Forgotten but not gone: the sequel

A few weeks ago, I showed the estimated effect of college education on Democratic vs. Republican voting in presidential elections from 1936-2012, using a combination of Gallup (1936-68) and GSS (1968-2012) data.  It occurred to me that the GSS also had a measure of income (in constant dollars).  Larry Bartels (Unequal Democracy) gives evidence from the 1952-2004 American National Election Studies suggesting that the association between higher income and support for the Republicans had become stronger over that period, although the association between higher education and support for the Republicans had become weaker.  I later found the same thing in an analysis that also controlled for occupation (the association between occupation and voting change in a more complex way, although in a sense it became weaker--read the paper for more detail than you probably want).

There's a good deal of sampling error in the ANES estimates, so the GSS is useful as a check.  The estimated effects (controlling for race, gender, education, and marital status) are shown in this figure (the scales are not the same but happened to be pretty similar):

There are some differences between ANES and GSS estimates for individual elections (especially 1980), but the general pattern is similar:  high in the 1970s and early 1980s, and then declining.  The decline is more evident with the GSS, since 2008 and 2012 remained low.  So it seems that Bartels' statement that "over the past half-century economic status [income] has become more important, not less important, in structuring the presidential behavior of white Americans" needs to be modified:  it  became more important from the 1950s through the 1980s, and then less important.  The general issue here is that the impact of education, income, and occupation don't follow the same path over time.  It might seem like they are all "indicators" of the same thing, your general position in society. so that when the impact of one moves in a particular direction, the impact of the others should as well.  But in fact, there are three different different things that need to be explained.