Publisher's note: This post, by Andy Jackson, was originally published in Civitas's online edition.
As the map drawing progresses in the General Assembly in the wake of the Harper v Lewis
preliminary injunction, all of the maps so far produced by members of the Joint Select Committee on Congressional Redistricting
have either an 8-5 or 7-6 likely split in favor of Republicans. A review of several attempts at creating congressional maps under neutral criteria reveals that an "8-5 map" is likely the best fit for our political geography.
Jowei Chen's "Plan 2-297"
As part of his work for the plaintiffs in Common Cause v Rucho, a case challenging North Carolina's congressional districts in federal court, University of Michigan political scientist Jowei Chen produced a map the plaintiffs submitted as an alternative to the map passed by the General Assembly. Chen's "Plan 2-297
" was a map with a 7-6 partisan split. That plan was held up by the plaintiffs as a model of what politically neutral redistricting would look like.
So, was that map the best that Chen found based on neutral criteria?
It turns out that the 7-6 split in the map Chen submitted was not the result of applying neutral standards, but because the Common Cause plaintiffs had told him to give them a 7-6 map
(from a disposition of Chen for Common Cause v Rucho):
- Q. (Attorney Phillip Strach) "All right. So the instruction to pick what I will call 7-6 maps, however set per the instructions, as opposed to 8-5 maps or 9-4 maps, i.e., maps that would elect nine Republicans, four Democrats, et cetera, did the idea to isolated maps that were 7-6 maps come strictly from counsel?"
- A. (Jowei Chen) "The instructions to follow these four criteria came strictly from League counsel. I don't know, you characterize it as an idea, and I don't know that it was an idea to me. It was simply an instruction."
So, while Chen's plan from Common Cause v Rucho demonstrates that a 7-6 map could be drawn, it does nothing to demonstrate that a 7-6 map is the best fit for North Carolina's political geography.
Other models show that an 8-5 split is a natural fit
So, without artificial constraints, what is the most probable result of districts drawn under neutral criteria. Another witness for Common Cause in Common Cause v Rucho was Duke University mathematician Jonathan Mattingly. His work found that the maps drawn by the General Assembly were statistical outliers when compared to sets of thousands of maps drawn by computers with politically neutral criteria, hardly surprising since Republicans in the General Assembly specifically stated that they had used political data for partisan advantage (and as the Democrats had done for decades before then).
So what would have happened under neutral criteria? A team of mathematicians led by Mattingly found that about 55% of their random maps
produced an 8-5 likely split in favor of Republicans based on the most recent presidential election (see chart on page 2). The next most likely outcome was a 9-4 split in favor of Republicans. An earlier election with more favorable results for Democrats produced about even odds for a 7-6 split favoring Democrats and a 7-6 split favoring Republicans, with the third most likely outcome an 8-5 split in favor of Republicans.
Of course, computer models suffer from only being able to blindly follow whatever criteria are plugged into them. Mattingly's team had the models account for the equal population of districts, avoiding splitting counties, compactness, and having two districts in which black voters could determine the outcome of the race (the latter to comply with the Voting Rights Act). However, the models did not comply with other good practices regarding redistricting, such as keeping communities together and avoiding unduly splitting municipal boundaries. The vast majority of the thousands of computer-generated maps would simply be inappropriate to use in the real world, which is why human judgment is still needed.
To put it another way, most of the maps randomly produced by computer models are junk.
So, in addition to comparing their randomly generated maps to the General Assembly maps, Mattingly's team also looked at nonpartisan maps generated by a bipartisan team of former judges
(labeled the "Judges" map). In analyzing the likely results of congressional races under that nonpartisan map, Mattingly's team found that three elections produced a likely 9-4 split, three produced a likely 8-5 split and two produced a likely 7-6 split: all in favor of Republicans (see chart on the top of page three
Those results are also consistent with the results found by the statistical website FiveThirtyEight. When they sought to produce a map "drawn to be compact while trying to respect county borders
," the result was a map with another 8-5 split favoring Republicans with one of the districts favoring Democrats being listed as "competitive."
So, if the General Assembly produces an 8-5 map, it would be appropriate for North Carolina's political geography.