Voter Turnout and Candidate Strategies in American Elections

2003. Journal of Politics 65(1): 161-189

Most spatial models of two-candidate competition imply that candidates have electoral incentives to present similar, centrist policies. We modify the standard Downsian model to include three observations supported by empirical research on American elections: that voters are prepared to abstain if neither competitor is sufficiently attractive (abstention due to alienation) or if the candidates are insufficiently differentiated (abstention due to indifference); that voters are influenced by factors such as education, race, and partisanship, that are not directly tied to the candidates' positions in the current campaign; and that voters' nonpolicy characteristics correlate with their policy preferences. We argue that all these conditions (except indifference) motivate vote-seeking candidates to present divergent policies that reflect the beliefs of voters biased towards them for nonpolicy reasons. We support our argument with applications to ANES data which suggest that in the 1988 presidential election, both the Democratic candidate Dukakis and the Republican Bush had electoral motivations to present policies that reflected the beliefs of their partisan constituencies, and that the threat of abstention due to alienation is crucial to that conclusion.

Our results suggest that voters' turnout decisions and their nonpolicy characteristics, even if the candidates in the course of a campaign cannot manipulate the latter, are nonetheless necessary for understanding candidates' policy strategies. Our conclusions have additional implications for the strategies candidates may feasibly employ when they pursue policy objectives other than maximizing support in the general election, such as achieving policy objectives or maximizing their joint probabilities of winning both the primary and the general election.

Assimilation and Contrast Effects in Voter Projections of Party Locations: Evidence from Norway, France, and the USA

2001. European Journal of Political Research 40: 199-221

In the standard Downsian model, voters are assumed to choose parties based on the extent of ideological proximity between the voter's own position and that of the party. Yet it is also well known that there are rationalization and projection effects such that voters tend to misestimate the policy platforms of candidates or parties to which they are sympathetic by overstating the correspondence between those positions and the voter's own preferences (see e.g., Markus and Converse, 1979; Granberg and Brent, 1980; Granberg and Holmberg, 1988; Merrill and Grofman, 1999). Here we follow insights in the psychological literature on persuasion (Sherif and Hovland, 1961; Parducci and Marshall, 1962) by distinguishing between assimilation and contrast effects. Assimilation refers to shortening the perceived ideological distance between oneself and parties whom one favors; contrast refers to exaggerating the distance to parties for whom one does not intend to vote. Using survey data on voter self-placements and party placements on ideological scales for the seven major Norwegian parties (1989), five major French parties (1988), and two major American parties (1984, 1988, 1992) we show that both assimilation and contrast effects are present in each country to a considerable degree. We also investigate the possible effects of randomness in party placement and scale interpretation - effects that can easily be confounded with assimilation but not so easily with contrast.

Click Assimilation to download the above paper.

Ecological Regression and Ecological Inference

Bernard Grofman

Samuel Merrill

In Ecological Inference, eds. Gary King, Ori Rosen, and Martin Tanner. 2004. Cambridge University Press. Pp. 123-143.

Adapted from a presentation at the Ecological Inference Conference, June 17-18, 2002, at Harvard University, Cambridge, MA

ABSTRACT

In ecological inference we seek to make use of data that is aggregated at the level of ecological units to make inferences about the behavior of individual actors. The ecological fallacy (Robinson, 1950) occurs when relationships between variables that obtain at the aggregate level are not found at the individual level (or conversely). Our primary focus in this paper is on proposing three "new" methods of ecological inference that are more "quick and dirty" in nature than the sophisticated statistical models offered by King and his colleagues. King's estimates the unknown parameters of interest using MLE methods on a truncated bivariate normal or Beta distribution overlaid on the tomographic plot lines in (b ^b, b ^w) phase space. The simplest of the three methods we offer replaces this MLE approach with a simple squared distance minimization algorithm on the tomographic plot lines. Specifically, for the district level solution, we choose that point on the district tomographic segment that minimizes the (weighted) sum of the squared distances to the tomographic line segments for the precincts. Suppose, for example, our problem is to estimate the proportions of blacks and whites who voted for a black candidate. These distances can be interpreted as the weighted differences between the proportion of voters for the black candidate in the ith precinct and what that proportion would be if the proportions voting for the black candidate broken down by race were the same as in the district as a whole. The precinct-level estimates for the parameters of interest are then obtained as the nearest point on the precinct tomographic line segment to the district solution. The analytic solution we obtain from this method is very similar in structure to that of the usual regression model -- but unlike regression -- our approach guarantees feasible solutions at both the precinct and district level.

The second and third methods we propose can each be thought of as forms of "constrained" Goodman ecological regression. The first of these latter methods generates a distance minimization in () phase space (where m and b are the slope and intercept of possible regression lines) rather than in () phase space. The latter operates in the original (x, y) space and finds the regression line that minimizes the sum of the areas between it and pairs of constraint lines segments that are generated by a variant of the Duncan-Davis method of bounds.

These methods demonstrate that the contrast between ecological regression in the form proposed by Goodman (1953, 1959) and ecological inference of the sort described in King (1997) is too easily exaggerated. Two of the three methods produce results that are, in general, very close to those produced by King's algorithms. Each uses either King's extension of the Davis-Duncan (1953) method of bounds or a simple variant thereof. Each operates without any assumptions about the distribution of parameters; but bootstrap standard errors can be obtained to assess the results.

Click Ecological Inference to download the above paper.

For a companion article investigating criteria for specifying a solution to the problem of ecological inference, select Criteria. See below for the title and abstract of this article.

ABSTRACT

We consider the general characteristics of "solutions" to the problem of making reliable ecological inferences in terms of eleven features: feasibility, completeness, coherence, sensitivity, substantive plausibility, parsimony, replicability, characterizability, diagnosticability, expansibility, explainability and calculability. We show that the family of methods in King (1997) and King et al. (1999) score high on all but one of these counts. However, our methods also score equally high on most features, and are also very easy to estimate using just an EXCEL spreadsheet.