Social Sciences Correction for “A field experiment on community policing and police legitimacy,” by Kyle Peyton, Michael Sierra-Arévalo, and David G. Rand, which was first published September 16, 2019; 10.1073/pnas.1910157116 (Proc. Natl. Acad. Sci. U.S.A. 116, 19894–19898).
The authors note a recently discovered coding error that affects the results for one of the four primary outcome measures reported in Fig. 1 of this paper. After correcting this error, the estimated effects of the community policing treatment on perceptions of police legitimacy are larger than previously reported. The substantive conclusions are unchanged. This affects the point estimates and confidence intervals in Fig. 1 (Legitimacy panel) of the paper and the corresponding point estimates and standard errors reported in column 2 of Table S13 in SI Appendix. None of the other results reported in the paper or SI Appendix are affected by this error. The originally published estimate for the intent-to-treat effect (ITT) on police legitimacy was 2.15 ( = 0.54) 3d after the intervention and 1.50 ( = 0.65) 21d after the intervention. The revised estimates are 3.60 ( = 0.57) and 2.41 ( = 0.71), respectively. The originally published estimate for the average treatment effect on the treated (ATT) was 5.17 ( = 1.29) 3 d after the intervention, and 3.67 ( = 1.59) 21 d after the intervention. The revised estimates are 8.69 ( = 1.33) and 5.85 ( = 1.71), respectively. The error arose because two of the eight Likert items used to construct the legitimacy measure were coded in the wrong direction in the follow-up surveys. Specifically, the statement for one of the seven-level items was worded such that respondent disagreement indicated greater legitimacy than agreement. Responses to this item should have been reverse-coded so that higher numeric values reflected greater legitimacy. This transformation was mistakenly applied to an adjacent item that should not have been reverse-coded. Consequently, two items in the legitimacy index were negatively correlated with the remaining six. The authors are grateful to Matthew H. Graham for discovering the coding error necessitating this correction. The corrected figure and its legend appear below, and the online version has been updated. The SI Appendix has been updated online.
Fig. 1.
Effect of community policing treatment on primary outcome measures. Primary outcome measures are indices of multiple questions and scaled to range from 0 to 100. Legitimacy is an index of responses to 8 statements about police legitimacy (e.g., “They make fair and impartial decisions”). Performance is an index of responses to 4 statements about police effectiveness (e.g., “I have confidence that the police in New Haven can do their job well”). Cooperation is an index of responses to 4 questions about willingness to assist police (e.g., “If the police were looking for a suspect who was hiding, and you knew where that person was, how likely would you be to provide the police with information?”). Compliance is an index of responses to 4 questions about willingness to comply with police directives (e.g., “If the police tell you to do something, you should do it”). Covariate-adjusted point estimates and 92% confidence intervals are constructed using the prespecified levels and estimation procedures described in the PAP (SI Appendix, Appendix C). Point estimates and SEs are presented in tabular form in SI Appendix, Table S13. SI Appendix, section 2 has all individual questions.