On the limits of ancestry predicting voting

I'm not sold on the idea that we can predict voting behavior by studying ancestry. It certainly has some predictive value, but a review of recent presidential elections doesn't fill me with confidence.

First of all, it's not totally clear why Appalachians would prefer a Clinton to an Obama. Is it simply race? Tilove says no:

With his Harvard pedigree, mellifluous voice and high-minded talk of moving beyond the politics of confrontation, [Obama] is totally out of place in Appalachia. It's like casting Hugh Grant instead of Mel Gibson as William Wallace in "Braveheart.''

Okay, so the Scots-Irish of Appalachia don't necessarily vote on bloodline; they're also looking for someone who epitomizes their values: tough, plain-spoken, independent, pugnacious, etc. But if that's the case, why did John Kennedy beat Humbert Humphrey by 21 points in the 1960 West Virginia primary? Not only was Kennedy a Catholic (a real outsider in Appalachia), but he totally matches the description of Obama above.

A review of the last 12 presidential elections in West Virginia doesn't help much, either.

• 1960 - JFK beats Nixon (JFK was an Irish Catholic Brahmin. Nixon was a scrappy, lower caste fighter with Irish, Scottish, and Welsh ancestry. I don't get this case.)
• 1964 - LBJ beats Goldwater (This one more or less works. Everyone thought Goldwater was nuts.)
• 1968 - Humphrey beats Nixon (Again, why not Tricky Dick?)
• 1972 - Nixon beats McGovern (McGovern was a commie, but he had the Scottish name!)
• 1976 - Carter beats Ford (Some of Appalachia runs through Georgia, so maybe Carter could appeal.)
• 1980 - Carter beats Reagan (The Carter appeal stuck, but Reagan had Scottish ancestry, and was certainly the more pugnacious and less intellectual of the two. What's up?)
• 1984 - Reagan beats Mondale (No shock.)
• 1988 - Dukakis beats Bush (So the Appalachians beat back an English blueblood... with a Greek?)
• 1992 - Clinton beats Bush (Clinton is of Scots-Irish ancestry and can mix it up.)
• 1996 - Clinton beats Dole (See above, although Dole was a plain-spoken war vet.)
• 2000 - Bush beats Gore (Both English bluebloods, but Bush does a better Scots-Irish impersonation)
• 2004 - Bush beats Kerry (Kerry has English and Eastern European heritage. Again, Bush does the better Scots-Irish impersonation.)

Now, we can't expect genealogy to predict elections perfectly. It's a multivariate world, not to mention a stochastic one. One of the interesting things about West Virginia is its secular shift from solidly Democratic to marginally Republican over time. Poor whites in labor unions used to exclusively vote Democratic; today they may vote Republican for cultural reasons. So party is having an influence here independent of genealogy. Campaign effects might exist, too. (I once heard an account, by someone in a position to know, that Joe Kennedy, Sr., bought every county Democratic chairman in West Virginia a new Cadillac prior to the 1960 primary.)

Still, the heritage explanation only seems to predict the election in about half of the above cases, by my count, which is no better than random chance. So I'm not totally sold on this. Maybe it works better in primaries.