with a standard, explicit brand equity scale as well as the IAT in order to spot instances when the two may diverge. Their findings suggest that for clearly distinguishable brands such as those of high versus low equity, both the IAT and the explicit brand equity scale were successful in capturing differences in perceptions. However, when two brands were less distinguishable in explicit terms, the IAT uncovered an implicit consumer preference for the pioneering brand that was not apparent from explicit brand equity measurement (Priluck & Till, 2009).
Early work on rumor processing and acceptance by Tybout, Calder, and Sternthal (1981) found that strategies other than refutation are more useful for quelling unfavorable brand rumors (such as the actual marketplace report at the time according to which McDonald's burger meat contained red worms). This is an important issue because despite explicit disbelief in the story, subsequent brand attitudes and purchase intent measures displayed significant declines. A storage rumor quelling strategy, for example, involved exposing consumers (simultaneously with the rumor) to novel information cues about the negative contaminant (e.g., red worms are used in high-end French cuisine). The strat-egy proved successful, although explaining the precise mechanism through which it operated was left unaddressed. For example, it could have been that the extra information cues interfered with the creation of a brand–contaminant association or that the positive valence of this information made the contaminant less objectionable (Tybout, Calder, & Sternthal, 1981). To disentangle these alternative explanations, Dimofte and Yalch (2010a) employed the IAT to demonstrate that the brand–contaminant automatic association is quick to emerge and cannot be suppressed, but the positive nature of the new information about the contaminant significantly improves implicit attitudes toward it (thus, worms are somehow not that bad after all and therefore the rumor is less damaging to the brand).
Finally, Dimofte and Yalch (2010b) introduced a mere association effect in the context of consumer information processing, which was driven by an inability to suppress automatically activated but irrelevant brand associations. In one of their studies, participants were asked to rate 20 academic institutions in terms of reputation as party or work-intensive schools, respectively. The two focal institutions were USC and UCLA, with the latter perceived to be more of a party school at statistically significant levels. Subsequent exposure to a series of brand logos that included that for Trojan condoms (vs. a control condom brand) was conducive to the emergence of an implicit association between USC and play (captured on the IAT), which in effect reversed prior explicit perceptions. Thus, the mere fact that the Trojan construct is associated with both condoms and the athletic teams of an academic institution produced an automatic transfer of attributes between the two that logically should not have occurred. The effect was also observed on evaluative judgments. In a different study, consumers exposed to the word frog were more likely to choose a wine bottle featuring a frog on its label, but the word warts (as a negative associate of toads and frogs) produced avoidance behavior for the same label instead (Dimofte & Yalch, 2010b). This research is informative regarding the unexpected and potentially damaging effects that may occur when specific primes are paired with brand names, despite the fact that the mere association effect should be consciously suppressed. Importantly, implicit measures are critical in demonstrating and explaining their underlying associative mechanisms.
The Predictive Power of Implicit Measures for Consumption Behavior
Fazio and Olson (2003) reviewed evidence for the predictive validity of implicit associations, in particular studies examining priming, the IAT, and other implicit measures (e.g., the word fragment completion task). They argued that individually these measures seem to predict subsequent behaviors. However, they also cautioned that implicit measures show surprisingly low correlations with each other, largely due to their low reliability and large measurement error (Fazio & Olson, 2003).
On the other hand, supporting the claim that, in general, explicit and implicit attitudes tend to be well aligned, Greenwald et al.'s (2009a) meta-analysis finds that the IAT correlates well with explicit measures, particularly in the area of brand preference and choice (though for a different perspective see Karpinski & Hilton, 2001). In specifically assessing the predictive power of implicit and explicit preferences on brand choice, Friese, Wänke, and Plessner (2006) found that the specific context underlying this choice plays an important role. In their research, participants were exposed to sets of branded or generic products, were then asked to provide preferences for either set by means of both explicit self-reports and via the IAT, and were later asked to choose one set to receive as a gift. Participants given ample time to consider their gift choice picked products (the branded or generic set, respectively) that were predicted exclusively by their explicit answers (regardless of whether their implicit and explicit preferences were congruent or not). However, those who had to decide under time pressure (5 seconds) and who had displayed inconsistent implicit and explicit preferences were less likely to make choices that converged with their explicit answers. While in general the authors found that explicitly measured preferences were reliable predictors of choice, their predictive power was significantly impaired in situations where consumers were pressured to rely more on highly accessible, automatic preferences (Friese, Wänke, & Plessner, 2006). In a similar vein, Friese, Hoffman, and Wänke (2008) demonstrated that in the context of individuals' food consumption choices, implicit attitude measures (i.e., the IAT) were better than explicit measures in predicting impulsive behaviors, whereas explicit attitude measures (i.e., self-reported ratings) were better predictors of controlled behaviors.
Maison, Greenwald, and Bruin (2001) examined the predictive ability of the IAT in specific consumption contexts such as those involving preferences for low- versus high-calorie foods and consumption of these food choices. They found that explicit attitudes toward low- versus high-calorie products correlated with implicit attitudes only for low-calorie foods. Furthermore, these implicit attitudes were related to respondents' self-reported consumption of low- and high-calorie products, respectively. Maison, Greenwald, and Bruin's (2001) results suggest that the IAT could thus be employed as a measure of implicit attitudes that may be predictive of behavior in situations where consumers hold ambivalent attitudes that may blur explicit self-reports. In other words, there are both positive and negative aspects related to low- and high-calorie foods (e.g., taste, weight control, etc.), and whichever is more salient during questioning may bias explicit responses. The use of the IAT may avoid this bias and seems to predict behavior reliably.
Steinman and Karpinski's (2008) research on consumer response to the clothing retailer brand GAP found that while the SC-IAT data did not correlate with explicit attitude measures, they were related to self-reports of GAP patronage and behavioral intentions. Whereas explicit attitude was a strong predictor of behavioral intentions, the authors proposed that “the SC-IAT could add to the specificity of prediction of consumer behavior” (Steinman & Karpinski, 2008). In related work employing the BART measure alongside more traditional explicit self-reports, Steinman and Karpinski (2009) evaluated the predictive value of this indirect instrument on consumer behavioral intent measures toward the Philadelphia Inquirer newspaper and Ben & Jerry's ice cream, respectively. The authors found that the BART was a significant predictor of behavioral intentions and added to the specificity of prediction above and beyond explicit attitude (Steinman & Karpinski, 2009). Similarly, Maison, Greenwald, and Bruin's (2004) limited meta-analysis of several studies employing both the IAT and explicit measures of consumer attitudes confirmed that the use of implicit measures enhances the predictive ability of consumption behavior relative to that of explicit measures alone.
Finally, Chan and Sengupta (2010) further qualify the correspondence between implicit/explicit attitudes and implicit/explicit behaviors. In their work on consumer flattery, individuals who were complimented by marketers in targeted communications reported awareness of the firm's ulterior motive but had difficulty adjusting for it (Chan & Sengupta, 2010). Instead of being replaced by a discounted explicit judgment, an implicit favorable reaction to flattery continued to exist along with it, which is in line with the dual attitude theory reviewed above (e.g., Wilson, Lindsey, & Schooler, 2000). Importantly, the attitude–behavior correspondence account was found to operate for immediate measurement, but a reversal occurred after a delay, such that implicit attitudes (measured similarly to explicit attitudes but with significantly longer time to respond) were in fact more predictive of behavior (store coupon choice from the ingratiating marketer or a competitor).
Research evidence reviewed in this paper highlights recent methodological advances in the area of implicit social cognition and their relevance to consumer psychology. Rooted in the conceptual distinction between implicit and explicit facets describing a variety of psychological constructs, a parallel dichotomy has been proposed relative to the specific measurement instruments to capture these constructs. The implicit–explicit distinction is of particular concern in two instances of consumer response to marketing stimuli. First is the case in which response management strategies are engaged and socially desirable or self-enhancing responses are provided in self-reports. In this case, individuals may be unwilling to provide the researcher with their true appraisals of the measured construct, rendering their explicit feedback invalid for purposes of assessing their underlying sentiments or intentions. Second is the case in which the implicit nature of the construct being measured or other related psychological processes makes these true ratings inaccessible for respondent introspection. When this occurs, the individual's bona fide efforts to provide accurate representations of these constructs or processes simply fall short on grounds of inaccessibility.
The value of implicit measures thus resides in their potentially superior ability to gather accurate construct measurement data despite consumers' reluctance or inability to provide them. In some of the work cited here, implicit measures were differentially qualified to capture cognitive processing effects that would otherwise be unobservable and left open to theoretical interpretation and debate in a manner reminiscent of behaviorism's black box paradigm. In other cases, implicit and explicit measures displayed unexpectedly low correlations, prompting consideration of more comprehensive theoretical frameworks that feature richer conceptual understandings. Finally, the fact that implicit measures were shown to display relatively high levels of predictive validity is an important consideration in a field concerned with understanding and predicting consumer behavior.
It also seems apparent that the IAT has emerged as the most preferred measure of implicit attitudes. The reasons for the IAT's attractiveness to researchers may have to do with its relatively good fit with the consumer research enterprise. The dual-category design is a great match for marketplace scenarios that juxtapose two direct competitors, while its general use in between-subjects designs avoids some methodological and interpretation issues raised in psychological research based on within-subject responses. It is also useful in situations featuring attitudinal ambivalence toward specific brands, as it allows for the emergence of more clearly defined automatic preferences once the burden of cognitive elaboration across a multitude of attributes is lifted.
Much of the research involving the IAT has been squarely focused on measuring implicit attitudes, at the expense of richer contexts, such as those involving assessment of memory or self-esteem, where the measure has significant potential as well. Moreover, future consumer research should expand the use of implicit measures beyond the IAT to perhaps the GNAT (uniquely suited to address single-brand implicit effects), the BART, or other, less common, methodological tools from social psychology (see Schellekens, Verlegh, & Smidts, 2010, on consumer use of language abstraction in word of mouth).
Other areas of great potential involve the application of classic effects from psychological research to the consumer domain in which implicit measures can be used to assess specific underlying mechanisms. For example, Dijksterhuis et al. (2006) have proposed that making optimal choices in complex situations entails nonconscious rather than conscious deliberation (e.g., deliberating internally in the absence of attention and effortful processing). The implications that this effect carries for consumer research are significant, and understanding why these choices produce better results and greater satisfaction provides a research opportunity that is both intriguing and appealing. If specific product attributes are perhaps erroneously overemphasized during explicit choice consideration, their reduced salience and importance in nonconscious deliberation invites researchers to consider other avenues of research, such as measurement of implicit responses.
In the end, the increasing popularity of implicit measures in mainstream consumer psychology and the emergence of findings based on their use that shed new light on a variety of consumer phenomena are encouraging. More work is needed, though, in order to demonstrate their usefulness to a wider constituency and highlight their incremental contributions in the field of marketing and advertising research. As the knowledge base on the topic widens, replication work, meta-analyses, and more comprehensive reviews will contribute to a better assessment of their future place in the field's methodological arsenal.