Jeffrey B. Lewis

Beyond the Median: Voter Preferences, District Heterogeneity, and Political Representation

Despite the centrality of the median voter prediction in political economy models, overwhelming empirical evidence shows that legislators regularly take positions that diverge significantly from the preferences of the median voter in their districts. However, all these empirical studies to date lack the necessary data to directly measure the preferences of the median voter. We utilize a unique data set consisting of individual‐level voting data that allows us to construct direct measures of voter preferences. We find that legislators are most constrained by the preferences of the median voter in homogeneous districts.

Ideological Adaptation? The Survival Instinct of Threatened Legislators

Do legislators remain at one point on the ideological spectrum for their entire careers, or do they update their ideological positions in response to the demands of constituents? Although theories of the electoral connection predict that legislators should adapt to voter demands, most empirical studies instead show that they maintain consistent positions in the face of changing political conditions. This article takes advantage of the natural experiment provided by Californias 2003 recall election—held at the midpoint of the states legislative session—to investigate the impact of a strong electoral signal that is isolated from other political changes. We show that after the results of the recall signaled a surge in support for the Republican Party, Democratic legislators, particularly those in the most competitive districts, moderated their voting behavior in an apparent case of ideological adaptation.

The Structure of Utility in Spatial Models of Voting

Empirical models of spatial voting allow legislators’ locations in a policy or ideological space to be inferred from their roll-call votes. These are typically random utility models where the features of the utility functions other than the ideal points are assumed rather than estimated. In this article, we first consider a model in which legislators’ utility functions are allowed to be a mixture of the two most commonly assumed utility functions: the quadratic function and the Gaussian function assumed by NOMINATE. Across many roll-call data sets, we find that legislators’ utility functions are estimated to be very nearly Gaussian. We then relax the usual assumption that each legislator is equally sensitive to policy change and find that extreme legislators are generally more sensitive to policy change than their more centrally located counterparts. This result suggests that extremists are more ideologically rigid while moderates are more likely to consider influences that arise outside liberal-conservative conflict.

Understanding King's Ecological Inference Model a Method-of-Moments Approach

he problem of inferring individual behavior from aggregate data is among the oldest problems in politT ical methodology. Simplest among these problems is the attempt to reconstruct the interior cells of a set of 2 x 2 tables from their marginal totals. An archetypal problem of this sort is the estimation of election turnout by race. In most cases, the number of whites and nonwhites who vote in an election is not tabulated. However, the white and nonwhite population and total turnout in a number of election reporting units (e.g., counties or precincts), is generally known. The problem of ecological inference, in this case (and in all cases considered in this article), is to use partially aggregated information on the marginal distributions of the variables of interest (e.g., percentage white and percentage turnout in each precinct) to infer the joint or conditional distribution of these variables across all reporting units (precincts). Ecological inference is generally considered to be at best a necessary evil. The glib solution often suggested is “Go collect individual-level data.” However, if anything, the need for reliable methods of ecological inference is perhaps greater than it has been at any time since the explosion of survey research in the early 1960s. First, as the tabulation of electoral returns becomes increasingly automated, very large sets of precinct-level election returns are becoming available (e.g., Lewis 1998 and King et al. 1997). A second source of demand is democratization in Eastern Europe and Latin America. Electoral competition in newly formed democracies is an active area of research. However, survey data are limited. Scholars in this field must rely on aggregate election returns to understand these elections (see Ames 1994). Finally, there is demand from an increasing number of scholars interested in quantitative approaches to historical questions. Unfortunately, most existing methods of ecological inference are perhaps most famous for their failures. The lack of robustness of the ecological regression model to violations of its assumptions, for example, represents a real impediment to its application and thus to the use of ecological data in the study of these important substantive problems. Against this background, Gary King offered A Solution to the Problem of Ecological Inference (1997). This new method has received considerable interest and attention. However, its workings and relation to existing methods are largely unexplored. In this article, I attempt to address these questions, and I develop King’s model (hereinafter El) by extending Leo Goodman’s (1 959) familiar ecological regression (ER) model. I consider analytically a good deal of the distance between what is generally referred to as Goodman’s ecological regression and King’s estimator. The empirical application that I use to exemplify the statistical issues inherent in King’s method (and ecological inference) is support for ballot propositions. The question considered is how to calculate the percentage support for propositions among those voting for each of two state assembly candidates. I have argued elsewhere that estimates of the support for propositions among those voting for winning assembly candidates can be used as a measure of the policy preferences of assembly members’ electoral coalitions (Lewis 1998). In particular, I consider the problem of using precinctlevel voting returns to estimate the support for Proposition 156 among those voting for the Democratic and Republican candidates for California’s 60th State Assembly District in 1992. Labeled the Rail Bond Act of 1992, Proposition 156 was the second in a planned series of three rail bonds that were part of a major transportation initiative taken by the state legislature in 1990. The proposition was narrowly defeated statewide by a margin of 49 to 51 percent. The 60th State Assembly District is carved out of Los Angeles County. The district has 301 precincts. Proposition 156 was supported by 40 percent of the 134,000 voters districtwide. The incumbent Republican assembly candidate, Paul Horcher, rather easily defeated his Democratic challenger, Stan Caress, taking 67 percent of the two-party vote.