Commentary on Gawronski, Ledgerwood, and Eastwick, Implicit Bias ≠ Bias on Implicit Measures

Abstract

The authors of the target article offer a definition of implicit bias as “unconscious effects of social category cues” (Gawronski, Ledgerwood, & Eastwick, this issue, p. 140), and, so defined, make a case for examining its causes, effects, and possible amelioration. We support this pursuit and offer some suggestions on how that might be accomplished. For years we have argued that however implicit bias might be defined, to equate it to the output of a measure that one happens to also call “implicit” or is a bad idea (e.g., Fazio & Olson, 2003; Olson & Fazio, 2009; Olson & Gill, 2022; Olson & Zabel, 2016). Indeed, we wrote nearly twenty years ago, “We would encourage researchers not to equate an implicitly measured construct with an unconscious one” (Fazio & Olson, 2003, p. 303). Since then, evidence has accumulated that bias on implicit measures is not implicit in the sense of being inaccessible to consciousness. For example, in 2007 we showed that implicitly measured self-esteem was consciously accessible and hence reportable on explicit measures when respondents were implored to be honest (Olson, Fazio, & Hermann, 2007). Similarly, implicitly-assessed antiBlack bias correlates with explicitly-assessed anti-Black bias under conditions of honesty and anonymity (Phillips & Olson, 2014; see also Hahn, Judd, Hirsh, & Blair, 2014). Nevertheless, we see the need for the authors’ treatise on the problems of conflating implicit bias with bias on an implicit measure, as prominent researchers in these domains persist in equating the two (Greenwald et al., 2022). However, and despite social scientists’ proliferation of near-synonyms, we also want to make a case that a focus on automaticity over implicitness with regard to implicit measures (and, as we will see, implicit bias) has a strong theoretical foundation and empirical support. Before the term “implicit” was popularized and applied to prejudice or attitude measurement, Fazio and colleagues (e.g., Fazio, Sanbonmatsu, Powell, & Kardes, 1986) were investigating the automatic activation of attitudes. The evaluative priming measure they developed is probably the second most-used implicit measure, after the IAT. On a given trial in a priming task, a prime (the attitude object, usually in image form) is presented briefly, followed by a clearly valenced target adjective (e.g., wonderful) participants are tasked with identifying as either good or bad by pressing one of two corresponding keys as quickly as possible. This seminal work found that for particularly strong attitudes, primes facilitate the identification of valence-congruent target adjectives: cake primes facilitated identification of positive targets, and death primes facilitated identification of negative targets (inhibition of valence incongruent prime-target pairs was also observed). This facilitation effect is automatic because the activation of participants’ attitudes occurred despite their goal to identify the valence of the targets, not the primes; it was thus unintentional but also efficient insofar as it happened very quickly. Fazio and colleagues replicated this essential effect with respect to racial prejudice in a later paper (Fazio, Jackson, Dunton, & Williams, 1995). A few years later, Greenwald and colleagues published their first article on the IAT (Greenwald, McGhee, & Schwartz, 1998). Much has been written about qualities of automaticity as applied to the IAT, but our sense is that IAT effects are not so much about the automatic activation of attitudes themselves as much as the controllability of those attitudes with regards to the forced categorization of two objects whose valence differ (e.g., for many Americans, Black and positive) but yet are assigned to the same response key. Nevertheless, and despite our suspicion that they tap somewhat different aspects of automaticity, priming and the IAT correlate substantially under specific conditions (see Olson & Zabel, 2016, for a review). Phillips and Olson (2014) also demonstrated that race bias as revealed on the IAT is experienced as spontaneously activated affect. Thus, at least with respect to the most commonly employed of them, implicit measures appear to reveal automatic properties of attitudes, which is why we have an affinity for the term “automatic prejudice” or, in keeping with fashion, “automatic bias.” Evidence of automatic activation of attitudes stimulated theoretical work that culminated in the MODE model of attitude-behavior relations, which articulates when and how attitudes with automatic properties exert impact—including impact about which people might be unaware—on judgments and behavior (Fazio, 1990; Fazio & Olson, 2014). According to the MODE model, and as demonstrated by researching employing the priming measure, attitudes of sufficient strength are likely to be automatically-activate upon perception of the attitude object (e.g., a member of a minoritized group). Once activated, such attitudes set in motion a cascade of processes that can culminate in a spontaneous attitudebehavior relation: an individual who harbors anti-Black bias