A holistic view of conditional American support for political violence.

We write to correct misrepresentations of our book (1) in Westwood et al.’s (2) article on the American public’s violent political views. The article strains to frame itself as a refutation, but, in fact, it counters a straw man.

The article’s (2) empirics are presented as a novel, necessary critique, but they largely replicate and extend the findings in our book (1), which we sent the authors several months before their publication. The article repeatedly cites our book but does not represent it. Instead, the article focuses on brief, necessarily simplified media reports published while our book was in production.

The timing of academic publishing means the article (2) rushed into print before our book (1), so its selective descriptions of our work left reviewers unable to fully weigh the article’s claims and contributions.

Like the article (2), our book (1) begins by showing that a small minority of the public holds violent political views on most questions, but levels of support depend greatly on details. In fact, most of the book focuses on how question wording and political context change observed support for political violence. That contingency undermines the article’s search for an exact level of support for violence.

The article (2) calls for specific questions to clarify “violence.” We agree. Our book (1) includes two dozen questions, general and specific, with a variety of response scales. We know what people meant by “violence” because we asked them in follow-up questions. In short, the article calls for measures that we already employed in the book.

Across all our items, violent responses range from 1% (self-report of physically aggressive political behavior) to 80% (US Revolution’s violence). That sensitivity to detail undermines the article’s (2) other major argument—that respondent inattention drives our results.

The article (2) uses an attention check to discard “inattentive” respondents, asking respondents to identify the state described in a vignette about violence—an insubstantial detail in a 100-word passage. Those who got it wrong endorsed violence more. Notably, the article’s SI Appendix fails its own check by mislabeling states for the vignettes. That’s a forgivable mistake, but it illustrates why excluding all wrong responses is risky.

The idea that inattentive respondents might inflate percentages is smart, and the truly inattentive should be discarded. However, measures must discern better to retain sincere responses, especially since close attention correlates with trait aggression—a strong predictor of violent political views. Aggressive respondents are more likely to miss insubstantial details of a vignette and to endorse violence. This does not automatically invalidate their stated attitudes.

Finally, the article (2) challenges our evidence linking violent partisanship to aggressive personality, but its results replicate our correlation. Likewise, our tests and theirs (ref. 2, figure 3) show that violent views are related to partisan contexts, confirming that politics matters beyond general aggressiveness, despite the article’s efforts to downplay partisan elements.

We welcome new measures and analysis on these challenging questions. Careful measurement and holistic interpretation of all evidence is essential. This is the main purpose—and contribution—of our work (1), but you would not know it from the article (2).