Introduction

The 2020 Presidential election was one I will remember vividly for the rest of my life. History will look back on the decision to elect Joe Biden as the 46th president as one of the most significant democratic choices of my lifetime. But as someone who identifies as liberal, the election results were more than shocking to me. Donald Trump received more than 73 million votes. I have plenty of friends who would identify as conservative, though I believe not one of them voted for Trump. So if I am living in such a big bubble, where are all the Trump voters? One of the obvious facts here is that the divide between the left in the right is largely a rural and urban divide. This data project tries to better understand that fact with a visual.

In order to better visualize the urban rural divide in the recent 2020 election results, I use a novel dataset that was introduced in the paper that I presented in class: The DMSP-OLS Nighttime Lights Time Series.

The DMSP-OLS Nighttime Lights Time Series is a series of lights at night data that are constructed at a yearly level with a mosaic of raw satellite images at a resolution of 30 arc-seconds, or roughly one km squares. Values for light intensity for each pixel are indexed and range from 0 to 62 and the data are available from 1992 to 2013. For this project, only the 2013 images are presented due to lack of computing power.

Light intensity is a unique and interesting measure of urbanization for this visual. Highly urbanized areas obviously radiate large amounts of light pollution while rural areas are clear. By combining the night lights with recent (unofficial) election results scraped from the New York Times, I am able to uniquely visualize the stark divide present in contemporary American politics.

Fact Construction

In order to construct this fact, I found a shapefile of the United State with polygons representing county borders. After limiting the extent of the shapefile to the contiguous US, the raw file plotted geographically looks like this:

For the raster image of the night time lights data, I similarly mask the extent of the global file to the contiguous US. Plotted it looks like this:

In order to use these data in conjunction, I had to aggregate the raster data up to the county level. With more time and some elbow grease, I would have preferred to use data at a more refined level. However, I could not acquire any election data at a finer geographic designation.

The pixel values in the raster are summed up for every point that lies within the county polygon. That sum is then divided by the total count of pixels within the polygon to construct the cell average light emission for each county. The following plot visualizes the aggregate data.

Finally, I merged the county level election data to the shapefile using county fips codes. (Note: I did not scrape these data myself, I found someone on github who did and used their code to construct my own election data - https://github.com/favstats/USElection2020-NYT-Results). Plotting the election results on the shapefiles looks like

The Fact

The fact is that there is a strong, positive correlation between counties with high average light emissions and counties that voted for Biden in the 2020 election. This is not surprising although it is novel in its own way. The blue line is just a line fit to minimize squared residuals and the red line is fixed at the 50% level to signify which candidate the county preferred.

## `geom_smooth()` using formula 'y ~ x'

Interpretation

This fact is not very significant in its own right. It is a symptom of the political polarization that has been growing in this country over the last decade and the economic downturn that has left particular groups behind. To me it is clear that this relationship exists. One way to push this fact further would be to compare this result with previous elections. It would be more interesting to measure the growth of this divide over time.

The reason I choose this for my project is that I really wanted to mess around with the night lights data. The fact that this was occurring during a historic election week likely led me to search for election data to use with it. Looking back over my project, I wish that I was able to find a more concrete, new fact that had more economic significance. Something that was easier to interpret and build upon into a project. But oh well. If I ever find a use for the night lights data in my own research, I have a starting place.

Also, I am heartbroken that I cannot embed a 3D render of the county map that I create. Unfortunately R Markdown does not seem to allow it without an error. So I included it as an .mp4