How These Polls Were Conducted

Here are the key things to know about these polls:

• Interviewers spoke with 713 registered voters in Arizona, 682 registered voters in Georgia and 682 registered voters in North Carolina from Sept. 17 to 21, 2024.

• Times/Siena polls are conducted by telephone, using live interviewers, in both English and Spanish. Overall, about 97 percent of respondents were contacted on a cellphone for these polls.

• Voters are selected for the survey from a list of registered voters. The list contains information on the demographic characteristics of every registered voter, allowing us to make sure we reach the right number of voters of each party, race and region. For this poll, interviewers placed more than 250,000 calls to more than 116,000 voters.

• To further ensure that the results reflect the entire voting population, not just those willing to take a poll, we give more weight to respondents from demographic groups that are underrepresented among survey respondents, like people without a college degree. You can see more information about the characteristics of our respondents and the weighted sample at the bottom of the page, under “Composition of the Sample.”• The margin of sampling error among likely voters is plus or minus 2.5 percentage points across the three states, plus or minus 4.4 percentage points in Arizona, plus or minus 4.6 percentage points in Georgia, and plus or minus 4.2 percentage points in North Carolina. In theory, this means that the results should reflect the views of the overall population most of the time, though many other challenges create additional sources of error. When computing the difference between two values — such as a candidate’s lead in a race — the margin of error is twice as large.

If you want to read more about how and why The Times/Siena Poll is conducted, you can see answers to frequently asked questions and submit your own questions here.

Full Methodology

The New York Times/Siena College polls were conducted in English and Spanish on cellular and landline telephones in Arizona, Georgia and North Carolina from Sept. 17 to 21, 2024; In all, 2,077 likely voters were interviewed. When all states are joined together, the margin of sampling error is plus or minus 2.5 percentage points for the likely electorate and plus or minus 2.4 percentage points among registered voters.

For each state poll, the margin of sampling error among likely voters is plus or minus 4.4 percentage points in Arizona, plus or minus 4.6 percentage points in Georgia, and plus or minus 4.2 percentage points in North Carolina.

Sample

The survey is a response rate-adjusted stratified sample of registered voters on the L2 voter file. The sample was selected by The New York Times in multiple steps to account for differential telephone coverage, nonresponse and significant variation in the productivity of telephone numbers by state.

To adjust for noncoverage bias, the L2 voter file for each state was stratified by statehouse district, party, race, gender, marital status, household size, turnout history, age and home ownership. The proportion of registrants with a telephone number and the mean expected response rate were calculated for each stratum. The mean expected response rate was based on a model of unit nonresponse in prior Times/Siena surveys. The initial selection weight was equal to the reciprocal of a stratum’s mean telephone coverage and modeled response rate. For respondents with multiple telephone numbers on the L2 file, the number with the highest modeled response rate was selected.

Fielding

The sample was stratified according to political party, race and region and fielded by the Siena College Research Institute, with additional field work by ReconMR, the Public Opinion Research Laboratory at the University of North Florida, the Institute for Policy and Opinion Research at Roanoke College, the Center for Public Opinion and Policy Research at Winthrop University in South Carolina and the Survey Center at University of New Hampshire. Interviewers asked for the person named on the voter file and ended the interview if the intended respondent was not available. Overall, across all three samples, 97 percent of respondents were reached on a cellular telephone.

The instrument was translated into Spanish by ReconMR. Bilingual interviewers began the interview in English and were instructed to follow the lead of the respondent in determining whether to conduct the survey in English or Spanish. Monolingual Spanish-speaking respondents who were initially contacted by English-speaking interviewers were recontacted by Spanish-speaking interviewers. Overall, across all three samples, 11 percent of interviews among self-reported Hispanics were conducted in Spanish, including 14 percent of weighted interviews.

An interview was determined to be complete for the purposes of inclusion in the ballot test question if the respondent did not drop out of the survey by the end of the two self-reported variables used in weighting — age and education — and answered at least one of the age, education or presidential election ballot test questions.

Weighting — registered voters

The survey was weighted by The Times using the R survey package in multiple steps.

First, the sample was adjusted for unequal probability of selection by stratum.

Second, each poll was weighted to match voter file-based parameters for the characteristics of registered voters.

The following targets were used:

• Party (party registration if available in the state, else classification based on participation in partisan primaries if available in the state, else classification based on a model of vote choice in prior Times/Siena polls) by whether the respondent’s race is modeled as white or nonwhite (L2 model)

• Age (Self-reported age, or voter file age if the respondent refuses) by gender (L2)

• Race or ethnicity (L2 model)

• Education (four categories of self-reported education level, weighted to match NYT-based targets derived from Times/Siena polls, census data and the L2 voter file)

• White/non-white race by college or non-college educational attainment (L2 model of race weighted to match NYT-based targets for self-reported education)

• Marital status (L2 model)

• Home ownership (L2 model)

• Turnout history (NYT classifications based on L2 data)

• State region (NYT classifications)

• Method of voting in the 2020 elections (NYT classifications based on L2 data)

• Party (NYT classifications) by age (L2), in North Carolina

• Census tract educational attainment, in Arizona and Georgia

Finally, the sample of respondents who completed all questions in the survey was weighted identically, as well as to the result for the general election horse race question (including leaners) on the full sample.

Weighting — likely electorate

The survey was weighted by The Times using the R survey package in multiple steps.

First, the samples were adjusted for unequal probability of selection by stratum.

Second, the first-stage weight was adjusted to account for the probability that a registrant would vote in the 2024 election, based on a model of turnout in the 2020 election.

Third, the sample was weighted to match targets for the composition of the likely electorate. The targets for the composition of the likely electorate were derived by aggregating the individual-level turnout estimates described in the previous step for registrants on the L2 voter file. The categories used in weighting were the same as those previously mentioned for registered voters.

Fourth, the initial likely electorate weight was adjusted to incorporate self-reported intention to vote. Four-fifths of the final probability that a registrant would vote in the 2024 election was based on their ex ante modeled turnout score and one-fifth based on their self-reported intentions, based on prior Times/Siena polls, including a penalty to account for the tendency of survey respondents to turn out at higher rates than nonrespondents. The final likely electorate weight was equal to the modeled electorate rake weight, multiplied by the final turnout probability and divided by the ex ante modeled turnout probability.

Finally, the sample of respondents who completed all questions in the survey was weighted identically, as well as to the result for the general election horse race question (including leaners) on the full sample.

The margin of error accounts for the survey’s design effect, a measure of the loss of statistical power due to survey design and weighting.

The design effect for the full sample is 1.43 for the likely electorate in Arizona, 1.51 for the likely electorate in Georgia, and 1.26 for the likely electorate in North Carolina.

Among registered voters, the margin of sampling error is plus or minus 4.1 points in Arizona, including a design effect of 1.27; plus or minus 4.3 points in Georgia, including a design effect of 1.34; and plus or minus 4.2 points in North Carolina, including a design effect of 1.26.

For the sample of completed interviews, among the likely electorate, the margin of sampling error is plus or minus 4.7 points in Arizona, including a design effect of 1.40; plus or minus 5.1 points in Georgia, including a design effect of 1.59; and plus or minus 4.6 points in North Carolina, including a design effect of 1.28.

Historically, The Times/Siena Poll’s error at the 95th percentile has been plus or minus 5.1 percentage points in surveys taken over the final three weeks before an election. Real-world error includes sources of error beyond sampling error, such as nonresponse bias, coverage error, late shifts among undecided voters and error in estimating the composition of the electorate.