Death, status, and party

When I was writing my recent posts on the alleged rise of "despair" I recalled that a recent paper by Siddiqi, So-Erdene, Hamilton, Cottom, and Darity found a correlation between changes in county-level white mortality rates and Republican voting in 2000-16:  places in which mortality rates increased shifted towards the Republicans.  Their interpretation was that both early deaths and Republican support reflect rising perceived "status threat."*  Oddly, they didn't consider an alternative interpretation, which is that changes in voting were influenced by changes in actual conditions.  A large part of the rise in premature mortality involves deaths from drugs.  It seems plausible that a rise in drug abuse would cause people to turn to the right, partly because people often react to social disorder by supporting "get tough" policies, and partly because they can blame drug users for bringing problems on themselves. 

In doing my post on "deaths of despair," I had noticed that, although drug-related deaths have increased pretty steadily over the last 50+ years, the rate of increase was unusually large in 2012-16, going from 11.7 to 18.4 per 100,000.  I had county-level data on votes and some other control variables in 2012 and 2016, so I added changes in drug related deaths.  The CDC does not report the exact number unless it is 10 or larger, so I omitted counties in with fewer than 10 in both 2012 and 2016--this eliminated most counties, bringing it down from over 3,000 to a little over 900.  For counties that had fewer than 10 in one year, I used 5 as the estimate for that year.  From that point, I computed the change in deaths per 100,000 and included it as an independent variable.  The estimates from one model (I experimented with different controls and the estimate for drug deaths was quite stable):



Term                 Est         SE      t-ratio  
Constant            0.400     0.0243      16.46       
% Rep 2012          0.922     0.0114      81.13    
% income under 10K -0.163     0.0584      -2.80   
% income over 150K -0.356     0.0324     -10.97   
Change drug deaths  0.075     0.0105       7.12   
log population     -5.93      0.380      -15.60   

The 25th percentile for changes in drug deaths is about 1.5 (per 100,000), the 75th percentile is about 15, and the 95th percentile is about 28.  So going from the 25th to 75th percentile would increase the expected Republican share by about 1%--going to the 95th by about 2%.  Of course, the counties aren't independent cases.  I didn't have time to do a sophisticated control for spatial correlated errors, but I did a crude and probably excessive one:  adding dummy variables for each state.  The estimated effect of drug deaths was only about a third as large, but still statistically significant.  

My analysis here doesn't refute the "status threat" interpretation--it's just meant to show that there's an alternative interpretation of the connection between changes in death rates and changes in voting patterns that has some promise.  I've had other posts arguing that the status threat idea doesn't fit the general pattern of changes in public opinion--particularly the recent shifts to the left on immigration and racial attitudes. 

*I was reminded of it by this blog post, which offers a different critique.