Understanding the relative importance of various influences on the behavior of elected representives is one of the central concerns of modern political-economy. Among the possible influences are the representative's constituency, campaign contributions and lobbying efforts by special interest groups, the representative's own preferences, and pressures from various party and nonparty leaders, such as the president and committee chairmen. Although aggregated vote data are often used to measure citizens' preferences in empirical studies of representation, and are also analyzed in studies of direct democracy, there has been little work carefully linking the statistical models to well-defined models of voter choice. As a result, the behavioral assumptions underlying these empirical studies are often vague. This is especially problematic for analyses of voting data, because of past work showing that many widely used statistical techniques are probably inconsistent with certain reasonable models of voter behavior. The contribution of the project comes from developing a framework for the statistical analysis of aggregated vote data of the sort that researchers often encounter in the study of ballot propositions and candidate elections, that is based on the standard spatial voting model. The framework is applied in a number of empirical studies of voting data on ballot propositions in California and other states over various periods since 1892. These studies utilize not only voting data on ballot propositions, but also demographic information, data on voting in presidential, gubernatorial and other elections, and data on roll-call voting in state legislatures and the U.S. Congress. One particularly important issue is the extent to which elected representatives act in the interest of a majority of their constituents.
