Who elected Trump? We started with a question: which demographic got Donald Trump elected? From the analysis I have done, it appears straightforward: disengaged voters, primarily noncollege, primarily white people who voted for a Democrat in the past (Obama), but didn’t show up to vote the same way for Hillary (or voted Trump). Often, these people reside in places hit hard by declining American manufacturing, where noncollege high-wage jobs (like mining) are disappearing. It is imperative for Democrats to figure out what wins those voters back. While it was trendy after the election to compare small loss margins to other small factors, like marginal changes in minority voting or likely voter suppression numbers, the vanished cohort of Obama voters dwarfs those other estimates across many battleground states. If a tiny factor (like marginal minority voting) is theoretically important, something 5-15x as big is much more important. Why does this matter? Read more

So what about the other states? From our original map, a big swath of the midwest is in the “more Republican shift than explained by education” category. We’ve already covered several states: Ohio, Wisconsin, and Minnesota. I’d like to focus on some of those states that made the most difference: Pennsylvania Michigan Iowa Between these and the states we have already covered, we will have analyzed 5 of the 6 states that flipped in 2016. These will be shallower dives, because you have already seen the broad strokes of major trends in states we have covered, and (honestly) because that level of deep dive takes a long time. Iowa Iowa, like its neighbor Wisconsin, saw an enormous downturn in Democratic turnout across the state. It really didn’t matter which year’s Republican vote total you used, or that Trump made gains, because the 2016 Democratic showing was the worst party showing by votes since at least 2004. Read more

TL;DR: Please answer these five questions. Feedback requested I started this piece as a personal project months ago, and have been working on it during nights and weekends with a budget of $0. As I dug into the question of who actually got Trump elected, it kept getting larger: “But what about religion effects? Military service?” You have seen the charts and story that made the most sense to me, with dead ends left out. This work is obviously not done; there are more states to cover, and we haven’t really touched “But why did noncollege whites vote like this?” However, after over half a year of work, I need to draw the line somewhere to start getting feedback from real people. My test readers have said encouraging things, but what do you think? Please give me feedback! It helps me improve and produce better work. I have a small Google form if you’d be kind enough to tell me what you think. Read more

This chart is easy to misread, because it doesn’t display differences. What if in 2012, Romney got 100% of the vote from people making $100K or more, and no votes from other income segments? Then Trump would have been wildly overperforming in the <$100K groups and underperforming in the >$100K group, but you can’t see that from this chart. In order to get a handle on what changed between 2012 (when the Democrat won) and 2016 (when the Republican won), we need to take the differences between those years. So let’s do that.

Where did Democrats have a large percentage of the vote? Predictably, in 2016 Dems won a belt of support from Mississippi through the Carolinas, (the Black Belt), and in coastal cities. Where did Republicans have a large percentage of the vote? Republicans won in rural areas in 2016. But how was turnout different from 2012? Where and how did the Democratic turnout change compared to 2012? In the first sign of trouble, Democratic turnout was sharply down almost everywhere in the American interior, with very few increases even in coastal states. Where and how did the Republican turnout change compared to 2012? By contrast, Republican turnout was somewhat higher in the interior. However, Republican support was down from 2012 levels in several Western states, both conservative (Utah, Idaho) and liberal (Washington, California). Where was the net Republican shift? Methods To understand what made 2016 different from 2012, we can just subtract. Read more

Education .centered-plotly{ width:600px; height:600px; margin: auto; background:url("/loading.gif") center center no-repeat; } (These charts are interactive. Mouseover to get values of nearby points, or click the legend to show/hide series.) We can start to investigate the drivers of 2012->2016 Republican voting shift by examining the characteristics of each county. Here we see what fraction of a county has a high school degree or less vs the 2016 shift toward Trump. We can see on the map that the low outliers are Utah and Idaho. This makes sense; McMullin was a spoiler candidate for Republicans in those states and especially popular among Mormons, who disliked Trump. .centered-plotly{ width:600px; height:600px; margin: auto; background:url("/loading.gif") center center no-repeat; } When we remove Idaho and Utah, the relationship becomes a bit clearer, but is still a drumstick instead of the line that would indicate a clean relationship. What if we consider the oft-cited influence of race in the election? Read more

If we glance at the race/education/Trump shift chart, it looks like there is a linear relationship in education except for (roughly) these red points: .centered-plotly{ width:800px; height:800px; margin: auto; background:url("/loading.gif") center center no-repeat; } (Remember that you can rotate 3D plots by clicking and dragging. I just eyeballed this slice, it was not scientific. It should be close enough for the broad-strokes analysis we’ll do though.) We can filter out all of the gray points and think of them as explained by the education effect. If we only plot these red points, where are they? .centered-plotly{ width:800px; height:500px; margin: auto; background:url("/loading.gif") center center no-repeat; } If we remove a bunch of states to zoom in on the midwest, we get this: .centered-plotly{ width:800px; height:500px; margin: auto; background:url("/loading.gif") center center no-repeat; } (Remember, you can toggle some of the buckets of data as invisible by clicking them in the legend. Read more

The outliers of Ohio are mostly concentrated in the east and southeast parts of the state. Change in Republican Presidential vote percentage points, 2016-2012 .centered-plotly{ width:920px; height:500px; margin: auto; background:url("/loading.gif") center center no-repeat; } (You can mouseover counties to get more information.) What is notable about that part of the state? That is where the coal seams are, and where coal mining has historically happened. That outlier area ~= coal country An initial guess might be that coal miners had exceptionally negative views of Hillary Clinton, and/or positive views of Donald Trump. Why might that be? Miners might have especially disliked Clinton “We are going to put a lot of coal miners and coal companies out of business. Right Tim?” - Hillary Clinton This clip makes a “miners dislike Clinton” intuition a little more substantial, trending toward plausible. I think the attack ad you make out of this clip if you run the Trump campaign in Ohio is pretty obvious. Read more

Your first intuitions in Wisconsin might be to check what was happening in Ohio to see if things are the same. On a couple of counts, they are. Education vs Shift, OH + WI .centered-plotly{ width:600px; height:600px; margin: auto; background:url("/loading.gif") center center no-repeat; } There is a very strong relationship between non-college rate and Republican shift by county, in Wisconsin as in Ohio. Taken together, these two series have an r-squared of 0.65 and a p value of 6.09 * 10^-38, which are very decisive “this is not random” metrics. Education vs Shift vs Whiteness, OH + WI .centered-plotly{ width:600px; height:600px; margin: auto; background:url("/loading.gif") center center no-repeat; } And this is the same stronger-than-nationally trend we saw in the Ohio case. But mining doesn’t appear to be a factor because Wisconsin mining employment is de minimis in a state of 5.8 million people. Read more