Blue City … Red City? A Comparison of Competing Theories of Core County Outcomes in U.S. Presidential Elections, 2000–2012

ABSTRACT:

The Red/Blue dichotomy describing presidential elections, while criticized, is ubiquitous: Red states vote Republican, Blue states Democratic. Locally, suburban and rural counties are often Red, urban counties Blue. This overgeneralization misses the Republican share of urban centers. This study analyzes the 2000–2012 presidential elections in core counties of metropolitan areas with populations over 250,000. Possible explanations for urban election outcomes cover three theoretical groupings: sociodemographics, culture, and economics. Several prominent explanatory variables from each are compared. Changes from 2000–2004 to 2008–2012 are highlighted given the 2008 economic crash and President Obama’s race and urban identity, which permitted him to cut President Bush’s core county share in half. Regression analyses find that sociodemographic and cultural features account for most variation for all elections, while economic indicators add little explanatory power. In contrast to conventional thinking, economics mattered most in 2004, culture increased in importance in 2008–2012, and urban foreclosures positively influenced McCain in 2008.

Notes

This study was planned and carried out without reference to Chinni and Gimpel’s () variables. However, after analysis and presentation, one reader noted a similarity to their 12 community types. Chinni and Gimpel create sociodemographic (Minority Central), cultural/religious (Evangelical Epicenters, Mormon Outposts), and economic categories (Industrial Metropolis) in which to fit all U.S. counties. It is a bit of a stretch to claim one single variable is the dominant force for each county. The 92 counties included in this study are spread across several of the 12 types and no substantial patterns are readily evident beyond simple generalizations (such as that Minority Central and Industrial Metropolis counties tend to vote Democratic).

In cases where the central city crosses two or more counties, the county with the largest share of the city population was chosen. Several cities (Baltimore, St. Louis, and Virginia Beach) are considered as counties by their state and thus included in the sample as cities. Volusia and Orange Counties, Florida, come closest to being contiguous of all counties in the sample. While the two do touch in a very small area, they are separated by Seminole County. Several additional counties were removed due to missing data on one or more of the cultural independent variables.

One option to diagnose the presence of spatial autocorrelation in this data set is to create a spatial weights matrix based on distance. Another option, which would allow the construction of a spatial model after diagnosis, is to convert the core counties to points, convert them again to Thiessen polygons, and then create a spatial weight file based on contiguity. The creation of the Thiessen polygons allows one to determine “neighbors” through first-order rook contiguity. See Appendix A for the resulting map. Both options produced the same results—Moran’s I and Lagrange Multiplier (both lag and error) tests produced statistics that were not significant at the 0.05 level. Moran’s I is close to 0 in each election model (for example, 0.017275 in 2012)—which, if 0, would signify complete spatial randomness. This compares to all-county studies with Moran’s I statistics between 0.5–0.6 (much closer to 1, which would mean perfectly clustered; see Kim et al., ). Because the LM (lag) was closest to significance (0.10 level), I ran a spatial lag model for each year and found that the spatially lagged dependent variable is not significant—nor is the likelihood ratio test.

Third party support is not taken into account. However, in cases where electoral fusion is present, such as New York state, all votes cast for the two major party candidates are tallied—including those cast for third parties nominating the major party candidates. For example, in 2012, the Romney/Ryan ticket appeared on both the Republican and Constitution Party ballots and the Obama/Biden ticket was nominated by both the Democratic and Working Family Parties in New York.

The 2007 data were collected prior to the 2008 election and analyzed before 2008 ACS data were released.

Updated 2010 RCMS data are available at the county-level online at the Association of Religion Data Archives’ website: www.thearda.com/rcms2010/.

Unemployment rate lagged one month is the standard practice in predicting election outcomes (e.g., Lewis-Beck & Stegmaier, ). One could, in theory, lag the measure of unemployment further—particularly if the emphasis of the analysis was to really understand the nature of unemployment’s effect on the outcome. Some have also used perceived unemployment and found even stronger effects (ibid). In this case, the focus is on capturing the general effect of economic conditions and, as such, this variable serves more as a proxy of economic conditions at the time of the election. In most of the election years under study, there were not any major demonstrated shifts in unemployment in the month or two immediately prior to the election. Thus, September unemployment is generally consistent with unemployment lagged by several additional months.

These manufacturing proportions of the labor force do include a degree of employees who work in the core county but do not reside there. An exploratory analysis of a diverse (in geography and size) subsample of 10 core counties, to assess the degree of differences between manufacturing employment by place of employment and, alternatively, place of residence, found that these two measures are very similar in most cases. The average absolute difference for these 10 was 0.38 of a percentage point in 2011 and they are thus highly correlated. Nonetheless, the hypothesized effect of this variable is not wholly based on the aggregated political preferences of manufacturing employees. Rather, it captures the electoral effect of communities having greater proportions of manufacturing employment—which may affect the aggregate political behaviors of the county beyond just the individuals employed in the sector (similar to how having significant employment in the coal industry affects the politics of Appalachian states like West Virginia).

The correlation between these two foreclosure measures is 0.690, significant at the .001 level.

Because this study employs OLS regression, predicted values less than zero or over 100 are possible. However, use of a logistic transformation of the DV to make it nonbounded [ln(y/1 – y)] does not alter the directions and significance levels of the variables and the predicted values using the ratio variable do not go below 0 or above 100 (nearly all proportions are in the linear part of the sigmoidal curve). Thus, for ease of interpretation of the coefficients, this study follows other county-level presidential election studies (e.g., Lacombe & Shaughnessy, ) that also leave the dependent variable in percentage (× 100) form. A related concern is that the ratio dependent variable does not distinguish between counties of different sizes. A population control variable added to the models is not significant and does not alter the findings.

The highest correlation and lowest tolerance scores amongst the independent variables shows that population density and city proportions come closest to a multicollinearity problem. However, the tolerance scores still exceed a conventional 0.20 cutoff point. Income per capita was removed from analysis due to its high correlation with education.

In 2008, the combination of manufacturing, unemployment, and foreclosures (when entered as the first block) accounts for about 15% of the explained variation in core county support for McCain.

Immigration and the rise of Latino voters are not addressed by this study (in order to keep the number of variables down) but deserve considerable attention in other national and urban-oriented studies.

Additional information

Notes on contributors

Joshua D. Ambrosius

Joshua D. Ambrosius is an Assistant Professor in the Department of Political Science and Master of Public Administration Program at the University of Dayton. His research interests include urban and housing policy, regional governance, and religious organizations. His academic work has appeared in such journals as Journal of Urban Affairs, Housing Policy Debate, American Review of Public Administration, Journal of Urbanism, Local Environment, and Interdisciplinary Journal of Research on Religion.