It s brief but is it better? An evaluation of the Brief Implicit Association Test (BIAT) Klaus Rothermund 1 & Dirk Wentura 2

Transcription

1 RUNNING HEAD: Evaluating the BIAT It s brief but is it better? An evaluation of the Brief Implicit Association Test (BIAT) Klaus Rothermund & Dirk Wentura Friedrich-Schiller-Universität Jena Saarland Universität ---- Experimental Psychology, in press --- Corresponding address: Klaus Rothermund Friedrich-Schiller-Universität Jena Institute of Psychology Am Steiger / Haus D-0 Jena, Germany Electronic mail: klaus.rothermund@uni-jena.de Phone: 00-(0)-

2 0 Abstract Sriram and Greenwald (00) introduced a new variant of the Implicit Association Test, which they termed the Brief Implicit Association Test (BIAT). The BIAT differs from a standard IAT by using less trials and by instructing participants to focus on only two of the four categories in each block (e.g., press the focal key for categories male+math and the other key for anything else [female+arts] in the compatible block, and press the focal key for categories female+math and the other key for anything else [male+arts] in the incompatible block). We argue that the focus manipulation of the BIAT does not suffice to fully control for focusing and recoding processes in the task. Compatibility effects in the BIAT are therefore still subject to influences that are unrelated to the conceptual relation between the target and attribute categories of the task (e.g., salience asymmetries, stimulus-based effects). Highlighting these non-associative influences, findings with the BIAT revealed strong asymmetries in compatibility effects, reliability, and convergent validity, depending on which of the two attribute categories was selected as a focal category in the BIAT. To eliminate these problems, we recommend the use of other, recently developed variants of the IAT that prevent recoding effects by eliminating the dual-block structure of the task. Keywords: Implicit Association Test, Associations, Recoding, Salience Asymmetries

3 0 It s brief but is it better? An evaluation of the Brief Implicit Association Test (BIAT) Sriram and Greenwald (00) recently introduced a new variant of the Implicit Association Test (Greenwald, McGhee, & Schwartz, ), which they termed the Brief Implicit Association Test (BIAT). This task deviates from the standard IAT protocol in several respects that lead to a substantial reduction in the length of the task (all practice blocks are omitted from the task, and the number of trials in each of the two combined blocks is drastically reduced). Apparently, shortening the task in this way does not substantially impair the psychometric quality and predictive validity of the BIAT. While these changes and results are clearly important for practical applications of the IAT, they have only minor implications with regard to the basic logic and functioning of the task. The theoretically most important innovation that was introduced in the brief IAT, however, is unrelated to its length: In the BIAT, participants have to focus on just two of the four categories (the focal categories ) that are assigned to one of two response keys (the focal key ). Participants have to respond with the focal key whenever an exemplar of one of the two focal categories is presented, and have to press the other key for anything else. This focusing manipulation eliminates the relevance of the non-focal categories for the task and is introduced in order to control for interindividual differences in the strategy that is used when performing a standard IAT (i.e., some participants might spontaneously invent such a focusing strategy whereas others do not). In the remainder of this article, we want to discuss whether introducing this focusing manipulation in the BIAT procedure also suffices to eliminate or control for those focusing processes that can be responsible for non-associative influences on the resulting compatibility effects. Before addressing this topic with regard to the BIAT, we first discuss the nature of non-associative influences that have been investigated with respect to the IAT. Non-associative influences in the IAT

4 0 The IAT has lately been criticized for being subject to influences that are nonassociative in nature, and which are unrelated to associations between the category concepts that define a specific IAT task. Among the most often mentioned non-associative influences are familiarity (Brendl, Markman, & Messner, 00), mere acceptance effects (Mitchell, 00), processing fluency (Chang & Mitchell, in press), polarity correspondence (Proctor & Cho, 00), salience asymmetries (Kinoshita & Peek-O Leary, 00, 00; Rothermund & Wentura, 00, 00; Rothermund, Wentura, & De Houwer, 00; Wentura & Rothermund, 00), perceptual similarity (De Houwer, Geldof, & De Bruycker, 00), and cognitive skills or task switching ability (Back, Schmukle, & Egloff, 00; McFarland & Crouch, 00; Mierke & Klauer, 00). The existence of these non-associative influences undermines the fundamental claim expressed in the very name of the task that an IAT effect is a pure measure of associations between the nominal categories of a specific IAT. All the different non-associative influences listed above affect IAT effects in a similar way: A common denominator of how these non-associative influences operate in the IAT is recoding (De Houwer, 00; De Houwer et al., 00; Rothermund, Teige-Mocigemba, Gast, & Wentura, 00). Recoding is a process that makes use of any kind of commonality between the target and attribute categories in order to reduce the complex -to- categorization task to a simple binary classification. In the compatible block of an IAT, the categories and/or stimuli that are assigned to the same key share certain semantic, structural, and/or perceptual features that can be used to select the correct response without having to process the relevant category membership of the presented stimuli. One example of recoding is to categorize stimuli of the attribute and target categories according to their salience. If both the target and attribute dimensions are characterized by a default figure/ground asymmetry, this structural feature can be used to select the appropriate response in the compatible block, in which both salient (and both non-salient) categories are assigned to the same key. Such a recoding in terms of salience reduces the task to a simple,

5 0 binary categorization task. Because such a recoding strategy is not possible in the incompatible block of an IAT, in which salient and non-salient categories are assigned to the same key, responding is facilitated in the compatible compared to the incompatible block, thus giving rise to substantial IAT effects. Importantly, selecting responses on the basis of salience (or any other feature that consistently helps to discriminate between the two responses for the target and attribute tasks) does not reflect associative relations between the nominal categories of the task (Rothermund et al., 00). Three sources of empirical evidence testify to the influence of non-associative influences and recoding processes in the IAT: First, substantial IAT effects can be found even in the absence of associations between the target and attribute categories (Mitchell, 00; Rothermund & Wentura, 00). Second, task-switch costs indicating difficulties to switch between the target and attribute categorization tasks are markedly reduced in the compatible block of an IAT, indicating that the two-dimensional task has been recoded to a simple binary decision task (Mierke & Klauer, 00, 00; Rothermund & Wentura, 00; Rothermund et al., 00). Finally, the existence of stimulus effects and stimulus-based cross-category associations in the IAT reveals that the stimuli are not only processed in terms of their relevant category membership but also with regard to the non-relevant dimension, indicating some kind of recoding of the task in terms of one underlying dimension (Bluemke & Friese, 00; Gast & Rothermund, in press; Rothermund et al., 00; Steffens & Plewe, 00). A reader might argue that a special kind of recoding that is, recoding in terms of valence (if valence connotations of the target categories are what should be assessed) might be intended. To clarify and to stay in line with our former work on the IAT (Wentura & Rothermund, 00), we should stress that the term recoding can be used in a broader and a more narrow sense. The broader sense focuses on any kind of commonality between the target and attribute categories, as introduced above. These commonalities might automatically be used to reduce task difficulty in the congruent block. If, for example, figure/ground

6 0 asymmetries are the commonalities, IAT effects cannot be taken as face-valid for what was intended to be measured (simply because it was not figure/ground asymmetry). We acknowledge, however, that if valence (and only valence) is automatically used for response selection during the processing of the target category membership information, IAT effects can be taken as a valid indicator of evaluative target connotations. Our salience account focused on the point that as long as there valence and salience are confounded, one cannot be sure which variation (salience or valence) is the driving force behind IAT effects. The more narrow sense of recoding is strategic recoding. That is, participants might strategically use, for example, the valence connotation of the target categories to simplify the congruent block. If, for example, ageism is assessed by using old vs. young names as the target categories and positive vs. negative words as attributes, participants who do not withhold their dislike of old people might simplify the compatible block by instructing themselves to press the negative key if anything negative appears, be it a negative attribute or an old name. Such a strategy makes the IAT more similar to an explicit measure than was originally intended, and it produces IAT effects that are strongly susceptible to stimulus-based effects rather than category effects (Gast & Rothermund, in press). To summarize, recoding effects are unwanted as long as one cannot be sure about the automatic and non-confounded nature of the resulting IAT effects. Recoding effects in the BIAT In the following, we want to discuss whether the focusing manipulation that was introduced in the BIAT suffices to eliminate the influence of recoding processes and strategies on compatibility effects in the BIAT. The basic structural feature of the IAT that is responsible for the emergence of recoding effects is its block structure (Rothermund et al., 00). A consistent mapping of categories onto responses for an extended block of trials is a prerequisite for the occurrence of efficient recoding (Shiffrin & Schneider, ; Strayer & Kramer, ). Seen from this perspective, by maintaining the basic dual-block structure of

7 0 the standard IAT, the BIAT may still be susceptible to non-associative influences and recoding processes. In essence, this susceptibility results from the fact that focusing on two of the four categories does not prevent additional focusing strategies (or recoding processes, as we prefer to call them) that create a unitary representation of the two focal categories. Specifically, in the compatible block of a BIAT, response selection may be governed by a single feature that is common to the two focal categories of the task and which distinguishes them from the remaining, non-focal categories and stimuli (e.g., high vs. low salience or familiarity, positive vs. negative valence, perceptual features). For example, in a flower/insect BIAT with good as the focal attribute category, valence and/or salience can be used to discriminate between the focal categories flowers and good, on the one hand, and the remaining stimuli in the compatible block. Such a simple recoding process is impossible in the incompatible block, in which the categories insects and good are assigned to the focal key. The difference in focusing or recoding processes between the compatible and incompatible blocks of a BIAT can explain (differences in) compatibility effects in this task. This example illustrates that focusing on two categories as it is prescribed in the BIAT does not suffice to control for or rule out additional focusing processes aiming at a further reduction of these categories in terms of a single attribute. Rather than being an indicator of associations between categories, compatibility effects might thus reflect differences in recoding processes between the compatible and incompatible blocks, in spite of the focus manipulation that is used in the BIAT. Sriram and Greenwald (00) did not systematically address the question of recoding processes in their study. Nevertheless, one aspect of their findings provides a hint that compatibility effects in the BIAT might result from differences in recoding. Attitude BIATs in which the category good was focal consistently produced better results than BIATs in which bad was the focal category (with regard to reliability and external validity). Similarly, self-concept BIATs using the category self as the focal category were superior to those

8 0 variants of the task in which other was the focal category. These differences can be explained with recoding processes in the broader as well as the more narrow sense. First, they can be explained by the existence of structural asymmetries between the attribute (and target) categories that are independent of associations. Focusing on exemplars of the familiar categories that are easily processed and low in salience is a default processing strategy for most participants. Using good or self in combination with the familiar target category as the focal categories in the compatible blocks of a BIAT (e.g., good + Kerry for Kerry supporters, self + female for women) thus corresponds to how most participants would have structured the task anyway, leading to a strong facilitation effect in these blocks that is based on a simple binary recoding of the task in terms of familiar vs. rest (M = ms vs. ms for the compatible / incompatible blocks of the good and self BIATs in Sriram & Greenwald s Exp.). In contrast, having to search for a combination of the two odd and unfamiliar categories in the compatible block of a BIAT when bad or other are focal does not correspond to the default processing strategy and does not as easily lend itself to a recoding in terms of unfamiliar vs. rest, because in this case, the instruction to actively search for the two unfamiliar categories interferes with the quality of these categories as being salient or standing out (M = ms vs. ms for the compatible / incompatible blocks of the bad and other BIATs in Sriram & Greenwald s Exp.). Second, certain focal categories might invite participants to a stronger degree than others to strategically recode the task in the compatible block. For example, it might be easier to adopt the strategy to use one key for anything that is liked by oneself (e.g., for Kerryrelated stimuli and positive attributes if one is a Kerry voter) than the reverse (i.e., to instruct oneself to use one key for anything that is disliked by oneself). For the self-concept BIAT it is self-evident that it must be easier to recode in terms of what is self-related than to recode in terms of what others are. This is so because for a given participant stimuli denoting own gender are non-ambiguously self-related and can easily be subsumed under the category

9 0 self. In contrast, it is not possible to subsume the category of the opposite gender under the category others because others can be male as well as female. Again, one can suspect that such versions of the IAT are more like an explicit measure than an implicit one. The previous arguments provide an explanation of the observed differences in the strengths of compatibility effects in BIATs depending on which attribute category is selected as focal. In addition, they also explain differences in reliability between the different variants of a BIAT. Given that it is more difficult to implement a recoding strategy in the compatible block of a BIAT with a focal category that is unfamiliar or heterogeneous, the resulting compatibility effect should be less strong and should be less susceptible to interindividual differences (e.g., different group memberships) that are related to the strategic use of recoding, thus producing less systematic variance. Of course, these empirical observations do not test the influence of recoding in the BIAT directly and cannot be taken as strong evidence for the claim that recoding processes are operative in the BIAT (e.g., the asymmetries could also be explained in terms of differences in associative relations regarding the attribute categories; see Sriram & Greenwald, 00). More stringent tests of the assumption that the dual-block structure of the BIAT invites recoding processes involve testing for compatibility effects of non-associated categories, for reductions of task-switch costs in the compatible block, and for stimulus-based effects in the BIAT. Conclusion The BIAT provides an interesting alternative to the standard IAT in research settings with strong time constraints (e.g., large panel studies) and can be recommended as an elegant and parsimonious short version of the IAT. Despite its practical advantages, the BIAT does not solve the fundamental problems regarding non-associative influences that also plague the IAT. By specifying two focal categories in each block of the BIAT, the authors wanted to reduce variation in the strategies that are used to perform this task. Restricting the attentional

10 0 focus to two categories, however, does not suffice to eliminate those focusing strategies and processes that reduce the two focal categories to one common attribute, which helps to separate them from the remaining, non-focal categories and stimuli. Due to its basic dualblock structure, compatibility effects in the BIAT are therefore still susceptible to differences in focusing or as we call it recoding processes between the compatible and incompatible blocks of the task. Preliminary evidence suggesting an influence of recoding processes based on structural properties of the categories was reported by Sriram and Greenwald (00). A direct and systematic investigation of these influences, however, is necessary and is an interesting question for further research. We want to stress that the BIAT is, of course, not the only implicit measure that is susceptible to recoding, but shares this problem with other tasks that are characterized by a dual-block structure (e.g., standard IAT [Greenwald et al., ], Go/No-go Association Task [Nosek & Banaji, 00], single category Implicit Association Task [Karpinski & Steinman, 00]). Our arguments were meant to show that prescribing a strategy to focus on two of the four categories as it was done in the BIAT with the aim to reduce differences in focusing does not suffice to eliminate or control the more far-reaching focusing processes that are crucial for the emergence of recoding effects. From our perspective, recent proposals of variants of the IAT that eliminate the basic dual-block structure are more promising (e.g., the recoding free Implicit Association Test, IAT-RF, Rothermund et al., 00; the single-block Implicit Association Test, SB-IAT, Teige- Mocigemba, Klauer, & Rothermund, 00; the Sorting Paired Features Task, SPF, Bar-Anan, Nosek, & Vianello, in press). For some of these implicit single-block measures of associations, it has already been shown that recoding processes do no longer have an influence on the resulting compatibility effects (e.g., task switch costs do not differ between compatible and incompatible trials, and stimulus-based effects are eliminated; Gast & Rothermund, in press; Rothermund et al., 00).

11 0 References Back, M. D., Schmukle, S. C., & Egloff, B. (00). Measuring task-switching ability in the Implicit Association Test. Experimental Psychology,, -. Bar-Anan, Y., Nosek, B. A., & Vianello, M. (in press). The Sorting Paired Features task: A measure of association strengths. Experimental Psychology. Bluemke, M., & Friese, M. (00). Do features of stimuli influence IAT effects? Journal of Experimental Social Psychology,, -. Brendl, M., Markman, A., & Messner, C. (00). How do indirect measures of evaluation work? Evaluating the inference of prejudice in the Implicit Association Test. Journal of Personality and Social Psychology,, -. Chang, B. P. I., & Mitchell, C. J. (in press). Processing fluency as a source of salience asymmetries in the Implicit Association Test. Quarterly Journal of Experimental Psychology. De Houwer, J. (00). The extrinsic affective Simon task. Experimental Psychology,, -. De Houwer, J., Geldof, T., & De Bruycker, E. (00). The Implicit Association Test as a general measure of similarity. Canadian Journal of Experimental Psychology,, -. Gast, A., & Rothermund, K. (in press). When old and frail is not the same. Dissociating category-based and stimulus-based influences on compatibility effects in four implicit measurement methods. Quarterly Journal of Experimental Psychology. Greenwald, A. G., McGhee, D. E., & Schwartz, J. L. K. (). Measuring individual differences in implicit cognition: The Implicit Association Test. Journal of Personality and Social Psychology,, -. Karpinski, A., & Steinman, R. B. (00). The single category Implicit Association Test as a measure of implicit social cognition. Journal of Personality and Social Psychology,, -. McFarland, S. G., & Crouch, Z. (00). A cognitive skill confound on the Implicit Association Test. Social Cognition, 0, -0. Mierke, J., & Klauer, K. C. (00). Implicit association measurement with the IAT: Evidence for effects of executive control processes. Zeitschrift für Experimentelle Psychologie,, -. Mierke, J., & Klauer, K. C. (00). Method-specific variance in the Implicit Association Test. Journal of Personality and Social Psychology,, -. Mitchell, C. J. (00). Mere acceptance produces apparent attitude in the Implicit Association Test (IAT). Journal of Experimental Social Psychology,, -. Nosek, B. A., & Banaji, M. R. (00). The Go/No-go Association Task. Social Cognition,, -.

12 0 Proctor, R. W., & Cho, Y. S. (00). Polarity correspondence: A general principle for performance of speeded binary classification tasks. Psychological Bulletin,, -. Rothermund, K., Teige-Mocigemba, S., Gast, A., & Wentura, D. (00). Eliminating the influence of recoding in the Implicit Association Test: The recoding-free Implicit Association Test (IAT-RF). Quarterly Journal of Experimental Psychology,, -. Rothermund, K., & Wentura, D. (00). Figure-ground asymmetries in the Implicit Association Test (IAT). Zeitschrift für Experimentelle Psychologie,, -. Rothermund, K., & Wentura, D. (00). Underlying processes in the Implicit Association Test (IAT): Dissociating salience from associations. Journal of Experimental Psychology: General,, -. Rothermund, K., Wentura, D., & De Houwer, J. (00). Validity of the salience asymmetry account of the Implicit Association Test: Reply to Greenwald, Nosek, Banaji, and Klauer (00). Journal of Experimental Psychology: General,, -0. Shiffrin, R. M., & Schneider, W. (). Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory. Psychological Review,, -. Sriram, N., & Greenwald, A. G. (00). The Brief Implicit Association Test. Experimental Psychology,, -. Steffens, M. C., & Plewe, I. (00). Items' cross-category associations as a confounding factor in the Implicit Association Test. Zeitschrift für Experimentelle Psychologie,, -. Strayer, D. L., & Kramer, A. F. (). Strategies and automaticity: I. Basic findings and conceptual framework. Journal of Experimental Psychology: Learning, Memory, and Cognition, 0, -. Teige-Mocigemba, S., Klauer, K. C., & Rothermund, K. (00). Minimizing method-specific variance in the IAT: A single block IAT. European Journal of Psychological Assessment,, -. Wentura, D., & Rothermund, K. (00). Paradigms we live by. A plea for more basic research on the IAT. In B. Wittenbrink & N. Schwarz (Eds.), Implicit measures of attitudes (pp. -). New York: Guilford Press.

13 0 Author Note Klaus Rothermund, Institute of Psychology, Friedrich-Schiller-Universität Jena, Germany, Dirk Wentura, Institute of Psychology, Saarland University, Saarbrücken, Germany. Preparation of this article was supported by the German Research Foundation (RO /-,). Correspondence concerning this article should be addressed to Klaus Rothermund, Friedrich-Schiller-Universität Jena, Institute of Psychology, Am Steiger / Haus, D-0 Jena, Germany. Electronic mail should be sent to Klaus.Rothermund@uni-jena.de.

14 0 Footnotes The standard IAT consists of seven blocks: In the first two blocks, simple binary categorizations of two target categories (e.g., flowers vs. insects) and two attribute categories (e.g., good vs. bad) are practiced. In a third (practice) and fourth (experimental) block, the two categorization tasks are combined, resulting in a -to- mapping of categories onto responses (e.g., press one key for flowers+good, press the other key for insects+bad). In a fifth block, the response assignment is switched for the target categories. The final blocks ( = practice, = experimental) again use a combined categorization of the four categories, now using the reversed assignment for the target categories (e.g., press one key for insects+good, press the other key for flowers+bad). A compatibility effect (IAT effect) is computed as the RT difference between the two combined blocks (e.g., ). Other variants of the IAT have been proposed that also used focusing strategies (e.g., the Go/No-go Association Task [GNAT] of Nosek & Banaji, 00, or the single category IAT [SC-IAT] of Karpinski & Steinman, 00). The focusing manipulations that were used in these measures, however, were developed in order to yield a measure of association for single target categories rather than eliminating variance in processing strategies, and differ from the focusing manipulation that is used in the BIAT (in the GNAT, one target category is combined with two different attributes whereas in the BIAT, one attribute category is combined with two different target categories; in the SC-IAT, a single target category is combined with two attribute categories in each block, sharing a response key with one attribute category in one block and with the other attribute category in the other block.). It may be argued that the BIAT was not developed with the aim to eliminate all kinds of non-associative influences. This may be true, but the focus manipulation was explicitly introduced by the authors with the intention of reducing spontaneous variation in subject strategy (p. ). The major aim of this article is to discuss which kinds of strategies and

15 0 processes can be controlled with this kind of manipulation, particularly with respect to an influence of recoding processes. This should be of interest for all potential users of the BIAT (or other paradigms using a focus manipulation) who might want to know to what extent the focus strategy helps to reduce the influence of processing strategies that are unrelated to associations. Recently, there has been a debate in the literature which valence category should be considered the salient one (Chang & Mitchell, in press; Kinoshita & Peek-O Leary, 00, 00; Proctor & Cho, 00; Rothermund & Wentura, 00). Though important in detail, at the abstract level of the present discussion it does not matter whether positive or negative is more salient. As pointed out by an anonymous reviewer, the finding of an asymmetry in compatibility effects depending on which attribute category is selected as being focal might also limit the advantages of using new variants of the BIAT: For each BIAT that uses new attribute categories, pretesting is necessary to identify which of the two attribute categories is the privileged one, which counteracts the advantage of a brief application of the task.