Patchwork Nation Methodology

The methodology for classifying America's 3,142 counties into 11 categories involved taking a large set of social and economic data and identifying the underlying factors that best explain how the counties differ from one another.

This is inevitably a somewhat subjective process, but, as explained below, the description of Patchwork Nation is rooted in rigorous analytical methods.

This project was funded by the Knight Foundation, a nonprofit philanthropic organization based in Miami, Fla. The project's principal consultants are James Gimpel, a political science professor at the University of Maryland, College Park, and Ann Cizmar, a PhD student there.

In this case, we started with a large set of population, demographic, economic, consumer expenditure, and religious adherence indicators. From them, we condensed all US counties into the 11 categories that we present on this website. The indicators selected for counties were chosen based on their relevance to American politics.

Included are several measures of income, local economic activity, and occupational mix; measures of racial and ethnic composition, and immigration; along with measures of religious adherence at each location for Catholics, evangelical Protestants, Jews, Mormons, and mainline Protestants.

Housing stock indicators were included along with population density, and whether the county was located within a major metropolitan area. We also captured the education level of the population along with recent population growth and migration figures.

We looked at consumer expenditure estimates (measured as a percentage of all household expenditures) for a variety of specific spending categories, including alcohol, tobacco, housing, new vehicle purchase, property taxes, and charitable contributions.

Source of data and analysis method

The majority of our data comes from the 2000 US Census and 2006 estimates of common census items at the county level. Data on religious adherence are from the Glenmary Research Center's Survey of Religious Congregations in America, 2000.

After we acquired the data, we converted all variables for the analysis to percentages or rates for purposes of factor analysis. A link to the data in Microsoft Excel format is here. These data are available to any person or organization that wishes to mine them. Our analysis can be used to focus on a single state or particular region and the categories can be mapped in those locations. Please feel free to contact Professor Jim Gimpel regarding the data or methodology.

We did the factor analysis of the data with standard statistical software, SPSS using varimax rotation. We identified 11 core components, which stood out statistically as best explaining the differences among counties across the wider spectrum of data. In statistical terms, these components explained correlations among the extensive set of county level indicators we used.

After several variables are found to indicate a single underlying component, the principal components procedure produces a factor score, derived from the weighted combination of the variables that are highly associated with that factor. (A factor is a composite index based on combinations of variables such as Hispanic, tobacco spending, and population growth.)

Not all factors accounted for the same amount of variation, classified locations equally well, or classified the same number of locations. For example, there were far more affluent metropolitan counties and agricultural and trade counties than there were counties closely tied to military bases.

After factor scores were extracted for each of the categories, we rescaled these scores to range from 0 to 100. All 3,142 counties receive a score on each of the 11 factors. The factor scores, therefore, are not exclusive. Many counties rank high on more than one factor because those counties are large and contain highly heterogeneous populations. For example, many large industrial cities are also diversifying. Locations with large retirement-age populations are also known for having trade and service economies.

In those cases where a county might fit one of several community types, we decided to categorize the county into one of the 11 categories by how high it ranked above the mean on each of the factors considered side by side. In some cases, researchers were required to make judgment calls about best classification based on additional detailed local information about the locations.

Limitations of research

We are mindful of the limitations and pitfalls of electoral research that is based on observing counties. First, measures of central tendency, absolute size, or county level percentages mask important ground-level details, such as variability internal to the county. Median income may be high in Chicago's North Shore neighborhoods, but lower on the South Side. A single median income figure for all of Cook County obscures this variability. The ecological fallacy also looms large if we use county level figures to make inferences about the behavior of particular places or individuals within the county.

However, we believe that an analysis of county level political behavior reveals the diversity of populations, lifestyles, and political preferences that underlie support for the two major political parties in contemporary elections. Two locations may cast equal proportions of their vote for the Democratic candidate, but do so for different reasons or from different perspectives. The meaning of a vote of support in one county may be different from the meaning of that same vote in another.

Our method is not the only possible, useful, or interesting one. Other classifications could equally analyze the ground-level variability in the nature of support for a candidate or political party. We would emphasize, however, that the local political environments that structure an individual's political life are made up of a complex of economic, social, and religious forces that combine in various ways to shape attitudes and political behavior. This is why our particular research strategy of analyzing data on all of these local characteristics is reasonable and defensible.