Some people have reacted to the shocking upset win by Bernie Sanders in Michigan on March 8 irrationally, as though we should never trust another poll or there’s some meta-phenomenon lifting Sanders to victory everywhere. As smart people stare at the numbers, though, we’re getting a picture of the Michigan upset that suggests some reason to wonder about Clinton’s polling margin in some states, but not necessarily in others. It’s worth a comparison of Michigan to the states that vote on March 15, since Clinton leads the polls in all of them.
According to the RealClearPolitics average of recent polls (which had Clinton up by 21 points in Michigan), she leads Sanders by 31 points in Florida and Illinois and by 20 points in North Carolina and Ohio. Remarkably, there have been no public polls in Missouri since August, but Clinton led Sanders in that one by 28 points. Recognizing that the p0lling numbers can change in the five days left before these states vote, it’s still logical to stipulate that Clinton’s leads in Florida and Illinois are probably large enough that even a once-in-a-lifetime polling error like that which occurred in Michigan isn’t going to obliterate it. The other states are within the political earthquake zone, in theory at least.
Now, if you look at what seems to have happened in Michigan, some additional comparisons can be made. The best comprehensive analysis we have so far is by Carl Bialik of FiveThirtyEight. He looks at multiple factors and ultimately concludes it was something of a perfect storm — itself pretty good grounds for doubting it’s going to happen again in any state in particular:
This is an outlier, a perfectly rotten combination of bad luck and bad timing. Several pollsters pointed out that they used the same methods in the Michigan Democratic primary as in other primaries — including Michigan’s Republican primary — with relative success.
Having said that, Bialik does look at some individual factors that helped produce the upset. Let’s look at them in terms of where they might be replicated on March 15:
1) Sanders benefited (a factor I emphasized in my own insta-take on March 8) from a really large spike in the youth vote, and then posted his customary (for northern states, at least) insanely high preference percentage among them. This is without question replicable in other states, depending on Sanders’s degree of campus organization and voter targeting. Replicable? Yes.
2) Sanders also posted — at 28 percent — his biggest share of the African-American vote yet (other than in his very white home state of Vermont). This c0uld in part be a function of the huge youth vote. But there’s been some speculation about northern black voters diverging from their southern counterparts due to factors ranging from religion to unionization to racial polarization. If there’s some truth to this, it might help Bernie in Ohio and Illinois, but not in Florida and North Carolina (border state Missouri is harder to categorize). Replicable? Probably only in the north.
3) There were far more independent voters (27 percent) participating in the Michigan primary than expected, and Sanders won more than 70 percent of them. This won’t happen in closed-primary Florida, but it could happen in the other March 15 states. Replicable? Yes.
4) There’s some evidence Clinton voters may have crossed over to the GOP primary to vote against Donald Trump. To the extent this phenomenon depended on overconfidence among Clinton supporters, the Michigan results probably ensure it won’t happen again. And Democrats cannot cross over in Florida or in North Carolina. Replicable? Probably not.
5) A potentially large factor that I also noted Tuesday night is that a lack of relevant Michigan precedents made it hard for pollsters to construct an accurate model of the state’s primary electorate. The state usually held caucuses until 2008, and that primary — held under the shadow of national party sanctions for violating the calendar rules — was marred by Barack Obama’s absence from the ballot. That should not be a factor in any of the March 15 states. Replicable? No.
6) And, finally, an upset the magnitude of Michigan’s has to make you wonder if there was a late trend pollsters failed to pick up — perhaps related, as most of the punditocracy quickly concluded, to the Sunday candidate debate in Flint. But as Bialik notes, the exit polls didn’t show some late-decider advantage for Sanders at all, and certainly nothing that would explain a 20-point polling error. But yes, it’s always smart to pay most attention to late polls so long as they are conducted by reputable pollsters.
7) There is another timing issue that Bialik doesn’t go into that should help avoid big polling errors in the March 15 states: Michigan is not an early-voting state at all; even absentee voting requires a defined “excuse.” All but one (Missouri) of the March 15 states allow for both in-person and no-excuses absentee balloting. That means pollsters have the opportunity to get an earlier look at actual voting trends, and a significant share of voters are by definition impervious to late trends. Replicable? Only in Missouri.
Adding all of this together, you’d guess Florida is as close to a lock for Clinton as you will find, and the state where Sanders is close enough and where many of the same factors that won Michigan for Sanders could be present again is Ohio. To be clear, this is not because of some midwestern rust-belt populism trend whereby the whole region is a ripe target for Sanders because he has voted against a hundred trade agreements or his democratic socialism is inherently attractive to the horny-handed sons and daughters of toil that older journalists remember from the distant past. If that were the case, Illinois could suddenly feel the Bern — but I just don’t see that happening. And while there’s not much information available, Missouri does seem ripe for an upset. Still, unless something changes fast, he’ll remain the underdog everywhere on March 15.