We illustrate the use of a class of statistical models, finite mixture models, that can be used to allow for differences in model parameterizations across groups, even in the absence of group labels. We also introduce a methodology for fitting these models, data augmentation. Neither finite mixture models nor data augmentation is routine in the world of political science methodology, but both are quite standard in the statistical literature. The techniques are applied to an investigation of the empirical support for a theory (developed fully by Hill and Kriesi 2001) that extends Converse's (1964) “black-and-white” model of response stability. Our model formulation enables us (1) to provide reliable estimates of the size of the two groups of individuals originally distinguished in this model, opinion holders and unstable opinion changers; (2) to examine the evidence for Converse's basic claim that these unstable changers truly exhibit nonattitudes; and (3) to estimate the size of a newly defined group, durable changers, whose members exhibit more stable opinion change. Our application uses survey data collected at four time points over nearly 2 years which track Swiss citizens' readiness to support pollution-reduction policies. The results, combined with flexible model checks, provide support for portions of Converse and Zaller's (1992) theories on response instability and appear to weaken the measurement-error arguments of Achen (1975) and others. This paper concentrates on modeling issues and serves as a companion paper to Hill and Kriesi (2001), which uses the same data set and model but focuses more on the details of the opinion-changing behavior debate.

Research in political science is often concerned with modeling dependent variables that are proportions. Proportions are relevant in a wide variety of substantive areas, including elections, the bureaucracy, and interest groups. Yet because most researchers rely upon an approach, OLS, that does not recognize key aspects of proportions, the conclusions we reach from normal models may not provide the best understanding of phenomena of interest in these areas. In this paper, I use Monte Carlo simulations to show that maximum likelihood estimation of these data using the beta distribution may provide more accurate and more precise results. I then present empirical analyses illustrating some of these differences.

Although there exist extensive results concerning equilibria in spatial models of two-party elections with probabilistic voting, we know far less about equilibria in multiparty elections—i.e., under what conditions will equilibria exist, and what are the characteristics of equilibrium configurations? We derive conditions that guarantee the existence of a unique Nash equilibrium and develop an algorithm to compute that equilibrium inmultiparty elections with probabilistic voting, in which voters choose according to the behaviorists' fully specified multivariate vote model. Previously, such computations could only be approximated by laborious search methods. The algorithm, which assumes a conditional logit choice function, can be applied to spatial competition for a variety of party objectives including vote-maximization and margin-maximization, and can also encompass alternative voter policy metrics such as quadratic and linear loss functions. We show that our conditions for an equilibrium are plausible given the empirically-estimated parameters that behaviorists report for voting behavior in historical elections. We also show that parties' equilibrium positions depend not only on the distribution of voters' policy preferences but also on their nonpolicy-related attributes such as partisanship and sociodemographic variables. Empirical applications to data from a recent French election illustrate the use of the algorithm and suggest that a unique Nash equilibrium existed in that election.

This paper illustrates one strategy for testing a theory of economic influences on voting. We use a competitive equilibrium model of the economy to determine the impact of an individual's economic position on his or her economic interests and, ultimately, political interests. We then test whether this impact is observed in voting behavior, addressing the resulting specification and estimation problems in the context of U.S. presidential election data. Our empirical results suggest that, despite these formidable problems, we can usefully connect political-economic models and discrete-choice (probit) models of voting.

Fixed-effects logit models can be useful in panel data analysis, when N units have been observed for T time periods. There are two main estimators for such models: unconditional maximum likelihood and conditional maximum likelihood. Judged on asymptotic properties, the conditional estimator is superior. However, the unconditional estimator holds several practical advantages, and therefore I sought to determine whether its use could be justified on the basis of finite-sample properties. In a series of Monte Carlo experiments for T < 20, I found a negligible amount of bias in both estimators when T ≥ 16, suggesting that a researcher can safely use either estimator under such conditions. When T < 16, the conditional estimator continued to have a very small amount of bias, but the unconditional estimator developed more bias as T decreased.

If the publication decisions of journals are a function of the statistical significance of research findings, the published literature may suffer from “publication bias.” This paper describes a method for detecting publication bias. We point out that to achieve statistical significance, the effect size must be larger in small samples. If publications tend to be biased against statistically insignificant results, we should observe that the effect size diminishes as sample sizes increase. This proposition is tested and confirmed using the experimental literature on voter mobilization.