Minimizing the influence of recoding in the Implicit Association Test: The Recoding-Free Implicit Association Test (IAT-RF)

Abstract

Recoding processes can influence the Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998) in a way that impedes an unequivocal interpretation of the resulting compatibility effects. We present a modified version of the IAT that aims to eliminate recoding, the IAT-RF (short for “IAT–recoding free”). In the IAT-RF, compatible and incompatible assignments of categories to responses switch randomly between trials within a single experimental block. Abandoning an extended sequence of consistent category–response mappings undermines recoding processes in the IAT-RF. Two experiments reveal that the IAT-RF is capable of assessing compatibility effects between the nominally defined categories of the task and effectively prevents recoding. By enforcing a processing of the stimuli in terms of their task-relevant category membership, the IAT-RF eliminates the confounding of compatibility effects with task switch costs and becomes immune against biased selections of stimuli.

Keywords:

• Implicit Association Test
• Recoding processes
• Task switch costs
• Consistent mapping
• Stimulus effects

Acknowledgments

We want to thank Johannes Kretzschmer for developing the programs for the experiments and Catharina Casper, Carina Giesen, Franziska Meißner, Bettina Seifert, and Lisa Schubert for collecting the data.

Notes

¹ The main point of the article by Gray et al. (2003) was that the IAT effect was significantly reduced for psychopathic murderers (to about 350 ms). It is unclear, however, whether the difference in IAT effects between the two groups of participants is due to a difference in the strength of associations, to a difference in recoding, or both.

² A similar variant of a blockless IAT has recently been proposed by Eichstaedt (2006), however, with a different theoretical purpose.

³ Another variant of the IAT-RF that is currently being tested in our laboratories (Teige-Mocigemba, Klauer, & Rothermund, 2007b) uses an additional feature of the stimulus (like word position) to indicate the category–response assignment for the respective trial (e.g., if the stimulus appears in the upper/lower half of the screen, responses have to be given on the basis of the compatible/incompatible mapping).

⁴ It should be noted that the IAT-RF introduces a new type of switch costs that refer to a switch between compatible and incompatible response mappings (compared to those trials in which the response mapping of the previous trial is repeated). These “mapping switch costs”, however, are unrelated to recoding processes because they do not refer to a switching between the target and attribute categorization tasks and thus are not indicative of whether the two tasks are reduced to a single task.

⁵ Values that were below 250 ms or that were more than three interquartile ranges above the median of the overall response time distribution were treated as outliers (Tukey, 1977).

⁶ We also computed the D measure for the compatibility effect of the standard IAT (Greenwald, Nosek, & Banaji, 2003) and found a large compatibility effect (M = 0.45, SD = 0.39), t(15) = 4.62, p < .01.

⁷ A similar Task Switch × Compatibility interaction also emerged if the D measure (Greenwald et al., 2003) was used as dependent variable instead of simple response time indicators, t(15) = 3.00, p < .01. D was more than twice as large for the task switch trials (M = 0.64, SD = 0.51) as it was for the task repetition trials (M = 0.31, SD = 0.42), indicating that the D measure does not eliminate effects of recoding in the standard IAT.

⁸ Trials of the IAT-RF in which response assignments were switched between the previous and the current trial (“mapping switch trials,” compatible–incompatible, incompatible–compatible) cannot be classified as being either compatible or incompatible with regard to task switch costs because the first trial of the task switch or task repetition sequence is always opposite in compatibility to the second. A categorization of the respective task switch or task repetition RT as referring to a sequence of either compatible or incompatible trials is logically impossible in this case.

⁹A similar reversal of compatibility effects for the standard IAT was also found for the D measure (Greenwald et al., 2003). The difference between the old-negative/young-positive version (M = +0.65, SD = 0.35) and the old-positive/young-negative version (M = –0.01, SD = 0.39) was highly significant, t(31) = 5.11, p < .01. Apparently, using the D algorithm does not eliminate recoding effects of biased stimulus sets in the standard IAT.

¹⁰ It has been suggested that biasing influences of stimuli in the standard IAT should be avoided by selecting sets of target stimuli that are balanced with regard to valence (Steffens et al., 2004). In our view, however, this strategy does not eliminate the problem of recoding completely. In another study, we found that recoding still had an influence on compatibility effects in the standard IAT, even if the target stimuli were balanced with regard to valence (Gast & Rothermund, 2007). Specifically, compatibility effects differed significantly for stimuli of opposite valence within a target category, indicating that the valence of the target stimuli had a substantial influence on response times. Apparently, balancing the target stimuli with regard to valence does not suffice to eliminate recoding, as long as the compatible and incompatible trials are presented in separate blocks. Furthermore, a recoding of the targets in terms of the attributes is just one possibility of how recoding can operate. Any feature that can help to reduce the complexity of the categorization task can be used for recoding (familiarity, salience, etc.). Balancing the targets with respect to valence thus does not rule out the possibility that other features are used for recoding.