Nate Cohn: Obama voters choosing Trump decided the election

(correction: I posted this momentarily before accounting for minors who are not eligible to register for voting. They accounted for 18.4% of the population in Ogemaw County in 2015. The figures and argument have been updated to account for this.)

Nate Cohn, the NYTs data guy, posted a few tweets about the election claiming "they lost this election because voters who supported Obama in 2012 voted Trump." I'm struggling to accept this as fact, rather than an assertion. And I don't doubt that it's possible, but I'm not sure whether it can be proven. I tried to do so. This post takes you through that effort.

Let's start with a few of Nate's tweets.

https://twitter.com/Nate_Cohn/status/796789450537861120

I am neither a stats guy, nor an election guru, so I may have a flaw in my understanding of this, but I'm not aware of any central repository of data collected on voters that could be used to validate whether individuals switched sides, or even whether the pool of voters was the close enough to being the same to validate the assertion. Nate offered the following Tweet as proof, and of the two counties displayed, I chose the first, Ogemaw County, Michigan, to look into.

https://twitter.com/Nate_Cohn/status/796844140784971776

Here's the Ogemaw County data for the past two presidential elections.

Nate Cohn's evidence?

In reviewing this, my initial reaction was that Nate appeared to have his smoking gun. A "small" county in which roughly 1,400 more people voted from Trump than Romney, and about 1,800 fewer voters came out for Clinton than for Obama, paints a picture of a group of voters that changed their mind, but I wanted to see what the population was, and their registration rates to ensure this story fit the data.

My assumptions were (1) that the area might be growing, thereby bringing a larger pool of voters to draw upon, or that (2) there might have been a large group of registered voters in both of these elections who may have voted in one election, but not the other, that may have helped account for the change in party vote totals.

(1) Population Change

As you can see, the population data was relatively flat during the time in question, so I'll cross (1) off the list. In fact, it's trending down a bit, which makes my case a little less likely.

(2) The pool of non-voting registrants

That leaves us to look at the number of people who were registered for both elections. Unfortunately, I wasn't able to find county-level registration data, but the Census Bureau site offers it at the state-level. This shows that 75% of Michiganders were registered to vote in 2012, while the national rate was 65.1%. This is interesting information for this discussion (and I did perform some calculations to see what Ogemaw county's electorate might look like given these rates), but I think we're missing a critical piece of the puzzle (the county-level registration rate) that may leave us short of a definitive answer.

2014 census bureau data

The most recent population estimate for Ogemaw County was 21,039 in 2014. If we use the Michigan-wide registration rate for 2012 (the most recent presidential election year that's available), we'd expect 12,995 people (75% of the population) to have been registered. If we used the national rate (65.1%), we'd get 11,280 registered voters. If we subtract the 9,857 votes cast for Republicans and Democrats in Ogemaw County in 2016 from the lower estimate of 11,280, registered voters, you're left with an estimate of 1,423 registered voters who voted 3rd party or abstained. That's just enough headroom to fit in the 1,396 additional Trump votes without necessitating a single switch form D to R. (Again, this is all supposition due to the lack of county-level registration data.)

That said, I'm sure there voters who crossed-over. It happens every election. People's circumstances change. Good and bad things happen to them. They learn; sometimes falsehoods. They "evolve their thinking."

Anyway, the matter at hand isn't whether some people switched sides, but whether they did so in sufficient number to decide the election. I'm not seeing where proper evidence was provided to prove the case. If county-level registration data is available, that might clear this up (as would individual voter data), and then the case would need to be proven across the states in question to show that Trump would not have won them without those cross-over votes. Until that's presented I think this case remains open. Hopefully, Nate Cohn will finish proving the case in an upcoming article. But if you see where I'm off the mark, I'd be happy to learn what I'm missing, or if there's a flaw in my logic.