Implicit Food Preference -1 - Running Heading: Implicit Food Preference

One of the issues concerning the application of implicit measures like the IAT is whether they can be successfully applied to consumer choices. Four studies (N=399) tested the predictive and incremental validity of an IAT towards fruits versus snacks on the spontaneous choice of a fruit or a snack at the end of the experimental session. Specifically, additive and interactive patterns to predict behavioral preference towards snacks or fruits were tested. The results show that the IAT has both predictive and incremental validity and support the additive pattern (i.e., both implicit and explicit measures independently predict the behavioral preference for fruits versus snacks). Implicit Food Preference 3 Since its introduction by Greenwald, McGhee and Schwartz (1998), the Implicit Association Test has been the most frequently used measure for assessing implicit attitudes. The IAT is a computerized task in which participants are requested to classify stimuli into two sets of two contrasted categories (e.g., two target categories: fruits and snacks; two attribute categories: positive and negative) with two different keys. The IAT measure is computed by comparing the relative response times among the two versions of the combined task in which stimuli from all categories are randomly presented. The assumption underlying the IAT is that if two concepts (e.g., fruits and positive) are highly associated, the categorization task will be easier (and the participants quicker to respond) when the two concepts share the same response key than when they require two different response keys (for a more detailed description, see Greenwald et al., 1998). Theoretically based on an associative network conceptualization of a Social Knowledge Structure in memory (see Greenwald, Banaji, Rudman, Farnham, Nosek, & Mellot, 2002) and on the definition of an attitude as the association between an attitude-object and a valence concept (Fazio, 1995), the IAT is assumed to reflect the relative strength of automatic associations between concepts. The IAT has been used as an implicit measure of attitudes, self-concepts, and stereotypes (for reviews, see Greenwald & Nosek, 2001; Hofmann, Gawronski, Gschwender, Lee, & Schmitt, 2005; Poehlman, Uhlmann, Greenwald, & Banaji, 2006). Numerous studies have shown the reliability of the IAT in various domains, with values generally hovering around .80 for the internal consistency and .60 for test-retest stability (e.g., Dasgupta & Greenwald, 2001; Greenwald & Farnham, 2000; Greenwald & Nosek, 2001; Perugini, 2005a). Predictive validity. One of the key issues concerning the IAT is its ability to predict relevant behaviors. In fact, one could argue that, both on practical and theoretical grounds, an acid test for a measure such as the IAT is whether it shows evidence of predictive validity (Perugini, 2005b; Perugini Implicit Food Preference 4 & O’Gorman, 2006). Poehlman et al.’s (2006) meta-analysis indicated significant relations of the IAT with several types of behaviors (e.g., self-reported, judgments, choices), thus demonstrating its ability to predict different criteria (e.g., nonverbal behaviors, impression formation, shyness, anxiety, consumer choices, voting) and to be used as an individual differences measure (see also Asendorpf, Banse, & Mucke, 2002). However, whereas the predictive validity of the IAT has been one of the main focuses in several papers, its incremental validity has not been studied as thoroughly. In other words, relatively few studies have tested whether there is some unique contribution from the IAT in the prediction of behavior, over and above the contribution provided by explicit measures. Indeed, in their meta-analysis reviewing 61 studies using the IAT, Poehlman et al. (2006) mentioned the possibility that the IAT has incremental validity but they did not test it statistically because of the lack of appropriate parameters in most of the studies reviewed. Some theoretical work has been devoted to understand better how implicit and explicit factors could underlie the execution of behavior. Dual-processes models theorizing the role of implicit and explicit processes (e.g., Fazio, 1990; Strack & Deutsch, 2004; Wilson, Lindsey, & Schooler, 2000) all assume the importance of both processes as basic determinants of behaviors. However, besides several similarities, these models differ in how implicit and explicit factors contribute in predicting behavior. Focusing on attitudes, Perugini (2005a) argued that three main categories of predictive models can be distinguished based on the pattern they refer to in describing the role of implicit and explicit processes: additive, interactive and double dissociation patterns. In the additive pattern, both explicit and implicit attitudes provide a unique prediction of behavior. Implicit and explicit measures are considered as distinctive aspects of the same underlying attitude, therefore assuming a single attitude representation and two different measures (Fazio & Olson, 2003). Thus, low correlations between the measures should not be Implicit Food Preference 5 taken as evidence of the existence of two independent systems but rather of the discriminant validity between two different types of measures (Perugini, 2005a). In predictive terms, this pattern would imply that an IAT measure should provide evidence of incremental validity for most types of behaviors. In the interactive pattern, implicit and explicit attitudes interact synergistically to predict behavior. This pattern is derived from a general theoretical framework of social behavior developed by Strack and Deutsch (2004). They postulate the interaction between a reflective and an impulsive system that activate the same behavioral schemata and usually operate in parallel. In this view, the correlation between implicit and explicit measures is in some sense irrelevant: implicit and explicit measures may therefore complement each other in predicting behavior (Perugini, 2005a, study 1; Brunel, Tietje & Greenwald, 2004; Maison, Greenwald, & Bruin, 2004). Empirically, this should typically be reflected in a significant interaction term between an implicit and explicit attitudinal measure over and above the unique individual contributions of either. Finally, in the double dissociation pattern, implicit attitude predicts solely spontaneous/automatic behaviors and explicit attitudes predict solely deliberative/planned behaviors. This pattern is derived from Wilson, Lindsey, and Schooler’s (2000) model of dual attitudes, postulating the independent co-existence of different evaluations, one implicit and one explicit, of the same attitude object. It has received some empirical support in a number of studies (e.g., Asendorpf, Banse, & Mucke, 2002; Dovidio, Kawakami, & Gardner, 2002; Perugini, 2005a, study 2). Food choice. A domain where the potentiality of a measure like the IAT can be exploited is the one of consumer choice (Brunel, Collins, Greenwald, & Tietje, 1999; Brunel, Tietje, & Greenwald, 2004) and, more specifically, food choice. However, the empirical evidence so far has been Implicit Food Preference 6 mixed, both at the level of eating habits and of specific behavioral choices. For example, Maison, Greenwald and Bruin (2001) used an IAT high vs. low calorie foods to successfully predict eating behavior that, however, was measured via self-report. In contrast, Roefs and Jansen (2002), using an IAT high vs. low fat, found that obese people had significantly more negative implicit attitudes towards high fat food compared to normal weight people, therefore implying a negative relation between IAT and eating behavior. At a more specific level of behavioral choices, a study by Karpinski and Hilton (2001, study 2) showed that an IAT did not predict the choice between a candy bar and an apple whereas an explicit attitudinal measure did predict it. In principle, such a choice should have been based on relatively more spontaneous processing and predicted also by the implicit measure. In support of this, Perugini (2005a, study 2) showed that the IAT significantly predicted a choice between snacks and fruits whereas the explicit attitudinal measure did not. The difference between the results could also be due to some procedural details. In fact, Perugini (2005a) considered attitudes toward the more general categories of snacks and fruits whereas Karpinski and Hilton (2001) considered candy bars and apples. Moreover, in Perugini (2005a)’s study, the behavioral choice was between different types of snacks and fruits, whereas Karpinski and Hilton (2001)’s behavioral choice was between one type of candy bar (Snickers) and a Red Delicious apple. Aim of the contribution This contribution is focused on the capability of the IAT to predict a behavioral choice between fruits and snacks. Four studies ran at different points in time tested the predictive and incremental validity of an IAT towards fruits versus snacks. Specifically, we tested whether both implicit and explicit attitudes towards snacks and fruits predict independently the behavioral choice between snacks and fruits (additive pattern) or whether implicit and explicit Implicit Food Preference 7 attitudes interact to predict the choice (interactive pattern). Testing of a double dissociation pattern was not possible due to the utilization of a single dependent variable.