There was a lot of hand-wringing about a “polling crisis” after 2016’s abysmal results, with some pundits worrying that we’d never be able to trust polls again. The polls in 2018, however, performed generally well. A post-mortem analysis by the New York Times’s Nate Cohn finds that final polls in aggregate were the most accurate they’ve been in a decade, and that the errors weren’t consistently in favor of one party or the other. On the whole, the results are pretty encouraging.
There was, however, some disparity in terms of which states had bigger polling errors than others. And that shows that the problems that pollsters had in 2016—which seemed, at its core, to be a problem with not anticipating how big a role educational attainment played in voter choices, and not adequately weighting for education (or, even when weighting for education, underestimating the size of the white, less-than-college voting bloc)—haven’t been totally solved.
The biggest errors were in states that are mostly white that are also disproportionately rural and non-college: Indiana, Missouri, and Tennessee (not coincidentally, some of the year’s biggest Senate races, where the Democratic candidates wound up losing), as well as Ohio. Indiana wound up being the “wrongest” in Daily Kos Elections’ own polling aggregation, finishing with a 2-point lead for Joe Donnelly. The losses in Florida and Missouri shouldn’t have been as surprising to people: our polling aggregates had both of those races tied heading into Election Day.
By contrast, the states that loom large in the public imagination as where the polling errors were most severe in 2016—Michigan, Pennsylvania, and Wisconsin—were much closer to being accurate this year. Again, though, that might be a factor of these states having larger suburban, college-educated blocs of voters when compared with, say, Indiana or Missouri.
And there were notable errors going the other direction in states with large Latino populations, where Democratic strength was underestimated: especially California, but also Nevada and to a lesser extent, Texas. Cohn points out, though, that in these states it seemed like less of a problem of reaching Latino voters, and more of the harder-to-fix problem of them remaining undecided until the end, when they broke heavily in the Democratic direction.
