《社会心理学》课程教学资源（文献资料）attitude&stereotype——Associative and Propositional Processes in Evaluation：An Integrative Review of Implicit and Explicit Attitude Change.pdf

Associative and Propositional Processes in Evaluation: An Integrative Review of Implicit and Explicit Attitude Change Bertram Gawronski University of Western Ontario Galen V. Bodenhausen Northwestern University A central theme in recent research on attitudes is the distinction between deliberate, “explicit” attitudes and automatic, “implicit” attitudes. The present article provides an integrative review of the available evidence on implicit and explicit attitude change that is guided by a distinction between associative and propositional processes. Whereas associative processes are characterized by mere activation independent of subjective truth or falsity, propositional reasoning is concerned with the validation of evaluations and beliefs. The proposed associative–propositional evaluation (APE) model makes specific assumptions about the mutual interplay of the 2 processes, implying several mechanisms that lead to symmetric or asymmetric changes in implicit and explicit attitudes. The model integrates a broad range of empirical evidence and implies several new predictions for implicit and explicit attitude change. Keywords: attitude change, cognitive consistency, dual-process models, evaluative conditioning, implicit measures If eras of psychological research can be characterized in terms of general ideas, a major theme of the current era is probably the notion of automaticity. Many aspects of human behavior that have previously been assumed to have their roots in higher order processes of deliberate reasoning are now viewed as resulting from automatic processes that may occur spontaneously and outside of people’s awareness or control (Bargh, 1997; Moors & De Houwer, 2006). This perspective is also dominant in contemporary research on attitudes, in which deliberate, “explicit” attitudes are often contrasted with automatic, “implicit” attitudes (Greenwald & Banaji, 1995; Petty, Fazio, & Brin˜ol, in press; Wilson, Lindsey, & Schooler, 2000; Wittenbrink & Schwarz, in press). Whereas the former are usually equated with deliberative, self-reported evaluations, the latter are typically inferred from people’s performance on response latency measures, such as the Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998) or sequential priming tasks (Fazio, Jackson, Dunton, & Williams, 1995; Wittenbrink, Judd, & Park, 1997). Even though previous models effectively account for the differential impact of implicit and explicit attitudes on spontaneous and deliberate behavior (e.g., Fazio & Olson, 2003; Strack & Deutsch, 2004; Wilson et al., 2000), theories of attitude change still face a considerable asymmetry in integrating the available evidence. Contemporary models of persuasion have been very successful in explaining the influence of different kinds of message cues on self-reported explicit evaluations (Chen & Chaiken, 1999; Kruglanski & Thompson, 1999; Petty & Wegener, 1999; for a review, see Visser & Cooper, 2003). However, changes in implicit attitudes are still largely unexplained. This shortfall has its roots in at least two conceptual issues. First, several models of attitude change do not distinguish between implicit and explicit attitudes in the first place but treat attitudes as a unitary construct (e.g., Chaiken, Liberman, & Eagly, 1989; Kruglanski & Thompson, 1999; Petty & Cacioppo, 1986; Van Overwalle & Siebler, 2005). As such, these models leave an explanatory gap for any dissociation that may emerge between explicit and implicit attitudes. For instance, some studies have demonstrated changes in explicit but not implicit attitudes (e.g., Gawronski & Strack, 2004), whereas others found changes in implicit but not explicit attitudes (e.g., M. A. Olson & Fazio, 2006). If explicit and implicit attitudes indeed represent a unitary construct, changes in one kind of attitude should usually be associated with corresponding changes in the other kind of attitude. Second, most attitude change models that do distinguish between explicit and implicit attitudes consider implicit attitudes to be stable evaluative representations stemming from long-term socialization experiences. Explicit attitudes, in contrast, are conceived as more recently acquired attitudes that coexist with the old, presumptively stable, implicit attitude (e.g., Petty, Tormala, Brin˜ol, & Jarvis, 2006; Wilson et al., 2000). These models can explain patterns that involve changes in explicit but not implicit attitudes (e.g., Gawronski & Strack, 2004). However, they leave an explanatory gap for the accumulating number of studies showing changes in implicit but not explicit attitudes (e.g., Dasgupta & Greenwald, 2001; Karpinski & Hilton, 2001; M. A. Olson & Fazio, 2006; for a review, see Blair, 2002). If implicit attitudes indeed reflect highly stable evaluative representations that have Bertram Gawronski, Department of Psychology, University of Western Ontario, London, Ontario, Canada; Galen V. Bodenhausen, Department of Psychology, Northwestern University. Preparation of this article was supported by Canada Research Chairs program Grant 202555, Social Sciences and Humanities Research Council of Canada Grant 410-2005-1339, and Academic Development Fund of the University of Western Ontario Grant 05-303 to Bertram Gawronski. We thank Andrew Becker, Riki Conrey, Julia Grafe, Wilhelm Hofmann, and Jim Olson for helpful comments on an earlier version of this article. Correspondence concerning this article should be addressed to Bertram Gawronski, Department of Psychology, University of Western Ontario, Social Science Centre, London, Ontario N6A 5C2, Canada. E-mail: bgawrons@uwo.ca Psychological Bulletin Copyright 2006 by the American Psychological Association 2006, Vol. 132, No. 5, 692–731 0033-2909/06/$12.00 DOI: 10.1037/0033-2909.132.5.692 692

their roots in long-term socialization experiences (e.g., Dovidio, Kawakami, & Beach, 2001; Petty et al., 2006; Rudman, 2004; Wilson et al., 2000), implicit attitudes should generally be more robust than explicit attitudes. The main goal of the present article is twofold. The first objective is to propose a new theoretical model for the study of explicit and implicit attitude change. The second objective is to provide an integrative, exhaustive review of the available evidence regarding implicit and explicit attitude change that is organized according to the implications of the proposed model (for a list of the reviewed studies, see the Appendix).1 For this purpose, the first section of this article outlines our theoretical conceptualization of implicit and explicit attitudes. In particular, we argue that implicit and explicit attitudes should be understood in terms of their underlying mental processes, which are associative and propositional processes. The second section uses the proposed associative– propositional evaluation (APE) model as an organizing framework to review the available evidence on explicit and implicit attitude change. This section specifies causes of implicit and explicit attitude change and how changes in one kind of evaluation may or may not be associated with changes in the other. Finally, the last section discusses the relation of our model to other theories of attitude change, the relative stability of attitude changes, limits of a single-process approach, directions for future research, and some methodological issues related to the application of the proposed model. Associative and Propositional Processes in Evaluation The systematic investigation of implicit and explicit attitudes began with the development of measures that were particularly designed to assess individual differences in automatic evaluations (Greenwald & Banaji, 1995). Research on implicit attitudes relies on a large variety of measures, such as the Implicit Association Test (Greenwald et al., 1998), affective priming (Fazio et al., 1995), semantic priming (Wittenbrink et al., 1997), the go/no-go association task (Nosek & Banaji, 2001), the extrinsic affective Simon task (De Houwer, 2003a), and the affect misattribution paradigm (Payne, Cheng, Govorun, & Stewart, 2005). These measures have been reviewed extensively elsewhere (De Houwer, 2003b; Fazio & Olson, 2003; Wittenbrink & Schwarz, in press), and thus we refrain from an elaborate discussion of them. Instead, we focus on a more detailed discussion of the proposed model that is used as the organizing framework of this review. The APE model builds on earlier dual-process theories of cognitive functioning that distinguish between two qualitatively different kinds of mental processes (e.g., Bazerman, Tenbrunsel, & WadeBenzoni, 1998; Kahneman, 2003; Lieberman, Gaunt, Gilbert, & Trope, 2002; Sloman, 1996; Smith & DeCoster, 2000; Strack & Deutsch, 2004). In particular, we argue that implicit and explicit attitudes should be understood in terms of their underlying processes, which can be described as associative processes for implicit attitudes and propositional processes for explicit attitudes. This conceptualization adopts the widespread definition of attitude as a psychological tendency to evaluate a given entity with some degree of favor or disfavor (see Eagly & Chaiken, 1993; Zanna & Rempel, 1988). However, the APE model goes beyond this definition by arguing that such evaluative tendencies can be rooted in two kinds of mental processes. Associative Processes The first source of evaluative tendencies resides in associative processes, which build the basis for what many researchers call implicit attitudes. Associative evaluations are best characterized as automatic affective reactions resulting from the particular associations that are activated automatically when one encounters a relevant stimulus. Such activation processes do not require much cognitive capacity or an intention to evaluate an object (Cunningham, Raye, & Johnson, 2004). The most important feature, however, is that associative evaluations are independent of the assignment of truth values. That is, associative evaluations can be activated irrespective of whether a person considers these evaluations as accurate or inaccurate. For example, the activation level of negative associations regarding African Americans may be high even though an individual may regard these associations as inadequate or false (Devine, 1989). Thus, associative evaluations are not personal in the sense that they are not necessarily personally endorsed (cf. Arkes & Tetlock, 2004). Instead, the primary determinants of association activation are feature similarity and spatiotemporal contiguity (Bassili & Brown, 2005; Smith & DeCoster, 2000). Another important aspect of associative processes is the notion of pattern activation (see Smith, 1996). Pattern activation refers to the idea that the activation of particular associations in memory is determined by the relative fit between (a) the preexisting structure of associations in memory and (b) the particular set of external input stimuli. For example, the associative pattern activated by the stimuli basketball and gym may include concepts such as “bouncing” but not concepts such as “floating.” However, the associative pattern activated by the stimuli basketball and water may include concepts such as “floating” but not “bouncing.” In other words, even though the concept “basketball” is associated with both “bouncing” and “floating” in memory, which of the two becomes activated depends on the particular context in which the stimulus basketball is encountered (see Barsalou, 1982). Thus, applied to attitudes, the same object may activate different associative patterns and thus different automatic affective reactions depending on the particular context in which the object is encountered. Still, automatic affective reactions are not purely context driven, as pattern activation generally depends on both (a) the preexisting structure of associations in memory and (b) the particular set of external input stimuli. Propositional Processes The second source of evaluative tendencies comes from propositional processes, which build the basis for what many research- 1 For the review part of this article, we used three general inclusion criteria: (a) A study must include at least one implicit attitude measure, (b) the implicit measure must tap a general evaluation rather than a semantic association, and (c) the study must include at least one experimental manipulation. Thus, the present review covers not only evidence on attitude change but also studies on attitude formation. In addition, the present review includes findings that may be considered as attitude shifts rather than attitude change. For the sake of simplicity, however, we generally use the term attitude change to refer to attitude formation, attitude change, and attitude shifts. IMPLICIT AND EXPLICIT ATTITUDE CHANGE 693

ers call explicit attitudes. Evaluations resulting from propositional processes can be characterized as evaluative judgments that are based on syllogistic inferences derived from any kind of propositional information that is considered relevant for a given judgment. Drawing on Strack and Deutsch (2004), such inferences are assumed to occur in a reflective system that is superordinate to an associative store. That is, the reflective system is assumed to transform inputs from the associative store into propositional format (e.g., a negative affective reaction is transformed into the proposition “I dislike X”). The resulting propositions are then subject to syllogistic inferences that assess their validity. Thus, the most important feature that distinguishes propositional from associative processes is their dependency on truth values. Whereas the activation of associations can occur regardless of whether a person considers the associations to be true or false, processes of propositional reasoning are generally concerned with the validation of evaluations and beliefs. Moreover, whether a given proposition is explicitly endorsed depends on its subjective validity, as determined by processes of propositional reasoning. Applying this reasoning to attitudes, we argue that people usually use their automatic affective reactions toward an object as a basis for evaluative judgments about this object. In other words, the default mode of propositional reasoning is affirmation of validity (Gilbert, 1991; see also Albarracı´n & Wyer, 2001), in this case the validity of the propositional implication of an automatic affective reaction (e.g., the proposition “I dislike X” implied by a negative affective reaction toward X). However, evaluative judgments can also be independent of automatic affective reactions, when the propositional implications of these reactions are rejected as a valid basis for an evaluative judgment (see Strack, 1992).2 Perceived validity of a proposition—and thus of the propositional implication of an automatic affective reaction—depends on the consistency of this proposition with other propositions that are momentarily considered to be relevant for the respective judgment. In the case of evaluative judgments, such propositions may include nonevaluative propositions referring to general beliefs about the world or propositional evaluations of other attitude objects (Bem, 1970; Jones & Gerard, 1967).3 If the propositional implication of an automatic affective reaction is consistent with other relevant propositions, it will most likely be considered as valid and thus serve as the basis for an evaluative judgment. If, however, the propositional implication of an automatic affective reaction is inconsistent with other relevant propositions, it may be considered invalid.4 For example, the propositional implication of a negative affective reaction to a minority member (e.g., “I don’t like this African American person”) may be inconsistent with the propositional evaluation of another attitude object (e.g., “It is bad to evaluate members of disadvantaged minority groups negatively”) and nonevaluative propositions about the world (e.g., “African Americans are a disadvantaged minority group”). Hence, the resulting inconsistency among the three propositions may lead to a rejection of the negative affective reaction as a valid basis for an evaluative judgment. However, the negative affective reaction may still serve as basis for an evaluative judgment if either the nonevaluative proposition about the world is considered invalid (e.g., “African Americans are not a disadvantaged minority group”) or the propositional evaluation of another relevant attitude object is rejected (e.g., “Negative evaluations of disadvantaged minority members are okay”).5 Another important aspect of propositional reasoning concerns hypothetical propositions and mere knowledge of propositions. From the perspective of the APE model, hypothetical propositions can be described as propositions that are assigned truth values with a preliminary character. That is, a proposition becomes hypothetical because of the tentative character of its truth value rather than because of something special about its content. This view has important implications for how hypothetical propositions affect information processing. Consistent with previous research, we argue that merely entertaining a particular proposition increases the momentary activation level of corresponding associations in memory (e.g., Gregg, Seibt, & Banaji, 2006). Because the default mode of propositional thinking is an affirmation of momentarily activated associations (see Gilbert, 1991), mere supposition should, in turn, increase the perceived validity of this proposition (for a review, see Koehler, 1991). Similar considerations can be applied to mere knowledge of a proposition. From the perspective of the APE model, mere knowledge of a proposition reflects cases in which a person knows about a proposition endorsed by other people irrespective of whether this person considers this proposi- 2 From the perspective of the tripartite model of attitudes (see Zanna & Rempel, 1988), one could argue that implicit attitudes (i.e., automatic affective reactions) represent the affective component attributed to attitudes, whereas explicit attitudes (i.e., evaluative judgments) are typically a joint product of both affective and cognitive components, with their relative influence depending on factors such as cognitive elaboration (e.g., Shiv & Nowlis, 2004) and propositional consistency (e.g., Judd & Lusk, 1984). 3 It is important to note that such other (evaluative or nonevaluative) propositions are also based on inputs from the associative store. The present model implies no independent storage of propositions in long-term memory (see also Deutsch & Strack, in press). 4 Wyer and Radvansky (1999) recently argued that people sometimes recognize the truth or falsity of a proposition spontaneously in the course of comprehending a proposition. In particular, these researchers argued that people spontaneously construct mental models of a proposition in the course of comprehending that proposition. If the constructed model is redundant with the content of a previously formed model, people spontaneously recognize the corresponding proposition as true. Moreover, if the constructed model does not match with a previously formed model, people assess the validity of the proposition by comparing the similarity of its mental model with previously formed models of the same predicate. Applied to the present question, it is important to note that Wyer and Radvansky postulated such spontaneous validation processes only for declarative propositions (e.g., events, episodes), not for evaluative propositions. Moreover, given that automatic affective reactions have their roots in spontaneously activated associations, these associations—by definition—represent a previously constructed model. As such, associative evaluations should spontaneously be considered as true, which is consistent with our assumption that people, by default, base their evaluative judgments on their automatic affective reactions, unless higher order processes of propositional reasoning lead to a rejection of the propositional implications of an associative evaluation (see also Deutsch, Gawronski, & Strack, in press; Gilbert, 1991). 5 These two cases are also reflected in contemporary theories of (explicit) prejudice, such that explicitly accepting the derogation of disadvantaged groups is considered a traditional form of prejudice, whereas denying discrimination is considered a modern form of prejudice (e.g., McConahay, 1986; Swim, Aikin, Hall, & Hunter, 1995; for a more detailed discussion, see Gawronski, Peters, & Brochu, 2005). 694 GAWRONSKI AND BODENHAUSEN

tion as true or false. We argue that mere knowledge of a proposition endorsed by other people can contribute to the activation of corresponding associations in memory even when a person does not believe in the validity of that proposition (Kawakami, Dovidio, & Dijksterhuis, 2003). For example, mere knowledge of a cultural stereotype may lead to automatic negative reactions toward the members of a disadvantaged minority group even when the stereotype is considered inaccurate (Devine, 1989). This assumption is supported by research showing that members of disadvantaged minority groups often exhibit automatic negative reactions toward their own in-group, particularly when they perceive the majority group to exhibit a high level of negativity against the minority group (e.g., Livingston, 2002). It is important to note that such automatic negativity against the in-group is often rejected on the propositional level, which leads to negative (rather than positive) correlations between personally endorsed evaluations and evaluations by others (cf. Heider, 1958). That is, higher levels of perceived negativity from out-groups lead to more negative evaluations of the in-group on the associative level but to more positive evaluations on the propositional level (Livingston, 2002). Cognitive Consistency According to the APE model, cognitive consistency is exclusively a concern of propositional reasoning (Gawronski, Strack, & Bodenhausen, in press). In particular, we argue that consistency results from a propositional process of consistency assessment that is based on the assignment of truth values and the application of syllogistic rules and logical principles.6 From a general perspective, two propositions are consistent with each other when both are regarded as true and one does not imply the opposite of the other. In contrast, two propositions are inconsistent when both are regarded as true and one follows from the opposite of the other (see Festinger, 1957). It is important to note that, because (in)consistency between two propositions cannot even be defined without an assignment of truth values, people have to resolve inconsistency by means of propositional reasoning—that is, either by changing the truth value of one proposition or by finding an additional proposition that resolves the inconsistency (Gawronski, Strack, & Bodenhausen, in press; Kruglanski, 1989). For example, if exposure to a minority member automatically activates negative associations, people either may reject the propositional implication of these associations because of its inconsistency with other accepted propositions (see above) or may find an additional proposition that resolves the inconsistency (e.g., “This African American person was unfriendly”). Whereas the former process refers to what has been described as negation (Gilbert, 1991), suppression (Wegner, 1994), and hierarchical inhibition (Bodenhausen & Macrae, 1998), the latter process may be described as rationalization (Festinger, 1957) or justification (Crandall & Eshleman, 2003; Dovidio & Gaertner, 2004). The important point is that whenever cognitive consistency issues arise, propositional reasoning is the mediating mechanism underlying the obtained evaluative outcomes. Cognitive Elaboration Cognitive elaboration, or the degree of active thought devoted to an attitude object, has long been assumed to be of paramount importance in attitude change (Greenwald, 1968; Petty, Ostrom, & Brock, 1981). As with other models of implicit and explicit attitudes (e.g., Fazio & Olson, 2003; Wilson et al., 2000), the APE model also implies a crucial role of cognitive elaboration. Fazio and Olson (2003), for example, argued that cognitive elaboration is a crucial determinant of people’s success in suppressing the influence of automatic attitudes on evaluative judgments. In a similar vein, Wilson et al. (2000) argued that cognitive elaboration is a crucial determinant of people’s success in retrieving their explicit attitudes from memory. Thus, both models imply that the correlation between implicit and explicit evaluations should decrease as a function of increasing cognitive elaboration. Consistent with this assumption, Florack, Scarabis, and Bless (2001) showed that implicit racial prejudice as measured by the IAT was more strongly related to deliberative judgments about members of the relevant racial out-group to the extent that respondents were low in need for cognition (i.e., had a low dispositional tendency to engage in deliberative analyses; see Cacioppo, Petty, Feinstein, & Jarvis, 1996). In a similar vein, a recent meta-analysis by Hofmann, Gawronski, Gschwendner, Le, and Schmitt (2005) showed that correlations between the IAT and explicit measures significantly increased as a function of increasing spontaneity in the course of making a judgment. The APE model similarly posits that increased elaboration often reduces the correlation between explicit and implicit attitudes. However, the APE model differs from previous accounts with regard to its assumptions about the underlying processes. As we have noted, these earlier accounts argued that cognitive elaboration influences success in suppressing automatic evaluations (Fazio & Olson, 2003) or in retrieving explicit attitudes from memory (Wilson et al., 2000). In contrast to these models, the APE model assumes that cognitive elaboration affects the complexity of propositional thinking by influencing how many judgment-relevant propositions one considers in addition to one’s automatic affective reaction (see Albarracı´n & Wyer, 2000; Kruglanski & Thompson, 1999). More extensive elaboration generally implies considering a greater number of propositions about the attitude object. To the extent that any of these additional propositions is inconsistent with the automatic evaluative response, the extra elaboration is likely to reduce the correlation between automatic affective reactions and evaluative judgments (e.g., Florack et al., 2001; Hofmann et al., 2005; see also Shiv & Nowlis, 2004). It is important to note, however, that increased cognitive elaboration does not inevitably reduce the relation between explicit and implicit attitudes. According to the APE model, cognitive elaboration should reduce the relation between explicit and implicit attitudes only if additionally considered propositions question the validity of one’s automatic affective reaction as a basis for an evaluative judgment (see Judd & Lusk, 1984). However, if additionally considered propositions do not question the validity of one’s automatic affective reaction, the relation between explicit and implicit attitudes should be unaffected by cognitive elaboration. Moreover, if additionally considered propositions confirm the subjective validity of one’s automatic affective reaction, the rela- 6 The term logical consistency is intended to refer more broadly to subjective consistency resulting from any kind of inferential rule that is considered to be valid, rather than to strict logical consistency in terms of normative syllogistic rules. IMPLICIT AND EXPLICIT ATTITUDE CHANGE 695

tion between explicit and implicit attitudes should actually increase (rather than decrease) as a function of cognitive elaboration. For example, if increased cognitive elaboration identifies an additional proposition (e.g., “This African American person behaved in a hostile manner”) that resolves the inconsistency among a propositionally transformed affective reaction (e.g., “I don’t like this African American person”); other, nonevaluative propositions (e.g., “African Americans are a disadvantaged minority group”); and propositional evaluations of other attitude objects (e.g., “It is bad to evaluate members of disadvantaged minority groups negatively”), the relation between explicit and implicit attitudes should actually increase rather than decrease as a function of cognitive elaboration. In other words, whether the relation between explicit and implicit attitudes increases or decreases as a function of cognitive elaboration does not depend on the amount of cognitive elaboration per se; rather, it is a function of a consistency assessment regarding the momentarily considered set of propositions. Attitudes as Online Constructions Although attitudes have traditionally been regarded as stable evaluative representations of an attitude object residing in longterm memory, some researchers have argued that attitudes are not merely retrieved from memory but rather are constructed on the spot (e.g., Schwarz & Bohner, 2001; Tesser, 1978; Wilson & Hodges, 1992; see also Bassili & Brown, 2005). A similar assumption is also implied by the APE model. However, the notion of attitude construction has substantially different meanings for associative and propositional processes. For propositional processes, attitudes can be considered to be constructed in that the propositional implication of one’s automatic affective reaction can be regarded as either true or false, depending on the other propositions that one simultaneously considers in forming an evaluative judgment. That is, changes in the set of momentarily considered propositions can influence whether the propositional implication of one’s automatic affective reaction is considered valid or invalid. As such, evaluative judgments resulting from propositional processes can be regarded as constructed because they are not determined a priori. Rather, the truth value of a given propositional evaluation depends on its consistency with the set of other propositions that are currently considered to be relevant, and this set may change as a function of various factors (e.g., cognitive elaboration). Notwithstanding these factors, however, there can nevertheless be a great deal of stability in evaluative judgments, to the extent that the set of relevant propositions that one considers in forming an evaluative judgment is consistent across contexts. Thus, even though evaluative judgments resulting from propositional processes can be regarded as constructed, the APE model does not preclude the possibility of attitudinal stability or rigidity on the propositional level. For associative processes, attitudes might also be considered to be constructed. However, in contrast to the active notion implied for the construction of evaluative judgments, the construction process for associative evaluations is rather passive. As we have outlined, we argue that automatic affective reactions depend on processes of pattern activation in associative memory. Moreover, these processes of pattern activation depend on (a) external input stimuli and (b) the preexisting structure of associations in memory. As such, different sets of input stimuli may activate different associative patterns for a given attitude object. Accordingly, the same attitude object may be evaluated differently as a function of the particular context in which it is encountered. This assumption is consistent with research showing that implicit attitudes are highly sensitive to the particular context in which an attitude object is encountered (e.g., Barden, Maddux, Petty, & Brewer, 2004; Schaller, Park, & Mueller, 2003; Wittenbrink, Judd, & Park, 2001). From this perspective, implicit attitudes, or associative evaluations, can be regarded as constructed on the spot because their activation depends not only on the preexisting structure of associations in memory but also on the particular set of input stimuli. As with evaluative judgments, however, associative evaluations can also be quite stable, such that they show a high level of temporal consistency if the activation patterns elicited when one encounters the attitude object are themselves robustly stable. A rotten egg will likely produce a negative affective reaction regardless of the particular context within which the offending egg is encountered. From a general perspective, temporal consistency of associative evaluations reflects the chronic or dispositional aspect that is often attributed to attitudes. We consider the issue of temporal stability of attitudes in more detail in the remainder of this article. Unconscious Attitudes A final foundational question concerns the potential unconsciousness of implicit attitudes. Some researchers have argued that implicit attitudes reflect unconscious evaluations of an attitude object, whereas explicit attitudes reflect their conscious counterparts (e.g., Banaji, Lemm, & Carpenter, 2001). This assumption has its roots in Greenwald and Banaji’s (1995) definition of implicit attitudes as “introspectively unidentified (or inaccurately identified) traces of past experience” (p. 5) that mediate overt responses. The APE model deviates from this contention. Although we do not rule out the possibility that certain affective reactions are below the threshold of experiential awareness (e.g., Winkielman, Berridge, & Wilbarger, 2005; see also Cleeremans & Jime´nez, 2002), we assume that people generally do have some degree of conscious access to their automatic affective reactions and that they tend to rely on these affective reactions in making evaluative judgments (Gawronski, Hofmann, & Wilbur, in press; Schimmack & Crites, 2005). To be sure, they also sometimes reject these automatic evaluations as a valid basis for an evaluative judgment. As we have argued, whether people consider their automatic affective reaction to be a valid basis for an evaluative judgment depends on the consistency of this evaluation with other propositions that might be relevant for an evaluative judgment. These assumptions about the consciousness of implicit attitudes—or automatic affective reactions—are consistent with recent findings by LeBel and Gawronski (2006), who found that correlations between explicit and implicit attitudes significantly increased (from r .19 to r .51) when participants were asked to focus on their feelings toward the attitude object in the course of making an evaluative judgment (for similar findings, see Nier, 2005). If implicit attitudes are truly unconscious, such introspection instructions should leave correlations between explicit and implicit attitudes unaffected. 696 GAWRONSKI AND BODENHAUSEN