The moment of truth

When journalists write stories about "working class whites" they may talk to factory or construction workers, but it they mention any survey data, it involves differences between more and less educated people.  That's because very few surveys now ask about occupation.  The American National Election Study still does, but it's an academic survey, so things take time.  The data for 2016 was released in May, but it didn't include the occupation variable.  But the occupation data has now been released, so now it is possible to distinguish between the effects of occupation and education. 

It's known that the gap between more and less educated (white) voters was substantially bigger in 2016 than it was in previous elections.  What about the gap between blue and white collar workers?  The ANES has a detailed classification, with about 90 different occupations.  I reduced this to six categories:  managers, professionals, other white collar or technical, protective services, supervisors, and manual workers.  "Protective services" is not usually distinguished as a separate class, but they are hard to place and Donald Trump often boasts about getting their support, so I made them a separate category.  Limiting the sample to non-Hispanic whites, here is support for Republicans in 2012 and 2016 relative to manual workers.

                         2012    2016

Managers          -.18     -.49

Professionals    -.34     -.78

White collar     -.12      -16

Protective          .49      .23

Supervisors       .24      .10

Manual                 0         0

To approximately translate these into percentage differences, multiply by 25.  For example, the Republican vote among supervisors was about 6% higher than among manual workers in 2012.  One thing to notice that the "New Deal" pattern, where Democratic support was higher among manual workers than among the middle classes, is pretty much gone; Democratic support is now highest among professionals.  The other is that there seems to have been a change between 2012 and 2016:  manual workers moved towards the Republicans relative to all other classes (or all others moved towards the Democrats relative to manual workers).  

These comparisons do not control for education.  If you do that, the effect of having a college degree is about -.23 in 2012 and -.72 in 2016--that is, about three times as large (these are close the gaps shown in the exit polls).  What about occupation after talking account of  education?  The estimated shift of manual workers is less than half as large as it was before including the control, and not anywhere close to statistical significance (a t-ratio of about 0.8).  That is, the major change in voting patterns involved education, not occupation.  Or in everyday terms, the big change was not that working-class whites turned to Trump, but that less educated whites turned to Trump.  You could say that this is just an academic distinction, because working class voters tend to be less educated.  But education and occupation are different things--of course they are correlated, but a lot of people without college degrees have white collar occupations (see this post).  I have argued that the changes had more to do with Trump's (and maybe Clinton's) styles and with views of the nation than with appeals to economic interests, and the fact that the change in voting patterns primarily involved education and not occupation supports this interpretation.