Corrigendum to “How pledges reduce dishonesty: The role of involvement and identification” [Journal of Experimental Social Psychology 113(2024) 104614] (Journal of Experimental Social Psychology (2024) 113, (S002210312400026X), (10.1016/j.jesp.2024.104614))

Eyal Peer*, Nina Mazar, Yuval Feldman, Dan Ariely

*Corresponding author for this work

Research output: Contribution to journalComment/debate

Abstract

After the publication of our paper, on the effects of involvement and identification of honesty pledges in reducing dishonest reporting, several issues have been brought to our attention by diligent researchers. Here, we offer corrections and explanations for these issues, including errors in reported analyses and several unreported deviations from preregistrations, which were either requested by the review team or made by us. In the following, we report each issue and explain how it may, or may not, affect the conclusions of our paper. Analysis errors In Study 3, section 7.2. Results. Exploratory 2 × 2 ANOVA Over All Problems, we report that the main effects for both involvement and identification, as well as their interaction, are significant. However, due to a coding error, the analysis also included the Self-report condition instead of analyzing only the four pledge conditions, which resulted in erroneous results. The corrected results show a significant main effect only for involvement, F (1, 1193) = 3.882, p = .049, eta2 = 0.01, and no significant effects for identification, F (1, 1193) = 0.029, p = .864, eta2 < 0.01, or the interaction, F (1, 1193) = 1.090, p = .297, eta2 < 0.01. Stemming from the same coding error, the results in section 7.2. Results. Exploratory analysis without exclusions are also erroneous and the correct analysis shows that main effects for involvement, F (1, 1591) = 1.833, p = .176, eta2 < 0.01, identification, F (1, 1591) = 0.845, p = .358, eta2 < 0.01, and their interaction, F (1, 1591) = 0.760, p = .383, eta2 < 0.01, are in fact all nonsignificant. These two ANOVA errors do not affect the subsequent Discussion of the Study 3 results in section 7.3 or the General Discussion in section 10 because we based those discussions on the pairwise comparisons of the pledges to the self-report condition. Writing errors We also found a few writing errors that should be corrected. In Studies 1, 2, and 3, the t-tests between the conditions should have been described as planned contrasts (as preregistered). In the following cases, the term “percent of problems” should have been “number of problems”: In Study 1, section 4.; in Study 2, section 5.2; and in Study 3, section 7.2. Additionally, in Study 3, section 7.2. Results. Exploratory 2 × 2 ANOVA Over All Problems, the analysis was actually preregistered and therefore not exploratory. Last, the paper refers to the OSF repository with the anonymized, view-only link. The public link is https://osf.io/hda7b Deviations from preregistrations requested by peer review During the review process, reviewers requested that we change some of our analyses and presentation of results to make them more intuitive, resulting in several deviations from our preregistrations that we failed to note in the published manuscript. First, all our preregistrations described the two dependent variables as “percent’ or “proportion” of problems reported as solved or actually solved, and our initial submission used that exact measure. Following the review team's requests, we changed the dependent variables to (raw) number of problems. This change does not affect any of the results because the percentages are a linear transformation of the number of problems. Second, our preregistrations of Studies 1 and 2 described analyzing the “cheating gap” (i.e., the differences between the percent of problems solved in each of the cheating conditions and percent of problems solved in the control condition), which our initial submission did. Following the review team's requests, we dropped the “cheating gap” and conducted the preregistered planned comparisons on the “number of problems” solved (i.e., self-report vs. each of the pledge conditions). Added Analyses. The review team requested additional analyses that we added to the paper without clearly specifying that they were not preregistered. In Study 3, section 7.2 Results. Mediator 1: Attention to the Content of the Pledge, we added a reverse mediation path by which attention to the content of the pledge (i.e., Recall Score) was the dependent variable and the number of all problems reported was the mediator. In Study 4, p. 8, section 8.2. Results and discussion, we added planned contrasts following the preregistered one-way omnibus ANOVA to be consistent with the preceding studies. Omitted Analyses. In some cases, we did not report some of the analyses that appeared in the preregistration. First, we preregistered two dependent variables: performance for all problems (main dependent variable) and performance for unsolvable problems only (secondary dependent variable)—a subset of all problems solved (Study 1: three out of 20; Study 2: two out of 10; Study 3: five out of 10; Study 4: seven out of 15 problems). Following the review team's advice that the analyses of the unsolvable problems distracted from the main message, we revised the paper to focus on performance for all problems solved and highlight only a small subset of the preregistered analyses of the unsolvable problems. We thus provide the omitted preregistered analyses for the secondary dependent variable below, which generally return weaker results that are less often significant but in the same direction as the results for the main dependent variable: (likely due to the lower measurement sensitivity afforded by fewer problems) Study 1, One-way ANOVA Over All Six Conditions. Difference for unsolvable problems statistically significant, as for all problems, F (5, 870) = 3.463, p = .004, eta2 = 0.02. Study 2, One-way ANOVA Over All Seven Conditions. Difference for unsolvable problems not statistically significant, unlike for all problems, F (6, 1367) = 1.86, p = .084, eta2 < 0.01. Study 3, 2 × 2 ANOVA. No significant main effect for involvement, F (1, 1193) = 2.395, p = .122, eta2 < 0.01, for unsolvable problems, unlike for all problems. No significant main effects for identification, F (1, 1193) = 0.850, p = .357, eta2 < 0.01, or their interaction, F (1, 1193) = 0.049, p = .824, eta2 < 0.01, for unsolvable problems, as for all problems. Study 3, Mediation Analyses. Because the ANOVA above did not return significant results, the mediation analyses for Mediator 1 (Attention to the Content of the Pledge) and Mediator 2 (Fear of Sanctions) would also show no significant results for unsolvable problems, as for all problems. Deviations from preregistrations made by the authors Added Analyses. In Studies 2 and 3, we added a one-way ANOVA over all pledge conditions only (i.e., excluding control and self-report). These revealed that the effects of the different pledges did not significantly differ. In Study 3, we mistakenly did not preregister that we would (consistent with our approach in Studies 1 and 2), follow the preregistered one-way omnibus ANOVA over all conditions with planned contrasts. Omitted Analyses. To simplify the paper, we did not replicate the ANOVA for Studies 1 and 2 using regression analyses. We had preregistered to use either the control or the self-report conditions as referents in these regressions (see Table S1 for the regression analyses with the control condition as the referent, and Table S2 for the regression analyses with the self-report condition as the referent). Note that the substance of the results of the regression analyses matches those from the pairwise comparisons reported on in the paper for all but one predictor in each of the two studies in Table S2 (noted with an asterisk). We also decided not to pursue examining differences in the distribution of the main dependent variable, performance for all problems, between conditions and in the share of “brazen” lies with Kolmogorov-Smirnov tests and follow-up chi-squared tests. We did so because it would have added an unnecessary layer of detail without changing the general conclusion of the paper. We likewise did not conduct any of the exploratory interaction analyses with demographic variables; researchers interested in conducting those analyses can do so by accessing our publicly posted data on OSF. Similarly, we had preregistered that we would compute and examine both of the dependent variables in Study 4 (i.e., their means and standard deviations) not just for each of the two matrix tasks separately but also overall. However, we ended up not including the examination of the overall means and standard deviations in the paper (i.e., they are not reported in Table 4) because we decided that they would obscure the separate examinations. Regarding those matrix tasks, we chose not to include the repeated-measures analysis that intended to examine the difference in reports between the first and second matrix task between conditions and replaced it with what we thought to be a better approach that considered the response to each problem over both matrix tasks (not preregistered, as disclosed in the paper). Finally, we only calculated the preregistered mediation analysis over the total number of problems reported as solved (i.e., across the two matrix tasks overall) in Study 4 and not for each matrix task separately. A moderated mediation analysis for the first matrix task showed that copying the pledge meaningfully reduced reports, b = −0.90, SE = 0.25, p < .01, 95% CI [−1.39, −0.42], but the 95% confidence interval around the indirect effect of recall included 0 whether the pledge was repeated, b = 0.14, SE = 0.13, 95% CI [−0.12, 0.41], or not, b = 0.13, SE = 0.12, 95% CI [−0.11, 0.35]; the confidence interval for the moderated mediation index likewise included 0, b = 0.02, SE = 0.024, 95% CI [−0.019, 0.076]. A moderated mediation analysis for the second matrix task also showed that copying the pledge meaningfully reduced reports, b = −0.75, SE = 0.26, p < .01, 95% CI [−1.26, −0.23], and the confidence interval around the indirect effect again included 0, whether the pledge was repeated, b = 0.038, SE = 0.14, 95% CI [−0.24, 0.32], or not, b = 0.033, SE = 0.13, 95% CI [−0.21, 0.28], as did the confidence interval around the moderated mediation index, b = 0.005, SE = 0.022, 95% CI [−0.041, 0.055]. Thus, the omitted separate analyses support the same conclusions as in the reported overall analysis: Attention to the content of the pledge did not significantly mediate the effect of a pledge's involvement on reducing self-reports.

Original languageEnglish
Article number104641
JournalJournal of Experimental Social Psychology
Volume114
DOIs
StatePublished - Sep 2024

Bibliographical note

Publisher Copyright:
© 2024 Elsevier Inc.

Fingerprint

Dive into the research topics of 'Corrigendum to “How pledges reduce dishonesty: The role of involvement and identification” [Journal of Experimental Social Psychology 113(2024) 104614] (Journal of Experimental Social Psychology (2024) 113, (S002210312400026X), (10.1016/j.jesp.2024.104614))'. Together they form a unique fingerprint.

Cite this