Political analysis and statistical applications for social policy research

This thesis presents three independent essays contributing to political science research in American social policy. Each essay addresses a significant substantive question regarding democratic institutions and their policy outputs, and provides a methodological solution to a quantitative political data problem. The first essay examines electoral support for old age pensions in the US states between 1914 and 1956. Using data from twelve states, this essay tests and finds evidence contrary to important models of the political economy of welfare state development. This analysis demonstrates the impact of asymmetric elite partisan cues which structure voter choices more strongly than do individual economic interests. Little individual-level survey data exists, so inferences are required from aggregate data. Original software is provided to implement methods appropriate for such data. The second essay involves a particular feature of experimental design. Increasingly commonly, political scientists have data on experimental units prior to the random assignment of treatment. The essay's first part argues that political science experiments, especially field experiments, have neglected to take advantage of the design feature called "blocking". This part details the advantages of blocking, explicating the real costs of ignoring this technique. The essay's second part enumerates several choices facing researchers who block, and suggests appropriate options for particular settings. This section also proposes the first formalized blocking solution to the problem of interference between units. Original software is provided to implement these methods; this software was developed in the context of perhaps the largest randomized health policy intervention ever undertaken. The third essay considers patterns of racial and ethnic voting in California's direct democracy. Previous work on these patterns either focuses exclusively on ballot propositions directly targeting racial minorities or makes restrictive statistical assumptions that yield incorrect conclusions. The first research agenda ignores the wide range of other issues on which racial groups express their electoral preferences. This essay shows how relaxing the second agenda's restrictive assumptions reveals that black, Latino, and Asian voters in California are less likely than whites to be on the winning side of ballot propositions. This finding holds across a variety of propositions not targeting minorities.