Evaluation is a Dynamic Process: Moving Beyond Dual System Models

Abstract

Over the past few decades, dual attitude ⁄process ⁄ system models have emerged as the dominant framework for understanding a wide range of psychological phenomena. Most of these models characterize the unconscious and conscious mind as being built from discrete processes or systems: one that is reflexive, automatic, fast, affective, associative, and primitive, and a second that is deliberative, controlled, slow, cognitive, propositional, and more uniquely human. Although these models serve as a useful heuristic for characterizing the human mind, recent developments in social and cognitive neuroscience suggest that the human evaluative system, like most of cognition, is widely distributed and highly dynamic. Integrating these advances with current attitude theories, we review how the recently proposed Iterative Reprocessing Model can account for apparent dual systems as well as discrepancies between traditional dual system models and recent research revealing the dynamic nature of evaluation. Furthermore, we describe important implications this dynamical system approach has for various social psychological domains. For nearly a century, psychologists have sought to understand the unconscious and conscious processes that allow people to evaluate their surroundings (Allport, 1935; Freud, 1930). Building on a model of the human mind rooted in classic Greek philosophy (Annas, 2001), many contemporary psychologists have characterized the mind as possessing discrete processes or systems: one that is evolutionarily older, reflexive, automatic, fast, affective, associative, and the other that is more uniquely human, deliberative, controlled, slow, cognitive, and propositional (see Figure 1). These dual process or system models have been highly influential throughout psychology for the past three decades (Chaiken & Trope, 1999). Indeed, a dual system model of the human mind permeates research in a wide range of psychological domains, such as attitudes and persuasion (Chaiken, 1980; Fazio, 1990; Gawronski & Bodenhausen, 2006; Petty & Cacioppo, 1986; Rydell & McConnell, 2006; Wilson, Samuel, & Schooler, 2000), stereotypes and prejudice (Crandall & Eshleman, 2003; Devine, 1989; Gaertner & Dovidio, 1986; Pettigrew & Meertens, 1995), person perception (Brewer, 1988; Fiske & Neuberg, 1990; Macrae & Bodenhausen, 2000), self-regulation (Baumeister & Heatherton, 1996; Freud, 1930; Hofmann, Friese, & Strack, 2009; Strack & Deutsch, 2004), moral cognition (Greene, Sommerville, Nystrom, Darley, & Cohen, 2001; Haidt, 2001), learning and memory (Smith & DeCoster, 2000; Sun, 2002), and decision-making (Kahneman, 2003; Sloman, 1996). Although dual system models provide generative frameworks for understanding a wide range of psychological phenomenon, recent developments in social and affective neuroscience suggest that the human evaluative system, like most of cognition, is widely distributed and highly dynamic (e.g., Ferguson & Wojnowicz, 2011; Freeman & Ambady, Social and Personality Psychology Compass 6/6 (2012): 438–454, 10.1111/j.1751-9004.2012.00438.x a 2012 Blackwell Publishing Ltd 2011; Scherer, 2009). Integrating these advances with current attitude theories, we review how the recently proposed Iterative Reprocessing Model (Cunningham & Zelazo, 2007; Cunningham, Zelazo, Packer, & Van Bavel, 2007) can account for apparent dual systems as well as discrepancies between traditional dual system models and recent research revealing the dynamic nature of evaluation. The model also address why the nature of evaluative processing differs across people (e.g., Cunningham, Raye, & Johnson, 2005; Park, Van Bavel, Vasey, & Thayer, forthcoming). Although we focus primarily on dual models of attitudes and evaluation due to space constraints, we believe the premises of our dynamic model can be generalized to other domains where dual system models of typically invoked (Chaiken & Trope, 1999), including social cognition, self-regulation, prejudice and stereotyping, and moral cognition. Therefore, we very briefly discuss the implications of our model for these other domains in the final section of this paper and encourage interested readers to read our more extensive treatment of these issues in the domain of stereotypes and prejudice (Cunningham & Van Bavel, 2009a; Van Bavel & Cunningham, 2011) and emotion (Cunningham & Van Bavel, 2009b; Kirkland & Cunningham, 2011, forthcoming). Attitudes and evaluation Attitudes are one of the most central constructs in social psychology, yet there has been considerable debate regarding the most fundamental aspects of attitudes (Fazio, 2007; Schwarz & Bohner, 2001). Allport (1935) defined an attitude as ‘‘a mental and neural state of readiness, organized through experience, exerting a directive or dynamic influence upon the individual’s response to all objects and situations with which it is related’’ (p. 810). Throughout the history of attitude research, theorists and researchers have attempted to provide a complete yet parsimonious definition of this construct. Wellknown examples include the one-component perspective (Thurstone, 1928), the tripartite model (Affective, Behavior, Cognition; Katz & Stotland, 1959; Rosenberg & Hovland, 1960), and more recently, a host of dual attitudes (e.g., Greenwald & Banaji, 1995; Rydell & McConnell, 2006; Wilson et al., 2000) and dual process models (e.g., Chaiken, 1980; Fazio, 1990; Gawronski & Bodenhausen, 2006; Petty & Cacioppo, 1986). It is widely assumed that attitudes are stored associations between objects and their evaluations, which can be accessed from memory very quickly with little conscious effort Figure 1 Illustrative example of the process and content of a dual system model (cited in Kahneman, 2003, p. 698). Evaluation is a Dynamic Process 439 a 2012 Blackwell Publishing Ltd Social and Personality Psychology Compass 6/6 (2012): 438–454, 10.1111/j.1751-9004.2012.00438.x (Fazio, 2001; Fazio, Sanbonmatsu, Powell, & Kardes, 1986; but see Schwarz, 2007). For example, people categorize positive and negative words more quickly when these words are preceded by a similarly valenced stimuli, suggesting that attitudes are automatically activated by the mere presence of the attitude object in the environment (Fazio et al., 1986). Moreover, people may have access to evaluative information about stimuli prior to their semantic content (Bargh, Litt, Pratto, & Spielman, 1989; but see Storbeck & Clore, 2007). Such research has led to the conclusion that the initial evaluative classification of stimuli as good or bad can be activated automatically and guide the perceiver’s interpretation of his or her environment (Houston & Fazio, 1989; Smith, Fazio, & Cejka, 1996). Dual attitudes and dual process models of attitudes The recent development of a wide variety of implicit attitude measures (Petty, Fazio, & Briñol, 2009; Wittenbrink & Schwarz, 2007), including measures of human physiology (Cunningham, Packer, Kesek, & Van Bavel, 2009), has fueled an explosion of research on dual attitude ⁄process ⁄ system models of attitudes and evaluations (see Table 1). Most of these models infer dual process architecture from observable dissociations between implicit and explicit measures of behavior (e.g., Dovidio, Kawakami, & Gaertner, 2002; McConnell & Leibold, 2001; Rydell & McConnell, 2006). Although many dual models generally share a common set of assumptions about the human mind, the specific features of each model differ. Therefore, we propose a rough taxonomy to characterize different classes of these models. ‘‘Dual attitudes models’’ tend to dichotomize the representations of attitudes into distinct automatic versus controlled constructs (Greenwald & Banaji, 1995; Rydell & McConnell, 2006; Wilson et al., 2000). In contrast, ‘‘dual process models’’ tend to dichotomize the processing of attitudinal representations into automatic versus controlled processes. There is considerable debate over whether these two types of processes are independent or communicate with one another (i.e., information from one system is available to the other system) (Fazio, 1990; Gawronski & Bodenhausen, 2006; Gilbert, Pelham, & Krull, 1988; Petty, Brinol, & DeMarree, 2007). In the latter case, interdependent dual process models have generally been proposed to operate in a corrective fashion, such that ‘‘controlled’’ processes can influence otherwise ‘‘automatic’’ responses (e.g., Wegener & Petty, 1997). Although dual attitudes models likely require dual processes to integrate different attitudinal representations into evaluations and behaviors, dual process models are less likely to require the assumption of dual attitude representations (e.g., Fazio, 1990). For the purpose of clarity, we use ‘‘dual system models’’ to capture models that assume dual attitudes and processes that do not interact (e.g., Rydell & McConnell, 2006; Wilson et al., 2000). There are, of course, many ways to hook up a dual system (see Gilbert, 1999 for a discussion). A complete discussion of all possible dual models and interconnections between these systems is beyond the scope of this article. Therefore, we focus on several core premises that many models have in common. Likewise, we focus on the core premises from our own model – rather than an exhaustive discussion (e.g., Cunningham et al., 2007) – in order to communicate key similarities and differences between dual models and our proposed dynamic model. Furthermore, we recognize that dual models and our proposed dynamic model do not exhaust all types of models of attitudes and evaluation – some extant models do include more than two processes (e.g., Beer, Knight, & D’Esposito, 2006; Conrey, Sherman, Gawronski, Hugenberg, & Groom, 2005) and many allow for interactive processes that operate in a post hoc, corrective fashion (e.g., Chen & Chaiken, 1999; Gawronski & 440 Evaluation is a Dynamic Process a 2012 Blackwell Publishing Ltd Social and Personality Psychology Compass 6/6 (2012): 438–454, 10.1111/j.1751-9004.2012.00438.x Bodenhausen, 2006; Strack & Deutsch, 2004; Wegener & Petty, 1997). However, few (to our knowledge) articulate how ‘‘controlled’’ processes might influence more ‘‘automatic’’ processes in an a priori fashion. In the following sections, we describe an alternative, dynamic model of evaluation in which a constellation of widely distributed ‘‘automatic’’ and ‘‘controlled’’ processes interact in a dynamic fashion to process evaluations. This model characterizes the human brain as a parallel system that generates evaluations by integrating the results of computations performed by a widely distributed network of component processes (Frank, Cohen, & Sanfey, 2009; McClelland & Rumelhart, 1986; O’Reilly & Munakata, 2000; Rumelhart & McClelland, 1986). We place the terms ‘‘automatic’’ and ‘‘controlled’’ in quotations Table 1 A sample of dual models cited in attitude research arranged in chronological order Model name Core characteristics Total citations References Heuristic versus systematic model Heuristic versus systematic Effortless versus effortful 2202 Chaiken (1980) Central and peripheral routes to persuasion Central versus peripheral 4192 Petty and Cacioppo (1986) Automatic and controlled components of stereotyping and prejudice Automatic versus controlled Unintentional versus intentional Spontaneous versus deliberative 3363 Devine (1989) MODE model Unconscious versus conscious Spontaneous versus deliberative ⁄ reasoned 1424 Fazio (1990) Implicit social cognition Implicit versus explicit Automatic versus controlled 2660 Greenwald and Banaji (1995) Two systems of reasoning Associative versus Rule-based Deliberative ⁄ analytic versus automatic 1627 Sloman (1996) Flexible Correction Model 302 Wegener and Petty (1997) Dual-process models Quick versus slow Effortless versus effortful Unconscious versus conscious Associative versus rule-based 610 Smith and DeCoster (2000) Model of dual attitudes Implicit versus explicit Automatic versus controlled 1209 Wilson et al. (2000) Reflective and impulsive determinants of social behavior Reflective versus impulsive 869 Strack and Deutsch (2004) Predictive model of implicit and explicit attitudes Implicit versus explicit Automatic versus controlled Unconscious versus conscious Effortless versus effortful 130 Perugini (2005) Dual system of reasoning Implicit versus explicit Fast versus slow Spontaneous versus deliberative Associative versus rule-based 131 Rydell and McConnell (2006) APE model Implicit versus explicit Automatic versus deliberative Associative versus propositional 566 Gawronski and Bodenhausen (2006) Total citations obtained from Google Scholar on February 9, 2012 represent a proxy for scientific impact (i.e., actual number of scientific citations may be more or less than the number of citations reported here). Evaluation is a Dynamic Process 441 a 2012 Blackwell Publishing Ltd Social and Personality Psychology Compass 6/6 (2012): 438–454, 10.1111/j.1751-9004.2012.00438.x because we believe they do not reflect bona fide attitudes, processes or systems, but merely a useful heuristic to describe the number and nature of the particular interactive component processes currently involved in evaluation (Cunningham & Johnson, 2007). The iterative reprocessing model of evaluation We argue that traditional conceptions of ‘‘automatic’’ and ‘‘controlled’’ processes as separable bona fide systems should be replaced by models that invoke the integration of multiple dynamic computational processes. In this paper, we describe one such approach – the Iterative Reprocessing (IR) Model (Cunningham & Zelazo, 2007; Cunningham et al., 2007). A fundamental assumption of the IR Model is that evaluative processes involve a series of iterative cycles: with every iteration, the current evaluation of a stimulus can be adjusted in light of additional contextual and motivational information to create an updated evaluation in line with finer stimulus detail, the context, and ⁄or current goals. As such, the IR Model shares two important features with dual system ⁄process models. First, stimuli evoke rapid perceptual and evaluative responses, and second, perceivers can become aware of these initial responses, and, with the right motivation and opportunity, modulate or elaborate upon them (for a more complete discussion of the similarities and differences, please see Cunningham & Zelazo, 2007; Cunningham et al., 2007). More pertinent to the current discussion, the IR Model differs from most dual process ⁄ system models in the following ways. First, in contrast to dual attitude models (e.g., Rydell & McConnell, 2006; Wilson et al., 2000), the IR Model does not assume distinct (implicit versus explicit) attitudinal representations stored in memory. Rather, the IR model characterizes evaluation as an emergent property of multiple processes that unfold over time. As such, differences in evaluation are largely due to differences in information processing, rather than qualitatively different attitudinal representations stored in discrete memory systems. Distinguishing between attitudes as relatively stable representations, and evaluations as the current state of evaluative processing, requires an explanation of how attitudes are transformed into evaluations. The IR model resolves this issue by proposing a connectionist framework in which attitudes are represented as (relatively stable) unit weights, whereas evaluations reflect the current pattern of activation of the units (Cunningham et al., 2007). In the context of evaluation, the weights can be conceptualized as having valence and intensity. Thus, what many dual process ⁄ system models term ‘‘controlled’’ processing may represent a change in the current activation pattern, and not necessarily in the unit weights (which can be activated easily, but generally change more slowly). This distinction allows for the apparent stability (i.e., the weights are relatively stable) and occasional flexibility (i.e., the patterns of activation are relatively variable) of implicit attitude measures without requiring distinct implicit versus explicit attitudinal representations. Imagine someone’s attitude towards members of a racial minority. The weights may be conceptualized as having a negative valence with modest intensity. When this perceiver encounters a racial minority member, his ⁄her evaluation (i.e., the current activation pattern) may be easily influenced by factors such as current context, motivation and goals. For example, even given relatively stable unit weights, the activation pattern regarding an other-race member may differ depending on whether the target belongs to the same team as the perceiver’s (Van Bavel & Cunningham, 2009a). As such, evaluations may shift radically even if the underlying attitude toward the other race has not. In sum, evaluations are constructions consisting of relatively dynamic activation patterns that are sensitive to 442 Evaluation is a Dynamic Process a 2012 Blackwell Publishing Ltd Social and Personality Psychology Compass 6/6 (2012): 438–454, 10.1111/j.1751-9004.2012.00438.x shifting contextual and motivational influences, and consist of a subset of input units with relatively stable weights, rather than being a veridical instantiation of our internal attitudes. Second, in contrast to most dual process ⁄ system models, the IR Model does not assume only two bona fide evaluative systems at work in the human brain (e.g., Gawronski & Bodenhausen, 2006; Smith & DeCoster, 2000; Strack & Deutsch, 2004). Instead, building on recent research on the functional anatomy of the human brain, the IR Model proposes that there are many highly interactive neural systems engaged in information processing (see Figure 1 for a simplified version of the model). Importantly, information can propagate forward or backward through the system, meaning that evaluative processes are part of an iterative cycle that unfolds in a dynamic fashion over time and is mutually constrained by so-called bottom-up and top-down influences (termed bidirectional excitation), until the network eventually settles into a stable state. As shown in Figure 2, brain regions such as the amygdala, ventral striatum, and posterior orbitofrontal cortex are normally engaged in initial evaluative processing of stimuli. The thalamus, sensory cortices, and bodily states provide input into initial and subsequent iterations – even within a time period that is typically considered ‘‘automatic’’ (i.e., a few hundred milliseconds). As information about the stimulus is reprocessed, higher-order brain regions, such as the PFC, influence the evaluation, and also reseed initial evaluative processing Amygdala Hypothalamus Physiological response Insular Cortex Thalamus OFC ACC DLPFC/ VLPFC RLPFC

Extracted Key Phrases

4 Figures and Tables

Cite this paper

@inproceedings{Bavel2012EvaluationIA, title={Evaluation is a Dynamic Process: Moving Beyond Dual System Models}, author={Jay J. Van Bavel and Yi Jenny Xiao and William A. Cunningham}, year={2012} }