To begin a tour of research on implicit bias, the construct must be defined conceptually and operationally, and Section 1 does just that. As we shall see, the accumulated literature has been characterized by definitional divergences that merit investigation and resolution.

A primary goal of prejudice and stereotyping research is to reduce intergroup disparities arising from various forms of bias. For the last thirty years, much, perhaps most, of this research has focused on implicit bias as the crucial construct of interest. There has been, however, considerable confusion and debate about what this construct is, how to measure it, whether it predicts behavior, how much it contributes to intergroup disparities, and what would signify successful intervention against it. We argue that this confusion arises in part because much work in this area has focused narrowly on the automatic processes of implicit bias without sufficient attention to other relevant psychological constructs and processes, such as people’s values, goals, knowledge, and self-regulation (Devine, 1989). We believe that basic research on implicit bias itself is important and can contribute to reducing intergroup disparities, but those potential contributions diminish if and when the research disregards controlled processes and the personal dilemma faced by sincerely nonprejudiced people who express bias unintentionally. We advocate a renewed focus on this personal dilemma as an important avenue for progress.

In this section, we reassess the value of explicit prejudice measures. P.J. Henry starts this discussion by reviewing critiques of implicit prejudice measures and points to the overwhelming evidence of the power of explicit measures to predict important outcomes. To date, implicit measures have not yet been shown to be similarly capable. Henry explains how the “implicit revolution” was founded on the claim that explicit measures are useless, yet this is clearly not so.

Researchers in cognitive psychology have proposed that there are two distinct cognitive systems or dual processes underlying reasoning: automatic (implicit) processing and effortful (explicit) processing. Multiple measures have since been developed to capture implicit attitudes. However, do these new measures truly capture implicit attitudes? And can these implicit measures be used interchangeably? To answer this question, we investigated the differences between two of the most popular implicit attitudes measures used in the study of political behavior, the Implicit Association Test (IAT) and the Affect Misattribution Procedure (AMP). We examined data from an original survey experiment investigating gender attitudes and a nationally representative survey that measured racial attitudes. We found that it is important to consider implicit measures alongside explicit measures, as they are not redundant measures. However, when implicit attitudes are measured with the IAT, our inferences are more consistent with predictions of dual process accounts. Moreover, the IAT picks up out-group bias in a way that the AMP does not. The two studies point to the presence of significant differences between different types of implicit measures, and a need to reconsider how implicit attitudes are measured.

The concept of unconscious bias is firmly entrenched in American society, yet evidence has accumulated in recent years questioning widely accepted claims about the phenomenon, including assertions that it can be measured reliably, influences behavior and is susceptible to intervention. We adopt a two-pronged approach to investigating the state of affairs: First, assessing claims made about unconscious bias in the public sphere; and second, conducting a national public opinion survey – the first of its kind, to the extent we can ascertain – designed to measure public understanding of unconscious bias. Results show that broad majorities of Americans think unconscious biases are prevalent, influence behavior and can be mitigated through training. Confidence in its accurate measurement is lower. The public sees unconscious biases as more prevalent than biases that are consciously held, and as worthy of mitigation efforts by businesses and government. Our chapter assesses these attitudes and understandings and compares them with the state of the science on unconscious bias.

In this chapter we identify scientific gaps research to date regarding the ability of IAT scores to explain real world racial gaps. We use the term “IAT scores” rather than “implicit bias” because, as we show: (1) Implicit bias has no consensual scientific definition; (2) A definition offered by Greenwald (2017) is shown to be logically incoherent and empirically unjustified; (3) Exactly what the IAT measures remains unclear. Nonetheless, meta-analyses have shown that IAT scores predict discrimination to a modest extent. Alternative explanations for gaps are briefly reviewed, highlighting that IAT scores offer only one of many possible such explanations. We then present a series of heuristic models that assume that IAT scores can only explain what is left over, after accounting for other explanations of gaps. This review concludes that IAT scores probably explain a modest portion of those gaps. Even if the IAT captures implicit biases, and those implicit biases were completely eliminated, the extent to which racial gaps would be reduced is minimal. We conclude by arguing that, despite its limitations, the IAT should not be abandoned, but that, even after twenty years, much more research is needed to fully understand what the IAT measures and explains.

On average, Black Americans’ health is poorer than that of White Americans. We examine three pathways by which implicit racial bias may contribute to racial health disparities. First, implicit and explicit racial bias cause racial discrimination, producing chronic stress and limited access to resources among Black targets of discrimination. This directly and negatively affects their health. This pathway has substantial empirical support. Second, physician implicit racial bias negatively affects treatment recommendations to Black patients, causing racial health disparities. Although intuitively appealing, currently there is little empirical support for this pathway. Third, physician implicit racial bias negatively influences the quality of healthcare interactions with Black patients, causing racial health disparities. This pathway has substantial empirical support. We conclude by highlighting differences in the ways social cognition and applied health disparity researchers study implicit racial bias, and make an argument for the benefits of dialogue and mutual collaborations between these two groups.

Recent findings show that it is possible in some cases to robustly and durably change implicit impressions of novel individuals. This work presents a challenge to long-standing theoretical assumptions about implicit impressions, and raises new research directions for changing and reducing implicit bias toward outgroups. Namely, implicit impressions of newly encountered individuals and groups are more amenable to robust change and updating than previously assumed, and some of the lessons from this work point to when and how we might try to change implicit bias toward well-known and familiar stigmatized groups and individuals.

There are widespread assumptions that implicit group bias leads to biased behavior. This chapter summarizes existing evidence on the link between implicit group bias and biased behavior, with an analysis of the strength of that evidence for causality. Our review leads to the conclusion that although there is substantial evidence that implicit group bias is related to biased behavior, claims about causality are not currently supported. With plausible alternative explanations for observed associations, as well as the possibility of reverse causation, scientists and policy makers need to be careful about claims made and actions taken to address discrimination, based on the assumption that implicit bias is the problem.

In this chapter, we examine the public’s understanding of implicit bias, a topic that has only recently caught the public’s attention. Given that political elites often set the contours of debate on political issues, we begin by conducting a systematic content analysis of newspaper headlines and cable news transcripts to assess the prevalence and nature of media coverage of implicit bias. We find that partisan media utilize starkly different frames regarding the scientific validity of the concept of implicit bias, the political intentions of those who use the phrase, and the requisite political recourses (if any). We then utilize two individual-level datasets to examine the mass public’s understanding of implicit bias. An original survey reveals a stark gulf in partisan understandings of implicit bias and an analysis of Project Implicit data highlights the interplay between personalized feedback from the IAT and ideology in shaping evaluations of the IAT. We conclude with a discussion of the challenges of science communication, particularly on issues relating to race, in a politically polarized age.

Despite twenty years of research, we have not yet reached a point of consensus about what might be considered the most important issue in the study of implicit bias: when and how strongly does it shape cognition and behavior? This section of this handbook reviews some of the relevant literature.

Recent decades have seen a series of attempts to further develop measures of implicit bias. Some observers have suggested drawing on lessons learned in the literature on optimal measurement of explicit bias to enhance implicit bias measures. Suggestions have also been made about how to improve meta-analyses of studies quantifying the strength of the link between implicit attitudes and behavior. For example, outdated statistical methods used in many meta-analyses of implicit bias may have led to incorrect inferences about the average effect sizes and can be avoided using newer techniques. Further improvement has been suggested to more effectively take into account omitted variables that may create spurious associations of implicit attitudes and behavior.

I highlight three issues pertaining to the Implicit Association Test (IAT). First, using the test’s documented validity estimates, I show that using the IAT to classify individuals can result in lower adherence to a benchmark of rationality than using a blatantly unfair categorization scheme. I also suggest that using base rates to classify people when negligible individuating information is available is rational. In fact, people use racial base rates when executing their own classification strategy but denigrate other people for doing so. Second, I emphasize the very tenuous relation between one’s IAT score and dependent variables such as medical therapy choices which can be influenced by multiple factors other than prejudice. Third, I question the use of the IAT as a basis for deeming a person to be implicitly racist and therefore ineligible to be hired or in need of “diversity training” whose benefits have yet to be established.