Analyzing Wisconsin Voter Participation by Population Centers in the 2018 Gubernatorial Election


Lab Reports, Maps

Introduction

Hoping to move outside of the city for this lab’s subject, I was excited to find interesting and manageable datasets for my home state of Wisconsin, made available by our Legislative Technology Services Bureau. With midterms and elections on the mind, data on state elections from 2012-2020 with information on voter racial identity, county, ward, and district, and breakdown by individual years and races was certainly intriguing and seemed an excellent resource to pull something relevant out of to visualize on a map. By combining it with the data on cities, towns, and villages across the state, I hoped to answer questions about voting rates according to population density. 

Methods 

Dataset

My datasets were found through the Legislative Technology Services Bureau Open Data website for Wisconsin: 2012-2020 Election Data with 2020 Wards and WI Cities, Towns and Villages (July 2022). The former provided me with the geographical data to visualize the state map as well as the voting demographic information while mapping the latter helped illuminate concentrations of population centers. 

Software

QGIS, an open-source geographic information application, was used in this section’s lab to produce my visualization of voter proportions to population in the 2018 governor’s race. 

Process & Reflections 

This course has steadily built on the difficulty of each platform from lab to lab, and QGIS was perhaps the least intuitive to me yet. Just finding data that made sense to me and came with a SHP file was difficult enough; LTSB’s Open Data Page was an exciting find with its numerous download options for state legislative and election data. I started with a map of the cities, towns, and villages data points overlaid with the state’s congressional districts. Because this did not necessarily provide any interesting information or answer any questions beyond how many municipalities exist in each district, I had to search further to find my final election dataset, which included helpful geographic information. Adding this as a SHP file while moving cities and towns to the top layer as centroids allowed for the latter to remain visible before the former’s data had its symbology edited. Under the graduated property, I decided to focus on the total vote count variable from the 2018 gubernatorial race between Tony Evers and Scott Walker. After classifying and applying, my map showed the total number of votes in the year’s governor’s race by municipal ward, with cities and towns mapped over via its centroid visualization. 

Better than just showing the raw numbers, I had to search how to possibly change this to instead a rate of voting by dividing votes by the population. After a lot of trial and error, I used the field calculator to express the variables GOVTOTAL18 / PERSONS, which divided the votes from that year’s race by population, creating a new column, which I simply named govvote. Changing my symbology to this new rate of voting instead shows the percentage of voters within each ward and thus highlights which city and town areas have higher and lower voter turnout against their total population (albeit in this one year and only for governor, though this was a significant year given Evers’ usurping of incumbent Walker). I did run this version past a friend who happens to be in politics, and her suggestions mainly included wanting to include more information from the dataset I had at my disposal, such as incorporating a racial breakdown of the voter turnout. Ultimately, I decided my visual was already rather busy, and I did not want it to feel weighed down by too much information. 

Once I added my legend, I did realize the number categories listed as proportional could be read as confusing in a few ways – it even momentarily confused me. While 0.375, the max for the smallest section, represents 37.5% proportionally to the whole population, it would be a lot simpler if the legend itself could just read as percentages instead. This is the most glaring element I wish I figured out how to fix and would pay closer attention to doing any further population work in QGIS going forward. Additionally, it took some time to realize what was happening in the last section, which goes from 0.574 to 8.125. Because this is especially concentrated in metropolitan areas like Madison and Milwaukee, I believe it may be influenced by a higher turnout because of an increased population of students, which would be outside residential population numbers. While that is one consideration, I am unsure if or what other factors may also be contributing to the skew, and I also don’t feel confident in how to isolate that issue within the QGIS platform. Despite these flaws, I believe my visualization is still fairly clear in its informational intent, and I would be interested to see the breakdown of similar data in the most recent election to compare differences in the state between midterm years.

References:

Legislative Technology Services Bureau Open Data Page (2012-2020). 2012-2020 Election Data with 2020 Wards. [Data set]. Shapefile. https://catalog.data.gov/dataset/donations-to-not-for-profit-organizations-affiliated-with-elected-officials

Legislative Technology Services Bureau Open Data Page (2022). WI Cities, Towns and Villages (July 2022). [Data set]. CSV. https://data-ltsb.opendata.arcgis.com/datasets/LTSB::wi-cities-towns-and-villages-july-2022/explore?location=43.548113%2C-88.604691%2C7.67