In this line of research we explore implicit attitude formation, change, and variation across groups. In these investigations, we use formal mathematical models to estimate the contribution of multiple processes to responses on implicit measures. In applying this process-level approach, we seek to guide the development of more effective implicit bias interventions and inform our theoretical understanding of between-groups differences in implicit bias.
When implicit measures were first developed, researchers assumed responses to the measures primarily reflect associations stored in memory. Subsequent research demonstrated that a variety of mental processes besides mental associations influences performance on implicit measures. However, the extent to which these non-associative processes were attitude-related remained an open question. The Quadruple process model (Quad model: Conrey et al., 2005) is well-suited to address this question, in that it estimates the contribution of four qualitatively distinct processes to responses on implicit measures. The four processes include: mental associations between objects and attributes; accuracy orientation; an inhibitory process that intervenes when mental associations conflict with accurate responding; and response biases.
We applied the Quad model to data from pairs of IATs that varied in conceptual overlap to investigate the extent to which implicit measures reflect attitude-related processes as opposed to domain-general processes (Calanchini, Sherman, Klauer, & Lai, 2014). Two processes (i.e., accuracy orientation, inhibition) correlated strongly across IATs, regardless of conceptual overlap, suggesting that these are relatively domain-general processes. In contrast, correlations for the other two processes (i.e., mental associations, response biases) decreased as conceptual overlap decreased, suggesting that they are relatively attitude-specific processes. These findings help to connect the IAT with executive functions, the collection of general-purpose regulatory processes that influence a wide variety of behaviors.
Having established that responses on implicit measures reflect both attitude-related and domain-general processes, we next investigated the extent to which each type of process is influenced by implicit bias-reduction interventions. Since the advent of implicit measures, researchers have devoted considerable resources to developing methods to reduce implicit bias. Interventions targeting attitude-related processes should only impact attitudes toward the relevant attitude object, whereas interventions targeting domain-general processes should be effective across multiple attitude objects. In an initial investigation, we demonstrated that a counter-prejudicial training intervention not only decreased the influence of biased associations, but also increased the influence of accuracy orientation (Calanchini, Gonsalkorale, Sherman, & Klauer, 2013). Building upon this evidence that bias-reduction interventions can influence both attitude-related and domain-general processes, we meta-analyzed IAT data from over 21,000 participants who had completed one of 18 interventions or a baseline control condition (Calanchini, Lai, & Klauer, in prep). Interventions that focused on counter-stereotypic exemplars or strategies to override biases consistently influenced both attitude-related and domain-general processes. Interventions that relied on evaluative conditioning or appeals to egalitarian values had inconsistent effects at the process level. Finally, interventions that focused on inducing emotion and perspective-taking largely did not influence any of the processes we examined. Taken together, these findings indicate that interventions can reduce bias through a variety of routes which, in turn, has important implications for developing new interventions. For example, researchers may be able to maximize the effectiveness of interventions by tailoring them to target cognitive processes that are specifically relevant to different domains (e.g., addiction), populations (e.g., the elderly), or other individual differences.
Not only are implicit attitudes changeable, but they can also vary across groups. For example, most members of high-status groups demonstrate implicit ingroup bias, whereas members of low-status groups are more variable in their implicit in- versus out-group biases. In a series of studies, we sought to understand the contributions of positive and negative evaluations to implicit intergroup bias (Calanchini, Schmidt, & Sherman, under review). Based upon the IAT responses of two participant populations (undergraduates, adults) across three content domains (race, age, sexual orientation), we divided participants into groups according to their demonstrated ingroup- or outgroup-favoring response tendencies. Quad process modeling revealed that the ingroup biases of relatively high-status group members (i.e., White, young, heterosexual) and the implicit outgroup biases of relatively low-status group members (i.e., non-White, old, homosexual) are both characterized by a greater contribution of positive than negative evaluations. In contrast, the implicit outgroup biases of high-status group members and the implicit ingroup biases of low-status group members revealed inconsistent patterns of positive and negative evaluations. These findings demonstrate that the same outcome (e.g., ingroup bias) can belie important differences at the process level, and highlights the importance of understanding phenomena at multiple levels of analysis.
Moving forward with this line of research, we seek to shed light on a puzzling finding in the implicit attitudes literature. Whereas most White Americans demonstrate pro-ingroup implicit racial bias, a third of Black Americans demonstrate pro-ingroup bias, a third demonstrate pro-outgroup bias, and a third demonstrate no bias (i.e., IAT scores near zero). Given the state of race relations and the saliency of race to minority group members, the absence of bias shown by such a large proportion of Black Americans is surprising. However, IAT scores near zero may reflect the presence of strong pro-ingroup and pro-outgroup attitudes rather than neutral attitudes. Because current measures and tools are poorly-suited to distinguish between ambivalence (i.e., strong positive and negative evaluations) and indifference (i.e., weak or neutral evaluations), my colleagues and I are working on a model to simultaneously estimate the influence of conflicting evaluations on implicit measures (Calanchini & Klauer, in prep).