We all want to know what will happen in November’s elections. This column won’t tell you.
We will get all the answers, y’all. In just under three weeks.
In the meantime, every indication is that Democrats have the advantage. What we don’t know is whether this will be a big year for Democrats or a really big one. Either way, Democrats will win the House majority. House Speaker Mike Johnson has to keep hope alive, but the rest of us can be realistic. Congressional poll results average about a 9-point generic ballot advantage for Democrats.
But if this is a really big year, Democrats should flip the Senate as well. Polls look like we’re headed in that direction, but what if they’re wrong?
Races in states like Nebraska and South Carolina look competitive. Texas might elect the first statewide Democrat in more than three decades. Sen. Susan Collins might finally succumb to political gravity in Maine. Democrats in Ohio and Iowa consistently lead in polls. Republicans reduced campaign spending in Georgia and North Carolina because they no longer look competitive. President Trump flew to Alaska because it’s a true toss-up.
Many of these races, however, still sit within the margin of error. The biggest question is whether the polls have the electorate right or are undercounting Republicans who aren't excited about voting this year but might still turn out. We’ve been to this rodeo many times, and pollsters have adapted our processes to account for nonresponse patterns that caused us to underestimate Trump’s support in the past. These problems have been less prevalent in midterm elections, but reaching Republicans is still a challenge, especially when the party is having a tough time, as it is now.
Some of our adaptations require doing what we used to say was bad practice. I’m old enough to remember when pollsters thought it was wrong to adjust for respondent education level because we don’t really know the electorate’s educational attainment. It’s now standard. Many pollsters now also weight by partisanship or the last election's outcome to adjust for Republican nonresponse, even though it’s difficult to pinpoint the true proportion of Republicans in a population, and the electorate often differs from what it was in the last election.
These help us get close to the right number of Republicans, but even if we do, it’s unclear whether we talk to the right kind of Republicans. If we only talk to the type of Republicans who are wavering on their vote, we may be missing base Republicans who will vote red even if they do think Trump is underperforming. Overestimating the former or the latter makes poll error worse, not better.
I saw these issues in my company’s Centerline National Poll. Republicans are less likely to say they are “certain” to vote this year, but more likely to say they “probably” will vote. While many pollsters define likely voters as only “certain” voters, I included both, since research suggests about 70 percent of “probably” voters do cast a ballot. That decision is the difference between a 12-point Democratic lead with “certain” voters and an 8-point Democratic lead that includes probable voters.
Another big red flag for nonresponse error was religious affiliation. Many polls don’t even ask respondents’ religion, but it's deeply intertwined with politics. Plus, we have excellent benchmarks for what religious affiliation should be in the population. White evangelicals, a staunchly Republican, high-turnout group, were dramatically underrepresented in my poll. Instead of 13 percent of the population, I had a meager 6 percent.
I adjusted religion, and Republicans gained a point on the congressional ballot even though Trump lost a point on his approval rating. The overall proportion of Republicans in the data was unchanged; my decision up-weighted a type of Republican who wasn't responding: consistent GOP voters who are unhappy with Trump.
The argument for not weighting them up is that people who don’t answer surveys won't vote. But according to exit polls, white evangelicals always turn out in high numbers, no matter how bad the year for Republicans. There is little reason to think they won’t now. Either way, they should have shown up in my survey of all adults at their population rate. This is survey nonresponse.
Survey error could go the other way, too, and Democrats could win by a greater margin than expected. I didn't see any red flags in the data warning of that scenario, but it could happen.
Ultimately, we can’t know until after the election, but pollsters shouldn’t wait for another big polling error before innovating and challenging our thinking. I opted to adapt by using the religion weight and a broader definition of likely voters, then being transparent about it to show clearly where poll error could exist this year.
Contributing editor Natalie Jackson is founder and principal of Centerline Research and Strategy.
