Abstract
People can behave in a biased manner without being aware that their behavior is biased, an idea commonly referred to as implicit bias. Research on implicit bias has been heavily influenced by implicit measures, in that implicit bias is often equated with bias on implicit measures. Drawing on a definition of implicit bias as an unconscious effect of social category cues on behavioral responses, the current article argues that the widespread equation of implicit bias and bias on implicit measures is problematic on conceptual and empirical grounds. A clear separation of the two constructs will: (1) resolve ambiguities arising from the multiple meanings implied by current terminological conventions; (2) stimulate new research by uncovering important questions that have been largely ignored; (3) provide a better foundation for theories of implicit bias through greater conceptual precision; and (4) highlight the broader significance of implicit bias in a manner that is not directly evident from bias on implicit measures.
Acknowledgements
We thank Galen Bodenhausen, Jan De Houwer, and Alex Madva for helpful comments on earlier versions of this article. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation.
Notes
1 Our use of the phrase effect of social category cues on behavioral responses should not be taken to mean that the source or cause of a dominant group member’s racist or sexist behavior is located within a marginalized group member (see Fields & Fields, Citation2014). Rather, by defining bias as a behavioral phenomenon, we leave room to ask questions about the explanations for that behavior (as we discuss further in the following pages). We also want to emphasize that the relevant causal force is social category cues at the stimulus level, which may not align with the personal identity of the target (e.g., when someone who identifies as White has stereotypical Afrocentric features).
2 We use the term bias to refer specifically to biases involving social category cues rather than biases in information processing more broadly, the latter of which includes numerous biases that are not directly related to the current question (e.g., hindsight bias, impact bias, etc.).
3 In line with the distinction between bias and error (Kruglanski & Ajzen, Citation1983), the proposed definition treats IB as a behavioral tendency rather than a deviation from a normative criterion of accuracy. A definition referring to accuracy seems problematic, because normative criteria for accurate social perceptions are inherently arbitrary (see Kruglanski, Citation1989). For example, if implicit race bias is conceptualized with reference to actual similarities and differences between Black people and White people, one would have to specify the relevant populations of Black people and White people, which is inherently arbitrary because there is no a priori basis to determine the relevant population (e.g., Black people living in a particular neighborhood, city, county, state, or country). Similar problems arise for normative conceptualizations in terms of rationality, in that (1) criteria for rational judgments involve a reference to goals and (2) normative propositions about goals are inherently arbitrary. For a more detailed discussion of problems with normative conceptualizations of implicit bias, see De Houwer (Citation2019).
4 A closely related issue is that, although the majority of implicit measures have been designed with the goal to measure responses arising from associative representations, a subset of implicit measures have been developed to capture responses arising from representations with propositional content (e.g., Barnes-Holmes et al., Citation2010; Cummins & De Houwer, Citation2019; De Houwer et al., Citation2015).
5 It is worth noting that our discussion of potential mechanisms underlying IB is not meant to be exhaustive, in that IB could result from multiple other mechanisms besides biased interpretation and biased weighting.
6 Consistent with this hypothesis, Hugenberg and Bodenhausen (Citation2003) found that biased interpretations of neutral facial expressions were associated with BIM but not BEM. However, their study did not include any data that could confirm unawareness, rendering interpretations in terms of IB premature.