Date of Completion


Embargo Period



exit polls, responsiveness, American politics, public opinion, Congress, MRP

Major Advisor

Samuel J. Best

Associate Advisor

Jeffrey W. Ladewig

Associate Advisor

Lyle A. Scruggs

Field of Study

Political science


Doctor of Philosophy

Open Access

Open Access


Do American political institutions comport with accepted norms of democratic representation? The three arguably most important American political scientists of the 20th century—Harold Lasswell, V.O. Key, and Robert Dahl—each held that if policymaking does not generally respond to the policy attitudes of constituents, then democracy in America is a fiction. The institutional design of the House of Representatives, which emphasizes frequent elections and small constituencies, makes it a likely locus of responsiveness in our political system. However, the vast literature investigating House responsiveness has failed to provide unambiguous answers about the phenomenon’s nature and extent. One reason for this is a data problem: measuring the constituency half of the constituent-representative dyad is difficult in the absence of large-n, geographically comprehensive social surveys of House district attitudes. Scholars have relied on survey disaggregation and proxy variables as substitutes, but these respectively proved unreliable and invalid. Multilevel regression and post-stratification (MRP), which calculates the attitudinal propensities of demographic groups in national surveys and adjusts these propensities by their demographic composition per jurisdiction, is a new and promising method for imputing district-level attitudes. A new generation of MRP responsiveness literature claims to, at last, precisely measure the attitudinal signals constituents are sending to legislators. I nonetheless take issue with three features of conventional MRP measures: they are cross-sectional (and assume public opinion is stable); they are derived from the general population (and not voters); and they are issue-focused (which poses problems when the issue agenda is unstable). In response, I generate longitudinal MRP measures of ideology using state and national exit polls from 2004 to 2010. Exit polls have larger biennial ns than widely-used social surveys like the Cooperative Congressional Election Study, and sample only voters. Employing my measures, I find evidence of heterogeneous ideological change and growing polarization in the districts. I also find that candidates and legislators do not respond to the preferences of median voters. Instead, and in violation of conventional democratic theory, they respond primarily to their party subconstituencies, particularly when district attitudes are widely dispersed. I offer directional voting as a formal explanation for this behavior.

Available for download on Thursday, September 05, 2019