Analyses

Every analysis run for this project, one document per entry — generated from docs/reference/analysis-index.yaml. Each AN-NNN records the question, script, design, results, and a confidence tag. Filter with the chips below.

Type
Cluster

Descriptive 37

Sample composition, balance, raw means, exploratory cuts — no causal claim.

AN-007Does a pollster's β depend on the share of their polls that are candidate-sponsored vs media-sponsored?yellow
AN-019AN-019: Does coverage_class cluster differently across sponsor types?yellow
AN-020AN-020: Does coverage_class cluster by specific sponsor route?red
AN-021AN-021: Does audit_pct distribution differ by sponsor type?yellow
AN-022AN-022: Does methodology completeness differ by sponsor type?yellow
AN-023AN-023: Pollster methodology fingerprint by candidate sharegreen
AN-024AN-024: Is deferral itself a sponsor-specific tell?green
AN-025Within race, do polls from high-volume firms get more media coverage than polls from low-volume firms? If yes, the media filter is the structural mechanism that gives reputation-by-volume its bite — and explains *why* big firms self-discipline (AN-018).yellowmechanism
AN-026Do leading candidates commission self-sponsored polls less than runners-up (selection), and does the within-candidate sponsor bias β vary by the sponsor's eventual rank under strict identification (race × week FE)?yellowmechanism
AN-027Does the AN-026 'winners over-commission' selection finding survive when rank is measured at commissioning time rather than at the eventual outcome, and does conditioning on race competitiveness recover the coordination-story prediction?yellowmechanism
AN-028Does the AN-027 rank-at-commissioning result (coordination signal in tight races; rank-1 commissioned polls show less slant) sharpen, attenuate, or hold when the temporal anchor moves from focal poll's date_end (AN-027 default) to date_start (the institutionally cleanest commissioning-latest-bound) or date_registered (literal regulatory filing moment)?yellow
AN-029Does the AN-027 / AN-028 two-mechanism interpretation of self-sponsor selection — safe-race leaders over-commission because they have money to spend; tight-race runners-up over-commission for strategic-voting demand — survive a direct campaign-finance-revenue control?yellowmechanism
AN-030Joint test of the AN-027/028/029 trilogy: does adding log(receita_2024) to the rank-at-commission selection regression (date_start anchor) attenuate the tight-race rank-2 coordination signal, the safe-race rank-1 dominance, or both?yellowmechanism
AN-032Within a sponsored × independent pair on the same muni × same candidate × ±14 days, does the sponsored poll's bairro selection over-represent the sponsor's party's prior-cycle (2020) strongholds relative to the matched independent poll? A positive paired contrast would be direct Channel A 'Bayesian cluster selection' evidence.yellowmechanism
AN-041AN-041: Does mode of data collection differ between sponsored and matched independent polls?green
AN-042AN-042: Does the sponsored side document interviewer training and supervision less than the matched independent?green
AN-043AN-043: Does nonresponse / undecided handling vary between sponsored and matched independent polls?yellow
AN-051AN-051: Does questionnaire-side name rotation and scenario filing differ between sponsored and matched independent polls?green
AN-053AN-053: Does the sponsor's candidate appear earlier in the headline roster on the sponsored side?green
AN-054AN-054: Does AN-051's rotation contrast map onto AN-018's firm-size discipline gradient?green
AN-055AN-055: Coverage class × candidate-base interaction (cheap Tier 2)yellow
AN-056AN-056: Within-pair quota distribution distance × bias contrastyellow
AN-057AN-057: Within-pair ponderação description × bias contrastyellow
AN-060AN-060: Race-level visibility shrinks the sponsor effectyellow
AN-061AN-061: Same-firm × same-race customer-discrimination testgreen
AN-062AN-062: Sponsor effect shrinks with independent-poll densitygreen
AN-065Do sponsored polls in firm-race cells with no media coverage show different bias than sponsored polls in covered cells?yellow
AN-069Does the within-poll sponsor-minus-opponent error gap widen between matched independent and sponsored polls? A DiD that complements digit forensics in narrowing the mechanism space.greenmechanism
AN-071AN-071: Per-firm accuracy versus sponsor biasyellow
AN-072AN-072: Candidate-level fraud-suit rate by sponsor exposure (2024)yellow
AN-072V2AN-072v2: Poll-level fraud-suit rate by candidate sponsorshipyellow
AN-073AN-073: Firm specialization by party in candidate-sponsored pollsyellow
AN-074AN-074: CPF cell repeat-dyad test — M1-individual vs M4 single-shotyellow
AN-078AN-078: Disclosure rate × sponsor route × dyad kind — M1 vs M5 on the committee gapyellow
AN-079AN-079: Cost × sponsorship × bias — is slant priced?green
AN-080AN-080: Slant unit — firm vs. statistician decompositionyellow
AN-081AN-081: Audit-request outcomes + follow-on linkageyellow

Causal 10

Causal estimates: main effects, heterogeneity, decompositions, formal tests.

AN-001H1: do Brazilian 2024 mayoral polls overstate the candidate that commissioned them?green
AN-004Does the position of peak sponsor-bias β shift at the 200k-registered-voters runoff-eligibility cutoff, as the coordination theory predicts?yellow
AN-005Does sponsor-bias slant decrease as election day approaches?yellow
AN-006Does β vary by which sponsor route (CPF / committee / party / party-name) maps to the candidate?yellow
AN-033AN-033: Does coverage deferral amplify sponsor bias?green
AN-040AN-040: Does coverage deferral amplify sponsor bias *only* for leaders?green
AN-045AN-045: Where does the rank-2 sponsor over-statement live? Race-margin decompositionyellow
AN-050AN-050: AN-045 sharpened with ex-ante rank (rank-at-commission)yellow
AN-075AN-075: CPF cell β decomposition — repeat vs singleton dyadred
AN-077AN-077: All-routes repeat-vs-singleton β decompositionyellow

Placebo 1

Falsification — effects that should be zero.

AN-003Does a candidate's poll number jump above their preceding independent-poll number when they commission a poll themselves?green

Robustness 22

Does the main result hold under alternative definitions, sample selection, and measurement validation?

AN-002Does the +7.75 headline survive when the comparator is restricted to contemporaneous independent polls in the same race?green
AN-008Does the within-candidate FE design work on the SP slice before the all-Brazil LLM extraction finishes?yellow
AN-009Is the within-candidate sponsor coefficient statistically heterogeneous across parties, or is the pooled +7.98 a sufficient summary?green
AN-010Does the +7.94 spec-3c sponsor coefficient survive an adverse referee's five highest-leverage attacks: pollster_self contamination of the comparator (K1), renormalization-denominator asymmetry (K2), Route B vice-prefeito mis-tagging (K3), drop_absorbed-induced sample selection (K4), and Route D party-regex false positives (K5)?green
AN-011Could the +7.94 spec-3c coefficient be generated by chance given the (race × week) FE structure alone? And is it concentrated in any single pollster firm or state?green
AN-012Does the spec-3c +7.94 coefficient survive (a) wild-cluster bootstrap inference (sharper than CRVE asymptotics at n=60 clusters), (b) sample-size weighting, and (c) alternative time-window definitions for the race × week FE?green
AN-013Do digit-frequency patterns in `poll_percent_raw` differ systematically between the sponsor's own candidate and other candidates? Differential within-poll anomaly would signal Channel B residual fabrication on top of the design-driven Channel A slant.greenmechanism
AN-013V2AN-013 v2: forensics targeting AN-013's blind spotsgreen
AN-014Is the K2 +2.88 pp renormalized-vs-raw gap driven by sponsors listing fewer candidates in their commissioned polls (red-team Channel-A design lever), or is it the mechanical multiplicative effect of renormalization on a non-zero raw within-candidate sponsor effect?greenmechanism
AN-015Could the +7-8 pp sponsor coefficient be driven by systematic LLM extraction-quality differences between sponsored and independent polls (e.g., sponsored-poll PDFs being professionally laid out and producing different extraction biases), rather than a real sponsor effect?green
AN-016Within each pollster firm — where PDF style, extraction-template, and house methodology are held fixed — does the within-candidate sponsor coefficient survive? Distribution range and per-firm significance are the strongest available test that the headline is not driven by between-firm extraction-style differences.greenmechanism
AN-017Does the AN-007 prediction — per-firm β increasing in the firm's candidate-sponsored customer-mix share — sharpen with the 3× larger sample from AN-016's within-firm β estimation?yellowmechanism
AN-018Does firm volume (log of total polls produced in 2024) explain AN-016's cross-firm β dispersion better than AN-017's candidate-share-of-customer-mix?greenmechanism
AN-031Does the AN-006 CPF route coefficient β = +19.12 pp survive wild-cluster bootstrap inference at n_self=18, and how do the other three route coefficients fare under the same test?green
AN-052AN-052: Does the AN-051 rotation contrast survive within firm?green
AN-059Of the headline +7 pp sponsor effect, how much is within-firm methodology slant vs sponsor selection of high-β firms? Add firm FE to the headline cand-FE spec and read off the difference.greenmechanism
AN-063Does Spec 3c survive when the comparator pool excludes pollster-self polls and is restricted to media-only?green
AN-064Do the spec-2 and spec-3c headline p-values survive alternative cluster choices and finite-sample-valid wild-cluster bootstrap inference?green
AN-066Do the within-pair Channel-A directional tests survive multiple-testing correction?green
AN-067Does the headline β depend on the choice of which estimulado scenario to use when a protocol reports multiple?green
AN-068Does the Spec 2 headline survive a random reassignment of sponsored_by labels — i.e., is the (sponsored_by, error) correlation a feature of the specific assignment, or could it arise under random pairings?green
AN-076AN-076: CPF β decomposition without institute FEyellow