Of two minds or one? A registered replication of Rydell et al. (2006)

ABSTRACT

Evaluative conditioning (EC) is proposed as a mechanism of automatic preference acquisition in dual-process theories of attitudes (Gawronski, B., & Bodenhausen, G. V. (2006). Associative and propositional processes in evaluation: An integrative review of implicit and explicit attitude change. Psychological Bulletin, 132(5), 692–731. doi:10.1037/0033-2909.132.5.692). Evidence for the automaticity of EC comes from studies claiming EC effects for subliminally presented stimuli. An impression-formation study showed a selective influence of briefly presented primes on implicitly measured attitudes, whereas supraliminally presented behavioural information about the target person was reflected in explicit ratings (Rydell, R. J., McConnell, A. R., Mackie, D. M., & Strain, L. M. (2006). Of two minds forming and changing valence-inconsistent implicit and explicit attitudes. Psychological Science, 17(11), 954–958. doi:10.1111/j.1467-9280.2006.01811.x) This finding is considered one of the strongest pieces of evidence for dual process theories (Sweldens, S., Corneille, O., & Yzerbyt, V. (2014). The role of awareness in attitude formation through evaluative conditioning. Personality and Social Psychology Review, 18(2), 187–209. doi:10.1177/1088868314527832), and it is therefore crucial to assess its reliability and robustness. The present study presents two registered replications of the Rydell et al. (2006) study. In contrast to the original findings, the implicit measures did not reflect the valence of the subliminal primes in both studies.

KEYWORDS:

• Evaluative conditioning
• subliminal influence
• implicit learning
• replication

Disclosure statement

No potential conflict of interest was reported by the author.

ORCID

Tobias Heycke http://orcid.org/0000-0001-6358-6713

Julia M. Haaf http://orcid.org/0000-0001-5122-706X

Christoph Stahl http://orcid.org/0000-0002-9033-894X

Notes

1. After the data of Experiment 1 was collected we were informed that a different set of target words was used in the original study.

2. The first three words were taken from a study by Bluemke and Friese (2006), while the last word was explicitly mentioned in the original article by Rydell et al. (2006).

3. Neither “laughter” nor “friend” were used as targets in the original Experiment, but were included in the material we initially received.

4. Additionally the words “Love” and “Peace“ were used as new words in the recognition test, but were also previously used as valent words in the IAT after block 1 and block 2. To avoid potential problems in the prime recognition check, we substituted the words “love” (Liebe) and “peace” (Frieden) with “nature” (Natur) and “cash” (Bargeld) in the prime-recognition task.

5. See table robustnessBF.pdf on osf.io/c57sr for a detailed overview of all Bayes Factors with different prior scaling factors.

6. In addition, during the final day of data collection, two participants instead of only one were collected accidentally, resulting in 51 (instead of the planned 50) complete data sets after testing a total of 53 participants. Results do not change when we excluded the 51st participant (largest deviation of d = 0.05; largest deviation of ). Two participants had to rate the target character also before the first learning phase, due to a programming error. Excluding these participants did not substantially alter the results (largest deviation of d = 0.07; largest deviation of ).

7. The data of the pretest and Experiment 2 were collected under a Born Open Data protocol (Rouder, 2016) in which they were automatically logged, uploaded, and made freely available as they were created (https://github.com/PerceptionCognitionLab/data1/tree/master/repRydell).

8. In the original study, some of the prime words, as well as some of the new words in the prime recognition task, also appeared as part of the behavioural statements during the learning phase. We slightly adjusted the behavioural statements (e.g. substituted single words in the behavioural description with similar but different words or rearranged the sentence and omitted the word in question) to remove this possible confound. Two independent raters confirmed that the behavioural statements were still clearly identifiable as positive or negative.

9. Additionally, we found no indications for an influence of the order of dependent measures or the key assignment during the IAT on the IAT results (as in Experiment 1 the Bayes factors were mostly not conclusive).

Additional information

Funding

The study was funded by a grant awarded to the fourth author by Deutsche Forschungsgemeinschaft Grant 1269/3-1.