Whenever someone claims that income affects people’s politics, I always ask, “Have you controlled for education?”  I’ve spent over a decade playing with public opinion data.  I know from experience that if you’re trying to predict people’s opinions using both income and education, education almost always crushes income.  As a rule, so-called income effects are education effects in disguise. 

As a result, I was skeptical when I first heard about Martin Gilens finding that American democracy primarily responds to the preference of the rich.  The first question on my mind was, “Did he control for education”?

Indeed he did.  Yes, Gilens main results simply show political responsiveness as a function of income.  As a sensitivity test, though, Gilens tried to predict political outcomes using the subset of survey questions where the 10th and 90th percentiles of the income or education distributions disagreed by at least 10 percentage points.  His results:


As you can see, Gilens found that both income and education mattered.  Yet income mattered more.  Holding income constant, American democracy pays more attention to the well-educated.  But holding education constant, American democracy pays even more attention to the high-income.  Interesting.

Still, I have one main reservation.  Since education has a much larger effect on policy preference than income does, the overall political effect of education could still easily exceed the overall political effect of income.  Suppose, for example, that politicians always heed the rich, but the rich only hold distinctive views 1% of the time.  The overall political effect of income on democratic outcomes would still be small.  In contrast, suppose politicians give college graduates double weight, but college graduates have distinctive positions on every issue.  The overall political effect of education on democratic outcomes would be massive, even though politicians often ignore educated opinion and slavishly obey rich opinion.

Measuring these net effects of income versus education would be labor intensive.  Just to ballpark things, however, I’ve emailed Gilens the following question:

In Table A3.4, you restrict analysis to the 1050 questions where the 90/10 income OR education split is 10 percentage points or more.

How many of those questions feature 10-point gaps for *both* income and education? How much for each separately?

If he responds, you’ll be the first to know.