714 GAWRONSKI AND BODENHAUSEN associations of such persons could be successfully influenced,they be observed (and sometimes even become stronger)when people would be expected to reject such associations as a valid basis for are unaware of the contingency between CS and US (e.g..Baeyens an evaluative judgment and to search for additional propositions to et al..1990).Second.in contrast to PC effects.EC effects do not maintain their committed views.Because strong attitudes are often depend on statistical CS-US contingencies.Instead,EC effects accompanied by extensive supporting knowledge (e.g..Abelson, seem to be primarily driven by spatiotemporal contiguity between 1988),it likely would not be hard to retrieve relevant propositions. the two kinds of stimuli independent of statistical ratios (e.g.. In contrast to the case for strong attitudes,ambivalent attitudes are Baeyens,Hermans,Eelen,1993).Third,in contrast to PC likely to produce much more pliable and unstable evaluative effects,EC effects are resistant to extinction.Whereas single judgments (e.g..Armitage Conner,2000),in part because they presentations of the CS typically reduce conditioning effects in PC. may elicit greater cognitive elaboration (Hanze,2001).Without such single presentations leave EC effects unaffected (e.g.,Baey- any particular bias to constrain the direction of elaboration,am- ens,Crombez,Van den Bergh,Eelen,1988).Taken together. bivalent evaluative judgments not only may be more unstable but these results suggest that,even though PC effects may be driven by may also be less consistently correlated with automatic evaluative higher order propositional processes (e.g..Holyoak et al.,1989; reactions (e.g.,Nosek,2005;see also Hofmann,Gschwendner, Lovibond,2003;Williams,1995),EC effects are better explained Nosek.Schmitt.2005). by a qualitatively different,associative mechanism(De Houwer et The most general claim of the APE model is that the temporal al.,2001). consistency of evaluative judgments should vary as a function of With regard to the role of propositional processes in human the temporal consistency of associative evaluations when associa- reasoning.we generally agree with the argument that many pro- tive evaluations are considered as a valid basis for evaluative cesses that have been described as associative may actually be judgments.However,the temporal consistency of evaluative judg- propositional (e.g.,Erb et al.,2003;Osman,2004).In fact,we ments can also be independent of the temporal consistency of claim that any kind of reasoning process is inherently proposi- associative evaluations when associative evaluations are rejected tional,given that reasoning is concerned with validation and the as a valid basis for evaluative judgments. assessment of truth values (Deutsch Strack,in press).To be sure,automatic affective reactions resulting from associative pro- One or Two Processes？ cesses are usually transformed into propositional format,thus representing a particular kind of information in propositional rea- Even though the distinction between associative and proposi- soning.However,this transformation does not imply that the tional processes is shared by several models of mental processing associative process that gives rise to automatic affective reactions (e.g.,Bazerman et al,1998;Kahneman,2003;Lieberman et al., itself is propositional.It is important to note that even when an 2002:Sloman,1996:Smith DeCoster.2000:Strack Deutsch. automatic affective reaction is rejected in the course of proposi- 2004).it also has been subject to criticism.With regard to the APE tional reasoning,the automatic affective reaction may still be model,the three most critical arguments in this debate are as unaffected,thus leading to a dissociation between evaluative judg- follows:(a)Effects of classical conditioning depend on higher ments and automatic affective reactions (e.g.,Gawronski Strack. order propositional rather than lower level associative processes 2004:see also Butler et al..2003:Gross.1998). (e.g.,Holyoak,Koh,Nisbett,1989;Lovibond,2003;Williams, With regard to the rule-based character of associative processes 1995),(b)reasoning processes that have been described as asso- (e.g.,Gigerenzer Regier,1996;Kruglanski et al.,in press),it is ciative are actually propositional (e.g.,Erb et al..2003;Osman, important to note that the primary criterion for distinguishing 2004),and (c)even associative processes follow rules,thus un- between associative and propositional processes in the APE model dermining the theoretical basis for a distinction between associa- is the notion of truth values rather than "if...then"rules.In fact. tive and rule-based processes (e.g.,Gigerenzer Regier,1996; we generally agree that associative processes follow a lawful Kruglanski,Erb,Pierro,Mannetti,Chun,in press).All of these manner,namely pattern activation in an associative network.How- arguments challenge the basic distinction between associative and ever,associative processes differ from propositional processes, propositional process,suggesting that a single process may be such that the activation of associations occurs independently of sufficient to account for the phenomena in question whether a person considers these associations to be accurate.Such In our view,the APE model is not vulnerable to any of these validation requires propositional processes,which assess the truth arguments.With regard to the role of propositional processes in or falsity of a given proposition by reference to their consistency classical conditioning (e.g..Holyoak et al..1989;Lovibond,2003; with other relevant propositions (Kruglanski,1989).Given that Williams,1995),it is important to note that the APE model is consistency assessment-and thus validation-is not possible primarily concerned with effects of EC rather than with signal without logical rules that define the relation between propositions learning effects resulting from Pavlovian conditioning (PC). an important feature of propositional reasoning is the notion of Whereas the CS in a PC paradigm typically acquires a predictive syllogistic reasoning.However,this feature is an implication of value for the US.the CS in an EC paradigm merely attains the our definition in terms of truth values rather than the defining affective guality of the US.This distinction is crucial,as the feature per se.Thus,even though the "rule-based"character of acquisition of affective quality differs from the acquisition of associative processes(e.g..Gigerenzer Regier,1996:Kruglanski predictive value in several important aspects (for a review,see De et al.,in press)may pose a problem to models that define asso- Houwer et al.,2001).First,in contrast to PC effects,EC effects do ciative and propositional processes in terms of similarity-based not depend on people's awareness of the contingency between CS processing versus application of syllogistic rules (e.g.,Smith and US.Whereas PC effects require that people are consciously DeCoster,2000),it does not affect the proposed definition in terms aware of the contingency between CS and US,EC effects can also of activation versus validation implied by the APE model.associations of such persons could be successfully influenced, they would be expected to reject such associations as a valid basis for an evaluative judgment and to search for additional propositions to maintain their committed views. Because strong attitudes are often accompanied by extensive supporting knowledge (e.g., Abelson, 1988), it likely would not be hard to retrieve relevant propositions. In contrast to the case for strong attitudes, ambivalent attitudes are likely to produce much more pliable and unstable evaluative judgments (e.g., Armitage & Conner, 2000), in part because they may elicit greater cognitive elaboration (Ha¨nze, 2001). Without any particular bias to constrain the direction of elaboration, am￾bivalent evaluative judgments not only may be more unstable but may also be less consistently correlated with automatic evaluative reactions (e.g., Nosek, 2005; see also Hofmann, Gschwendner, Nosek, & Schmitt, 2005). The most general claim of the APE model is that the temporal consistency of evaluative judgments should vary as a function of the temporal consistency of associative evaluations when associa￾tive evaluations are considered as a valid basis for evaluative judgments. However, the temporal consistency of evaluative judg￾ments can also be independent of the temporal consistency of associative evaluations when associative evaluations are rejected as a valid basis for evaluative judgments. One or Two Processes? Even though the distinction between associative and proposi￾tional processes is shared by several models of mental processing (e.g., Bazerman et al., 1998; Kahneman, 2003; Lieberman et al., 2002; Sloman, 1996; Smith & DeCoster, 2000; Strack & Deutsch, 2004), it also has been subject to criticism. With regard to the APE model, the three most critical arguments in this debate are as follows: (a) Effects of classical conditioning depend on higher order propositional rather than lower level associative processes (e.g., Holyoak, Koh, & Nisbett, 1989; Lovibond, 2003; Williams, 1995), (b) reasoning processes that have been described as asso￾ciative are actually propositional (e.g., Erb et al., 2003; Osman, 2004), and (c) even associative processes follow rules, thus un￾dermining the theoretical basis for a distinction between associa￾tive and rule-based processes (e.g., Gigerenzer & Regier, 1996; Kruglanski, Erb, Pierro, Mannetti, & Chun, in press). All of these arguments challenge the basic distinction between associative and propositional process, suggesting that a single process may be sufficient to account for the phenomena in question. In our view, the APE model is not vulnerable to any of these arguments. With regard to the role of propositional processes in classical conditioning (e.g., Holyoak et al., 1989; Lovibond, 2003; Williams, 1995), it is important to note that the APE model is primarily concerned with effects of EC rather than with signal learning effects resulting from Pavlovian conditioning (PC). Whereas the CS in a PC paradigm typically acquires a predictive value for the US, the CS in an EC paradigm merely attains the affective quality of the US. This distinction is crucial, as the acquisition of affective quality differs from the acquisition of predictive value in several important aspects (for a review, see De Houwer et al., 2001). First, in contrast to PC effects, EC effects do not depend on people’s awareness of the contingency between CS and US. Whereas PC effects require that people are consciously aware of the contingency between CS and US, EC effects can also be observed (and sometimes even become stronger) when people are unaware of the contingency between CS and US (e.g., Baeyens et al., 1990). Second, in contrast to PC effects, EC effects do not depend on statistical CS–US contingencies. Instead, EC effects seem to be primarily driven by spatiotemporal contiguity between the two kinds of stimuli independent of statistical ratios (e.g., Baeyens, Hermans, & Eelen, 1993). Third, in contrast to PC effects, EC effects are resistant to extinction. Whereas single presentations of the CS typically reduce conditioning effects in PC, such single presentations leave EC effects unaffected (e.g., Baey￾ens, Crombez, Van den Bergh, & Eelen, 1988). Taken together, these results suggest that, even though PC effects may be driven by higher order propositional processes (e.g., Holyoak et al., 1989; Lovibond, 2003; Williams, 1995), EC effects are better explained by a qualitatively different, associative mechanism (De Houwer et al., 2001). With regard to the role of propositional processes in human reasoning, we generally agree with the argument that many pro￾cesses that have been described as associative may actually be propositional (e.g., Erb et al., 2003; Osman, 2004). In fact, we claim that any kind of reasoning process is inherently proposi￾tional, given that reasoning is concerned with validation and the assessment of truth values (Deutsch & Strack, in press). To be sure, automatic affective reactions resulting from associative pro￾cesses are usually transformed into propositional format, thus representing a particular kind of information in propositional rea￾soning. However, this transformation does not imply that the associative process that gives rise to automatic affective reactions itself is propositional. It is important to note that even when an automatic affective reaction is rejected in the course of proposi￾tional reasoning, the automatic affective reaction may still be unaffected, thus leading to a dissociation between evaluative judg￾ments and automatic affective reactions (e.g., Gawronski & Strack, 2004; see also Butler et al., 2003; Gross, 1998). With regard to the rule-based character of associative processes (e.g., Gigerenzer & Regier, 1996; Kruglanski et al., in press), it is important to note that the primary criterion for distinguishing between associative and propositional processes in the APE model is the notion of truth values rather than “if . . . then” rules. In fact, we generally agree that associative processes follow a lawful manner, namely pattern activation in an associative network. How￾ever, associative processes differ from propositional processes, such that the activation of associations occurs independently of whether a person considers these associations to be accurate. Such validation requires propositional processes, which assess the truth or falsity of a given proposition by reference to their consistency with other relevant propositions (Kruglanski, 1989). Given that consistency assessment—and thus validation—is not possible without logical rules that define the relation between propositions, an important feature of propositional reasoning is the notion of syllogistic reasoning. However, this feature is an implication of our definition in terms of truth values rather than the defining feature per se. Thus, even though the “rule-based” character of associative processes (e.g., Gigerenzer & Regier, 1996; Kruglanski et al., in press) may pose a problem to models that define asso￾ciative and propositional processes in terms of similarity-based processing versus application of syllogistic rules (e.g., Smith & DeCoster, 2000), it does not affect the proposed definition in terms of activation versus validation implied by the APE model. 714 GAWRONSKI AND BODENHAUSEN