In Carey et al. (Reference Carey, Nyhan, Phillips and Reifler2023), we presented an experiment embedded within a panel survey assessing whether alerting Americans to either national, co-partisan, or opposing partisan descriptive norms of mask-wearing affected their own mask-wearing intentions. This descriptive norm was presented as the percentage of Americans, Democrats, or Republicans who claimed to wear masks at least ‘most of the time.’ We found that alerting participants to the national norm, but not the co-partisan or opposing partisan norms, increased mask-wearing intentions. This was particularly the case for Republicans and those who previously underestimated Americans’ levels of mask-wearing. Before proceeding in greater detail, we want to extend our extreme gratitude to Katharina Isser for uncovering some issues and bringing them to our attention. We regret the errors in the original.

Descriptive statistics on the percentage of respondents that under- and over-estimate Americans,’ Democrats’, and Republicans’ mask-wearing presented in Table 1 were incorrect. Part of this was due to coding error. The thresholds used for under-/over-estimation for Americans matched the full sample (70% and 90%, respectively), rather than the general population sample as pre-registered (64% and 84%, respectively). The threshold for under-/over-estimating Democrats’ mask-wearing needed to be slightly adjusted from 80%/98% to 79%/99%. Finally, instead of over-estimation being coded as 66% for Republicans, it was erroneously coded at 56%. However, even with the original code produced, discrepancies existed, for which the source is unknown. In the interest of full transparency, Table 1 presents the rates of under-/over-estimation reported in the original paper, by the replication code, and by the corrected code. The substantive interpretation of relative differences between the full sample, Democrats, and Republicans generally doesn’t change (though Republicans no longer underestimate national mask-wearing more than others), but absolute percentages do change.

Table 1. Comparison of code-returned; originally reported; vs. corrected results

Under-/overestimation thresholds used in the original code: 70%/90% (Americans); 80%/98% (Democrats); 47%/56% (Republicans). Corrected under-/overestimation thresholds: 64%/84% (Americans); 79%/99% (Democrats); 46%/66% (Republicans).

Changing the thresholds for under-/over-estimation also means that the results presented in Figure 2 slightly differ. The effects of being told about co-partisans’ or out-partisans’ mask-wearing on respondents’ mask-wearing intentions remain indistinguishable from zero, and the effect of being told about Americans’ levels of mask-wearing still has a positive and significant effect on mask-wearing intentions among those who previously underestimated it. However, in the original analyses, a positive and significant effect was observed (p<.05) among those who accurately estimated the national norm. This effect is no longer statistically significant (p=.08) using the new thresholds. The figure below (labeled Figure 2) depicts the altered version of this figure.

Figure 2. Effects of descriptive norms treatments among those who previously underestimated, accurately estimated, and overestimated descriptive norms of mask-wearing.

Changes flowing from these corrections also affected Tables A3, A4, A13, A14, and A15, which also have similar substantive interpretations as the original results so the new tables and figures are provided in an updated supplementary materials section.

Regression results depicted in Table A9 were slightly incorrect as a result of updating this particular table in LaTeX manually. Specifically, in Model 2, the following were incorrect:

• • The coefficient of the Wave 2 Fact-Check treatment is −.078 instead of −.026, and remains non-significant.

• • The standard error for the Out-Partisan Norms Treatment × Wave 2 Fact-Check interaction is .095 instead of .097, with the overall effect remaining non-significant.

• • The coefficient for the constant is 3.436 instead of 3.435, remaining positive and statistically significant.

• • The R² for the model is .289 instead of .290.

In Model 3, the following were incorrect:

• • The Co-Partisan Norms Treatment’s standard error is .089 instead of .088, with the overall effect remaining non-significant.

• • The standard error for the Out-Partisan Norms Treatment × Wave 2 Fact-Check interaction is .097 instead of .090, with the overall effect remaining non-significant.

Predicted probabilities depicted in Table A12 slightly differ from what was reported for the full sample, though the interpretation of the predicted probabilities remains similar: those in the American Norms Treatment condition display higher mask-wearing intentions than those in the control condition.

In Table A25, the percentage aged 18-34 in the control condition is incorrect as a result of a typographical error – it was reported to be 109.0%, but is actually 10%.

In Table A26, missingness levels in the control condition were incorrect and lower than originally reported. Rather than being more common in the control condition, missingness on key variables is more comparable to other conditions.