We are the local milk people.

Tuesday, January 3, 2017

Is There Any Value to the Exit Polls?

I've wanted to put together a post on the exit polls for quite a while but have struggled to really come up with a focus for it.

Rather than identifying how and why they were so wrong (which would take a bit of speculation), I'm going to lay out a case that they cannot be trusted at all. This flies in the face of a lot of political commenter who seem to agree that they are flawed but still use portions of the exit poll data to inform their conclusions. I liken this to cutting steak with a spoon.

My strategy is to conduct an analysis of the demographics of this election with real numbers to see just how far off the exit polls really were in terms of race and gender. I'm utilizing demographic data from Pew, the Exit Polls, voter turnout demographic numbers, and the final voting numbers.

So let's get started:

Clinton: 65,844,694

Trump: 62,979,616

Others: 7,804,203

Total: 136,628,513 (out of 146 million people who are registered to vote).

Yes, 218 million people are voter-eligible but about a quarter of those aren't registered. If we look just at registered voters, we're over 93% for this election.  Whether or not someone is registered is by far the best indicator of whether or he or she will head to the polls in November, so voter registration is one of the best tools for GOTV.

STEP 1: Assigning values to the demographic percentages

Whites made up about 77% of the voters in 2016. That makes the white share more comparable to 2010 or 2014 than 2008 or 2012. Non-white races and ethnicities, therefore, made up 23% of the voters. Men made up about 48% of voters. Women 52%. While the Current Population Survey (CPS)has yet to release actual data on voting numbers we can extrapolate here: 

Whites made up 105.2 million voters. Non-white 31.4 million.
Men made up 65.6 million voters and women 71.0 million.

Unfortunately, without CPS data (which comes out in May), we cannot dig any deeper without making some very big assumptions about gender and race breakdowns.

STEP 2: Applying the exit polls

According to the exit polls Trump won 53% of men and 42% of women. He won 58% of whites. Clinton won 41% of men and 54% of women. She won 37% of whites. Unfortunately while we have data on the non-white share of the vote, we only have exit polls on individual racial groups and thus cannot compare those.

Breaking down those percentages we can thus estimate that Trump won:
Men = 34.8 million out of 65.6 million votes cast
Women = 29.8 million out of 71 million votes cast
Total = 64.6 million

And Clinton won:
Men = 26.9 million out of 65.6 million votes cast
Women = 38.3 million out of 71 million votes cast
Total = 65.2 million

Note the discrepancy here. The Exit Poll results overestimated Trump's totals by nearly 2 million votes. Let's now look at it across race.

Trump took 58% of whites. So that would give him 61 million votes among whites alone. Clinton took 37% giving her 38.9 million votes. While we cannot dig any deeper here, for the exit polls to be correct, they would imply that Hillary had to have received 85.7% of the nonwhite vote and Trump just 6.4%. So we can see that there was clearly an overstatement of white support for Trump in the Exit Poll.

What is interesting here is that most critiques of the exit poll were that the poll did not identify enough white male voters. But as I've shown here, if anything it actually oversampled white males.

In sum, we really can't trust it for whites or nonwhites. And we cannot trust it for men or women. So it's effectively a useless tool. And that means conclusions found here, here, and here from Nate Cohn are meaningless. And it's not just Cohn who overstates the value of the exit polls. Take a look at fivethirtyeight and NBC among others. In fact the only source that seemed skeptical enough of the validity to warn against drawing conclusions was Vox. Too bad their advice was largely ignored.