POLLS OCTOBER 31, 2013
Only a couple weeks have elapsed since the shutdown ended, but there’s already a few bits of evidence that the Democratic wave has subsided. Last night, an NBC/WSJ survey showed Democrats ahead by just 4 points on the generic ballot among registered voters, down from an 8 point lead at the heart of the shutdown. Earlier in the day, Democracy Corps released a survey showing that Republican incumbents in competitive districts led by the same margin that they did in June.
These numbers aren't good enough for the Democrats. I don’t think anyone seriously believed Democrats were poised to retake the House in June, so it should be pretty disappointing for Democrats if they haven't made any gains in the horse race. The survey also highlights the GOP's advantage in the battlegrounds. Incumbent Republicans in the 25 most competitive battlegrounds have higher approval ratings and favorability ratings than the president. The numbers in the Democratic-held battlegrounds are just as bad for Democrats as the numbers are for Republicans.
A 4 point Democratic advantage among registered voters just isn’t impressive—it’s only slightly larger than the 3 point Democratic advantage in the NBC/WSJ survey conducted before the shutdown. Historically, a 4 point Democratic advantage among registered voters is not consistent with a tsunami. Ahead of 2006 or 2008, for instance, the NBC/WSJ survey never showed Democrats with such a slim advantage. Democrats need a 2006, 2008-esque wave to expel Republicans incumbents from their lean-red districts.
It’s important to remember that these are just two polls. If you see a shift in two polls, and declare it to be a real sign of a change, you’re going to get burned with some regularity. Put differently: the next 5 polls could show Democrats ahead by 8 on the generic ballot. But it shouldn’t be too surprising if these findings are ultimately confirmed by other surveys. It’s not hard to imagine why the Democratic advantage would have faded, as the shutdown is over and the Obamacare website is a catastrophe.
There’s an even more intriguing possibility: The Democratic wave was never as large as the polls suggested. It’s possible that the polls were affected by differential non-response bias, where one group (in this case Republicans) is less likely to complete a survey than other voters. The heart of the problem is that if the composition of the new sample is different than the last one, changes in the results could be due to change in the sample, not preferences of the target population. It’s possible that no one changed their mind, except about whether to take the poll. So it’s somewhat harder to trust the polls when there’s a big news event that might be discouraging certain types of voters from participating. Many believe that this is at least partially responsible for the vaunted, but short-lived post-convention “bounce.”
There’s growing evidence, in my view, that lower response rates are increasing the risk of fluctuations due to differential non-response. In 2012, it appears that response rates got so low, and started fluctuating so wildly, that Pew Research decided to weight by the average of past respondent’s self-reported vote in their final pre-election survey. After the first presidential debate, Romney led in many national surveys, but we can say in retrospect that it is highly unlikely that Romney ever held a lead at all. Meanwhile, the panel based surveys, like RAND, YouGov showed the president maintaining a lead throughout the month. The panel-based surveys could compensate for differential non-response because they were re-contacting respondents and could see whether voters were changing their mind.
In the past, I wouldn’t have given much thought to whether differential non-response was affecting the polls. There was solid evidence that it wasn’t an issue. Now I’m not so sure. The next five polls, after all, might show Democrats ahead by eight. If the polls confirm NBC/WSJ, it doesn’t mean the polls are wrong, but it’s a reason why I’m increasingly interested in looking at data over a longer period of time, especially in the aftermath of high-profile events.