The vertical position of the circle represents the percent of votes that went to Donald Trump on Tuesday in that county. The higher the circle, the more votes for Trump. The size of the circles represents the size of the population.
(I left out Minnesota — sorry Rubio fans! — and Alaska because there weren't county-level voting results for me to compare to death rates.)
The red line on each plot summarizes the relationship between white deaths and Trump support. In every state except Massachusetts, the counties with high rates of white mortality were the same counties that turned out to vote for Trump.
We're focusing on middle-aged whites because the data show that something has gone terribly wrong with their lives. In a study last year, economists Anne Case and Angus Deaton pointed out that mortality rates for this group have actually been increasing since the '90s.
That fact becomes more alarming when you look at the context. Over the past decade, Hispanic people have been dying at a slower rate; black people have been dying at a slower rate; white people in other countries have been dying at a slower rate.
What is going on, then, with death rates for middle-aged whites? Case and Deaton aren't sure, but they argued that alcohol abuse, suicides and the opioid epidemic have something to do with it. The rate of fatal "poisonings" for instance — a category that includes drug overdoses — more than tripled among middle-aged whites since 2000.
Economic struggles have likely contributed as well. Case and Deaton also found that the increase in the death rate has been driven by people with less education. For those without a college degree, the economy in recent decades has been increasingly miserable. This may explain why some have turned to self-destructive behaviors, such as drug and alcohol abuse.
The people I've been describing — this distressed, dying demographic slice of America — are similar to the people who tend to vote for Trump, according to phone and exit polls. Trump supporters are mostly white; skew older; and are less likely to have college degrees than other Republicans.
[I]t is nonetheless striking that Trump’s promise to "Make America Great Again" has been most enthusiastically embraced by those who have seen their own life's prospects diminish the most — not [only] in terms of material wealth, but in terms of literal chances of survival.
It's true that life was once better in many parts of America. In the late '90s, not only was the death rate for middle-aged whites lower, but median wages for non-college workers were higher. Since then, globalization sucked away many more manufacturing jobs, and the Great Recession gave an extra kick to places that were already in decline.
To investigate the relationship between white mortality, economic anxiety and Trump support, I looked at which economic characteristics in a county made people more likely to vote for Trump on Super Tuesday. These included:
- The middle-aged white death rate (non-Hispanic whites, ages 40 to 64, in 2011-2014)
- The change in the middle-aged white death rate from early 2000 to present
- The death rate for middle-aged Hispanics and nonwhites
- The percent of people employed in manufacturing
- The change in the percent of people employed in manufacturing
- The median income
- The growth (or decline) in the median income since 1999
- The percent of people working
- The change in the percent of people working since 1999
- The percent of people with a bachelor's degree or more
- The urban or rural character of a county
Even after controlling for these other factors, the middle-aged white death rate in a county was still a significant predictor of the share of votes that went to Trump.
Here's how to think about it: Say that County A and County B are roughly identical, except for the death rate of their white, middle-aged residents. County A's death rate is in the 75th percentile — meaning that the middle-aged whites there are dying faster than in three quarters of other counties — while County B's is in the 25th percentile. Between these two economically similar places, County A would give an extra percentage point to Donald Trump in the primary.
Three other characteristics stood out as highly statistically significant:
- The fraction of people with bachelor's degrees. All else held equal — including the death rates — places where people were more educated were less likely to vote for Trump. This effect was large. About a seven percentage-point increase in the fraction of people with BAs (the difference between the 75th and 25th percentiles) predicts about a four to five percentage point decrease in Trump's vote share. In other words, more educated counties don't vote Trump. (They tend to vote Rubio, polls show.)
- The fraction of people in the county who are working. After controlling for other factors, the percentage of people with jobs was a significant predictor of the Trump vote share. If an additional 12 percent of adults had jobs (which is roughly the gap between the 75th and 25th percentiles), Trump would have lost about two percentage points of the vote in the primaries.
- The decline in manufacturing. In the early 2000s, increased trade with China delivered another whammy to American manufacturers. The data show that the places that lost a lot of manufacturing jobs since 1999 were also more likely to vote for Trump on Super Tuesday.
The data deliver a clear message: Economic distress in many parts of the country is driving voters toward Donald Trump. But there remains this mystery about white death rates.
One factor that stood out as not being predictive of the Trump vote is the rate at which Hispanic and nonwhite people die. There only seems to be a relationship between Trump support and the middle-aged white death rate. And economic factors don't fully explain it.
We still don't know what exactly is causing middle-aged white death rates to rise, but it seems that Donald Trump has adeptly channeled this white suffering into political support.
Understanding why this part of America is so unhappy — why some white people are literally dying faster — may help explain how Trump became such a powerful force in this election.
(Note: If you want to see the regression table, I put it on Twitter. I included fixed effects for each state and urbanicity. Data came from the 2010-2014 five-year American Community Survey, the 2000 Census, the Centers for Disease Control and Prevention, and the Associated Press.)