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 bias measures were first developed, researchers assumed that responses primarily reflected associations stored in memory. However, subsequent research demonstrated that a variety of cognitive processes besides mental associations influence performance on implicit measures. Capitalizing on this insight, I use a class of formal mathematical models – multinomial processing tree models (MPTs) – to estimate the contribution of multiple processes to responses on implicit measures in order to better understand the cognitive processes related to implicit bias change.
Decades of dual-process social cognitive theory posits an interplay between associative and control-oriented processes in producing judgments and behaviors. Implicit measures were designed to minimize the influence of control-oriented processes, which would suggest that interventions to reduce implicit biases by strengthening control will be ineffective. To test this possibility, my colleagues and I applied an MPT model to implicit association test (IAT) responses from over 21,000 participants who had completed one of 18 bias-reduction interventions (Calanchini, Lai, & Klauer, 2020). Our findings challenge perspectives that assume that implicit biases are uncontrollable: interventions that relied on evaluative conditioning influenced control-oriented processes, and interventions that relied on counterstereotypic exemplars or strategies to override biases influenced both associations and control-oriented processes. These results illustrate the variety of routes through which implicit bias can be reduced.
In addition to identifying the cognitive processes underlying implicit bias change, Calanchini et al. (2020) revealed that interventions that changed associations were much more likely to reduce positive White associations than negative Black associations. This finding converges with existing theory positing the primacy of positive ingroup evaluations versus negative outgroup evaluations in intergroup biases. However, Calanchini et al. (2020) focused only on the implicit biases of White people in the context of race. Building on these findings, my colleagues and I focused on the role of group status in intergroup bias by examining the implicit biases of normatively lower-status social groups across a variety of intergroup domains (Calanchini, Schmidt, Klein, & Sherman, under review). MPT modeling indicated that the outgroup biases of normatively lower-status group members (i.e., Asian, Black, homosexual, older participants) consistently reflected greater contribution of positive outgroup (i.e., White, heterosexual, younger) evaluations than negative ingroup evaluations. This pattern of results challenges traditional assumptions that outgroup favoritism among normatively lower-status groups primarily reflects internalized inferiority, but converges with my own recent research into the cognitive processes related to outgroup favoritism. Based on a sample of over 700,000 participants, spanning eight intergroup domains and 14 nations, my colleagues and I found that the extent to which social groups demonstrate outgroup favoritism depends on how much stigma they experience from the society in which they reside (Essien, Calanchini, & Degner, 2020).
MPT models are more than just a statistical approach; instead, they reflect theoretical assumptions articulated in equation form. Consequently, research using MPT models is positioned to advance the dual-process social cognitive theory upon which they are based. For example, even though initial assumptions proved to be false that responses on implicit measures solely (or even primarily) reflect associations stored in memory, an open question remained whether the processes that jointly contribute to responses are all attitude-related. To test this question, I applied an MPT model to the responses of participants (total N > 52,900) who completed pairs of IATs that varied in conceptual correspondence (Calanchini, Sherman, Klauer, & Lai, 2014). In some cases the two IATs measured constructs that corresponded closely (e.g., race and skin tone evaluations), but in other cases the constructs were conceptually distant (e.g., disability evaluations and gender stereotypes). We correlated MPT parameters across pairs of IATs, and found that two of these parameters – reflecting control processes related to accuracy and inhibition – correlated strongly across IATs, regardless of conceptual correspondence, which suggests that these parameters reflect relatively domain-general processes. In contrast, correlations across IATs for two other MPT parameters – reflecting activated associations and response bias – varied as a function of conceptual correspondence, which suggests that these parameters reflect relatively attitude-specific processes. In related work, I have investigated the associative nature of the processes assumed to underlie implicit social cognition. Early theoretical perspectives posited that responses on implicit measures reflect evaluative associations formed through repeated experience over time and, thus, are slow to form and slow to change. However, my colleagues and I have demonstrated that implicit attitudes can be changed quickly (Cone & Calanchini, 2021) and without direct experience (Smith, Calanchini, Hughes, Van Dessel, & De Houwer, 2019). In both cases, implicit attitude change manifests on MPT parameters assumed to reflect activated associations. Taken together, this work sheds light on the qualitative nature of the processes underlying implicit social cognition, and in doing so is positioned to continue to advance social cognitive theory.
Many people view MPT modeling as an arcane and difficult statistical procedure. In reality, MPT modeling is no more complicated than other modeling approaches, such as structural equation modeling or multi-level modeling. However, unlike these other, better-known modeling approaches, there are very few courses or texts dedicated to MPT modeling. Recognizing this gap in the literature, I have written two articles aimed at making MPT modeling more accessible and encouraging their use. I wrote Calanchini (2020) for a non-expert scientific audience, with the goal of demonstrating how MPT modeling has advanced, and can continue to advance, the field of implicit social cognition. In another article (Calanchini, Rivers, Klauer, & Sherman, 2018), my colleagues and I proposed that MPTs can be used as theoretical bridges to connect research in cognitive and social psychology. Scientists in both research traditions often rely on tasks that are structurally similar – such as the go/no-go task and go/no-go association task – but interpret performance on such tasks very differently. Whereas cognitive psychologists tend to focus on the contributions of control-oriented processes, social psychologists generally focus on the contributions of activated mental associations. In reality, responses on these and many other tasks reflect the joint contributions of both types of processes. MPTs provide a means to quantify the contributions of distinct processes, and in doing so provide a common framework to connect otherwise disparate research traditions.