How CivicLens surveys are built, fielded and weighted.
Mode
CivicLens surveys are conducted by SMS text-to-web. Sampled voters receive a text message inviting them to complete the survey, and the questionnaire is administered online to their cell phone. Respondents answer at their own pace, without an interviewer present.
We use this mode consistently rather than switching between modes across surveys. Mode affects results — people answer some questions differently to a live interviewer than to a screen — so holding it constant means that movement between waves reflects the electorate rather than a change in how we asked.
Sample
Sample is drawn from the appropriate state’s voter file. Every respondent is a matched, registered voter with a known registration record and vote history, not an anonymous panelist recruited from the open internet.
This matters most in the elections we survey. Primaries and runoffs turn out a small, unrepresentative slice of registered voters, and a sample of the general public tells you very little about who will actually decide them. Starting from the voter file lets us build the universe around real participation history rather than around self-description alone.
Sample source for each survey is named in that survey’s methodology block.
Screening
Every CivicLens survey opens with a turnout intention question, and the universe is defined from the combination of that answer and the respondent’s vote history on file. The screen is published as the first question on every topline, so readers can see exactly how the likely-voter universe was constructed rather than being asked to accept the label.
Weighting
Completed samples are weighted to age, gender, region and vote history. Targets are built to the expected composition of the electorate being surveyed rather than to the registered population as a whole.
We weight on these four variables consistently across surveys. Where a particular survey requires anything beyond them, that is stated in its methodology block.
Margin of error
Margins of error are reported at the 95% confidence level and apply to the full sample.
Question design
Ballot tests name every candidate on the ballot and offer an explicit undecided option. We report undecided voters as undecided.
Full question wording is published for every question, in the order asked. Where candidate order is rotated, or a question is asked of a split sample, that is disclosed on the topline.
What we publish
Every release includes the field dates, mode, universe, sample size, margin of error, sample source and weighting variables, along with downloadable toplines containing every question as it was read.
Corrections
If we publish an error, we correct it in place and note what changed and when. We do not quietly edit a published figure. If you believe you have found a mistake in one of our releases, write to civiclensresearch@gmail.com and we will look at it.
Questions about methodology, including requests for detail not covered here, go to the same address.