Unequal Democracy is an ambitious book that takes on many aspects of inequality in American politics. Its theme, to oversimplify a bit, is that the growth of inequality in America is a product of politics. Not globalization, not a neutral result of the excesses of a market economy, the growing inequality is seen by Bartels as the expected result of persistent biases in the way that economic outcomes are translated into political outcomes—with a policy bias that further enhances inequality. I find myself in agreement with the general theme and with most of the supporting analyses. In the balance of this brief essay, I wish to turn to one specific piece of the whole, the representation analysis of chapter 9, “Economic Inequality and Political Representation.”
Bartels's basic story in chapter 9 is that U.S. senators—and by implication, other representatives—are highly responsive to the views of upper-income constituents, somewhat responsive to middle-income constituents, and not at all responsive to the poorest third. His analysis shows that the views of some citizens are reflected in Senate votes, but not all.
Bartels's approach is to divide the state samples in the Senate Election Study into three groups of roughly equal size by income. He then associates the average views of each income group on liberal and conservative self-identification and on abortion policy with various measures of Senate votes. The typical statistical model allows each of the three income groups to have a separate regression coefficient predicting the votes of the senators from each state. The generic finding from several analyses is that the high income group has a big coefficient, the middle income group a considerably smaller one, and the low income group has a coefficient that is usually wrong signed and statistically insignificant. From these findings flow the conclusion of unequal representation. I wish to question that conclusion.
1. A Multivariate Conception
To the degree to which low-income constituents have the same identifications and preferences as do middle- and upper-income types, they would automatically be represented by a senator responding to, say, only upper-income constituents. Bartels considers this incidental responsiveness not the real thing. Instead, he has income groups competing so that the measure of responsiveness is response to each group's views controlling for the views of the other two. So his actual finding in these repeated analyses is that there is no discernible influence of low-income constituents while controlling for the views of the middle- and upper-income constituents.
We can offer two explanations of such a finding. One is Bartels's preferred interpretation, that lower-income groups have no influence because senators do not, in fact, respond to their views. (The conclusion is hedged by an assertion that causal analysis is not possible with the static research design.) The alternative is a multicollinearity interpretation. Insofar as the views of lower, middle, and higher are correlated within each state, then (a) we would not expect all of them to reach the significance threshold when competing with one another, and (b) we would expect those best measured to dominate the competition. Thus, to the degree that upper-income constituents are also better educated, more involved, and so forth, we would expect their views to be the most precise, with those of the middle less so, and those of the lower-income constituents least. This would produce the reported finding, even in the absence of any discriminatory behavior on the part of senators. Footnote 1
There is a peculiar issue that afflicts the analysis using liberal–conservative self-identification as the measure of the public's ideology. That is that almost one fourth of all Americans, a group I call “conflicted conservatives,” combine a conservative self-identification with net policy preferences that are left of center. Since this involves a failure to understand the dominant frame of conservatism as advocacy of small government and low taxes, it is naturally more prevalent among less educated and informed citizens, who we would expect to find disproportionately among the lower income groups. The problem for representation is that any representative who represented what these people think they are (conservative) would fail to represent their (liberal) preferences, and vice versa. This concern would not apply to the abortion analyses, which produce somewhat less distinctive findings.
Bartels, as always, is careful in stating his conclusions. But it would be easy to misread him to be saying that low-income constituents are not represented, when what he has, in fact, shown is that their views lack an independent ability to predict Senate votes once the views of the other two income groups in each state are taken into account.
If it were the case that views were sharply divided along income class lines, then it would matter a great deal whether there were differential responsiveness to income classes. If, as we often think, the poor were relatively liberal, middle-income groups moderate, and upper-income groups conservative, then a responsiveness disproportionately to the upper-income group would violate the preferences of the poor. But they are not. A similar analysis of the 2004 American National Election Study shows all three groups quite similarly located on average, with the poor almost exactly in the center (4.01 on the 1 to 7 scale, where 7 is most conservative), the middle most conservative (4.44), and the rich very slightly less so (4.41). Abortion attitudes are similarly undifferentiated by income, with all three groups on the pro-choice end of the scale, the poor least so (2.65, where 4 is strongly pro-choice) and the rich most (3.00), with the middle group undifferentiated from the poor (2.67). In such an undifferentiated public, it matters whose views are best represented, but not much.
2. Static and Dynamic Models of Representation
The Bartels analysis is static, showing association between the views of income groups and the voting of senators. Imagine instead a dynamic conception of representation, in which representation is about responding to changes in preferences. Then we can imagine electorates stratified along income lines that are either divisive—different groups moving in different directions—or parallel—everybody moving in tandem. Since Page and Shapiro (Reference Page and Shapiro1992), the dynamic public opinion literature has been moving toward a consensus view that parallel movements are the norm.Footnote 2
If the “parallel publics” thesis is true, and I believe that it is, the implication is that a representative responding to any group of constituents is necessarily responding to all. As in the static case, one may argue that such representation is inadvertent, and therefore less meaningful than deliberate responsiveness. But the case is much less compelling when everybody's changing preferences are mirrored in the changing behavior of representatives. Without divisiveness of movements of preferences, there are no oxen to be gored. In the static case, we have seen that the views of income groups are not very different at any particular time. Here we have a stronger case, where there is strong correlation across income groups in the direction of changes in preferences.
It may be in the nature of social scientists to go looking for divisiveness and to frame our analyses around the question of who wins and who loses. It is probably in the nature of ambitious politicians to go looking for consensus (on the direction of policy change) and to respond to it. Responding to divisiveness adds some votes and subtracts others. Responding to a consensus about direction of change adds only. Why leave part of your constituency satisfied and part dissatisfied when you can satisfy all?
How you interpret the Bartels finding, in the end, depends on what representation means. If it means winning a contest with other groups when views conflict, then the Bartels result that the poorest Americans are ignored holds. But there is no such conflict on the liberal/conservative self-identification or abortion preferences that Bartels examines. The three income groups are more similar than different. And the changes in identification and preferences are even more similar.
What we would really like to know is what happens when self-interest moves the rich and poor to have opposite preferences, for example, on tax progressivity. When issues become complex, the break between self-interest and the representative response may occur—as we see in the case of the estate tax—in a failure of poor people to understand their self-interest, rather than in a failure of representatives to heed their views. The Bartels result implies that if ever there were a clean case where the views of the poor conflicted with those of the rich, the poor would lose. Whether such a case ever exists is more in question.
In the end, I concur with Bartels that politics has played a very large role in building a much less equal society over the last four decades. I disagree only in pointing the finger at a failure of representation as a cause.
