DRA 2020


2010 Voting Districts Shapes

  • 2010 Census data (except CA, MT, OR, RI use Census Block Group shapes.)
  • 2008 Presidential election data: all states
  • 2016 Presidential election data: AK, CA, CO, DE, FL, GA, HI, MA, MD, MI, NC, NM, RI, SC, TN, TX, VA, WI
  • 2016 Senate election data: AK, CO, FL, GA, HI, MD, NC, SC, VA, WI
  • 2016 Governor, Lt Gov, and/or AG election data: DE, NC, WI
  • 2017 Governor, Lt Gov, and AG election data: VA
  • 2018 Senate election data: VA
  • 2012/2016 PVI data: states with more than 1 congressional district

2016 Block Groups Shapes

  • 2016 American Community Survey data (Census population estimates)
  • 2008 Presidential election data: all states except KY
  • 2016 Presidential election data: MI, PA, VA


  • 2008 Presidential election data for IN, NC, NM, TX: Stephen Ansolabehere, Jonathan Rodden. 2010 Census Shapefile and Precinct-Level Election Data. Release 2011.
  • 2008 Presidential election data for AK, AL, AR, AZ, CA, CT, DC, DE, FL, GA, HI, IA, ID, IL, KS, KY, LA, MA, MD, ME, MI, MN, MS, MO, MT, ND, NE, NH, NJ, NV, NY, OH, OK, OR, PA, RI, SC, SD, TN, UT, VA, VT, WA, WI, WV, WY: Steve Gerontakis with assistance by John Mifflin.
  • 2008 Presidential data for CO: Tyler Chafee.
  • 2016 Presidential data, plus 2016/17/18 election data for other contests for VA: Princeton Gerrymandering Project
  • 2016 Presidential data, plus 2016 election data for other contests for AK, CA, CO, DE, FL, GA, HI, MA, MD, MI, NC, NM, RI, SC, TN, TX, WI: Florida Election Science Team, 2018, "2016 Precinct-Level Election Results"
  • PVI data was obtained from Decision Desk HQ by Cook Political Report and FiveThirtyEight.
  • PA 2010 Voting Districts are the originals, not updated.

Disaggregation and Aggregation

To map election results and demographic data from a source geometry (shape set) to a destination geometry (shape set), we disaggregate the data from the source geometry to census blocks and then aggregate from those to the destination geometry.

For example, 2016 election data for many states is presented in terms of each state's 2016 precincts (or voting districts), which typically differ from their 2010 precincts. The representation of those precincts on a map is the geometry with which that election data is associated. Our algorithm determines the geometric mapping between the 2016 precincts and 2010 census blocks and then distributes the results for each precinct among the blocks in that precinct, according to the 2010 voting age population of each block. (That's disaggregation.) Then we determine the geometric mapping between census blocks and the 2010 precincts and add the results from all the blocks in each precinct. (That's aggregation.)

We compared the aggregated data to the original to see how many votes were "lost" in the process. For most states less than 1 vote was lost for most contests. But a few had larger losses, which are noted below.

  • CA: up to 0.15% lost
  • FL: up to 0.3% lost
  • MD: up to 0.01% lost
  • NC: up to 0.01% lost
  • WI: up to 0.2% lost