Redistricting and Gerrymandering Effects

What the evidence says

Institutional design determines who draws lines, and who draws lines determines the direction of partisan bias. Across a simulated set of reforms, McCartan (2024) finds that changing map-drawing authority from a single-party legislature to an independent commission reduces a party's excess seats by about 0.5 and raises the share of competitive seats within a state from 25% to 38%. The effect scales with how completely the reform removes partisan control: reforms that restrict partisan actors in multiple ways, such as removing veto points (the source cites Michigan), produce a larger swing (an estimated 6.2 additional Democratic seats under nationwide implementation) than commission structures where parties retain some control (the source cites Ohio). All three commission styles tested increase electoral responsiveness to shifts in voter preference.

At the national level, Kenny (2023) quantifies what enacted 2020-cycle maps did relative to a nonpartisan simulated baseline: a net Republican advantage of about 2.3 seats (8.6 Republican-favoring seats against 6.2 Democratic-favoring seats), a drop in competitive districts from 50 to 34 (a 32% reduction, using a 47.5–52.5% baseline vote-share definition of competitive), a drop in responsiveness from 9.2 to 7.8 seats gained per additional percentage point of national vote, and a rise in the vote share Democrats need to win a House majority from 50.9% to 51.1%. State-level case studies point the same direction. In Wisconsin, Herschlag (2017) finds the enacted plan more Republican-biased than over 99% of an ensemble of alternative compliant plans, and shows Republican seat counts holding constant as the Republican vote share fell from 51% to 48%. In North Carolina, Best (2017) finds a 5.7 percentage point swing toward Democrats across two statewide races produced only a 1-seat gain for Democrats, and that about two-thirds of state senate districts favored Republicans regardless of whether the statewide Democratic vote was 44.2% or 49.9%.

Mechanistically, packing is the lever these studies point to: concentrating opposition voters into a small number of districts while distributing the remainder as minorities across many others (Best, 2017). Simulation work by Gatesman (2018) quantifies this: the Friedman-Holden packing algorithm increases the districts won by the advantaged party from 2.42–2.56 to 3 out of 5 (a 9–12% relative increase) but, without spatial constraints, produces highly disconnected districts (3.7–15.5 connected components on average); adding spatial restrictions cuts that to 1.5–2.4 components. Separately, Deshpande (2023) models "votemandering," where gerrymandering is paired with targeted campaign spending: the seat bonus from this combination grows with campaign budget (with diminishing returns) and shrinks as voter turnout rises or compactness constraints tighten, with little measured effect from voter spatial clustering.

Downstream health and social outcomes tied to redistricting are measured but thin. Schnake-Mahl (2025) reaggregates the same mortality data to pre- and post-2010-redistricting congressional boundaries and finds within-district disparities in infant mortality widen (median MLD from 0.059 to 0.075) while within-district disparities in deaths of despair narrow slightly (median MLD from 0.19 to 0.18); spatial autocorrelation of deaths of despair falls from Moran's I 0.20 (p = 0.04) to 0.08 (p = 0.17), losing its statistical support entirely. Rushovich (2024) reports an association between a state's gerrymandering score and between-district variation in uninsurance rates (coefficient 0.25, 95% CI −0.04 to 0.53, p = 0.10) — a confidence interval that crosses zero. No claim here quantifies a causal pathway from district lines to health outcomes; the associations are measured, the mechanism is not.

Judicial involvement in redistricting is noted but not quantified: Levitt (2008) cites unspecified prior studies finding that judges supervising line-drawing tend to adopt plans favoring their own party, without reporting a magnitude. And a portion of the evidence base is about measurement itself rather than gerrymandering's effects: Becker (2020) finds that different flood-fill algorithms used to generate ensembles of redistricting plans for comparison do not sample uniformly — a bounding-box variant produces more compact plans (fewer cut edges) than a standard flood fill, and drawing all districts simultaneously produces more symmetric shapes than drawing them one at a time. This bears on how confidently any of the ensemble-based outlier findings above (Wisconsin's 99th-percentile ranking) can be read, since the ensemble-generation method itself shapes what counts as typical.

Where the evidence splits

No contradictions in the current sources.

Thresholds and dose-response

In Wisconsin, Herschlag (2017) reports the enacted plan yields results "highly skewed to the Republicans" specifically when the statewide Democratic vote share exceeds 50–52% — below that range the plan behaves like a typical plan in the comparison ensemble, but above it the plan becomes an outlier, described as acting like a "firewall" that holds the Republican majority in place. The same source reports Republican seat counts remaining fixed as the Republican vote share fell from 51% to 48%, a 3-point swing with no seat response. In North Carolina, Best (2017) reports a 5.7 percentage point swing toward Democrats (44.2% to 49.9%) converting into a gain of only 1 seat.

Open empirical questions

• Whether the association between state-level gerrymandering and between-district uninsurance variation holds up outside a cross-sectional design with a confidence interval that crosses zero — no causal pathway is established.
• Whether the mortality and deaths-of-despair disparity shifts tied to district reaggregation reflect a health effect of redistricting itself, or simply an artifact of how populations are regrouped — the source does not test a causal mechanism.
• Whether the magnitude of judicial partisan bias in line-drawing, cited but not quantified by Levitt (2008), holds up under direct empirical measurement.
• Whether votemandering's seat bonus, demonstrated in simulation and one Wisconsin case study, generalizes to other states or election cycles.