Abstract
<p>Turnbull-Dugarte and López Ortega's "Still Instrumentally Inclusive" (the Response) answers my replication (the Comment) of their 2024 American Political Science Review paper (the Article) with, among other things, a multiverse of 2,970 analyses based on eighteen different weighting schemes, randomization inference, and a trichotomous moderator. As I show in this rejoinder, each is undermined by bona fide errors in their code, their writing, and their interpretation. First, the multiverse is just many correlated variants of the same test. This is in large part due to the weighting schemes which suffer multiple problems (some due to code errors, some intrinsic), and are largely correlated variations on one weighting scheme. Second, due to code errors the randomization inference compares a weighted statistic against an unweighted null and, in the two heterogeneity tests, permutes the interaction column rather than treatment assignment column. Every permutation p-value the Response reports is therefore computed against the wrong reference distribution; when they are recomputed correctly, two of the five it reports as significant are no longer so, and they are the two the Response emphasizes. Third, the trichotomous moderator is erroneously described as terciles, but it is in fact based on non-tercile cutpoints. The key subgroup effect reported in the Response thus rests on comparing just thirty high-weight control respondents against thirty-one high-weight treated respondents, and ends up reaffirming my original point that the heterogeneity in study 2 is with respect to weights, and not immigration sentiment as their theory predicts. I close by documenting seven claims in the Response that are untrue or misleading.</p>