Playing with 2016 Electoral Data - Part 2

Firewall Break

This is the second part of a series looking at 2016 electoral data. If you haven’t read the first part, check it out!

I was interested in testing some questions about the 2016 election that came up.

The first, most obvious thing that I could look at from the existing notebook was to get a feel for what happened in the “firewall” states that changed the outcome of the election. As a recap, the more famous swing states like Florida, North Carolina, Ohio and Iowa were not “must-wins” for Clinton. The states that were “must-win” were dubbed the “Clinton firewall”, and consisted of states that had typically historically voted for Democrats. This consisted of Michigan, Wisconsin, Virginia, Colorado, Pennsylvania, and New Hampshire. In the end Clinton won Colorado, New Hampshire, and Virginia, but despite polls that strongly suggested otherwise, lost Michigan, Wisconsin and Pennsylvania. Particularly strange is that before the elections, all six states looked just as likely to vote for Clinton, at 60-80% margins, according to FiveThirtyEight. So what happened in the three where she lost?

If you visually compare the county breakdowns in those three states; you begin to see the breakdowns. Here’s a version that is zoomed in more:

Firewall Break Zoomed

Just speaking qualitatively, you see a lot more red in 2016 on the right compared to 2012 on the left. But in fact, the difference between winning and losing came down to a margin of less than 1% in each state. Here’s what we know about what the total vote differences ended up as:

State Clinton Trump Diff
Wisconsin 1,382,536 (46.45%) 1,405,284 (47.22%) 22,748 (0.77%)
Michigan 2,268,839 (47.27%) 2,279,543 (47.50%) 10,704 (0.23%)
Pennsylvania 2,926,441 (47.85%) 2,970,733 (48.58%) 44,292 (0.73%)

For a total of 77,744 votes, the course of the election went rather differently than predicted. Those 77K votes, representing less than 1% in each state, ended up swinging 46 electoral votes differently than predicted, and handed the margin of victory to Trump. It is these votes that I am the most curious about. Why were they so hard to see in the polls? Who were these people and why were they so hard to appeal to? What lies at the core of those differences?

I want to see some additional demographic information about the counties. After a little googling, it seems like we have such data:

  • Poverty estimates per county for 2015
  • Population estimates per county for 2015
  • Unemployment and median income for 2015
  • Education levels from 1970-2015

I’ll take a look at incorporating this into the data notebook next to see if this sheds some light on these crucial 77K votes.