Yet another rehash of the factoid that “states that went Republican in November accounted for eight of the 10 states with the highest divorce rates in 2006” causes John Sides to commit social science.
Welcome to another episode of “The Ecological Fallacy“! Once again: you cannot infer the behavior of individuals — Democrats and Republicans — from data at an aggregate level, such as states.
What happens when we look at individual-level data? Blow’s story falls apart.
See the post for a discussion of methodology and coding. Here’s what the results look like graphically:
About 29% of Democrats, 30% of independents, and 26% of Republicans are or have been divorced or separated.
About 19% of Democrats, 19% of independents, and 15% of Republicans admit to having an extramarital affair.
If anything, Republicans are slightly less likely than both Democrats and independents to get divorced or mess around. This is the opposite of what Blow suggests — which, yet again, reveals the problems of using aggregate data to make individual-level inferences.
Adding some additional controls for “age, sex, race, educational attainment, and year of survey” he finds “Republicans are 2 percentage points less likely to be or have been divorced. They are 4 points less likely to admit to an extramarital affair.” He admits that there may be factors that he’s not considering that would make the differences even smaller. The bottom line, though, is right:
But I think the basic finding is likely robust: partisanship has a very weak relationship with either divorce or infidelity, and the relationships that do exist suggest that Republicans are less, not more, likely to get divorced or be unfaithful. Those, like Blow, who want to decry Republican “hypocrisy” on issues of family and sexuality may want to focus their ire on Sanford, Ensign, et al., and not on Republicans in the mass public.
I would have frankly guessed this result even without doing the statistical analysis, just given the unlikelihood that choice of catch-all party would have much to do with something so unrelated. When one considers that, at the aggregate level, Alabama Democrats are much more like Alabama Republicans than they are to California or Massachussets Democrats, the realization that party preference has little predictive value at the individual level should be obvious.
