Rich Frank @richwfrank.bsky.social · 25/08/2026Finding #3. Level of analysis matters more than we usually admit. Of the 13 robust predictors, only two (population size and domestic conflict) work at both levels. A competitive election predicts violence nationally but not sub-nationally. 9/n 100
Rich Frank @richwfrank.bsky.social · 25/08/2026Sub-nationally the question shifts from where to whom. Specifically, younger citizens and female candidates face higher risk. But most studies do not test the whom. 81% of subnational models never say who the perpetrator was, and 97% of national models never say who the victim was. 10/n 100
Rich Frank @richwfrank.bsky.social · 25/08/2026My takeaways for future research: - Focus more on election-specific triggers, not just slow-moving structural factors, - Better id perpetrators & targets, - stop generalizing across levels of analysis, - fix measurement. Findings from 157 measures from 40 sources are unlikely to generalize. 11/n 100
