Salience, Cognitive Control, and Social Behavior

The interaction of salience and cognitive control is an enduring area of interest in the SCANlab, going back to Dr. Bartholow’s undergraduate days. In his undergraduate senior honors thesis, Dr. Bartholow found that participants asked to read résumés later recalled more gender-inconsistent information about job candidates. This general theme carried through to Dr. Bartholow’s dissertation research, in which he used event-related brain potentials (ERPs) to examine the neurocognitive consequences of expectancy violations. In that study, expectancy-violating behaviors elicited a larger P3-like positivity in the ERP and were recalled better compared to expectancy-consistent behaviors (Bartholow et al., 2001, 2003). Back then, we interpreted this effect as evidence for context updating (the dominant P3 theory at the time). As theoretical understanding of the P3 has evolved, we now believe this finding reflects the fact that unexpected information is salient, prompting engagement of controlled processing (see Nieuwenhuis et al., 2005 ).

Our research has been heavily influenced by cognitive neuroscience models of the structure of information processing, especially the continuous flow model (Coles et al., 1985; Eriksen & Schultz, 1979) and various conflict monitoring theories (e.g., Botvinick et al., 2001; Shenhav et al., 2016). In essence, these models posit (a) that information about a stimulus accumulates gradually as processing unfolds, and (b) as a consequence, various stimulus properties or contextual features can energize multiple, often competing responses simultaneously, leading to a need to engage cognitive control to maintain adequate performance. This set of basic principles has influenced much of our research across numerous domains of interest (see Bartholow, 2010).

Applied to social cognition, these models imply that responses often classified as “automatic” (e.g., measures of implicit attitudes) might be influenced by control. We first tested this idea in the context of a racial categorization task in which faces were flanked by stereotype-relevant words (Bartholow & Dickter, 2008). In two experiments, we found that race categorizations were faster when faces appeared with stereotype-congruent versus –incongruent words, especially when stereotype-congruent trials were more probable. Further, the ERP data showed that that this effect was not due to differences in the evaluative categorization of the faces (P3 latency), but instead reflected increased response conflict (N2 amplitude) due to partial activation of competing responses (lateralized readiness potential; LRP) on stereotype-incongruent trials. A more recent, multisite investigation (funded by the National Science Foundation) extended this work by testing the role of executive cognitive function (EF) in the expression of implicit bias. Participants (N = 485) completed a battery of EF measures and, a week later, a battery of implicit bias measures. As predicted, we found that expression of implicit race bias was heavily influenced by individual differences in EF ability (Ito et al., 2015). Specifically, the extent to which bias expression reflected automatic processes was reduced as a function of increases in general EF ability.

Another study demonstrating the role of conflict and control in “implicit” social cognition was designed to identify the locus of the affective congruency effect (Bartholow et al., 2009), wherein people are faster to categorize the valence of a target if it is preceded by a valence-congruent (vs. incongruent) prime. This finding traditionally has been explained in terms of automatic spreading of activation in working memory (e.g., Fazio et al., 1986). By measuring ERPs while participants completed a standard evaluative priming task, we showed (a) that incongruent targets elicit response conflict; (b) that the degree of this conflict varies along with the probability of congruent targets, such that (c) when incongruent targets are highly probable, congruent targets elicit more conflict (also see Bartholow et al., 2005); and (d) that this conflict is localized to response generation processes, not stimulus evaluation.