Technical Documentation

Survey Methodology

A full account of how Pollfinity Research designs, fields, weights, and reports public opinion surveys — and why we believe methodological transparency is non-negotiable.

Overview

Pollfinity Research employs an SMS-to-web panel methodology anchored in probability-based sampling from the L2 national voter file. Our approach combines the reach and cost efficiency of digital outreach with the statistical rigor of a known sampling frame — a combination that, when properly executed, produces results comparable in accuracy to traditional telephone surveys at a fraction of the cost.

Every design decision we make — from sample selection to weighting scheme to margin-of-error calculation — is documented and published alongside our topline results. We believe polling consumers deserve to evaluate the technical underpinnings of any survey before placing weight on its conclusions.

How We Build a Survey

Sampling Frame: L2 Voter File

Our sampling universe is drawn from the L2 national voter file — a commercially maintained database of registered voters that aggregates official state voter registration records with commercially sourced consumer and demographic data. L2 provides individual-level records including name, address, party registration, vote history, age, and in many states, gender and race/ethnicity.

Drawing from a registered voter file rather than random-digit dialing (RDD) or opt-in internet panels provides a defined, probabilistic sampling frame with verifiable demographic characteristics — a significant methodological advantage over non-probability approaches.

Outreach: SMS-to-Web Invitation

Selected respondents receive an SMS message from Pollfinity Research containing a unique survey link. The link directs respondents to a mobile-optimized web survey hosted on Typeform. Outreach is conducted in compliance with the Telephone Consumer Protection Act (TCPA); respondents may opt out at any time by replying STOP.

SMS outreach rates among registered voters are substantially higher than cold-call telephone response rates, and the web survey format allows for richer questionnaire design including visual ballot-style question presentation.

Likely Voter Screening

Raw sample draws from registered voter records. We apply a likely voter screen based on a combination of respondent self-report ("How likely are you to vote in the November 2026 election?") and verified vote history from the L2 file. Respondents with low self-reported probability and no recent vote history are excluded from topline results.

The specific likely voter model applied to each survey is documented in that survey's technical report.

Demographic Weighting

Completed interviews are weighted to match the demographic and geographic characteristics of the likely voter population. Weighting targets are derived from the L2 voter file and, where applicable, from state Board of Elections registration data.

Standard weighting variables include: party registration, geographic region, age cohort (18–34, 35–49, 50–64, 65+), and gender. Additional variables (race/ethnicity, education) may be incorporated on a survey-by-survey basis where sufficient cell counts permit reliable estimates.

Survey Platform: Typeform

Surveys are hosted on Typeform, a GDPR-compliant web survey platform with strong mobile responsiveness. Typeform's one-question-at-a-time format reduces question-order effects and improves completion rates relative to traditional grid-style survey interfaces.

Response data is stored in Pollfinity Research's private account and is not shared with or accessible to Typeform for any marketing or third-party purpose.

AI-Assisted Automation

Pollfinity Research uses artificial intelligence tooling to automate routine survey production tasks: sample selection and stratification, SMS outreach sequencing, data cleaning, weighting computation, and results formatting. AI automation enables us to operate at lower cost than traditional survey organizations without sacrificing statistical rigor.

All AI outputs are reviewed by a human researcher before any results are published. We do not use AI to impute, synthesize, or fabricate survey responses.

Margin of Error and Sampling Error

The margin of error (MoE) for a simple random sample of n = 1,500 at the 95% confidence level is approximately ±2.5 percentage points for a 50%/50% proportion — the most conservative estimate. For proportions further from 50% (e.g., 20% or 80%), the margin narrows.

The formula for a proportion's 95% confidence interval margin is:

MoE = 1.96 × √( p(1−p) / n )

Reported margins of error apply to the full sample. Subgroup analyses (e.g., results among Republicans only, or among voters in New York City) carry larger margins owing to smaller cell sizes. Pollfinity Research reports subgroup margins in its full technical crosstabs.

Important caveat: Sampling error is one source of total survey error. Other sources include non-response bias, coverage error (voters not in the voter file), question wording effects, and weighting model assumptions. We endeavor to minimize each of these and to document them honestly.

Independence and Funding

Pollfinity Research LLC does not accept funding from candidates, campaigns, political action committees, political parties, or advocacy organizations. All survey work is self-funded. This independence is structural: it is enforced by our operating policy, not merely stated as an aspiration.

We publish results regardless of which candidate or position leads. We do not offer private or proprietary polling services and have no commercial relationship with any electoral actor.

Transparency Standards

For every published survey, Pollfinity Research provides:

  • Full questionnaire wording, in order of administration
  • Field dates (start and end)
  • Sample size and completion rate
  • Sampling methodology and population definition
  • Weighting variables and targets
  • Margin of error for full sample and major subgroups
  • Funding and independence disclosure

Full crosstabs are available upon request to journalists and academic researchers. Contact polls@pollfinity.com.

Quality Control

Prior to publication, each dataset undergoes the following review steps:

  1. Data cleaning: Removal of straight-line responses, implausibly fast completions, and respondents who fail embedded consistency checks.
  2. Weight trimming: Individual weights are capped at 3.0× and floored at 0.25× the unweighted cell proportion to prevent extreme observations from distorting results.
  3. Internal consistency check: Topline figures are cross-validated against subgroup breakdowns to identify anomalies.
  4. Human review: A Pollfinity researcher reviews all outputs before publication and signs off on the final release document.

Contact and Data Requests

Journalists, academics, and researchers may request full crosstabs, raw (anonymized) data, or technical documentation for any published Pollfinity Research survey. Contact: polls@pollfinity.com.