Indirect measures as a signal for evaluative change

Abstract

Implicit and explicit attitudes can be changed by using evaluative learning procedures. In this contribution we investigated an asymmetric effect of order of administration of indirect and direct measures on the detection of evaluative change: A change in explicit attitudes is more likely detected if they are measured after implicit attitudes, whereas these latter change regardless of the order. This effect was demonstrated in two studies (n=270; n=138) using the self-referencing task whereas it was not found in a third study (n=151) that used a supraliminal sequential evaluative conditioning paradigm. In all studies evaluative change was present only for contingency aware participants. We discuss a potential explanation underlying the order of measure effect entailing that, in some circumstances, an indirect measure is not only a measure but also a signal that can be detected through self-perception processes and further elaborated at the propositional level.

Keywords:

• Evaluative change
• Evaluative learning
• Measures order
• Indirect
• Direct

We would like to thank Jan De Houwer, Yoav Bar-Anan, and one anonymous reviewer for insightful comments and helpful suggestions on a previous version of this manuscript and Jennifer Annacoreta, Eleonora Paratore, and Angelo Pirrone for help in data collection.

We would like to thank Jan De Houwer, Yoav Bar-Anan, and one anonymous reviewer for insightful comments and helpful suggestions on a previous version of this manuscript and Jennifer Annacoreta, Eleonora Paratore, and Angelo Pirrone for help in data collection.

Notes

¹ Following De Houwer and Moors (2010), in this contribution we will use the terms direct and indirect to refer to the measures and explicit and implicit to refer to the constructs.

² Note that the indirect measures used in these studies were affective liking (gut-feeling, cf. Olson & Fazio, 2001) rather than semantic differentials. When semantic differentials have been used instead, the typical result is that the evaluative learning effect is detected also with the direct measure when administered after an indirect one (Perugini et al., 2012). The more likely reason for these differences in results is that semantic differential measures are generally superior to affective liking measures both psychometrically and theoretically. This point is developed more in detail in Perugini et al. (2012).

³ More specifically, the contingency awareness measures that we used should be considered as reflecting identity memory rather than valence memory (cf. Stahl, Unkelbach, & Corneille, 2009).

⁴ In this as well as the next studies the results were not affected substantially if the counterbalanced factors were not included as covariates (e.g., all significant effects remained significant and there were no additional significant effects without the covariates). Considering all analyses for all groups in all studies, there were only two significant covariate effects. Specifically, in Study 1 there was an effect of the order of groups in the learning task/keys for the SR, F(1, 193)=5.00, p=.026, , and of order of blocks in the IAT, F(1, 193)=19.13, p<.001, , concerning the IAT for the contingency aware participants.

⁵ We do not report the specific statistical results for the subsample of contingency unaware participants as the sample size was too small (n=9) to provide reliable results. The results for this subsample should be considered with due caution.

⁶ Valence of the 20 words used in the Attitude Induction Task was established in a pre-test in which 39 students (21 women, 18 men, M _age=22.77 years, SD=3.05) evaluated 92 positive and negative words on 9-point scales from 1 (Very negative) to 9 (Very positive). From this set we selected 10 moderately positive words (M=6.32, SD=0.70), t(38)=11.70, p<.001, and 10 moderately negative words (M=3.81, SD=0.81), t(38)=−9.15, p<.001. Words were matched for mean length (6.4 vs. 6.8 characters) and were reasonably similar for frequency (158.4 vs. 76).

⁷ We also analysed the data in Study 3 separately for three subsamples of participants: Contingency aware (positive scores), contingency unaware (scores equal to 0), and incorrect contingency aware (negative scores). For contingency aware participants, main EC effects were present for both IAT, t(54)=4.31, p<.001, Cohen's d=1.17, and explicit attitudes, t(54)=3.49, p=.001, Cohen's d=0.95. For contingency unaware participants, no EC effects were present for either IAT, t(50)=−0.48, p=.631, Cohen's d=0.14, or explicit attitudes, t(50)=−0.13, p=.899, Cohen's d=0.04. Finally, for the incorrect contingency aware subsample, there was a tendency towards a significant EC effect for the IAT, t(36)=−1.76, p=.086, Cohen's d=0.59, and a significant EC effect for explicit attitudes, t(36)=2.26, p=.032, Cohen's d=0.75.