Here are the nuts and bolts of how our poll works. Read about the thinking behind the choices we made, and the trade-offs involved.
The New York Times Upshot/Siena College Research Institute poll is a response-rate-adjusted, probability-proportionate-to-size survey of the likely electorate.
Telephone numbers were selected from an L2 voter file stratified by age, region, gender, party, race and turnout in 2014. The probability of selection was inversely proportionate to telephone coverage in each strata and the probability of response, weighted by a modeled turnout score.
Voters were contacted on cellular and landline telephones. Interviewers asked for the person named on the voter file, and ended the interview if the intended respondent was not available. Interviews were conducted in English, and also in Spanish in districts where at least 10 percent of registered voters were Hispanic, per L2 data.
The sample is weighted to match Upshot estimates for the composition of the likely electorate by age, race, region, turnout, party, gender and education.
The exact categories used for weighting may vary from poll to poll, depending on the availability of voter file data or the importance of weighting on the interaction between two weighting targets, like between region and party, or race and education.
The party weight is partisan primary vote history or party registration, or, in the states where neither is available, a model of likely support for Donald J. Trump in the 2016 presidential election.
Voter file data is used for weighting, except for education.
We derived estimates for the composition of the likely electorate by age, race, turnout, party, gender and region from a vote-history-based model of turnout in the 2014 midterm election, adjusted by the partisan and geographic turnout patterns of 2017 and 2018 special and general elections in Arizona 8, New Jersey, Virginia, Georgia 6, Pennsylvania 18 and Ohio 12. The estimates are then adjusted to match estimates for turnout by district, based on a model of off-year, regularly scheduled election turnout from 2005 to 2017.
The estimates for education are based on a model of turnout in the November 2014 voting and registration supplement to the census Current Population Survey, adjusted for changes in turnout by education in the Virginia and Ohio 12 elections. The adjustment is based on a model of validated turnout among Upshot/Siena poll respondents that controls for the variables in the C.P.S.-based model.
The final survey weight adjusts the likely electorate weight to account for the self-reported turnout of poll respondents. The final survey weight is equal to the likely electorate weight, divided by the initial probability of voting and multiplied by the final probability of voting, which incorporates self-reported turnout. The final turnout score is 56 percent turnout model and 44 percent self-report, based on a model of validated turnout in prior Upshot/Siena polls.
The “live” data is weighted gradually, beginning at N=50. The maximum survey weight is equal to 2 percent of the sample, and the minimum survey weight is equal to 1/2 percent of the sample^2. At n=50, the minimum and maximum weight equals 1. The likely voter adjustment begins at N=1.
The margin of error is adjusted to account for the loss of statistical power due to weighting, known as the design effect, including the final likely voter adjustment.