Tuesday, November 11, 2008

Why did Obama Win? Money and Income?

I don't feel much like writing today; or this week at all for that matter. So I copped something I ran across that's election/politics related in, of all places, Science Magazine. This statistical analysis (actually a book review) perhaps looks at how Americans vote from a perspective not normally examined by the main stream pundits.

Published in Science magazine, October 31, 2008, 322:676.
Many of us discuss elections, often with concepts like blue and red states. Andrew Gelman is a creative Columbia statistician who joined with four former political science students to dig deeper. The most creative analyses in the book use Gelman’s multilevel methods (pp. 179-181). But the technical background is near-invisible: here there are no equations and few numbers—rather, dozens of revealing graphics, all of which are very clear.

The book is unusual in its aim toward the general lay reader, not the political scientist. It is clear and crisply written. It quotes and refutes many widespread views of journalists and political pundits, even if also building on political science.

Three main findings illustrate their approach. First, rich individuals vote more Republican. Second, rich states (like Connecticut, with high average-income residents) vote more Democratic. These conclusions are contrasted with a third, very intriguing finding: In poor states like Mississippi income strongly predicts Republican voting, while in rich states like Connecticut, the rich and poor differ little in their voting. Such results are illustrated with scatter plots and line charts showing income versus level of Republican voting for citizens, but then repeating the same analysis for different types of states.

These three key ideas are refined by systematically adding other factors. Time: State voting differences have mostly emerged in the last 20 years. Religion: Surprisingly, church attendance seems more tied to voting Republican for the rich than for the poor. Class: State differences are stronger for upper income persons. Race: A great deal of party by income voting may be due to race, especially in the South. Polarization: The parties have grown more deeply divided on issues since the 1960s.

The elegance of the Gelman et al book comes first from the clarity of the income-party model, but also from its methodology. By consistently repeating similar analyses contrasting state and individual effects, they refute the “ecological fallacy” of stereotyping individuals’ behavior on the basis of data about where they live. For example, many have compared state income with state voting and falsely concluded that rich individuals vote more Democratic. The authors are able to quickly dispel this myth while simultaneously navigating the intricacies of the actual relationship between income and voting.

So what is missing? Most obviously there is little time devoted to the complicating effect of social issues. Past work has shown that while income explains many fiscal preferences, education is more powerful for social issues like abortion, the environment, gender roles, or minority tolerance. When these several variables combine in party voting, the results are not always linear or elegant. Gellman et al thus convey a view of income-based elites, more than an issue-specific view of multiple group activists. Income and presidential voting are hard to link neatly to such issue politics. Gelman et al thus continue an old tradition of class politics, despite decades of work stressing its decline—as in Terry Clark and Seymour Lipset, eds., The Breakdown of Class Politics. Gelman et al say almost nothing about political culture, although many analysts use culture to revise class-based models. Indeed many argue that a new political culture is emerging built on combing social with income-related issues (like David Brooks’ bohemian-bourgeois or Bobos, who are fiscally conservative and socially liberal).

Example: The culture wars debates have been probed by Morris Fiorina in Culture War? The Myth of a Polarized America and Bill Bishop in The Big Sort: Why the Clustering of Like-Minded America Is Tearing Us Apart. These bring in a wide range of other variables and processes. Gelman et al include many brief reviews of other sub literatures (like friends and neighbors p. 134), but do not use them seriously. They are especially critical of David Brooks’ books on Bobos, and Thomas Franks’ What’s the Matter With Kansas? Both Brooks and Franks’ stress how combing multiple dimensions (like abortion and taxes) generate new results; but Gelman et al mostly address each dimension separately, finessing the complex arguments. Or they present data on more complex points (like pp. 128-129) without joining them to their core model.

Unfortunately, while Gelman et al recognize that their models have only some of the answers, they don't tell the reader how strong their models are. Work by others shows that using Gelman et al’s variables explain some 10-20 percent of the variance in voting, which makes clear that 80 percent is not explained. The strength of their work is that they built a tight, parsimonious argument. It resembles math model building, sans equations. Read Gelman et al along with others like Brooks or Bishop and you will find a richer interpretation by two smart journalists. Or Polsby and Wildavsky’s Presidential Elections is the political science classic. If future work extends the power of Gellman et al’s methods by adding a few more critical processes, this would enrich the study of American politics.

Conclusion: the book is fun to read, and even while too simple, it might become a minor classic.

No comments: