The end result is that we know less about the opinions of this decisive slice of the electorate, which is demographically and politically distinct from the country as a whole. To help fill this gap, The Post teamed up with George Mason University’s Schar School for Policy and Government to conduct a survey exclusively of voters living in battleground congressional districts. The 69 competitive districts were defined as those rated as “tossup,” “leaning Republican” or “leaning Democratic” by the Cook Political Report on Aug. 24, as well as a few additional districts identified by Post staff.
In designing the survey, we identified that our standard approach of randomly sampling cell and landline phone numbers with professional interviewers would have significant limitations in sampling voters in specific congressional districts. Telephone area codes have never been an accurate indicator of phone users’ congressional districts, especially given that many cellphone users move to a different state or region while keeping the same area code. In addition, voters may be unable to accurately report the congressional district in which they are registered, which could lead us to survey the wrong set of voters.
To address this challenge, in the Battleground District Poll we used registration-based sampling, which relies on drawing a sample from official databases of registered voters produced by each state and compiled by L2, whose database was used to draw this survey’s sample. L2’s database and others like it indicate the congressional district in which each voter is registered, making it possible to draw a sample of voters registered in a specific set of congressional districts.
Most registration-based surveys are conducted by telephone, though this carries a challenge of its own, since for a sizable percentage of voters listed in such files the phone number is either missing, inaccurate or nonworking.
Mailing addresses, however, are available for nearly all registered voters, offering the ability to reach a sample of the entire registered voter population. The survey research firm SSRS of Glen Mills, Pa., worked with The Post and the Schar School to design a survey that contacted a random sample of voters living in one of the 69 battleground congressional districts identified by The Post. The sample was stratified by frequency of voting in recent general elections; more regular voters were given a greater chance of selection to increase the sample size of likely voters. The eventual sample was weighted to correct for differences in chances of selection by historical turnout.
Voters were contacted in mid-September with a letter invitation to participate in the survey online using a computer or mobile device and were also sent a second reminder letter halfway through the survey period. Voters who did not have Internet access were invited to take the survey by phone. The selected voters were given a unique passcode to take the survey and security measures were put in place to ensure no individual could take the survey multiple times. Respondents were given a small monetary incentive or gift card when the survey was completed.
Respondents completed the survey from Sept. 19 to Oct. 5, a longer field period than most political surveys, but a fairly short field period for mail-recruited surveys. The period included the final weeks of the Supreme Court nomination period for Brett M. Kavanaugh and started a few days after the publication of Christine Blasey Ford’s allegations that Kavanaugh sexually assaulted her when both were teenagers. A total of 3,407 registered voters across the 69 districts completed the survey.
Once all responses were completed, the sample was divided into districts won by Donald Trump and Hillary Clinton in 2016, with each set weighted to its actual proportion of battleground districts. Respondents within each set were weighted to match the gender, age, race/ethnicity and educational makeup of registered voters in these combined districts. Demographic targets for age and sex were derived from L2’s database. Targets for race/ethnicity and education were drawn from analysis of the Census Bureau’s large-scale American Community Survey and Current Population Survey, which measures voter registration in even years.
The sample was also weighted to match the share of voters who are registered as Democrats and Republicans in states that allow registration by party. In states without party registration, voter-file indicators of turnout in a recent Democratic primary or Republican primary were used as a weighting benchmark. Voters whose L2’s files indicated they voted in 2016 were weighted to match the percentage of voters who actually supported Clinton and Trump, according to final vote tallies in battleground districts.
To identify likely voters, the sample of registered voters was analyzed to assess each respondent’s likelihood of voting. The L2 voter database includes records of validated turnout in past elections, while the survey included two measures of current enthusiasm to vote, including respondents’ self-described certainty of voting in the election and how closely they said they are following the election. The turnout and enthusiasm indicators were combined in accordance with how well they predicted voting in 2017 elections for Virginia governor and U.S. Senator in Alabama.
The analysis assigned each voter a probability of turning out in the 2018 election, which was used to produce results among likely voters. In the end, the results for congressional vote preference among likely voters are quite similar to registered voters overall.