Barriers to Voting: A Look at Mohave County, AZ, Post-Shelby
Introduction/Background:
For a country that heavily focuses on the promotion of democracy around the world, the United States has a long history with one of the basic tenets of democracy: the unfettered ability to cast a vote. In the earlier days of the Republic, individual states decided who had the right to vote - for many states, this meant white men who owned a certain amount of land. In 1856, all white men could vote. In the 100 years following the Civil War, other major groups of people began to be able to vote as well: non-white men in 1870 with the 15th Amendment; women in 1920, after a long battle leading up to the 19th Amendment; Chinese immigrants in 1943 by the Magnuson Act, to name a few. Finally, in 1971, people over 18 were granted the right to vote through the 26th Amendment.
However, these Amendments did nothing to stop covert methods of disenfranchisement. Following the passing of the 15th Amendment, the former slave states used Jim Crow laws to discriminate against African Americans, using tactics such as literacy tests and poll taxes. Literacy tests prevented racial minorities and immigrants from voting and poll taxes disenfranchised the poor (again, typically racial minorities and immigrants). Legally and politically, it was very difficult for more liberal actors to overturn the decision of Plessy vs. Ferguson. After much pressure from the Civil Rights Movement (including people like Martin Luther King, Jr. and Malcolm X) and the attacks on non-violent protesters across Alabama, President Lyndon Johnson called for the creation of the Voting Rights Act of 1965.
The Voting Rights Act (VRA) is a landmark piece of legislation by all accounts, but the most relevant part is Section 5. Known as the “preclearance provision”, this section forced “covered jurisdictions” (i.e. Alabama, Alaska, Georgia, Arizona, etc.) with a history of discrimination to submit any and all changes to the U.S. Department of Justice or the U.S. District Court for D.C. so that the Attorney General can decide if it has a discriminatory purpose or effect.[1] By implementing Section 5,
“Congress had found that case-by-case litigation was inadequate to combat widespread and persistent discrimination in voting, because of the inordinate amount of time and energy required to overcome the obstructionist tactics invariably encountered in these lawsuits. After enduring nearly a century of systematic resistance to the Fifteenth Amendment, Congress ... decide[d] to shift the advantage of time and inertia from the perpetrators of the evil to its victims.”[2]
This part of the VRA made it extremely difficult for states to enact discriminatory changes and most likely deterred many possible changes. However, in 2013, Shelby County, Alabama filed a suit in the district court, seeking judgement that section 5 and 4(b)[3] of the VRA are “unconstitutional and a permanent injunction against their enforcement.”[4] The Supreme Court agreed, 5-4, and struck down the “coverage formula” that dictated which jurisdictions had to submit to preclearance, allowing previous Section 5 states to make voting changes unfettered until Congress can create a new formula. According to a report by The Leadership Conference Education Fund, these states have since “shortened voting hours and days, enacted new barriers to voter registration, purged millions of eligible voters from the rolls, implemented strict voter identification laws, reshaped voting districts, and closed polling places.”[5]
Polling location closure is particularly vicious. It often occurs at quietly and at the last minute, leading to longer lines at other locations, transportation issues, denial of poll-site assistance, and a general increase in the cost of voting. In addition, polling locations close for a variety of legitimate reasons, making it impossible to attribute to discrimination. As the institutional cost of voting increases, turnout decreases -- and as turnout decreases, the administrative cost to the state of holding extra polling locations increases, providing a convenient reason for more locations to close.[6] Diverting Democracy finds that, although there has been an increase in voter turnout, there were 1,173 fewer polling places in formerly covered Section 5 jurisdictions between the 2014 and 2018 midterm election and 1,688 overall closures between 2012 and 2018.[7]
One of these counties is Mohave County in Arizona. According to Diverting Democracy, Mohave County closed about half of its 2012 polling stations (~34 locations), from 70 locations to 36. Of these 34 closures, 30 of them closed after the 2014 midterm elections when Republicans gained control of both chambers of Congress. Since these closures occurred after Shelby, Mohave County was not required to justify their actions - as a result, it is very difficult to find any releases or statements on the motivation behind these closures.
With this paper, I hope to address two issues. First, I will analyze the effect of the polling location closures in Mohave County on both racial and political demographics. I chose Mohave County because of the magnitude of its closures (49%) relative to the overall number of polling locations (70 to 36). If the statements made by activists and the Leadership Conference Education Fund are true, then we will see that minority voters are more affected by these closures than the white majority. In addition, I will test the common argument that the closure of these locations were driven by a low voter turnout.
Data:
The analysis will be done in ArcGIS with some data cleaning in Excel and Stata. The 2016 polling location and voting precinct boundaries shapefile was downloaded from the Mohave County GIS webpage.[8] However, data for Mohave County’s polling locations prior to Shelby was not available publicly. This could have derailed the project before any analysis was to begin, but luckily, there was a natural proxy at the precinct level. In 2015, Mohave County approved a consolidation of voting precincts, going from 73 precincts to 24. According to the U.S. Election Assistance Commission (EAC), each voting precinct must be represented with a polling location.[9] By consolidating the voting precincts, the state can offer less voting locations. It could also behave as a form of gerrymandering - the absorption of demographics means that the citizens are less accurately represented. While this is not the same as having the voting locations pre-Shelby, this change should still allow us to analyze the effects on certain demographics.
Data is much more easily found at the voting precinct level. Stephen Ansolabehere (Harvard University) and Jonathan Rodden (Stanford University) had compiled an Arizona shapefile for election results in 2010.[10] This includes the 73 precincts before consolidation and includes the percentage of votes, by party, in the 2008 presidential general election and the 2010 senatorial midterm election.
The biggest bulk of information comes from the Public Mapping Project, of which the principal investigators are Dr. Michael McDonald (George Mason University) and Dr. Micah Altman (Harvard University).[11] The team at the Public Mapping Project disaggregated precinct level election data with census block data to create a dataset containing demographics and election results at the census block level. The issue with combining census block information with voting precinct boundaries (the Census calls them Voting-Tabulation Districts, or VTDs) is that the geographies of the two levels of data do not add up. McDonald and Altman used 3 different methods of disaggregating precinct level data to the census block level.[12] While error is present as a result of the disaggregation, this dataset was the most well-documented and complete for the analysis that I had hoped to run.
When interpreting the results created by this data, it is important to remember its limitations. We must incorrectly assume that the population and its demographics are evenly distributed across the state, due to the aggregated nature of the data. The creation of a more nuanced spatial analysis is beyond the constraints of this paper - thus, its conclusions must be taken as an introductory measure which hopes to serve as a foundation for other projects.
Methodology:
In order to prepare the Public Mapping Project data for merging with the relevant shapefiles, I used the identifier data to restrict and rename all 73 voting precincts in Mohave County in Excel. I then re-aggregated the data, taking the sum of every variable by their unique identifier in Stata. After importing this re-aggregated data into Excel, I calculated the percentages of each demographic and election results in order to better understand the magnitude of each variable. The data was imported into ArcMap 10.7.1 and joined with the Arizona shapefile provided by the Harvard Dataverse. The Arizona and Mohave County shapefiles were projected into “NAD_1983_NSRS2007_StatePlane_Arizona_West_FIPS_0203_Ft_Intl”, after which I restricted the Arizona shapefile to only show Mohave County through feature selection.
Figure 1 is a simple visualization of the percentage of white, non-Hispanic, voting age population per district. This is an introductory map, meant to show the distribution of white voters in Mohave County during 2010. I used a geometric interval classification due to the skewness of the data. To make Figure 2, the percentages in Figure 1 were rasterized to a half mile cell block. Zonal statistics was used to take the mean of each cell within the boundaries of the consolidated precincts, providing us with the data for Figure 2. In order to better visualize the magnitude of absorption, Figure 3 was created by taking Figure 1 and showing the consolidated precinct boundaries rather than the pre-consolidation precinct data. This map is essentially the intermediate step between Figures 1 and 2 and allows us to see where and to what degree demographics are being absorbed.
The political voting data is based on the Public Mapping Project data. The Harvard and Public Mapping Project data are very similar, but since the Harvard data uses the Public Mapping Project data as well, it seemed more fitting to use this data instead. It also contains the fraction of votes up to 2 decimal places, which is much more precise than the rounding method that the Harvard data does. To create Figures 4 and 5, I used the same process as creating Figure 3. This is the more succinct way to visualize the effect of consolidation. The colors on the map correspond to the American political parties - red for Republicans, blue for Democrats, and purple for close races. As a form of robustness check, I have also included the midterm elections - often the second largest (in terms of attention and voter turnout) election.
Figure 6 shows where the polling locations are relative to precincts that have both a high number of voting aged population (VAP) and voter turnout (total votes cast) in the 2008 presidential election. Presidential general elections tend to have a higher voter turnout than Senate midterm elections and gives us a better estimate of people who care some bit about voting but not necessarily enough to participate in the midterms. The process that made Figure 2 was used until there were 2 rasters that had been calculated to fit the 2016 boundaries - 1 for VAP and one for voter turnout. Using a conditional statement in the raster calculator, precincts in 2016 were given values of 0 if VAP was below the mean VAP across the county OR 1 if voter turnout was below its mean percentage across the county. The individual above mean categories were created by excluding values below the mean.
Lastly, Figure 7 is a basic Euclidean distance calculation to the nearest polling location from any point in Mohave County, ignoring geographical and topological constraints. The Euclidean distance tool was first used on the polling locations, resulting in an output that did not process past the extent of the points itself. The tool was run once more on that output, this time specifying the extent of Mohave County, resulting in an output that was wider than the county itself. To achieve the results, the values of the raster were extracted by using the county boundaries as a mask. The distances were adjusted to be in intervals of 10 miles each.
Results:
Figures 1, 2, and 3 were created to visualize the demographics of voters in the pre- and post-consolidated precincts. While Mohave County is almost overwhelmingly white, there remains pockets of precincts where the total of ethnic minorities is highly present. For example, the dark green precinct to the east-northeast part of Mohave County is almost entirely populated with Native Americans. The precincts where there are majority minority voters (i.e. where voting age minorities outnumber voting age whites) are not affected by the changes. Rather, it is areas where there is a (relatively) middling number of minority voters (50% - 30%) that get absorbed if one looks at the west-southwest border. The light green areas in Figure 1 have become darker green in Figure 2, indicating an increase majority of white voters. It seems as if the consolidation forces racial lines to meet somewhere in the middle. While the representation of minority voters in the area has decreased by an average of 5%, the percentage of white voters has also decreased by about 5%, ensuring an equal “loss” on both sides of the racial line. In Figure 3, we can also see specifically that the 2016 polling location is located close to the previously high minority area, which should improve minority voter turnout in that area regardless of consolidation.
Next, Figures 4 and 5 looks to determine if the consolidation of polling locations is politically motivated. For the general election in 2008, it is unsurprising that a county in Arizona is deeply red, especially since the Republican nominee, John McCain, served as their long-time senator. We find that the same precincts show up as areas of interest. In these figures, the western border, particularly in the zoomed area, tended to vote more Democrat than the rest of the county. In addition, the other highlighted areas in the center of Mohave county get absorbed, but these are relatively small, considering our spatial resolution is a half-mile block. The midterm election follows the same trend, with small areas in the center and the west border of Mohave being consolidated into larger ones. Once again, we find no evidence that the consolidation had any effect on the political distribution of the county.
Figure 6 hopes to provide some insights on the polling locations as they stood in 2016. By highlighting the precincts with a high voting age population and a high voter turnout, we test 2 counterfactuals of the administrative benefit argument: (1) that the precinct’s voting age population is too low, so they should create an inter-precinct polling location; and (2) that counties with low turnout incur a net negative benefit for having the polling location open. We find that the opposite is true – some counties with a population and turnout over the mean have only one precinct for people to report to and others that are below the mean have multiple. We see that the large county in the north has a large voting age population, but below-mean voter turnout. Toward the southeast, we see that while the voting age population is below the mean, there is a high amount of turnout.
Figure 7 finds that most of Mohave County falls under 30 miles of a polling station. Given the overall population size of the county, this seems acceptable, but could use some improvement. The parts that are over 30 miles of the nearest polling location fall under a national park and is no cause for concern.
Interpretations and Policy Recommendations:
Overall, we cannot prove, visually or otherwise, that the consolidation of voting precincts was more harmful to minorities or to a specific political party. On the contrary, my data shows that Mohave County covered their bases when redrawing the precinct lines, accounting for possible instances of discrimination or partisanship.
Instead, the primary complaint from this paper comes from the possible inefficiencies of and at polling locations, which is shown by Figure 6. First, consider the areas of high voting age population and low voter turnout, as shown in light blue. The northern precinct is mostly national park land, with presumably very little to no people living in that area. It must be the case then, that the population is clustered around the northern area of that precinct itself. If people are highly clustered in a location, the use of a small number of polling location makes sense when considering ease of delivery, but one must also consider the case that the low turnout rate is explained by difficulties at the location itself, such as long lines, inadequate access, etc. This is beyond the scope of this paper, but it is important to consider such possibilities and warrants a deeper investigation by election officials in Mohave County.
Second, the southeast portion of the county has high voter turnout, but low population size. Combined with the characteristics of the other maps, it is a reasonable assumption that the area is populated by older, possibly retired, white and Republican voters. While there is no indication in my data of precinct quality, the county should proactively ensure the accessibility of these polling locations, following the guidelines laid out by the EAC. Barreto et al. found that a “ten-unit shift in precinct quality from lowest to highest quality should result in a boost of 3.4 points to that precinct’s voter turnout rate”, which is significant at the 95% confidence level.[13]
Lastly, while the polling locations are acceptable in the voting precincts that were consolidated vis-a-vis racial discrimination, other polling locations are inefficient based on the distance it takes to get to one’s assigned location. Arizona enforces a one precinct, one location rule, meaning that one’s precinct of residency determines the ONLY location one can vote. Therefore, when reading Figure 7, one should remember that the boundaries are important when determining voting location, and that the real distance to the polling location is often longer due to travel conditions. Cantoni (forthcoming) finds that, in Massachusetts and Minnesota, a 0.245-mile increase in distance reduces ballots cast by 2 to 5 percent across all elections. This effect is 3 times higher during non-presidential elections in high-minority areas than in low-minority areas.[14] In an expanded paper, one could look at the density of population using census block level data to further optimize the polling locations, but for now, the combinations of Figures 6 and 7 provide targeted areas of improvement for Mohave County.
Future Works:
As previously stated, this paper hopes to serve as a starting point on a discussion of the fallout of Shelby County v. Holder. While my analysis supports the idea that Mohave County acted thoughtfully with their precinct consolidation regarding discrimination, it has not ruled out the possibility that the closure of half of its polling locations adversely affected minorities disproportionately. I have listed a few continuations of this research that would give us a more robust understanding of the effects of Shelby – not just in Mohave County, but in other affected states, particularly in states like Georgia and Alabama.
Replicate research on data with historical polling locations for previously labelled Section 5 states, of which are listed in Diverting Democracy.
Analyze different demographics such as age or income.
Use closer year demographics (eg. 2014 ACS) or more direct methods, such as registered voter rolls, to improve the accuracy of this analysis.
[1] https://en.wikipedia.org/wiki/Voting_Rights_Act_of_1965
[2] South Carolina v. Katzenbach, 383 U.S. 301, 328 (1966)
[3] Section 4(b) defines the eligible districts as ones that had a voting test in place as of November 1, 1964 and less than 50% turnout for the 1964 presidential election.
[4] Oyez. “Shelby County v. Holder.” (2013)
[5] The Leadership Conference Education Fund, Diverting Democracy: Polling Place Closures and the Right to Vote. September 2019. http://civilrightsdocs.info/pdf/reports/Democracy-Diverted.pdf
[6] Barreto, et. al, “Are All Precincts Created Equal?”, Political Research Quarterly Vol.62, Number 3 (2009)
[7] Diverting Democracy, pg. 10.
[8] http://www.arcgis.com/home/item.html?id=de45bf32947f4927afc111a2e13e8ae9
[9] https://www.eac.gov/assets/1/6/Chapter_9_Polling_Place_and_Vote_Center_Management.pdf
[10] Stephen Ansolabehere; Jonathan Rodden, 2011, "Arizona Data Files", https://doi.org/10.7910/DVN/HTUGFD, Harvard Dataverse, V1, UNF:5:rVBxfNvXPiyymE4TepqjMQ== [fileUNF]
[11] http://www.publicmapping.org
[12] They describe their methodology here: http://www.publicmapping.org/resources/data#TOC-Methods-of-Merging-Election-and-Census-Data
[13] Barreto et al (2009), pg. 454
[14] Cantoni, Enrico. “A Precinct Too Far: Turnout and Voting Costs.” American Economic Journal: Applied Economics. https://doi.org/10.1257/app.20180306.
