Bucking trends

     After last November's election, I heard more than one TV pundit remark upon a certain correlation: densely-populated areas tend to vote, in modern US presidential elections, for the Democratic candidate, and less densely-populated areas tend to vote for the Republican. That this seemed to be news to them puzzled me, but it did get me thinking: what are the counterexamples? Where are the rural liberal places, and where are the urban conservative places?

     All it took was election returns by county, census data on population density, and R. I ended up with this plot:

The ringed points on the plot represent the most positive and most negative residuals, i.e. the data with the greatest difference from the line of best fit.

     Obama's vote share is on the y-axis, and the logarithm of population density is on the x-axis (because adding, say, 100 people to a county could have a huge political effect if the county only has 100 people to begin with; adding 100 people to Los Angeles County, on the other hand, wouldn't change much).

     Clearly, the correlation we assumed holds up, although it's not as strong as I thought it'd be; as density increases, Obama's vote share increases. But what of the outliers—those data points down and to the right, or up and to the left?

     The red points are those where Obama's share was more than 26 percentage points lower than the model predicts; for the blue points, Obama's share was more than 34 percentage points higher than expected (those cutoff points aren't really significant). Here are the locations of those counties:

     The greatest outlier on the conservative side is Utah County, Utah, just south of Salt Lake City. It contains Provo, Utah's third largest city. The greatest left-leaning outlier is Shannon County, South Dakota, which lies within the Pine Ridge Indian Reservation. Over half of its population is below the poverty line.

     There are some obvious geographical patterns. Interestingly, the Deep South has outliers in both directions, located very near each other. A range of sparsely-populated, Democratic-leaning counties stretches from southwest Colorado to southern Texas, while there's a cluster of relatively dense conservative counties in northern Utah and southern Idaho. There's also a group of rural, blue counties in the Dakotas.

     It might not be a surprise that race is an important factor. Most of the red counties are 90% or more white; all are at least 80% white. The westernmost group is in the heart of Mormon country, so Mitt Romney's candidacy may have made these counties vote even more Republican than usual (I only used 2012 presidential returns). Other red areas include Randall County, Texas, which skirts the southern edge of Amarillo; Montgomery County, Texas, containing northern suburbs of Houston; Livingston Parish, Louisiana, between Baton Rouge and New Orleans; and Leslie County, Kentucky, a mountainous, coal-mining county with a community named "Hell For Certain".

     The blue areas in Montana, North and South Dakota, Wisconsin, and Arizona are majority-American Indian. The blue Texas counties and most of the blue counties in New Mexico are majority Hispanic/Latino, and the blue counties in the Southeast and Maryland are majority-black. Hawaiʻi County, Hawaii, on the other hand, is very diverse and doesn't seem to be majority-anything, although Obama's Hawaiian childhood may have played a part here. Finally, Windham County, Vermont, and two of the three blue counties in Colorado are majority-white (non-Hispanic). Who are these rural white liberals? Vermont hippies, maybe, but I'm not sure about Colorado. San Miguel County, in particular, had the greatest residual of any of the majority-white counties. It contains the town of Telluride, which I know nothing about, except that it's apparently an excellent ski resort. It's interesting that it's right across the border from conservative Utah.

     So, race probably accounts for much of the aberration from the trend. On the other hand, we could say that the correlation is only "normal" for the Midwest and the coasts, and reflects the perspective of the people who live there, not everywhere in the United States.