Exit polls have long been used as not only a closer-to-real-time means of tracking the state of political races but also as a statistical basis from which to compare to official counts in order to identify the existence of potential election fraud. Let me stop to note right there to say that while that latter use may suggest election fraud, in no way does it prove election fraud. Discrepancies may be explained by other means.
We’ve become so accustom to the mechanic that we even tend to rely on it to judge the credibility of other countries’ democratic processes.
But the 2016 Democratic Primary is throwing a bit of a wrench into the statistical value of exit polls. The most important limitation of exit polls to note is that they’re based exclusively on those who go to the polls on election day to vote, either immediately or by provisional ballots that may be counted later. This is incredibly important to note because states’ early voting options are continuously evolving and becoming a more attractive option, and especially to those who decide who they’re going to support early in a race and who don’t or can’t attend a polling location and deal with the potential wait time.
Because I live in Maricopa County, Arizona and because it was one of the worst locations for voting problems, I’ll be using that as my primary example moving forward. Sources are provided below.
First, a data dump for Maricopa County:
Eligible Voters: 1,238,792
Total Votes: 621,959 (50.2% voter turnout)
Election Day Turnout: 88,187 (14.17% of total votes)
Early Voting Turnout: 533,772 (85.82% of total votes)
So far we see that the sampling available for exit polling is inherently limited to just 14.17% of those voting. Normally that doesn’t present a significant problem, however this election cycle has a trend that throws it off the rails, and here’s where it gets a bit more complicated:
As this chart shows, there’s an incredible generational divide between those who take advantage of early voting options. To the point: millennial really don’t. Only a relative handful of millennial cast early ballots.
A bit more data dump:
Democrat Total Votes: 250,488 (40.24% of total votes)
Democrat Hillary Votes: 141,441 (56.46% of democrat total votes)
Democrat Bernie Votes: 102,962 (41.10% of democrat total votes)
Here’s the rub. When I researched this just after the election in March, the County Recorder’s website had nice, pretty statistics breaking down candidate support by election day vs. early voting, and now it’s been updated and doesn’t have that breakdown. I can tell you that among Democrat Early Voting, Hillary was favored by 12%, but until I find another source for that to cite you only have my word on that number. And I’m not going to pretend random people are going to place value in that.
That leaves me with the less scientific argument, but one I think is still credible, and that’s that Bernie’s key demographic is known to be millennials and the data provided above does demonstrate that that age group doesn’t have much interest in voting by early ballot, which means they’re more inclined to vote in person. Which means, finally, that the sampling of voters available to exit polls is going to inherently slant towards Bernie. If that slant is anywhere near what I saw in early voting of 12%, that’s so far beyond accepted error margins as to make exit polls valueless.
I’d also point out this reddit page that includes several valuable replies covering the statistics of the Maricopa County race, including various speculative models if disenfranchisement hadn’t been so prominent.
Sources:
MSNBC: What does the early vote tell us about who will do well in the Arizona primaries?
Maricopa County Election Results: 03-22-2016 Final Summary Report.txt