IDENTITY

Our understanding of identity descends largely from social identity theory (SIT) (Tajfel Reference Tajfel1969, Reference Tajfel1974, Reference Tajfel1982a, Reference Tajfel1982b; Tajfel et al. Reference Tajfel, Billig, Bundy and Flament1971; Tajfel and Turner Reference Tajfel, Turner, Hatch and Schultz2004; Turner Reference Turner1975), and self-categorization theory (SCT) (Turner et al. Reference Turner, Hogg, Oakes, Reicher and Wetherell1987, Reference Turner, Onorato, Tyler, Kramer and John1999; Turner Reference Turner and Tajfel1982, Reference Turner, Ellemers, Spears and Doosje1999; Turner et al. Reference Turner, Oakes, Haslam and McGarty1994).Footnote ¹ Identity, according to Tajfel (Reference Tajfel1974, 69) is “that part of an individual's self-concept which derives from his knowledge of his membership of a social group (or groups) together with the emotional significance attached to that membership.” Building upon SIT, SCT focuses more directly on the complex interaction between the self and group identities. As Turner and Onorato (Reference Turner, Onorato, Tyler, Kramer and John1999, 20–21) describe it: “The basic process postulated is self-categorization, leading to self-stereotyping and the depersonalization of self-perception.”

A more recent entrant, BIT was developed with the rather lofty goal of creating a “unified theory of implicit attitudes, stereotypes, self-esteem, and self-concept” (Greenwald et al. Reference Greenwald, Banaji, Rudman, Farnham, Nosek and Mellott2002, 3). BIT presents social knowledge as a complex collection of associations. An individual's self features associational links to many group or category objects. Self-esteem and assessment of the qualities of these objects (including the self) are represented by associations with other concepts or categories (e.g., mother, scientist, 49ers fan) and evaluative properties (negative and positive valence). The associations vary in strength, and these strengths are expected to correlate in ways that reflect consistency throughout the social knowledge structure (SKS). For example, the strength of association between “male” and positive or negative valence and “male” with “self” should be consistent with the extent to which one associates “self” with positive or negative valence. The self is the centerpiece of BIT's social knowledge structure. All three approaches to identity, but especially SCT and BIT, focus on the centrality of the self-concept in social cognition. While it should not be thought of as a departure from SIT and SCT, BIT does differ in key ways that are important to the topic of this article:

These distinctions make BIT a rich source of theoretical and methodological leverage for scholars assessing identity. The practical definition of identity that emerges, which is the feature of BIT most relevant to this article, is far less burdensome than some others. Simply put: when one associates the self with a group or category, that is an identity. The strength of that association is the intensity of identification. It should be noted that this parsimonious definition, while most relevant to the approaches and methods discussed here, is but one among many features of BIT.

The explicit scales traditionally used to measure identity often feature survey items that depend on rather specific definitions and measure self-reports of downstream consequences of identity. For example, the identification with a psychological group (IDPG) scale developed by Mael and Tetrick (Reference Mael and Tetrick1992), includes items such as: “When someone criticizes this group, it feels like a personal insult,” and “I have a number of qualities typical of members of this group,” and “The limitations associated with this group apply to me also.” But, as with many outcomes, these may be subject to substantial heterogeneity in their expression. One might imagine, for instance, an individual who identifies with a group, but does not necessarily internalize the limitations attributed to that group. Should this be taken as an indication of weaker identity? Should we really treat each such self-reported outcome of identity as part of the way we measure its presence in the respondent? And, should we weight each of these items equally? These are questions of measurement that one avoids when thinking of identity as the simple association between a group and the self in the way that BIT does. This association is, after all, what we believe to be at the heart of the phenomena measured by explicit scales. If we have the ability to measure the association directly, why not do so? This is where implicit measures enter the discussion. The IAT, in particular, is ideally suited to measure this sort of association. As Greenwald et al. (Reference Greenwald, Banaji, Rudman, Farnham, Nosek and Mellott2002, 8) point out: “First, some of the associative links of SKS may not be available to introspection and may therefore not permit accurate assessment by self-report measures (cf. Greenwald and Banaji Reference Greenwald and Banaji1995). Second, self-report measures are susceptible to artifacts (such as impression management and demand characteristics) that can distort reporting even of associations that are introspectively available.”

THE IMPLICIT ASSOCIATION TEST AND IDENTITY

The IAT is rooted in two relatively simple premises: (1) it takes more in the way of processing to perform a task that conflicts in some way with the associations already established in one's mind, and (2) tasks that are more processing intensive take longer (Donders Reference Donders1969). In fact, these statements are the basis of all implicit measures based on response latency.Footnote ² Use of the IAT, in particular, has exploded in social psychology and many other fields in the last decade.Footnote ³

The IAT generates a measure of relative association by having respondents rapidly classify stimuli presented to them on a monitor. The computer-based task typically includes attribute (e.g., good and bad) and target concepts (e.g., black and white). Each of these is represented by related words or images that serve as exemplars. The instructions are the key to the task and define a series of blocks. Each block has its own instructions, which ask the respondent to categorize the attributes and targets in different combinations. Using the good/bad and black/white example, a given block may ask respondents to press one key with their left hand for any good or white exemplars and to press another key with their right hand for any bad or black exemplars. In this case, good is associated with white and bad associated with black. Another block instructs respondents to press one key when presented with bad or white exemplars and press another key for good or black exemplars. The exemplars appear in the middle of the screen in rapid succession and respondents are asked to press the assigned buttons for each block accordingly. Typically, a red “X” indicates to a respondent that he or she has made an incorrect classification. After a series of such blocks, researchers have response latency averages for each paired comparison. The presumption, again, is that respondents act more quickly when the instructions match the associations in their minds. Subtracting the average response time for blocks with one type of association from the response time for blocks with the other as part of a bounded version of Cohen's d generates a single relative measure of associational direction and intensity.

The identity IAT differs from the application just described in that it measures identity rather than an attitude or evaluation. To do this, we replace the attribute concept normally used in IATs with “self,” and we make the identity of interest (e.g., Democrat or Republican) the target concept (Devos and Banaji Reference Devos and Banaji2005; Greenwald and Farnham Reference Greenwald and Farnham2000; Nosek, Banaji, and Greenwald Reference Nosek, Banaji and Greenwald2002). Respondents are instructed to associate terms such as “I,” “me,” “mine,” and “they,” “theirs,” and “them” with images or words representing the groups in question. Identity IATs are also very compatible with use of the brief IAT (BIAT) approach (Sriram and Greenwald Reference Sriram and Greenwald2009). The underlying principles of this abbreviated version of the method are the same as the standard IAT, but the procedure is altered to substantially decrease the length of the task. In the case of identity, this means that the instructions ask respondents to focus on associations with “self” as opposed to associations with “other.” This omission is not especially costly from a measurement perspective because associations of “self” with a group or social category have proven more reliable than similar associations with “other” (Sriram and Greenwald Reference Sriram and Greenwald2009). Figure 1 gives an example of a BIAT block in which respondents are told to associate “self” pronouns with Democratic Party images.

Figure 1

Example Democratic Identity IAT Block

These screen captures show examples of the images presented to respondents during a brief IAT block in which they are instructed to associate “self” with Democratic images. Figure 1(a) shows the instructions provided to subjects as they begin the task.

IMPLICIT PARTY IDENTITY

Party identification (PID) in the United States is an attachment that matches up nicely with the identity IAT because of both its theoretical roots and the fact that it has been increasingly conceptualized as a social identity (Green, Palmquist and Schickler Reference Green, Palmquist and Schickler2002; Greene Reference Greene1999, Reference Greene2000, Reference Greene2004; Huddy, Mason, and Aarøe Reference Huddy, Mason and Aarøe2010; Nicholson Reference Nicholson2012). When it comes to PID, the standard measurement has become the de facto definition of the concept for political scientists. That definition is the seven-point “Michigan” scale emerging from the traditional two-item survey measure. In the end, voters are broken down into “strong” partisans, “not so strong” partisans, “leaners,” and “pure independents.” But what is the underlying concept? For guidance, recall the objective of those who developed the measure. Campbell et al. (Reference Campbell, Converse, Miller and Stokes1960, 121) came about as close as one could to describing PID in terms of social identity theory before that theory had been put forward:

The identity IAT discussed here is well suited to measure the psychological, affective orientation proposed by the Michigan scholars. The differences in latency measured by the IAT are influenced by consistency (or lack thereof) between fast affective reactions and slower conscious reactions. When these two do not match, the expectation is that the task in question takes slightly longer to complete. Smith and Nosek (Reference Smith and Nosek2011, 300) “suggest that, although explicit evaluations can be meaningfully parsed into affective and cognitive components, implicit evaluations are more related to affective than cognitive components of attitudes.” Indeed, the key feature of implicit measures that distinguishes them from explicit ones is that they do not require introspection on the part of the respondent. Introspection may be the entry point for social desirability bias or reconceptualization of partisan intensity in terms of spatial proximity as opposed to visceral identification. Stated in terms of the Cunningham, Zelazo, Packer, and Van Bavel (Reference Cunningham, Zelazo, Packer and Van Bavel2007, 748) iterative reprocessing model, implicit measures may tap into initial (often valence based) iterations in neural processing: “when a Democrat is conflicted about his marriage to a Republican (or visa [sic] versa), he can re-represent the relationship at a higher-level of analysis—they both share a passion for the American political system. As the computations become increasingly complex, additional explanatory factors can be created to organize and make sense of the factors at the first level.” Each succeeding iteration has the potential to add noise if what you want to measure is the first-level “affective orientation.” Given that this is our purpose, as handed down from both Campbell et al. (Reference Campbell, Converse, Miller and Stokes1960) and later explicit conceptualizations of PID as a social identity, the IAT provides a new way of excluding later iterations. In the future, neuroimaging may provide even more effective ways of doing this.Footnote ⁴ But, neuroscience has not yet reached a level of familiarity with structure and function to allow the necessary reverse inferences, and the technology is not yet suitable for the necessary large-N studies (Theodoridis and Nelson Reference Theodoridis and Nelson2012).

In the case of political party, the identity IAT measures the extent to which a respondent's conceptualization of self is cognitively linked to a party. Not only is this link interesting from a measurement perspective (when compared to traditional explicit measures), but it is a key microfoundational element of the self-esteem based ingroup/outgroup cognitive biases observed for social identities. The existence of such an association might be considered a requirement for the conceptualization of PID as a social identity. This measure should allow us to build on existing work examining the microfoundations of PID (Burden and Klofstad Reference Burden and Klofstad2005; Greene Reference Greene1999, Reference Greene2000, Reference Greene2004; Huddy, Mason, and Aarøe Reference Huddy, Mason and Aarøe2010).

There may well be social desirability considerations pushing respondents toward claiming status as political independents, and this measure would be an effective method for overcoming that. However, this type of discrepancy between the implicit and explicit is not the only benefit of such a measure. This represents an important departure from many other applications of the IAT. For example, when measuring the implicit association between good/bad and black/white or math/humanities and male/female, the enterprise is largely designed to address the fact that respondents may not wish to admit to holding those associations or may not even be aware that they hold them. In the case of party, the standard two-item measure leaves us with a percentage in the low teens of pure independents. While the partisan associations of these respondents may be worth examining (for an example, see Hawkins and Nosek (Reference Hawkins and Nosek2012)), the relatively small number of pure independents in the electorate would make the measure discussed here of limited use if the primary objective were to discern a binary partisan association among those unwilling or unable to admit to it. A broader use is in (1) providing a pure measure of identity defined in its most basic form and (2) applying that measure to learn more about intensity of partisan attachments.Footnote ⁵ The first use relates to basic ways in which identity affects cognition. If we conceptualize identity as the association between self and a group label, the IAT measure presented here can be thought of as a behavioral measure of identity. Response latency is used to determine precisely that relative association in the case of political party. To what extent is my conceptualization of “self” associated in my brain with one party or the other? Furthermore, an association such as this is likely at the heart of many of the mechanisms behind ingroup favoritism and bias. If party and self are closely associated at an affective level in a partisan's processing, an attack on one amounts to an attack on the other. Success for one is tantamount to success for the other.

The identity IAT (especially as applied to partisanship in the United States) is also notable in that it appears less susceptible to some common critiques of the IAT generally. In particular, critics have questioned the interpretation of IAT scores as indicators of racial bias or discriminatory attitudes (Arkes and Tetlock Reference Arkes and Tetlock2004). The use of the IAT to measure the brand of associative identity described in BIT requires far less in the way of assumptions. Also, questions have emerged regarding whether the IAT measures the subject's own attitudes or those he or she has perceived in his or her environment (Karpinski and Hilton Reference Karpinski and Hilton2001). This is often an important distinction when assessing things like racial attitudes or gender stereotypes, but it is more difficult to imagine that the self-to-group associations measured by the identity IAT are largely reflections of environmental perceptions. For the issue of environmental associations to present a problem in this case, one would need a significant number of respondents to believe that others associate them with a group or category with which they do not associate themselves. In the case of party, at least, this is likely not a common phenomenon.

Because of conceptual simplicity, BIT and the identity IAT offer an appealing new approach for those hoping to study the microfoundations of politically relevant identities. Party identity is one newly developed area of application, but there are numerous others in which methods like those discussed here are likely to see increased usage going forward.