A larger and more solid gap than Hillary Clinton ever enjoyed.
by Matthew Yglesias, Vox
A Monday morning CNN poll showed Joe Biden with a staggering 14-point lead over President Trump as the electorate’s stated level of concern with “race relations” soars and the former vice president is seen as much better equipped to handle the issue.
Winning the popular vote by such a large margin would likely mean Democrats overperformed in battleground states and in places like Georgia, Iowa, and Texas that would put the Senate clearly in play.
And while the CNN poll is just one poll, and something of an outlier at that, there is now a very clear trend in national polling — Biden was winning before the outbreak of massive national protests in the wake of George Floyd’s death, and that lead has gotten bigger.
Maybe not 14 points bigger, but bigger than it was before and clearly larger than any lead Hillary Clinton ever held in the 2016 campaign.
The polls are getting better for Biden across the board
As Nate Cohn, the New York Times’s polling and data guru, explains, every single high-quality national poll with proper education weighting had Biden leading two months ago. And almost all of them have him leading by more today. The exception is the NBC/WSJ poll, which has him flat at a 7-point lead.
That basically gives us a full ‘wave’ of the major national polls for the first time since 3/15-4/15, and Biden’s lead has grown by an average of 4 points since then pic.twitter.com/PV2MTuXKSy— Nate Cohn (@Nate_Cohn) June 8, 2020
Some analysts, like Economist data journalist G. Elliott Morris, suspect this may have something to do with nonresponse bias. This is a phenomenon that is now fairly well-documented whereby when a given candidate has a bad news cycle, his supporters don’t actually turn against him but do become less likely to speak to pollsters. A nonresponse theory, in other words, would posit that Trump supporters have become less proud to be supporting him and less likely to respond to polls — thus biasing the sample toward Democrats.
Hard to believe there’s not a ton of partisan nonresponse going on right now. Would like modelers/political data journos to at least engage with that possibility, if not adjust for it explicitly. The trend in polls that correct for partisan balance is flat (same w online polls). https://t.co/EPZOE8XIoV— G. Elliott Morris (@gelliottmorris) June 8, 2020
It is always a good idea to have some skepticism about large apparent shifts in the polls. But things like weighting polls for partisan balance can overcorrect for nonresponse bias by ruling out the possibility that people are switching out of a party with an unpopular leader (you may recall this as the “unskewing” controversy from the 2012 presidential cycle).
One way or the other, CNN polling analyst Harry Enten assures us that the share of Democrats in their sample didn’t change — Biden’s lead got bigger because he consolidated Democratic support and is doing better with independents.
This is different from Clinton’s polling lead
For reference, Biden’s lead in the polls is larger than Hillary Clinton’s was in early June. He’s up by about 8 points in the current RealClearPolitics average while Clinton’s lead was closer to 5 points. Clinton did break out to about a 7-point lead in mid-October after widespread discussion of a recording in which Trump can be heard talking about sexually assaulting women.
But Biden’s lead is not only larger than that, it’s more secure.
Even on October 18, Clinton was only at 46 percent in the polls with Trump doing terribly at 39 and plenty of undecided and third-party voters. Today’s average, by contrast, has Biden at 50 percent with Trump at 42. That much smaller number of undecided or third-party voters augurs a more stable race and a more difficult road for Trump to a come-from-behind victory.
Of course, people who remember confident forecasts of a Clinton win may have their doubts. But it’s important to recall that this was more a case of bad forecasting than of bad polling. In the 2016 cycle, the non-FiveThirty Eight forecasters generally modeled each state separately. That meant that if Clinton had a narrow polling lead in Pennsylvania, and a narrow lead in Wisconsin, and a narrow lead in Michigan, and a narrow lead in Florida, and a narrow lead in North Carolina, the models treated it as reasonably likely that she might in fact lose one of those states but incredibly unlikely that she would lose all five.
A coin toss is 50-50, but the odds of losing five coin tosses in a row are about 3 percent. Many models treated Clinton’s odds in each swing state as somewhat better than a coin toss, thus creating the result that it was extraordinarily unlikely that she would lose all five.
A different way to think about it is that if polls in Michigan are wrong because they undercounted the number of non-college white Trump voters, this should be a factor in every state with large numbers of non-college whites. That’s almost every state in the union, and certainly every swing state. When you understand polling error that way, then Clinton’s real but narrow lead in the polls should have been understood as making it unlikely that she would lose but only moderately so.
This Biden lead is different. If you’re up 10 points in the polls, then the polls could be off by 4 points (which would be a big but not mind-blowing error) and the Electoral College could have a 4-point skew toward Trump (which is at the upper range of plausibility but not totally out there) and Biden would still win. Basically, if you are up 10 in the polls, you are almost certainly going to win.
None of this means that Biden will win in November, simply because a lot of things will happen between now and then. But we have very good reason to believe that if the election happened tomorrow, Biden would win — for it to be otherwise would involve a massive breakdown in polling, not a “normal” kind of statistical error.