Automatic evaluation isn't that crude! Moderation of masked affective priming by type of valence

Abstract

In two experiments, the automatic processing of evaluative information was investigated using a masked affective priming paradigm, varying valence (positive vs. negative) and relevance (other-relevant traits vs. possessor-relevant traits; Peeters, 1983) of prime and target stimuli. It was found that under specified conditions, valence-congruency effects were only found if prime and target matched with regard to relevance type (i.e., were both either of the other-relevant or possessor-relevant type). These results suggest that automatic processing of affective information conveys not only the positive–negative differentiation, but also the relevance type of valence. Consequences for research on automatic attitudes, especially prejudice, are discussed. For current research on masked priming, it is important to highlight that the subliminal effect was found even for non-practised prime stimuli.

Keywords:

• Affective priming
• Evaluation
• Prejudice
• Implicit attitudes
• Unconscious

Acknowledgements

The research reported in this article was supported by a grant from the Deutsche Forschungsgemeinschaft (DFG) to DW (WE 2284/3).

Notes

¹Of course, other tasks (e.g., pronouncing the target, see, e.g., Bargh, Chaiken, Raymond, & Hymes, 1996; Hermans, De Houwer, & Eelen, 1994; Spruyt, Hermans, De Houwer, & Eelen, 2002, 2004; deciding whether it is a legal word, see, e.g., Wentura, 2000; non-affective semantic categorisation, see, e.g., De Houwer et al., 2002; Spruyt, De Houwer, Hermans, & Eelen, 2007; or naming the colour in which it is presented, see, e.g., Hermans, van den Broeck, & Eelen, 1998; Rothermund & Wentura, 1998) were employed as well (see, e.g., Klauer & Musch, 2003, for a review). However, the evaluative decision task is the basic task of affective priming research. It is associated with unequivocal results, a process theory that cannot be applied to the other tasks, and is supposedly the only one yielding masked (subliminal) effects that are the focus of the present research.

²This point should not be overemphasised, though. A simple consideration shows that the comparison of mean RTs for congruent and incongruent trials with the neutral baseline is ambiguous: assume that mean RT for congruent trials is 500 ms and mean RT for incongruent trials amounts to 530 ms. If the mean RT for the neutral baseline is M=530 ms, it seems as if priming was entirely due to facilitation. Now assume that negative primes have an additional main effect (e.g., besides facilitating negative responses for most of the trials they capture attention in some trials, thereby increasing RTs for positive as well as negative targets). This main effect will increase mean RTs for the congruent as well as for the incongruent condition. To extend our example, if this increase amounts to 30 ms, we face a result suggesting that priming is entirely due to interference, which, however, is not the case.

³We decided on this version of test because it constitutes what was recently termed a test of “partial awareness” (Abrams & Grinspan, 2007; Kouider & Dupoux, 2004): Any remnant of surface features of the prime (e.g., a letter or a letter pair) helps participants to select the correct answer even if the masked prime was not processed on a semantic level.

⁴There were three missing cases for the direct test because their responses were not adequately recorded (probably due to using the wrong keys during the direct test).

⁵Because an F-test with one degree of freedom is equivalent to a t-test (with differences as the dependent variables) and given our specific predictions, a one-tailed test is allowed (see Maxwell & Delaney, 1990, p. 144).

⁶For the emphasis-on-accuracy sample, both intercepts were not significantly different from zero (see Table 2).

⁷We refrained from statistical tests because assumptions of independence are likely violated due to the use of the same words in different pairings.

⁸The two kinds of analyses would provide exactly the same result if there had been no missing trials. That is, any differences are only due to minor differences in error rates and outlier rates for the different conditions. The correlation between the individual difference value between priming for pairs matching in relevance and those non-matching in relevance and the individual regression weight for the interaction term is r=.996.

⁹A t-test for trimmed means (see Wilcox, 1997, 1998) with a trimming of γ=.11 was done to adequately account for one outlying regression weight.