What Really Happened in Virginia in 2017, Part Deux

The exit polls were quite misleading on what really happened in Virginia in 2017. That was my argument, based on estimates we have done at CAP for our Voter Trends in 2016 project, which indicated that 2016 exit polls in VA had practically reversed the correct proportions of white college and noncollege voters, dramatically underestimating the share of white noncollege voters in that  state (and pretty much every other state). My extrapolation, based on these data, was that the 2017 exits once again got it wrong in Virginia.

Confirmation of my assessment comes from big data gurus Catalist in a ground-breaking analysis of election and polling data from Virginia by their Chief Scientist, Yair Ghitza. Combining pre-election turnout scores, precinct-level 2017 election returns and polling data, Ghitza's analysis shows that the exit polls were indeed incorrect in the portrait they painted of 2017 Virginia voters.

Here's the basic story. The 2017 Virginia exits claimed that white college educated voters vastly outnumbered white noncollege voters by 41-26. They further claimed that Northam carried the white college vote by a narrow 51-48 margin, while losing white noncollege voters by 26-72.

The Catalist analysis is quite different. Their estimate is that Virginia voters in 2017 were 42 percent white noncollege while just 34 percent were white college graduates. Furthermore, they find that Northam carried white college voters by double digits (as I predicted earlier), 56-44, while Gillespie's margin among white noncollege voters was significantly less (65-33) than shown by the exits.

As these data imply, the Catalist analysis finds that, overall, the voting electorate was more white and less minority than shown by the exit polls. Blacks were 18 percent of voters not 20 percent; Latinos were 3 percent of voters, not 6 percent. On the other hand, Catalist finds that voters from these two groups were more strongly Democratic (93-6 and 73-25, respectively) than indicated by the exit polls.

Ghitza notes that there are some uncertainties to this analysis but they are likely to have had only modest effects (a point or two) on the estimates they produced. The large and very important differences with the exit polls are very unlikely to go away no matter how much they refine their analysis.

I agree. It is time to face up to the fact that the exit polls are just not right in important respects and that people should treat their findings with more skepticism. On a hopeful note, Ghitza reports that Catalist may be able to deploy their basic methodology very quickly after coming elections to generate more accurate estimates of vote share and vote preference by demographic group. Let us hope! The exit polls clearly need some real competition.