When old and frail is not the same: Dissociating category and stimulus effects in four implicit attitude measurement methods

Abstract

It is not always clear whether implicit attitude measures assess the attitude towards single stimuli or the attitude towards categories. Nevertheless, this is important to know—both for interpreting implicit attitude effects and for selecting the test that is most appropriate for individual research aims. We investigated this for four implicit measures: the standard Implicit Association Test (IAT), the IAT-recoding free (IAT-RF), and two versions of the Extrinsic Affective Simon Task (EAST, identification (ID)-EAST). Effects in the standard IAT reflect evaluations of categories and single stimuli, whereas the IAT-RF measures attitudes towards categories only. Both versions of the EAST measure evaluations of single stimuli independently from the evaluation of categories. Three different effect sources are distinguished: attitudes towards single stimuli (IAT; EAST and ID-EAST), attitudes towards target categories (IAT and IAT-RF), and processes of recoding (IAT), which do not necessarily reflect attitudes.

Keywords:

• Implicit measures of attitudes
• Implicit Association Test
• Extrinsic Affective Simon Task
• Categorization
• Stimulus effects
• Recoding processes
• Affective valence

Acknowledgments

We thank Catharina Casper, Carina Giesen, Franziska Meißner, Bettina Seifert, and Lisa Schubert for critical comments in the planning phase and help with selection of the material and Joris Lammers for helpful comments on an earlier version of the manuscript.

Notes

¹ Implicit measures are not restricted to measuring associations to the categories “good” and “bad”. Different attribute dimensions are possible. As we focus on implicit attitude measures, we restrict the discussion to valence as attribute dimension throughout the article.

² The more conventional analysis including both attribute and target stimuli does not allow the analysis of stimulus effects. It shows strong general category compatibility effects for both the IAT and the IAT-RF. In the IAT, reaction times are shorter in the category-compatible (M = 833, SD = 209) than in the category-incompatible block (M = 979, SD = 272), t(18) = 4.91, p < .001, d = 1.13. In the IAT-RF, reaction times are shorter in category-compatible (M = 912, SD = 158) than in category-incompatible trials (M = 948, SD = 172), t(19) = 3.52, p < .01, d = 0.79. Of growing importance in research on and with the IAT are D measures (Greenwald, Nosek, & Banaji, 2003)—alternative scoring algorithms that were shown to improve results according to different criteria. However, the application of the D measure to our research question and data is somewhat problematic for a number of reasons: (a) The D measure includes both attribute and target stimuli but only target stimuli can be used for our research question, (b) the D measure is based on and validated for a block structure but to compare D values to investigate stimulus effects different D values have to be calculated for trials within the same block, and (c) the D measure includes practice trials. Due to a small number of practice trials in our experiments, the exclusion of attribute trials and the separate calculation of D measures for different stimulus types, the number of usable practice trials was very low. Taking these considerations into account we calculated amended D3 measures for category-consistent and category-inconsistent stimuli without practice trials. D values over stimulus conditions are clearly positive (M = 0.37, SD = 0.33), F(1, 19) = 25.28, p < .001, η² _p = 0.57, which indicates a strong effect of category compatibility. D values tended to be larger in the condition with category-consistent stimuli (M = 0.47, SD = 0.43) than in the condition with category-inconsistent stimuli (M = 0.27, SD = 0.42), but not significantly so, F(1, 19) = 2.88, p = .11, η² _p = 0.13. These findings might suggest that the D measure leads to a reduction of stimulus effects, but the small sample size does not allow any strong interpretation of this nonsignificant effect. D measures are not developed and validated for the IAT-RF.

³ RTs that were below 250 ms or were more than three interquartile ranges above the third quartile of individual response time distributions (“far out values”; Tukey, 1977) were treated as outliers.

⁴ To keep analyses maximally comparable across the different experiments and paradigms, we decided to analyse compatibility effects in the EAST and ID-EAST in an analogous fashion as in the IAT and IAT-RF (i.e., compatibility effects refer to category valence and were computed separately for target stimuli with a category-consistent and category-inconsistent valence). It should be noted that compatibility effects in the EAST commonly refer to the valence of the target word rather than to its category membership (categories are not part of the task). These two ways of analysing the data, however, are mathematically identical: Standard stimulus-based compatibility effects correspond to the interaction of category–response compatibility and stimulus–category consistency.

⁵ Specifically, responses were faster for stimuli with a category-consistent valence (negative–old, positive–young) if the valence of the key press corresponded to the valence of the category (old → “negative”, young → “positive”); for the category-inconsistent stimuli (positive–old, negative–young) responses were faster if the valence of the key press was opposite to the valence of the category (old → “negative”, young → “positive”).

⁶ In its first published version, the ID-EAST comprised only two different target stimuli (“beer” vs. “sprouts”; De Houwer & De Bruycker, 2007a).

⁷ The ID-EAST is more complex than the other categorization tasks because it involves a hierarchical categorization process in each trial: In a first step, stimuli have to be classified as belonging to the category old/young or not, which determines the type of task for the second categorization (upper case vs. lower case for the targets, evaluation for the attributes). Errors can occur at both steps of the categorization process, which increases the chance of an erroneous response.

⁸ The IAT and the IAT-RF studies took place at the same occasion with random assignment of persons to experiments. The other two experiments were run on two different occasions a few months later in the same laboratories.

⁹ We thank Christoph Klauer for pointing out that finding differences in average effects for different types of stimuli does not necessarily imply differences in the interindividual rank orders of the effects.