Per-pollster β ranges from −10 to +30 across 33 firms

🟡 Per-pollster β across 33 institutes with ≥ 5 self-sponsored polls ranges from −10.5 (Paraná Pesquisas) to +30.3 (Intenção Instituto, n = 5 — noisy). Mid-volume institutes (Verita, IIP, Census) sit near zero.

The "Brazilian pollster" average of +8 disguises substantial firm- level heterogeneity. Several established firms are essentially unbiased on this metric. The reputation-equilibrium prediction (higher candidate-share → higher β) has the right direction (+13.6 unweighted slope, p = 0.40) but is underpowered at n = 11 institutes.

The IIP/Census exception — large firms with high candidate-share but near-zero β — is consistent with the secondary "volume → discipline" prediction.

Sources.

Own analysis: build/table/per_pollster_beta.csv; build/table/pollster_customer_mix.csv; build/figure/pollster_beta_vs_customer_mix.pdf; AN-007
Cross-refs: H10 pollster-reputation; theory.md § Pollster reputation and customer-mix sorting

Cited analyses

AN-007 yellow descriptive

Per-pollster β regressed on candidate-sponsored share gives slope = +13.58 (unweighted) / +6.28 (n-weighted), both positive but underpowered (p=0.40 / 0.47, n=11). Direction matches the reputation-equilibrium prediction; the two highest-volume firms (IIP, Census) sit near β=0 against the strict monotone version.