A lot of political analysis focuses in near-obsessive detail on demographic groups. The strategy, for example, that Mark Penn outlined for Hillary Clinton before the 2008 Democratic primaries identified 25 distinct “swing” demographic categories, including such relatively narrow ones as women aged 35 to 49 and single men.
Demographics, certainly, are a part of the picture — and they are something that we discuss around here on occasion. But with one clear exception, the impact that demographics have on voting behavior is a lot more fluid than you might think from all the talk of “soccer moms” and “Nascar dads.”
On Saturday, for instance, we noted that membership in a labor union increased the likelihood that a person voted for Barack Obama by about 12 percentage points. Is that meaningful? Sure: elections take place on the margins.
Viewed another way, though, it doesn’t really tell you all that much. If you think that being in a union determines 12 percent of someone’s vote, that means that other factors determine 88 percent of it.
The truth is that none of us is just one thing. We are all members of any number of different demographic categories — and the voting tendencies associated with those categories often point in different, or even conflicting, directions. For instance, I am a non-unionized white male who makes an above-average income, all things that predict Republican voting — but I’m also college-educated, relatively young, and live in the urban Northeast, all things that predict Democratic voting. To the extent that my political interests are dictated by my demographics, I have a lot of competing priorities.
One way we can generalize this idea is to return to the data set that we used on Saturday — the National Annenberg Election Survey — and create a personalized probability profile (sorry if I sound like an eHarmony commercial) for each of the roughly 15,000 respondents in the survey.
So, for instance, we can estimate the likelihood that a married Asian-American woman living in a Milwaukee suburb, or a childless, churchgoing military veteran living in Boise, would vote for Barack Obama or John McCain, based on their responses to the poll. Then we can combine these individual estimates into a probability distribution to see what the country as a whole looks like:
For most of the country, the prediction falls somewhere in the middle of the scale. About 52 percent of the people in the model, based on their demographics, had no less than a one-in-three chance, but no more than a two-in-three chance, of voting for Barack Obama. For only 19 percent of voters could the voting preference be predicted with at least 80 percent certainty by demographics.
If you looked at political variables for each respondent, as well as demographics, you could improve on those projections quite a bit. If you knew, for instance, that someone was a Democrat, you’d know that they were very probably going to vote for Mr. Obama. But how do you know if someone is a Democrat? Based on their demographics alone, you usually wouldn’t have more than a fuzzy idea . . .
. . . unless, of course, that person happens to be African-American.
You may have noticed that the chart above contains two distinct peaks. There’s a big glut of people who are somewhere around 50-percent likely to vote for Mr. Obama — as you would expect — but then there is a smaller but still easily discerned cluster of people who are about 90 percent likely to be in the Obama column. This smaller group is made up almost entirely of African-Americans:
As I mentioned on Saturday, even if every other demographic characteristic was favorable to the Republicans — say, for instance, that a voter was 72, wealthy, extremely religious and a veteran who lived in the rural South — he was nevertheless at least 80 percent likely to vote for Mr. Obama if the voter was also black. (For many other African-Americans, the probability approached 100 percent.)
Some of this, of course, is because Mr. Obama himself is black. But the phenomenon of overwhelming black preference for Democratic presidential candidates is not new: even George McGovern got 82 percent of the black vote in 1972.
Still, it is very much the exception. If we look at the same data for Hispanic voters, for instance, we see that although they’re generally more inclined to vote Democratic, there is a much wider spread of probabilities. Some Hispanic people, based on their other demographic characteristics, were near-locks to vote for Mr. Obama, but others were more likely to vote Republican:
A candidate would do herself few favors either writing off “the Hispanic vote” or taking it for granted. Many Latinos are swing voters, and they care about many issues besides immigration.
Meanwhile, the pattern for voters who describe themselves as evangelical Christians is quite interesting:
While this is obviously a Republican-leaning group overall, that doesn’t guarantee anything: only about 3 in 10 evangelicals could be considered at least 75 percent likely to have voted for John McCain, according to the model. This group contains its share of swing voters, and there are a sizable number of black evangelical Christians, who were very likely to vote for Barack Obama.
Finally, here is the chart for voters who were members of a labor union or who lived in a household with someone else who was:
While being in a union is a statistically significant predictor of Democratic voting, this characteristic competes with, and is often outweighed by, any number of other factors. That means that, on the one hand, there is significant risk for Republicans in alienating union members: a lot of them are the very epitome of swing voters. On the other hand, it means that while Democrats are likely to do fairly well with union voters, those votes are far from certain. A union man is a union man, and also a lot of other things: he may be a father and a veteran and Hispanic and a churchgoing Catholic and a resident of Denver— and all of those things will tug in different ways on the choice he will get to make as a voter.