Stephen A. Jessee’s Ideology and Spatial Voting reports on an innovative research strategy designed to test the spatial theory of voting. Spatial theory posits that voters and candidates have placements in a space, such as a line that goes from liberal to conservative. In two-candidate elections, a voter would choose the candidate that is closer to the voter’s placement, known as the ideal point. After three chapters of literature review and “theory,” Jesse very nicely presents his results in four chapters that cover basic spatial voting and how spatial voting interacts with partisan identification, information, and the perception of policy positions of presidential candidates. Scholars who use spatial models will definitely want to read this book.
There is a clear dependent variable—the respondents’ choices between the major party American presidential candidates—between Bush and Kerry in 2004 and between McCain and Obama in 2008. (Although Jessee uses the word dynamic, the analysis is purely static, with the two years considered separately albeit with comparison.) The 2004 election was assessed over a year after the election, the 2008 election a month before. There is enough similarity to the results that the differences in timing do not appear to be important.
The primary independent variable is a respondent’s position on a liberal-conservative ideological dimension. This is measured not in a conventional way (say from the familiar seven-point liberal-conservative scale found in election surveys), but from a set of binary policy choice items placed on Polimetrix internet surveys.
An example of a policy item is “Tax cuts for those making over $250,000 should be reversed.” In addition to the positions of the respondents, the positions of the presidential candidates become part of the analysis. For the 2008 election, the campaign positions of the candidates were coded as if they were survey responses. Jessee’s innovation has two big advantages. First, ideology is probably better measured by using a substantial number of policy choices than by self-placement on a scale. Second, the research design forms a bridge between candidates and voters and places them on the same scale.
The results are impressive. Scaled ideology is a strong “predictor” of presidential voting. There is some “valence” bias. Bush, for example, had a better than even chance of getting the votes of those who were ideologically indifferent. An ideological effect is found even after controlling for partisanship. Partisan bias does coexist with ideology and in fact accounts for most of the simple valence bias. Ideology has a greater effect on more informed voters while those less informed are more influenced by partisanship. Voter perceptions of the ideological positions of the candidates, as measured by voter perceptions of the policy positions of the candidates, do exhibit some dispersion, but on average the perceptions are accurate and do differentiate the candidates.
In the concluding chapter, Jessee is admirably upfront about the reverse causality problem of candidate preference potentially influencing policy preference and, hence, ideological position. This caveat is homage enough to the current causality fetish in the profession. Jessee has made an important contribution in showing that vote choice and ideology are correlated.
There is, however, a methodological objection to the bridge Jessee has constructed between the candidates and the voters. The objection concerns item selection. The problem is most easily illustrated by the 2008 election, where only 10 policy items were used to do the scaling. For nine of these, the liberal and conservative positions can easily be established, to quote the author, “on common understandings of the ideological direction of each proposal” (p. 204). The tenth, which was a “bailout bill” or TARP (Troubled Asset Relief Program) question, is trickier. Roll call voting on TARP in the House of Representatives did not follow standard liberal-conservative lines, but House conservatives were the most strongly opposed, so let’s agree that being against the bailout is the conservative position. Now, Obama and McCain were coded as agreeing on three of the ten policy items. These agreements make them moderates of sorts. In the scaling, Obama and McCain were used to pin down the space, being stuck at −1/4 and +1/4, respectively. Any respondent who made the same choices as Obama would have to also be assigned a position of −1/4, or else something would be terribly wrong with the Bayesian method. Similarly, any respondent who made the same choices as McCain would be at +1/4. Those who gave 10 liberal responses would be given the most negative, extremist scores.
Say Jessee had chosen ten items on which McCain and Obama had disagreed. For example, get rid of the TARP item where they agreed and use Obama’s campaign position on a three-month moratorium on mortgage foreclosures. Then McCain and Obama would have been extremists with all the voters likely to fall between them. The illustration shows that the common placement of candidates and voters is very sensitive to item selection, particularly when the number of items is as small as ten. For the 2004 election, there were 27 items, constructed to parallel Senate roll calls. Bush’s positions are taken from his announced positions on the roll calls. Kerry’s positions are presumably taken, although the book does not clearly state so, from his actual roll call votes. Even 27 items represents a weak bridge. I would be reluctant to estimate NOMINATE liberal-conservative positions from just 27 roll calls.
Bridging is now the latest cottage industry in ideological analysis, but it’s not new. Keith Poole and I, in our 1997 book, bridged the House and Senate by making interest groups voters, just as Jessee made Bush a voter on 2004 roll calls. But researchers now take whatever bridge they can find, such as NPAT survey responses, conference report votes, or common space regressions, and declare victory. Ideology and Spatial Voting is just one of several projects that have left me skeptical.
Other methodological objections concern the estimation model. Although the estimated presidential voting models are nuanced with respect to partisanship and information, the ideal point estimation is definitely one quadratic utility model with error fits all. That is, not only is the shape of the utility function the same for all but the error variance is also the same.
The quadratic model assumes that a small difference in position between, say, Rick Perry and Rick Santorum, would make a big difference in utility for progressives, in contrast to the quasi-concave model of NOMINATE, where the utility differences would be smaller. With respect to roll call voting, Royce Carroll, Jeffrey B. Lewis, James Lo, Keith T. Poole, and I have shown that the quasi-concave model is much preferred to the quadratic even though ideal point estimates are highly similar under both models. (“The Structure of Utility in Spatial Models of Voting,” American Journal of Political Science 2013). These authors also show that extremists have more sharply defined utility functions than moderates. If a common spatial utility function, quadratic or quasi-concave, is nonetheless a workable assumption for congressional elites, can we assume the same is true for the mass public? More work is needed here. For example, poorly informed voters may have very flat utility functions to start with, as suggested by the information proxy in Jessee’s vote choice analysis. Certainly it is a stretch to assume that voters and presidential candidates have identical error variance.
I would also complain about the lengthy déjà vu literature review and the chasing down of such straw men as the median voter theorem and directional voting, perhaps all de rigeur in the standard process of turning a dissertation into a book. Yes, but one can skip to page 38, read the rest and learn a lot.
The spatial model studied by Jesse was first introduced by Harold Hotelling in the 1920s and elaborated in its modern form by the late Melvin Hinich and his several coauthors in the 1960s and 1970s. Jesse’s work testifies to the model’s strength as a vehicle for empirical research in political science.
