UC Berkeley

Political scientists, journalists, and astute political observers agree that American political parties, at both the mass and elite level, have become more partisan over the past fifty years. The two national parties have increasingly moved apart on various social and economic issues, and elected representatives have become more ideologically divided over time. In response to party polarization, rank-and-file party identifiers have further sorted themselves into the Republican and Democratic parties. Their evaluation of political figures, as well as their positions over a range of political issues, is more likely to be influenced by their party identification now than decades ago. Yet when one closely examines the overall distributions of ideology and public opinions, one can hardly detect any changes in them over time. The electorate remains largely ideologically centrist and moderate.

Given frequent elections and the prevalence of watchdog groups and a savvy media, one would expect that representatives who deviate ideologically from the preferences of their constituents would get voted out. In principle, one would expect any electoral disconnect to diminish, if not completely disappear in the long run. Yet in reality, the opposite is true --- the electoral disconnect between the overall attitudes of the mass public and those of elites appears to have widened over time. The empirical puzzle is: How can party polarization be sustained when the constituents who elect them are not ideologically divided?

The answer is geography. Federal elections are geographically-based. There are four hundred and thirty five separate electoral districts for House seats, and fifty districts for Senate seats. Congressional members are single-minded re-election seekers who are held accountable to their home constituents---not to the national electorate. In order to understand what contributes to party polarization and electoral disconnect, one must begin by studying the spatial composition of voters across geographic regions. I argue that the increasingly skewed spatial distribution of partisan preferences, which I refer to as `geographic polarization of partisan preferences', holds the key to the empirical puzzle.

There are two ways in which geographic polarization of partisan preferences can occur. One is through electoral behavioral change; another is through spatial compositional change.

By electoral behavioral change, I refer primarily to party sorting that began after 1980. As the two national parties pull apart ideologically, voters can easily differentiate between the major parties and align themselves with the political party that lies closer to their political preferences. Because of party sorting, the connection between a person's socio-demographic characteristics and his or her partisan preference strengthens over time.

By spatial composition change, I refer to the condition in which the demographic make-up of geographic regions (or electoral districts) gets altered over time in ways that are politically relevant. There are two mechanisms that can induce spatial compositional change. The first mechanism is selective migration. When individuals' politically relevant socio-demographic characteristics and lifestyle preferences are correlated with both the migration decision and residential choice, then voters become geographically sorted over time in ways that matter for their political choices.

The second mechanism is place-varying generational replacement. The conventional notion of generational replacement suggests that cohorts coming of age in different time (or political) periods might develop distinct partisan preferences or beliefs. As younger cohorts replace older cohorts, the overall composition of the electorate might then change. While such conventional generational replacement continues to take place, I show that, after 1980, there is an additional form of generational replacement that is spatially dependent. In addition to the time in which one comes of age, the place in which one comes of age also matters. In California, I demonstrate that younger voters coming of age in pro-Democratic regions (the Bay Area and Los Angeles County) are systematically more likely to identify as Democrats than those growing up in pro-Republican regions. As these younger cohorts age, the spatial disparity in term of partisan preferences widens.

I use California as a case study because it is a very large and diverse state. There are many unique historic data, including individual-level opinion polls, voter registration data and yearly county-level demographic data. By assembling and examining various datasets, I show that selective migration began long before elite polarization resumed in Congress in the 1960s. Migration patterns have largely been stable over the past few decades. They are mostly driven by economic considerations and not by religious preferences. Over time, the accumulation of selective migration results in an increasingly skewed spatial distribution of voters of various socio-demographic characteristics. When the two national political parties began to pull apart on social and economic issues during the Reagan administration, voters took the cues and became more sorted by partisanship. Party sorting had two political impacts. First, it led to the onset of place-varying generational replacement. Second, it further accentuated the connection between the skewed spatial distribution of social-demographic characteristics and aggregate partisan preferences. Consequently, geographic polarization began to emerge in 1980 and continued to increase over time.

Prior to 1980, counties in California used to have fairly similar partisan preferences. The moderate, centrist distribution of ideology among voters was reflected by a bell-shape distribution of partisan preferences at the county level. By 2000, the distribution of partisan preferences at the county level had become bimodal --- counties either had gone more pro-Democratic or pro-Republican, with very few electorally competitive counties in between. Through the interaction of electoral behavioral changes and spatial compositional changes, the moderate, centrist electorate is now spatially arranged in partisan polarized districts. These districts perpetuate party polarization at the elite level as Congressional candidates must appeal to either strongly Democratic or strongly Republican electorates within their districts. Lastly, I argue that since party sorting and the pattern of selective migration are deeply entrenched in the electorate, geographic polarization and electoral disconnect are likely to be sustained in the long run.