« links for 2008-05-15 | Main | "How Do EITC Recipients Spend Their Refunds?" »

Thursday, May 15, 2008

"Obsessed with Demographics"

Slicing the demographic pie for political analysis:

Polling's fuzzy math, by Crispin Sartwell, Commentary, LA Times: American "political analysis" has become obsessed with demographics.

For example, pundits and pollsters held that the Democratic contests in Ohio and Pennsylvania between Hillary Rodham Clinton and Barack Obama turned on the vote of "white working-class men,"... Those primaries supposedly showed Obama's problem for the general election.

I suggest to you that this kind of analysis ... is both fundamentally non-empirical and fundamentally non-explanatory.

Take an election, for example, that finishes 54% to 46% in Clinton's favor. Now say that white working-class men constitute 12% of the vote, and 10 of every 12 of them (10% of the overall vote) go for Clinton. Obviously, white working-class men were the pivot on which the election turned. If Obama could have broken off half the vote that went to Clinton, he would have won: He would have increased his vote by 5% and reduced hers by 5%, and won 51% to 49%.

But notice that the vote of any like-sized segment is equally explanatory. If most "soccer moms" or most "people ages 35 to 44" or most people "with annual incomes between $50,000 and $70,000" or most "people in the southeast corner of the state" voted for Clinton, we can say that had they voted for Obama, he would have won.

So the assertion, for example, that the result turned on the votes of white working-class men is completely unsupported by the demographics. It no more turned on that group than on any other substantial group that supported Clinton. ...

The way that polling and demographics slice up the population is, ultimately, a matter of preference; it does not derive from, but is a presupposition of, the "science." Searching for segments of the electorate that vote as a bloc, demographers split the population up into groups they decide are important or salient. And their decisions don't necessarily reflect empirical results -- they are more an index of their own social attitudes, presumptions and prejudices.

It would be nearly as scientific to rig up any segment of the population and regard it as decisive: blue-collar women, black and white, under 35; black men plus Latino women; left-handed divorcees.

The results might be striking; the voting habits of such groups might be as, or more, strongly correlated than race, income and gender grouped in the conventional ways. But even if the results were not striking, even if the groups were evenly split, they would be decisive by the standards of this sort of demographic analysis.

When you bring a set of racial or gender-based categories to the data, the divisions these attitudes represent will always be confirmed as the most important divisions in our society. That just reinforces the problematic divisions that infested the attitudes of the pollsters in the first place. And then, at the end of each election, our divisions of race, gender and class are, in our imaginations, stronger.

The right response to the notion that "scientific polling" shows that the election outcome turns on white men or black women or soccer moms is a shrug of the shoulders and the arch of an eyebrow.

This has been bugging me, so let me try to put down some thoughts. To answer the question "which groups and which issues were decisive in the election," one way to proceed would be to find groups of people that you believe are particularly sensitive to a political message, a message that differs across the candidates, and then see if the groups are swayed one way or the other. If they are, then it's fair, I think, to say that the particular political stance was a determining factor.

But it's not enough to just define large groups, or at least groups large enough so that if they had moved substantially one way or the other, then the race would have come out different. The groups have to be sensitive to the message, it has to be possible to move people across the line by adopting a particular stance (or having particular characteristics, some of which you may not be able to change). In addition, this group needs to be more sensitive, significantly so, than any other possible grouping of people to the issue. If everyone is equally sensitive to the issue, then any grouping is arbitrary.

So I think the real objection is in how the groups are defined. To say white males are sensitive to race, or populist stances, etc. may or may not be true, but this linkage is assumed when the groups are defined - it is often simply asserted. If it then turns out that the votes are skewed as hypothesized, then causality is attributed to the factor in question. If white male votes are skewed to the protectionist candidate, and if we have assumed white males are sensitive to the trade issue, then we will conclude that trade is the determining factor, and white males the determining group (this doesn't have to be unique, there could be another issue and another group that was also decisive in the same sense).

Given this, the assertion that a group is sensitive to a particular message needs to be established, it can't just be asserted. If it's true that certain groups care deeply about certain issues, and if the candidates differ on these issues, and if the vote is weighted strongly in one direction, then I would accept this as evidence that this issue and this group was a decisive factor in the election. But if I'm reading the above correctly, that would be an invalid conclusion.

What am I missing?

    Posted by on Thursday, May 15, 2008 at 12:15 AM in Economics, Politics | Permalink  TrackBack (0)  Comments (14)


    TrackBack URL for this entry:

    Listed below are links to weblogs that reference "Obsessed with Demographics":


    Feed You can follow this conversation by subscribing to the comment feed for this post.