Georeferencing democracy, one precinct at a time.
Metric Geometry and Gerrymandering Group
Ruth Buck | Katie Jolly | Katya Kelly
To analyze the performance of electoral districts, you need to understand the distribution of voters. Across the country, votes are reported at the precinct level, but only a small number of states provide that data in a GIS format suitable for spatial analysis.
The Voting Rights Data Institute (VRDI) was a 2018 summer intensive sponsored by the Metric Geometry and Gerrymandering Group at Tufts and MIT, with major support from a Bose Research Grant at MIT and from the Jonathan M. Tisch College of Civic Life at Tufts.
We set out to find out what type of data we could obtain from the individual counties in Ohio on their precinct boundaries, and to work towards turning that into a unified statewide shapefile. We began by calling officials in all 88 counties. Some were able to provide county-wide precinct shapefiles, but many could not. Most of the counties that did not provide shapefiles sent PDF maps by email or sent hard-copy precinct maps by postal mail. The quality of these maps varied greatly. Some of the maps show rough boundaries drawn over highway maps in magic marker. Certain counties did not know where their precinct boundaries were at all.
Below is a breakdown of tasks with the number of counties in each category.
- cleaning and merging shapefiles that were publicly accessible or were provided by counties (46) ;
- digitizing and merging the precinct maps provided by Ohio county officials, whether PDF (27) or paper (18) ;
- using Ohio's public voter file to approximate boundaries for the ones that provided no maps (7 counties).
The first thing to do after receiving a precinct map in the form of a PDF (after converting it to a JPEG) is to georeference the image. We manually picked points on the JPEG that visually match to a point in an OpenStreetMap in QGIS. This allows us to give lat/long coordinates to the map image. The next step in the process is to digitize the boundaries shown in the georeferenced image. This means turning the boundaries in the image (which is a raster file) into a usable vector layer (shapefile). We aggregated 2010 census blocks to create the precinct shapes in the georeferenced images.
For counties that did not have maps available of their precincts, we needed to find another way to estimate the boundaries. Ohio's voter file is public, and it includes both addresses and precinct assignments for the registered voters in a county. We used a Census API to find 2010 census block GEOIDs for each voter. Using these GEOIDs, we developed a methodology to draw approximate precincts from census blocks. First, we classified the census blocks that already had addresses assigned to them, based on the precinct designations of those voters. Then we can classified (rook) adjacent blocks. We ran the algorithm multiple times in order to account for blocks that initially had no automatically detected rook neighbors. While this is imperfect, we have found that it provides a workable approximation when tested on counties for which we already have shapefiles. When we tested this method against Monroe County's precincts (chosen for its relatively small voter file), we found that only 2.9% of the area in the county was assigned to the wrong precinct; we consider that to be within an acceptable margin of error. For more information about the algorithm we wrote, read the blog post by Katie.
We obtained the data for 2016 election returns from the Ohio secretary of state website. We then created a lookup table for precincts by manually matching precinct names and codes from the election data to the precinct information from the shapefiles we merged. After filtering out invalid precincts from the merged shapefiles, we performed a full join on both precinct and county names to obtain a complete shapefile with the election data attached. We are still working on joining all the counties.
- Ethan Ackerman (digitizing)
- Emilia Alvarez (digitizing)
- Patrick Barnacle 1 (data collection, election data consultant)
- Assaf Bar-Natan (data collection, digitizing)
- Eion Blanchard (data collection, digitizing)
- Ruth Buck (project leader)
- Sophia Caldera (digitizing)
- Eduardo Chavez-Heredia (digitizing)
- Coly Elhai (digitizing)
- Michelle Feng (digitizing)
- Natalia Hajlasz (digitizing)
- Mallory Harris (digitizing)
- Max Hully (digitizing)
- Amara Jaeger (digitizing)
- Katie Jolly (project leader)
- Katya Kelly (project leader)
- Samir Khan (digitizing)
- Zach Levitt (data collection, digitizing)
- Heather MacDougall 1 (voterfiles)
- Bryce McLaughlin (digitizing)
- Everett Meike (digitizing)
- Sloan Nietert (digitizing)
- Cara Nix (voterfiles)
- Nathaniel Poland (digitizing, voterfiles)
- Adriana Rogers (data collection, digitizing)
- Sarita Rosenstock (digitizing)
- Kaki Ryan (digitizing, voterfiles)
- Anna Schall (digitizing)
- Lily Wang (digitizing)
- Hannah Wheelen (data collection, digitizing, voterfiles, joining)
- Cory Wilson (digitizing)
1 Not affiliated with VRDI
This shapefile is the most up-to-date as of 11/24/2018. We are still working on matching a few of the counties to election results and aggregating some results. Most of the geometry has been cleaned, including removing dangles, (most) gaps, and (most) polygon slivers. If you find other errors, please let us know! Some of the larger gaps are still in the process of being fixed. We recommend using the geopackage over the shapefile because it is a more flexible and lightweight filetype to work with. We include the shapefile because it is still the industry standard, though.
Below are listed all of the variables in the attribute table for the precinct shapefile and a brief explanation of each one:
PREC_SHP: name of the precinct as displayed in the merged shapefileCNTY_NAME: name of the county in which the precinct is locatedCNTY_GEOID: GEOID of the county in which the precinct is locatedPREC_ELEC: name of the precinct as displayed in the election dataPREC_CODE: a three-letter code, unique only within each county, assigned to the precinct in the election dataPREC_MIS: contains comma-separated names and codes of mismatched precincts between the shapefile and election dataREGION: broader region of Ohio in which the precinct is locatedMET_AREA: metropolitan/media area in which the precinct is locatedNUM_REG: number of registered voters (as of 2016 general election)TRNOUT: number of people who actually voted (as of 2016 general election)TRNOUT_PCT: percentage of registered voters who actually voted (as of 2016 general election)PRES_DEM16: number of votes for the 2016 Democratic presidential candidatePRES_IND16: number of votes for the 2016 Independent presidential candidatePRES_GRN16: number of votes for the 2016 Green Party presidential candidatePRES_REP16: number of votes for the 2016 Republican presidential candidateSEN_REP_16: number of votes for the 2016 Republican senatorial candidateSEN_DEM_16: number of votes for the 2016 Democratic senatorial candidateGOV_GRN14: number of votes for the 2014 Green Party gubernatorial candidateGOV_DEM14: number of votes for the 2014 Democratic gubernatorial candidateGOV_REP14: number of votes for the 2014 Republican gubernatorial candidatePRES_LIB12: number of votes for the 2012 Libertarian presidential candidatePRES_DEM12: number of votes for the 2012 Democratic presidential candidatePRES_REP12: number of votes for the 2012 Republican presidential candidatePRES_GRN12: number of votes for the 2012 Green Party presidential candidate
The precinct shapefile is currently displayed using the NAD83(HARN) UTM zone 17N projection, which has an accuracy of better than 1m in the contiguous United States. We chose this projection because the extent of the data is only the state boundary of Ohio, which makes a localized projection such as a UTM zone a good choice.
The products of this project should be considered public and freely shareable. Please cite this repo and credit the Metric Geometry and Gerrymandering Group.