Methodology is a feature, not a footnote. Every PolitoPolls release leads with how the data was collected, who was interviewed, and how the numbers were weighted — in plain language anyone in the room can defend.
Defaults, not constraints. Every project is scoped to the question — but these are the principles we don't bend on.
Modeled likely-voter universes drawn from voter file data, segmented by partisanship, region, age, race/ethnicity, gender, and turnout score. Quotas set against known electorate composition; weights applied to known demographic and turnout targets post-field.
Live phone (landline & cell), interactive voice response, SMS-to-web, and online panel — blended in proportions calibrated to the universe. The mode mix is documented on every release.
Instruments pre-tested for clarity, balance, and order effects. Message-testing modules use randomized exposure with control groups so we can isolate true effect from acquiescence bias.
Iterative proportional fitting (raking) on demographics and political behavior. Margin of error reported including design effect. Sub-group reads disclosed with appropriate cautions on sample size.
English and Spanish field as standard. Mandarin, Vietnamese, Haitian Creole, Tagalog, and additional languages are available depending on universe and budget.
Field dates, sample size, mode mix, weighting variables, and margin of error released with every topline. For public-release work, full questionnaires and crosstabs are posted alongside the report.
The biggest difference between a poll that holds up and one that doesn't usually isn't the question — it's the sample frame. Self-reported "I will vote" misses on both ends: it includes habitual non-voters who say yes anyway, and excludes irregular voters who don't see themselves as political.
We build likely-voter universes from registered-voter file data using turnout score, vote history, and modeled propensity. The universe is then weighted post-field to known targets on age, race/ethnicity, gender, education, region, and projected turnout composition — the variables most predictive of actual electoral behavior.
Statewide benchmark study, n=600 likely voters:
Mix calibrated by universe demographics. Older, rural, and lower-income populations weight more heavily to live phone; younger and urban populations weight to text and web.
PolitoPolls work aligns with the AAPOR Transparency Initiative — the disclosure standard developed by the American Association for Public Opinion Research and adopted by reputable firms across the industry.
Who paid for the research. Public-release work names the sponsor in the topline. Confidential client work names the sponsor internally and to anyone redistributing the data.
Who was interviewed — registered voters, likely voters, primary electorate, full adults — and the sample size, including effective sample size after weighting.
Exact start and end of fieldwork, and the proportion of interviews completed by each mode (live phone, IVR, SMS-to-web, online panel).
The full list of demographic and behavioral variables used in post-stratification weighting, with target distributions from a named source (Census, voter file, prior election).
Reported including design effect — the more honest number — not just the textbook formula. Sub-group margins disclosed where reads are reported.
Full questionnaire (with question order and randomization noted) released alongside any publicly-released study. Available to clients on request for confidential work.
The only honest accuracy test for a polling firm is how its pre-election numbers compare to certified results. We benchmark our work against final election outcomes and publish the deltas — including the misses.
PolitoPolls follows the disclosure standards advanced by the American Association for Public Opinion Research. For confidential client work, the same standards apply internally — every project file includes the same documentation, even when the results are never published.
Request a Sample Topline See Services