Behavior-Based Assessment in Psychology - Going Beyond Self-Report in the Personality, Affective, Motivation, and Social Domains

[14][15]Chapter 2

Implicit Association Tests, Then and Now

Marco Perugini, Giulio Costantini, Juliette Richetin, and Cristina Zogmaister

Department of Psychology, University of Milan-Bicocca, Italy

One of the ways to understand the importance of a scientific contribution is by looking at how many times it is cited in the scientific literature. The original paper by Greenwald, McGhee, and Schwartz (1998) that presented the Implicit Association Test (IAT), published in the Journal of Personality and Social Psychology (JPSP), has so far been cited 1,900 times (as retrieved from Web of Science, March 26, 2012). Putting this figure into perspective, it is the most cited paper published in JPSP, the second most cited being a subsequent paper by Greenwald and colleagues on an improved scoring algorithm of the IAT (Greenwald, Nosek, & Banaji, 2003), and the fifth most cited paper in the whole field of psychology between 1998 and 2012. There is therefore little doubt that the IAT represents one of the most important developments in the field of psychology during the last 15 years. In this chapter we will first define direct and indirect measures, then present the IAT, discuss some cognitive processes behind its functioning, and briefly review some variants that have appeared in recent years. Adopting a psychometric perspective, the second part of this chapter will deal with issues such as the scoring of the IAT and its reliability and validity. The last part will focus on methodological issues relative to the development and the use of an IAT in a research context. Throughout the chapter our review will provide an overview of what has been done (then), what is the current state of knowledge (now), and what are the potential interesting developments (future).

Direct and Indirect Measures

In this chapter we use the terms direct and indirect to refer to the measures, and explicit and implicit to refer to the constructs. We should, however, clarify that we have modified the definitions provided by De Houwer and Moors (2010). According to the authors:

This definition of a direct measure is problematic from a psychometric perspective because a direct self-assessment of a construct is never possible given that multiple items (questions) are[16] by definition needed to measure a construct. Therefore, criterion 2 can never be respected apart from the trivial, and psychometrically deficient, case of using a single question to measure a construct1. Using this definition virtually no measure in psychology can be classified as direct from a psychometric perspective and the distinction put forward by De Houwer and Moors (2010) would be of little utility. We think that the taxonomic distinction by De Houwer and Moors (2010) is very important but, to increase its usefulness, we propose to modify the definition of a direct measure. We define a direct measure as a measurement procedure that is characterized by (a) a personal evaluation (e.g., questions such as “do you start conversations?” or “do you like chocolate?” requiring answers such as “very often” or “very much”) that is targeted to (b) an attribute (c) that could be included in the definition of the construct that the measurement outcome is assumed to capture (e.g., extraversion, attitude toward chocolate).

The first property (personal evaluation) helps to differentiate a direct measure from a measure such as the IAT. The third property (could be included in the definition of the construct2) helps to differentiate standard questionnaires from measures such as the Name–Letter Task (NLT; Nuttin, 1985) that rely on a personal evaluation but that capture an attribute that would not be used to define the construct. In fact, starting conversations very often or affirming that one likes chocolate very much could be included in the definition of the constructs of extraversion and attitude toward chocolate, respectively. On the contrary, no one would include in the definition of self-esteem the preference for the letter of one’s name. In other words, the critical question here is to ask oneself whether one would use the measured outcome as a potential defining element of the construct: If the answer is no, the measure is indirect. Of course, often this is a continuum that we are dichotomizing only as a means to clarify the property. The second property (an attribute) helps to accommodate the fact that psychological measurement is generally characterized by two levels of abstraction, items and construct (e.g., Edwards & Bagozzi, 2000). Therefore, the measurement outcome is an element (an attribute) related to the construct rather than the construct itself (the attribute).

Using this definition as a benchmark, all measures should ideally have the second property (i.e., they are multi-items or stimuli), direct measures have all properties, whereas indirect measures do not have at least one among the first and the third properties. Moreover, this definition could be useful to further distinguish between different types of indirect measures depending on which of the two differentiating properties are missing. For instance, one could argue that the IAT does not have the first and the third property whereas the NLT has the[17] first but not the third property. In fact, as we will detail later, a typical IAT is a task that is not characterized by a personal evaluation (e.g., it does not require one to express a personal opinion), similar to indirect measures such as the Affective Evaluative Priming (AEP; Fazio, Sanbonmatsu, Powell, & Kardes, 1986). The Affective Misattribution Paradigm (AMP; Payne, Cheng, Govorun, & Stewart, 2005) and NLT instead rely on a personal evaluation (e.g., evaluate as positive or negative Chinese ideograms; evaluate alphabet letters) but the attributes they capture (e.g., preference for Chinese ideograms; preference for a letter) would not normally be used to define the construct (e.g., related to the primes in the AMP, self-esteem in the NLT).

What Is the Implicit Association Test?

The IAT is a paradigm that has been developed for the measurement of psychological constructs through the strength of associative links between concepts. It has been implemented to investigate a broad range of constructs (see the meta-analysis by Greenwald, Poehlman, Uhlmann, & Banaji, 2009). Unlike traditional interviews and questionnaires, in the IAT respondents are not requested to describe their own opinions or attitudes (e.g., by selecting their agreement to a question among several response options) but, rather, these are inferred based on their performance in a series of categorization tasks. Respondents see a series of stimuli appearing on a computer monitor (words or images) that represent two different (typically opposite) concepts and the two polarities of an attribute dimension. For each stimulus, they are required to press one of two different keys of the keyboard, depending on their category membership. For instance, in an IAT aimed to measure prejudice against Blacks, the two concepts could be the social categories Black and White, represented respectively by photographs of Black and White faces, and the attribute could be the positive–negative evaluation, represented by words (e.g., rainbow, rotten). The IAT is structured in different blocks. In the simple categorization blocks, the participants’ task would be to press one key for White and the other for Black faces, or to press one key for positive and another key for negative words. Each stimulus belongs univocally to one category; the categorization task is therefore easy and the presence of an unambiguous relationship between each stimulus and its category is one of the prerequisites of a good implementation of IAT. The task is made more complex by the presence of two double categorization critical blocks, namely, blocks of trials in which exemplars representing the concept or the attributes are to be categorized. Continuing the previous example, in one of the critical blocks one key would be used for White faces and negative words, and the other for Black faces and positive words. The association between concepts and attributes is counterbalanced in the other critical block and therefore respondents would use one key for White faces and positive words and the other key for Black faces and negative words. The critical block in which the associations in response between a concept and an attribute is consistent with the cognitive associations of the respondent is called a compatible block, and the other is called an incompatible block. Based on speed and accuracy of performance in the critical blocks, the IAT score can be computed (Greenwald et al., 2003), which is typically interpreted as an indirect measure (Greenwald et al., 1998).

Which Cognitive Processes Underlie the IAT Effect?

Various theoretical explanations of the IAT effect have been proposed. According to De Houwer (2001), the IAT relies on a response compatibility effect. After repeated categorizations of exemplars of the attribute dimension by pressing two different keys of the keyboard, these acquire a specific meaning. If, for instance, the attribute dimension is evaluative, the key used[18] for negative words acquires a temporary negative evaluation and the key for positive words a temporary positive evaluation....