Brazilian presidential runoff
Final-share uncertainty
Scenarios
Every adjustment reconditions the behavioral simulations on the same polling data.
Sources of uncertainty
Variance allocation between current voter-state uncertainty, future changes and model assumptions. Individual transfer, turnout, geographic and measurement effects overlap.
Methodological sensitivity
Evidence
Institute-by-institute reliability audit
All reported modes come from the prior frozen polling dataset. Original sampling plans, response rates and weights were not independently accessible. Missing disclosure is not treated as proof of defective sampling. The exclusion effect below measures influence, not quality.
| Institute · reported mode | Surveys | Exclude: Δ F mean |
|---|
Potential bias differs by mode: online panels face recruitment/selection uncertainty, phone surveys coverage and nonresponse uncertainty, and face-to-face surveys location/interviewer-selection uncertainty. These are risks, not established errors for a named institute.
Model and identification
First-round regional candidate totals constrain correlated voter-transition simulations for all eliminated candidates, retention/switching among finalists, abstention, blank/null ballots, and re-entry into valid voting. Geographic allocation of minor electorates is partially pooled, not directly observed.
The polling likelihood fits multiple dated waves with correlated house, survey-mode and time-specific errors; latent trend is regularized. Four competing measurement-error models are averaged with explicit assumed weights of 40%, 25%, 20% and 15%. These are not validation-derived probabilities.
National polls cannot identify each transition-matrix cell or the sign of a common survey-selection bias. Remaining campaign shocks use broad heavy-tailed priors. Interactive changes are counterfactual assumptions, not new observations.
Posterior numbers are weighted Monte Carlo estimates. Model calibration on the unobserved second-round outcome is impossible at the forecast cutoff. Published polling percentages have not received an extra correction for first-round polling misses.
Source status
First-round totals and October polling inputs were retained from the previous dashboard. Primary institute methodological documents could not be retrieved in this session. The baseline therefore does not claim a completed independent methodological audit.