Forgotten but not gone

In 1996, I compiled information on education, occupation, and vote in American presidential elections from 1936 to 1992.  The 1936-68 data was from Gallup polls.  In the early 1970s, Gallup stopped asking about occupation, but the General Social Survey appeared, so it was possible to continue the series.  I never published anything from this, and had forgotten about it until during this election campaign.  Remarkably, I not only still had the data, but had it in a form that could be read.  So here, at long last, is the estimated effect of higher education (college graduate=2, some college=1, high school or less=0) on Democratic vs. Republican voting, 1936-92:

The black line includes a control for race (and its interaction with election), but not occupation; the red includes a control for occupation (and its interaction with election).  The estimates with and without controls for occupation are not very different:  in both cases, the effect of education moves in a generally pro-Democratic direction over time.  At the beginning of the period, people with more education were more likely to Republican; by the end, there was little difference.  The 1972 election (Nixon vs. McGovern) stands out as the only one in which more educated people were substantially more likely to vote Democratic than less educated people.  We can take the history on from 1992 using exit polls (see this post).  In 1996-2012 the effect of education didn't change much, but in 2016 in moved in a pro-Democratic direction again.  The scales aren't really comparable, but the 2016 gap was much bigger than 1972.  In 1972, about 40 percent of whites with college degrees supported McGovern, against 33% of whites without college degrees, for a percentage difference of 7; in 2016, the percentage difference was about 18.
         The data I compiled also included non-voters.  Here is the estimated effect of college education on non-voting (vs. voting for any candidate):  again, black is without controls for occupation and red is with controls.

In 1936, college graduates were only slightly more likely to vote than people with no college, but the difference grew steadily.  Of course, you can't learn about non-voting from exit polls, but I believe that the difference has grown since then.  I may update the data using the GSS and see.

PS:  I just show the effect of higher education--the effect of differences in education up through high school graduate hardly changed at all over the period.

[Data from the Roper Center for Public Opinion Research; I should also acknowledge the Center for Advanced Study in the Behavioral Sciences, where I was a fellow when I compiled the data.]