US Congress

How gerrymandered are the maps coming out of the TXLege this session?

We report...you decide! See our table below for basic metrics and links to detailed analysis.

Metrics

MM = mean-median score (expressed as difference in vote share needed for majority)

MM (%ile) = percentile of MM score in ensemble

PB = partisan bias score (expressed as difference in seats won at 50% vote share)

PB (%ile) = percentile of PB score in ensemble. 

Favors = Does the map favor one party relative to another, compared with the ensemble? Possible values: D/R/N (neutral)/NA (not applicable)

2D (# of total) = number of ensemble plans that score worse on both MM and PB, or are equal on PB and worse on MM. We only assessed plans which favored one party over another on both PB and MM; e.g. in contrast S2100 favors Republicans slightly on MM, but is neutral on PB.

For integer-valued quantities such as partisan bias, the percentile value is sensitive to the exact definition. We used an average between the between the "weak" and "strict" definitions (i.e. "less than or equal to" and "strictly less than"), as implemented in the SciPy stats package (stats.percentileofscore(kind='mean') ).

Methodology

We generated an ensemble of 1,500,000 US Congressional plans using Markov Chain Monte Carlo algorithms as implemented in GerryChain (MGGG, 2018). All plans are legal plans with the exception of having a larger than acceptable population imbalance (up to 1%; this can cause at most a 1% deviation in the statistics of interest). Data was compiled from the Census Bureau (population data and geographic units) and the Texas Legislative Council (election returns from the 2020 Presidential and US Congress races). Proposed plans were obtained from the Texas Legislative Council. Each plan was assessed on a variety of partisan metrics; the range of observed values in the ensemble is then used to assign a percentile value to each metric as computed for each proposed plan.