Modeling the Effect of Evaluative Conditioning on Implicit Attitude Acquisition and Performance on the Implicit Association Test

Modeling the Effect of Evaluative Conditioning on Implicit Attitude Acquisition and Performance on the Implicit Association Test Boon-Kiat Quek (boonkiat.quek@northwestern.edu) Andrew Ortony (ortony@northwestern.edu) Department of Psychology, Northwestern University, Evanston, IL 60208, USA, and Institute of High Performance Computing, Agency for Science, Technology and Research, Singapore Abstract Using a previously proposed computational model of human performance on the Implicit Associations Test (IAT), we ex- plore how evaluative conditioning could inform attitude ac- quisition and formation of automatic associations in memory, and demonstrate the effects of such learning on implicit task performance on the test. This is achieved by augmenting the model with a learning mechanism based on a modified Heb- bian learning rule that adapts associative strengths between concepts depending on the temporal proximity of their activa- tion. By manipulating the frequencies at which different stim- uli are paired and presented as input to the network, we demonstrate how virtual subjects could acquire associative strengths that were subsequently reflected in simulated IATs as stronger relative preferences in favor of targets that were more frequently presented with positively-valenced stimuli. The model predicts that associations that are already strong have limited prospects for continued reinforcement. Keywords: Hebbian learning; implicit attitudes; simulation; localist-connectionist networks. Introduction Much discussion over the emergence of automatic associa- tions between concepts and their evaluations in memory has taken place within the context of evaluative and classical conditioning (e.g., De Houwer, 2007; De Houwer, Baeyens & Field, 2005; Olson & Fazio, 2001; 2002). Evaluative conditioning is defined as a change in the extent of liking or disliking towards a stimulus that is caused by the frequent pairing of that stimulus with other liked or disliked stimuli (De Houwer, Baeyens & Field, 2005). The interest in evaluative conditioning research is fueled by the fact that it has the potential to explain the emergence of attitudes and account for the ways in which people’s atti- tudes and beliefs, and consequently their behavior, could be influenced. Thus, it has wide implications especially with regards to consumers’ preferences, tastes, and purchasing habits. For instance, Gibson (2008) recently demonstrated the effect of evaluative conditioning in influencing implicit attitudes towards mature brands (e.g., Coke and Pepsi). It was shown that the consistent pairing of positive stimuli with a particular brand could help create and strengthen positive attitudes towards that brand, although the effect was observed only for subjects who had relatively neutral atti- tudes towards both brands to begin with. Olson and Fazio (2001; 2002) reported similar conditioning effects in which frequent pairings between novel conditioned stimuli (CS) and valenced unconditioned stimuli (US) could result in the acquisition of implicit attitudes towards novel target con- cepts that were created a propos for the experiments, and consequently influence subjects’ behaviors and responses on Implicit Association Tests (IAT; Greenwald, McGhee & Schwartz, 1998) involving those novel targets, even though subjects reported no explicit memory of the CS-US pairings. However, the causal mechanisms by which the evaluative conditioning effect could emerge have yet to be satisfacto- rily uncovered, owing in part to conflicting empirical data about the conditions under which such effects might occur (De Houwer, Baeyens & Field, 2005). Many controversies revolve around whether associations were learnt as a result of automatic as opposed to conscious controlled processes, whether evaluative conditioning effects were due to a reper- toire of processes (as opposed to a single mechanism) or contingent on subjects’ awareness of stimuli pairing, and whether the learning is resistant to extinction (De Houwer, 2007; Walther, Weil & Dusing, 2011). This paper represents our attempts at providing a compu- tational account of the effect of evaluative conditioning on the acquisition of automatic associations between concepts in memory. Through simulations, we examine the impact that frequent pairing of target stimuli with various positively or negatively valenced stimuli would have on implicit task performance, such as on the Implicit Association Test. This is done with a number of goals in mind. First, to provide ad- ditional support for the cognitive plausibility of a previously proposed computational model of implicit task performance on the IAT (Quek & Ortony, 2011). Our approach is to augment the localist-connectionist model with a cohesive explanatory account of how automatic associations between concepts in memory could be formed or acquired through experience, a process analogous to how various attitudes are acquired throughout an individual’s lifetime. A second goal is to determine if we could make use of the computational model to address some of the research gaps identified by De Houwer, Baeyens and Field (2005), espe- cially in view of what they see as a lack in the availability of detailed accounts for the processes and mechanisms that un- derlie evaluative conditioning, and the conditions under which it could occur. More generally, and as pointed out by Van Overwalle and Sieber, (2005), there appears to be lim- ited theoretical advancement in the “understanding of the storage or strengthening of attitude-object associations in human memory.” Before more empirical insights are made available, computational approaches such as modeling and simulation could provide an interim but effective means for understanding various candidate processes underlying atti- tude acquisition or formation (e.g., Eiser, Fazio, Stafford &