IMPLICIT AND EXPLICIT ATTITUDE CHANGE 715 mplicit and explicit attitude ch This limitation is most evi nediated b (b)an irect influenc uch as cha es in plicit but not (e Ga mpliciLanitud ski&Strack. 2004).As vide an in t supplement to standard analyses in trated s are well expla about the change.M the particular interplay of ass (a)t h other prediction still remain to be mediated by changes in the other mplications and Future Directions ance and ass ciative self-ancboring on postdecisional atti ude change.or the 1 of proactive const the attitude cha e APE ture re APE on th ological Methodological Issues implici des (e o rolled uds .the APE the disun te aluation equate affectiv ante to the h ch as the Implicit Ase Test (Greenwald eta How efit from a riori of the applying the APE modc nable sho only im Implic Test (Greenwald ct 1998 ance of consider may be respo ible for changes nberg.Banse.Zukova. Klauer,2003:Gawror for example.it s s im sider whether Banse 1999:Kawakami Dovidio 2001d h ges in t 001: vated by a given stimul In a similar vein. s in exnlici d on correlations betw d加en that are d for low reha udgment,o in th 2001 nn Gawronski,et al.,2005).He erent natterns of implicit and explicit attitude change it s portant to consider the intemal co only which the the tw s before interpreting the results of correlation or mediation icroa nt kinds of implicit attitude ng or and prope Rathe tha iust of the particul used as prim timu Brewer.200 the valen arch s oth the nd implicit attit Finally, we argue that any single-process model faces problems in explaining the multitude of patterns obtained in research on implicit and explicit attitude change. This limitation is most evi￾dent when it comes to explaining dissociations in attitude change, such as changes in explicit but not implicit attitudes (e.g., Gawron￾ski & Strack, 2004), changes in implicit but not explicit attitudes (e.g., Karpinski & Hilton, 2001), or antagonistic changes in im￾plicit and explicit attitudes (e.g., Strack & Deutsch, 2004). As we have illustrated, such dissociations are well explained by the APE model. However, it is less clear how a single-process model would account for such dissociations in attitude change. Moreover, the APE model implies specific predictions about (a) the conditions that should lead to changes in explicit attitudes, implicit attitudes, or both explicit and implicit attitudes and, given that corresponding changes in explicit and implicit attitudes are predicted, (b) the particular pattern of how changes in one kind of attitude should be mediated by changes in the other. Implications and Future Directions In addition to providing an integrative framework of the avail￾able evidence on explicit and implicit attitude change, the APE model also has several implications for future research. First, the APE model highlights the importance of focusing on the particular psychological processes that underlie implicit and explicit atti￾tudes. Rather than using properties to describe implicit and explicit attitudes (e.g., automatic vs. controlled, conscious vs. unconscious, old vs. new), the APE model stresses the distinct psychological nature of two qualitatively different processes: associative and propositional processes. Fundamental to the APE model is the assumption that the two kinds of processes are influenced in very different ways. Thus, future research on attitude change may benefit from a priori considerations regarding which of the two processes are influenced by a particular variable under investiga￾tion, thus allowing for specific predictions regarding whether a given variable should influence only explicit attitudes, only im￾plicit attitudes, or both explicit and implicit attitudes. Second, the APE model highlights the importance of consider￾ing different processes that may be responsible for changes in explicit and implicit attitudes. With regard to changes in implicit attitudes, for example, it seems important to consider whether a particular variable leads to changes in the preexisting structure of associations or to changes in the associative pattern that is acti￾vated by a given stimulus. In a similar vein, changes in explicit attitudes may be due to changes in associative evaluations, changes in the set of propositions that are considered for an evaluative judgment, or changes in the strategy to achieve propositional consistency. Because these processes can be associated with dif￾ferent patterns of implicit and explicit attitude change, it seems important to consider not only which of the two processes is affected in the first place but also how the respective process is influenced by a particular variable. Third, the APE model highlights the importance of focusing on mediating mechanisms as well as the particular interplay of asso￾ciative and propositional processes. Rather than just testing whether a particular variable influences only explicit attitudes, only implicit attitudes, or both explicit and implicit attitudes, future research should address the mutual relation between explicit and implicit attitude change. Such investigations seem particularly relevant in the case of corresponding attitude changes, which may reflect either (a) an indirect influence on explicit attitudes that is mediated by implicit attitudes or (b) an indirect influence on implicit attitudes that is mediated by explicit attitudes. Because the particular interplay between associative and propositional pro￾cesses seems crucial for a sufficient understanding of explicit and implicit attitude change, mediation and correlational analyses pro￾vide an important supplement to standard analyses in terms of general effects on mean values. Finally, the APE model implies several assumptions about the particular interplay of associative and propositional processes that have empirical implications for explicit and implicit attitude change. Some of these predictions have already been confirmed in previous research, whereas other predictions still remain to be tested. Most of these predictions refer to mediating mechanisms, such as the mediating role of associative evaluations in minimal group settings, the role of contingency awareness for EC effects on explicit and implicit attitudes, the joint influence of cognitive dissonance and associative self-anchoring on postdecisional atti￾tude change, or the impact of proactive construction versus retro￾active rejection of evaluations in persuasion. Thus, the APE model may stimulate future research designed to test these predictions. Methodological Issues Throughout this review, we have largely equated evaluative judgments resulting from propositional processes with self￾reported evaluations and equated automatic affective reactions resulting from associative processes with performance on indirect measures, such as the Implicit Association Test (Greenwald et al., 1998) or affective priming (Fazio et al., 1995). However, the latter equation implies a number of methodological issues that one needs to consider when applying the APE model. First, measures of implicit attitudes have shown large variation with regard to their internal consistency. Whereas many studies using the Implicit Association Test (Greenwald et al., 1998) showed quite satisfying reliability scores between .70 and .90 (e.g., Asendorpf, Banse, & Mu¨cke, 2002; Egloff & Schmukle, 2002; Gawronski, Ehrenberg, Banse, Zukova, & Klauer, 2003; Gawron￾ski, Geschke, & Banse, 2003), several other studies exhibited only moderate (e.g., Banse, 1999; Kawakami & Dovidio, 2001) or relatively unsatisfying scores (e.g., Banse, 2001; M. A. Olson & Fazio, 2003). Because many of our hypotheses about mediation processes depend on correlations between associative evaluations and evaluative judgments, low reliability of implicit attitude mea￾sures poses a serious problem (see also Cunningham, Preacher, & Banaji, 2001; Hofmann Gawronski, et al., 2005). Hence, when testing predictions derived from the present framework, it is im￾portant to consider the internal consistency of the implicit mea￾sures before interpreting the results of correlation or mediation analyses. Second, different kinds of implicit attitude measures may tap different kinds of evaluative associations. For instance, whereas affective priming tasks seem to be primarily influenced by auto￾matic evaluations of the particular exemplars used as prime stimuli (e.g., Livingston & Brewer, 2002; M. A. Olson & Fazio, 2003), the Implicit Association Test seems to be influenced by the valence of both the individual exemplars (e.g., Blu¨zmke & Friese, 2006; Govan & Williams, 2004; J. P. Mitchell et al., 2003) and the IMPLICIT AND EXPLICIT ATTITUDE CHANGE 715