David Frum's statistical mistake
David Frum’s recent piece in the New York Times Magazine is fascinating and worth reading. His central contention is that wealth and income inequality cause voters to be, on average, more likely to be Democrats. The only problem with this elegant thesis is that the data suggest it is statistically meaningless. I had meant to write a letter to the NY Times Magazine outlining the flaws in Frum’s argument, but the birth of my son intervened and kept me busy! I post these thoughts here in hopes of starting an online conversation about the issue.
The first flaw of Frum’s analysis is simply that the relationship between wealth and party affiliation is rather weak. Frum mentioned wealthy zip codes as being “a roll call of Democratic strongholds.” I couldn’t find data on voting patterns by zip code, so instead I analyzed data at the county-level. Out of the twenty richest counties in the US, the majority (13) voted for Bush more than Kerry. This is another kind of roll call Douglas County (CO), Loudon County (VA), and Hunterdon County (NJ) all were very wealthy and solidly Republican. It is true, on average that wealthier countries vote more Democratic, but just barely: each $10000 increase in median family income increased the vote to Kerry by 0.8%. What’s more, this relationship only explains 0.3% of the total variation in voting pattern, making it essentially worthless for prediction.
The situation is similar for Frum’s assertion that places with great income inequality tend to vote Democratic. Out of the 20 countries with the greatest income inequality, Bush won 7, places like Greene County (GA), Summit County (UT), and Lake County (IL). For each 10% increase in the share of income controlled by the rich, the vote for Kerry increased on average by the 5%. While I recognize that in politics this is perhaps a significant trend, more than 90% of the total variation in voting patterns remains unexplained.
Most grave is Frum’s assertion that inequality somehow causes people to vote Democratic, making Republican policies that tend to increase income inequality rather damaging to Republican Party interests. There simply is little support for this assertion. Correlation is not causation. Bigger cities have both more churches and more bars than small cities, but that does not mean churches drive people to drink! Similarly, cities may on average have greater income inequality than rural areas, simply because the very rich and the very poor move there in search of opportunity, and cities may vote Democratic more, but that does not necessarily mean the two are related. By the same logic, one could argue that since less education is correlated with voting Republican, a lack of schooling causes people to vote Republican, an argument I’m sure Mr. Frum finds repugnant.
