Scoring North Carolina
North Carolina got new State House and State Senate district maps this week. At PlanScore, we helped a number of observers and experts watching the rapid map drawing process evaluate the maps under consideration and develop an informed opinion.
This was a very rapid redistricting process, starting with a court ruling less than three weeks ago throwing out North Carolina’s state legislative maps as an unconstitutional partisan gerrymander and ordering lawmakers to draw up new ones in two weeks. September 17 was the deadline to propose and accept new maps for both state legislative chambers.
Here’s a summary of the maps created by last week’s redistricting process, summarized by Nicholas Stephanopoulos in Election Law Blog:
For the state house, the old plan had an efficiency gap of 9%, a partisan bias of 7%, and a mean-median difference of 5% (all in a Republican direction, and based on a model using 2016/2018 data). On the other hand, the new map has an efficiency gap of 5%, a partisan bias of 3%, and a mean-median difference of 3% (again all pro-Republican). So the new map is about half as skewed as the old plan.
For the state senate, the old plan had an efficiency gap of 11%, a partisan bias of 6%, and a mean-median difference of 4% (all pro-Republican). By comparison, the new map has an efficiency gap of 3%, a partisan bias of 2%, and a mean-median difference of 3% (all pro-Republican). So the new map is approximately one-third as skewed as the old plan.
To calculate these estimates, we needed to be able to predict how the new maps would behave in future elections even though they’d never been used by real voters.
Predicting Elections
While new maps were prepared, we kicked off the process of building an updated N.C. predictive model thanks to help from friends Hannah, Sam, Hope, and James at Princeton Gerrymandering Project. The first step in building PlanScore’s model requires raw election results at the precinct geographic level, compact geographic areas comparable in population size to zip codes or census tracts. Next we use these to simulate plausible future election results, and finally apply these simulations to completely new district plans. We’ve described in detail PlanScore’s model process before, and followed this simple data collection recipe for North Carolina:
- We collected precinct-sorted data from the State Board of Elections and build a complete spreadsheet of 2016 and 2018 results for the Senate and House races we’re modeling plus the 2016 U.S. Presidential results. Fortunately data for North Carolina is easily downloadable from official sources, but this is not always true. We’ve argued in the past that open precinct-level data is vital for predicting how a plan will behave.
- We processed votes data through Eric McGhee’s ordinary least squares regression model to simulate future election outcomes, using party vote shares for the chamber we’re modeling regressed on vote shares for major, national contests like the U.S. Presidency. Although updated with new data, this is the same model we used to score Pennsylvania plans last year when their Congressional districts were invalidated by the State Supreme Court.
- Finally, we formatted model data for PlanScore so we can apply it to new and old plans for an apples-to-apples comparison of vote behavior.
Who Used This Model
Three elections experts have used our North Carolina model to review the new State Senate and State House maps:
- “As with the state house revised maps, the revised 2019 senate maps saw a decrease in the efficiency gap, partisan bias, and mean-median differences from the 2018 senate maps. Now, we await the three judges to review and make their own determination, likely with the assistance of a special master.” – Michael Bitzer, Old North State Politics
- “Here at the Princeton Gerrymandering Project, we have noticed algorithmic biases in the process used to generate the remedial map. Using the PlanScore.org engine and additional analysis, we furthermore find that the map still contains between one-half and two-thirds of the partisan advantage that was present in the illegal gerrymander.” – Sam Wang
- “Given a baseline of perfect symmetry, one would find the remedial plans better than their predecessors but still reasonably far from treating both parties equally (especially the House plan). But given a baseline of randomly generated maps, one would find the remedial plans satisfactory.” — Nicholas Stephanopoulos
