H8: β is larger for first-time / lesser-known candidates (information cue)

H8: β is larger for first-time / lesser-known candidates (information cue)

The headline sponsor effect (H1) is an average over very different candidates. If voters treat poll standing as a quality signal and Bayes-update from prior to posterior, the amount of belief movement scales inversely with the precision of the prior. Slant inserted into a poll about an entrenched incumbent should do little; slant inserted into a poll about a first-time candidate with no prior public record should do a lot. The quantity of interest is whether β interacted with a candidate-experience indicator is positive — that is, whether the sponsor effect is heterogeneously larger where voters' priors are diffuse.

Evidence strength: Not tested. No analysis (AN page) has yet estimated β heterogeneity by candidate experience. The framework is taken seriously because it is the distinguishing prediction of one of the three voter-side theories in theory.md; the empirical pass is queued behind the rank/runoff heterogeneity work in H6 and H7.

Theory

The framework is Polls as quality cues (Bayesian information) (theory.md §"Polls as quality cues"). Voters hold a noisy prior over candidate c's quality θ_c and observe poll standing s_c as a public signal of θ_c. Bayes-updating implies that posterior movement per unit of signal scales inversely with prior precision: a well-known incumbent has a sharp prior and the poll barely moves the posterior; an unknown challenger has a diffuse prior and the same poll shifts the posterior a lot. The cue-taking / low-information-rationality tradition gives the cognitive microfoundation [cite:lupia1994shortcuts; cite:popkin1991reasoning].

If voters then treat slanted polls the same way they treat honest polls — i.e., they do not fully discount sponsor identity (see theory.md §"Why bias survives voter discounting") — the sponsor captures a return that scales with how diffuse the prior is. β should therefore be largest where priors are weakest.

Prediction

Two-way β heterogeneity by (final-rank-quintile × first-time-candidate) on the analysis panel. The information-cue prediction is that β is highest among inexperienced candidates regardless of rank — the row "first-time = 1" dominates uniformly across rank quintiles. Concrete candidate-level priors to operationalize:

The cleanest contrast is the first-time-candidate indicator: a binary variable derived from the 1998–2024 candidato panel.

Competing predictions

Coordination (H6). β peaks at the M+1 viability cutoff regardless of candidate experience. If H6 is the true mechanism, the two-way interaction should show β-by-rank dominating and the first-time-candidate dimension dropping out conditional on rank.

Bandwagon (H7). β peaks just below the front-runner regardless of experience. Same prediction structurally: the rank dimension dominates the experience dimension.

Volatile-sponsor confound. First-time candidates may attract a systematically different sponsor mix — e.g., smaller regional firms, new entrants, less-vetted contratantes. Any β difference along the experience dimension could then reflect sponsor heterogeneity rather than voter inference. Pollster-firm fixed effects (cf. AN-016 within- firm β, AN-059 firm FE decomposition) are the natural absorbing control.

Open testability concern (carried from theory.md §"Polls as quality cues"). The "quality signal" channel is a posited cognitive process — voter inference is not directly observed. The heterogeneity test is indirect; even a clean experience-axis difference does not by itself pin down belief-updating as the mechanism.

Prior research

The cue-taking tradition is well established [cite:lupia1994shortcuts; cite:popkin1991reasoning], but there is no direct test on the slanted-poll subset. Polls have been studied as information shortcuts; sponsor-induced slant has been studied separately (online opinion-survey experiments [cite:leeper2019sponsorship; cite:crabtree2020sponsorship]). The intersection — does slanted poll information move belief more when priors are weaker — is novel.

Evidence

Analysis Bearing Key takeaway
(none yet) — No AN page has tested the experience-heterogeneity prediction. Closest neighbors are the rank-based heterogeneity analyses (AN-026, AN-027, AN-045, AN-050), which test coordination/bandwagon predictions on the rank axis but do not split on candidate experience.

Open tests

Two-way β heterogeneity: rank × first-time-candidate

The headline test. Estimate β separately by (rank-quintile, first-time-candidate) cell on the analysis panel and read off where the peak sits. Information cue predicts a uniform shift along the experience dimension; H6/H7 predict the rank dimension dominates. The β cube (rank × runoff-eligibility × first-time) is summarized in theory.md §"Discriminating among the voter-side theories" as the within-paper test that separates the three voter-side theories.

Robustness to sponsor mix

Re-estimate the same heterogeneity absorbing pollster-firm FE (cf. AN-016, AN-059) to rule out the volatile-sponsor confound. A first-time-candidate effect that survives firm FE is much harder to read as sponsor heterogeneity.

Endorsement / coverage gradient

Theory.md prediction 2 (quality cues): candidates with strong endorsements or heavy prior media coverage should show smaller β. Requires an external candidate-coverage measure; blocked on coverage-extraction infrastructure (the news pipeline used in references/news/stories.csv is a starting point but does not yet yield per-candidate coverage counts).

Data requirements

Status. Queued. Empirical pass blocked behind the Cluster C rank/runoff heterogeneity work in H6 and H7; the two-way cube is the joint output.