Slant is 2.3× larger in tight races than in landslides

🟡 β by race-margin tertile: tight = +10.07 (p < 0.001), mid = +8.19 (p < 0.001), wide = +4.39 (p = 0.094). Strong monotonic gradient.

Consistent with both coordination (M+1 cutoff matters more when margins are tight) and bandwagon (more room to flip apparent leader when leader's lead is small). Doesn't discriminate between voter-side theories on its own.

Sources.

Own analysis: build/table/heterogeneity.csv (margin_split rows); AN-004 § Follow-ups
Cross-refs: coordination-shift finding; H6 coordination-peak; H7 bandwagon-peak

Cited analyses

AN-004 yellow causal

Peak β shifts from rank 2 in small munis (+10.14) to rank 3 in runoff-eligible munis (+9.43). Direction matches Cox's M+1 rule (M+1=2 vs M+1=3); bandwagon would predict rank 2 in both. The cleanest design-level discriminator the project supports with current data.