ABSTRACT
In two experiments, the impact of faking on the affect misattribution procedure (AMP) was examined. Results revealed that faking influences both the overall means and the convergent validity of AMP effects in terms of correlations with self-report measures. Faking effects were very selective in that they affected fake-prime trials only, for which AMP effects were significant, but reversed in direction, while AMP effects for non-fake trials remained intact. Importantly, neither strategic advice nor prior task experience was a necessary prerequisite for successful faking. The discussion focuses on possible processes underlying successful faking in the AMP.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1As Arabs, we chose portraits of men who wore turbans and had long beards, and thus were characterised by features that are typically negatively associated in Western societies. As liked celebrities, we selected portraits of Jürgen Klinsmann (coach of the German football national team in 2006), Günther Jauch (a famous German television presenter), Claus Kleber (a famous German newsreader), and Ulrich Wickert (also a famous German newsreader). All materials can be obtained from the authors.
2Sample size per group was approximately doubled in Experiment 2 as compared to Experiment 1 because we examined a correlational research question here that requires a larger sample size to permit sound conclusions (cf. Teige-Mocigemba & Klauer, Citation2013).
3Note that we re-ran all correlation and moderated regression analyses with the average of (a) the z-transformed difference of the thermometer ratings for the Greens and for the CDU, (b) the z-transformed political standpoint, and (c) the z-transformed difference of the ratings for the fake primes of the Greens and the CDU as the measure of the self-reported political attitude (see Teige-Mocigemba & Klauer, Citation2013). Analyses with this composite score revealed the same pattern of significant and non-significant results as analyses with the thermometer rating with the exception that in the CG, the correlation of the AMP effect and the self-reported political attitude measure as well as the corresponding simple slope in the moderated regression analysis moved from one-tailed significance to non-significance (r = .18, p = .20; β = .28, t(96) = 1.16, p = .25).
4Note that we re-ran all correlation and moderated regression analyses after excluding bivariate outliers that were identified by computing two outlier indices per correlation and participant. Following Bar-Anan (Citation2010), participants with an absolute studentised residual larger than two or with a Cook's D-value above the threshold of the relevant sample size were excluded. Analyses without bivariate outliers revealed the same pattern of significant and non-significant results as analyses including all participants.
5Note that we did not explicitly ask participants whether or not they had prior experience with an AMP in the past. However, to the best of our knowledge, only our group conducts AMP research at the Institute of Psychology and the AMP is not part of the regular teaching (in psychology). Furthermore, we excluded participants who had participated in a previous AMP experiment in our lab. For these reasons, we consider it unlikely that participants were acquainted with the AMP before participating in the present experiments.
6To test this assumption, we restricted the correlational and moderated regression analyses to the subsample of supporters of the Greens (n = 87) because this subsample was considerably larger than that preferring the CDU (n = 13). If the constant shift account was correct, we would expect positive correlations in both the EG (n = 41) and the CG (n = 46); if the constant shift account was false, we would expect positive correlations only in the CG, but considerably lower correlations in the EG. Results were not conclusive and only descriptively contradicted the constant shift account. The interaction term in the moderated regression analysis did not reach significance, β = −.23, t(83) = −1.04, p = .30, and neither the simple slope tests (CG: β = .26, t(83) = 1.36, p = .18; EG: β = .03, t(83) = .32, p = .75) nor the correlations were significant in either group (CG: r = .20, p = .18; EG: r = .05, p = .75; for similar findings, see Teige-Mocigemba & Klauer, Citation2013). Similarly, the zero correlation in the EG did not support the interpretation according to which participants’ faking attempts are a function of their self-reported political attitudes. Given the restrictions of both sample size and variance, however, we hesitate to derive strong conclusions from these re-analyses.