File size: 2,737 Bytes
0369477 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 |
FOR THE EDITOR / REPLICATOR:
In assembling the replication materials, we identified a handful of small changes to the empirical results. Of these, only one appears in the body of the paper and is a vanishingly small difference from the previous draft that we submitted. The remainder appear in the Supporting Information and are attributable to setting a random seed for replication, identifying an error in the code for one of the tables, identifying an error in one of the DMA shapefiles, and identifying an error in a function of a third party package we used. Importantly, none of these changes impact the substantive conclusion of our results. We describe each in detail below.
MANUSCRIPT:
- Figure 4: We identified an error in one of the DMA shapefiles which affected the vanilla diff-in-diff specification. The substantive interpretation of the results hasn't changed, and the interaction coefficient estimate has moved from -6.75 (2.14) to -6.97 (2.15).
SUPPORTING INFORMATION:
- Table SI-1: We identified an error in one of the DMA shapefiles which affected the comparison between dropped and kept counties in the FE specification. The substantive interpretation that these differences should work against our findings hasn't changed. Specifically, the Sanders 2016 voteshare differences have shifted from 39.9 / 46.2 / 0 / -8.9 to 39.7 / 46.1 / 0 / -8.9.
- Figure SI-3: We hadn't set a seed for these permutation results in our previous submission. We have now done so, resulting in insignificantly different distributions than what was originally submitted. The substantive interpretation of these permutation tests hasn't changed.
- SI page 15: We mistakenly reported that our choice of covariates left us with 152 observations for inference in the March subset. This number is actually 304 in the updated estimates contained in Table SI-5 (see note below) and should have been 370 in the previous draft.
- Figure SI-13 & SI-14: We noticed an error in the calculation of the dashed lines for these plots in the original package function. We have fixed the error and notified the package author. It has no impact on our findings or conclusions (and indeed is almost imperceptible to the naked eye).
- Figure SI-10: The inclusion of the caucus measures influenced the results displayed in this plot. The changes are very small (with the exception of the matched specification, they are basically identical) and do not influence the substantive conclusions.
- Table SI-5: A mistake in our previous submission failed to add Sanders 2016 and the caucus indicators as controls. We have fixed this mistake, resulting in slightly different coefficients. None of the substantive interpretations are affected. |