A recent independent audit (https://error.reviews/reviews/lades-2020/) of our 2020 paper made us aware of four issues with this paper:
i) The line that referred to the multiple comparisons should have made clear that these were additional to the main analysis in the paper to avoid confusion. Table S2 in the Supporting Information (which is presented again below as Table 2) is transparent about the use of the Benjamini-Hochberg adjustments.
ii) Alongside our fixed effects, random effects, and fixed effects with multiple comparisons tables, we could have included a fixed effects model clustering for standard errors at the individual level. While we think it is not unconventional or misleading to present fixed effects and random effects models in the way we did, we agree that clustering has analytical advantages and would have been preferable. We added the regression table from a fixed effects model with standard errors clustered at the individual level below in this letter as Table 1. To facilitate the direct comparison between the regression table in the Supplementary Information of our paper and the regression table with clustered standard errors, we present both tables below as Table 1 and Table 2.
iii) Table S2 in the Supplementary Information of the paper does include an error (also highlighted in red below in Table 2). We included a significance flag “†” indicating that the association between “Schooling children” and “Positive affect” is significant after Benjamini-Hochberg adjustment, which it is not. We are grateful to the external reviewers for highlighting this error.
iv) The external reviewers noticed six duplicate ids (essentially three cases in which two people were assigned the same participant id) in the raw data-file. The external panel provider gave us the data with these duplicate ids. Basic demographics that we got directly from the panel provider are the same across these sets of observations, but they differ in the demographic data we elicited ourselves. We believe it is correct to include all 6 observations and assume they are independent of each other in the analysis.
We appreciate the opportunity to reflect on potential improvements to our analysis and the presentation of the results. None of the above issues change the substantive conclusions of the paper.