The survey is a size-adjusted probability stratified sample of the L2 electorate. The sample was selected by The New York Times in several steps to account for differential telephone coverage and non-response.
To adjust for noncoverage bias, the L2 voter pool for each congressional district was first stratified by state district, party, race, gender, marital status, household size, voting history, age, and homeownership. For each stratum, the proportion of registrants with a telephone number and the average expected response rate were calculated. The average expected response rate was based on the unit nonresponse model of previous Times/Siena surveys. The initial selection weight was equal to the reciprocal of the stratum’s average telephone coverage and the modeled response rate. For respondents with multiple phone numbers in the L2 file, the number with the highest modeled response rate was selected.
Second, the probability of selection was weighted by the respondent’s probability of voting in the 2022 election, based on a model of turnout in 2018 and 2010 as a function of the respondent’s political and demographic characteristics.
The sample was stratified by party, race, and region and housed at the Siena College Research Institute, with additional fieldwork by ReconMR, IPOR at the University of Roanoke, and PORL at the University of North Florida. Interviewers asked for the person listed on the voter list and ended the interview if the intended respondent was not available.
In Kansas, Nevada, and New Mexico, the tool was translated into Spanish by ReconMR. Bilingual interviewers initiated the interview in English and were instructed to follow the respondent’s lead in deciding whether to conduct the survey in English or Spanish. Monolingual Spanish-speaking respondents initially contacted by English speakers were re-contacted by Spanish-speaking interviewers.
The survey was weighted by The Times using the R survey package. Survey weights were trimmed to the 99th percentile.
First, the sample was adjusted for unequal probability of selection by strata.
Second, the sample was weighted to match parameters based on voter files for the characteristics of the likely electorate. The sample was also weighted to match census estimates of the educational attainment of the likely electorate.
Estimates of the characteristics of the likely electorate were based on several models. First, the model estimated likely turnout by state based on recent turnout and midterm race competitiveness. Second, an individual-level turnout model in 2010, 2018, and 2021 was used to estimate the probability that registrants will participate in the midterm elections depending on their demographic and political characteristics. Finally, individual-level estimates were adjusted to match expected participation by state.
Third, the intention to vote by itself was included in the estimate of the respondent’s probability of voting. This was based on a model of verified voter turnout in Times/Siena polls from 2016 as a function of one’s voting intention and the probability of turnout modeled before the survey, including adjustment for the higher turnout of survey respondents than non-respondents.
Fourth, the final probability voter weight was equal to the original probability voter weight multiplied by the final turnout probability including one’s voting intention divided by the initial modeled turnout probability.
The following voter-based targets were used to weight the sample to match the characteristics of registered voters and the likely electorate:
• Party (NYT classification based on L2 data and in states without party registration or primary voting history partisanship model based on previous Times/Siena polls)
• Age (age given in the self-report or age in the electorate if the respondent refuses)
• Plant (model L2)
• Gender (gender stated in the self-report or gender in the voter register, if the respondent refuses)
• Marital status (model L2)
• Home ownership (model L2)
• Region (defined by NYT)
• Turnout history (NYT classification based on L2 data)
• How to vote in the 2020 election (NYT classification based on L2 data)
The following census-based objectives were used to weight the sample to match the characteristics of registered voters and the likely electorate:
• Educational attainment (NYT model based on ACS and CPS data)
• White/nonwhite x four-year college graduation (NYT model based on ACS and CPS data)