What's the Matter with "What's the Matter with 'What's the Matter with Kansas'"
Here's the abstract from the Bartels article (warning pdf):
What’s the matter with “What’s the matter with Kansas?, by Larry Bartels: Abstract: Thomas Frank’s What’s the Matter with Kansas? asserts that the Republican Party has forged a new “dominant political coalition” by attracting working-class white voters on the basis of “class animus” and “cultural wedge issues like guns and abortion.” My analysis confirms that white voters without college degrees have become significantly less Democratic; however, the contours of that shift bear little resemblance to Frank’s account. First, the trend is almost entirely confined to the South, where Democratic support was artificially inflated by the one-party system of the Jim Crow era of legalized racial segregation. (Outside the South, support for Democratic presidential candidates among whites without college degrees has fallen by a total of one percentage point over the past half-century.) Second, there is no evidence that “culture outweighs economics as a matter of public concern” among Frank’s working-class white voters. The apparent political significance of social issues has increased substantially over the past 20 years, but more among better-educated white voters than among those without college degrees. In both groups, economic issues continue to be most important. Finally, contrary to Frank’s account, most of his white working-class voters see themselves as closer to the Democratic Party on social issues like abortion and gender roles but closer to the Republican Party on economic issues.
Bartels clearly wrote a very important article. However, there is a problem, 1952 was an anomalous year. As noted by another commenter at Krugman's site (before I got there) it was a huge Republican landslide. Losing 1% is not much, but Kerry got a much larger fraction of total votes than Stevenson in 52. If Bartels had looked at the difference between White men without college degrees and all voters the change would be much larger than 1%.
This would still be true (though less so) if one corrected for increased numbers of Black voters, increased numbers of hispanic voters and increased numbers of college educated voters (break up into cells by race, education and gender and then make a weighted 2004 average with weights based on 1952 votes (for both candidates summed) by category).
I might add that Stevenson and Eisenhower were special. Stevenson was notoriously high brow. Once in a rally someone shouted "every thinking man is for you" and he replied "That's not enough. I need a majority." It's amazing he got any votes in Kansas at all.
US voters in 1952 were notoriously anti intellectual. Frank might reply that what was the matter with Kansas was even more extreme in 1952. Eisenhower campaigned promising to increase the generosity of social security pensions (as he did). When he left office, the top marginal tax rate was huge (Kennedy cut it). There was no compelling reason for poor Whites to vote against him, he wasn't a pro rich class warrior like Reagan or Bush Jr.
Also he was not upper class. In 2004 two absurdly upper class candidates ran for President. I mean both were members of Skull and Bones. In 1952 one candidate was of humble class origens. That would be Eisenhower.
I have no doubt that data from 1948 would be completely different. For one thing Truman won. For another Dewey was as aristocratic as an American can be (think of how many different Dewey's you know of and their jobs, admiral, philosopher and the inventor of the Dewey decimal system).
Such data are not available, but one can avoid so much reliance on 1952 by using regressions on a trend. This is not black magic statistical analysis and someone ought to do it.
Oh also regress on interaction of a trend and Bartles' variables in regressions including year dummies.
This would be easy and is likely to get a lot of attention from Paul Krugman (hint hint).