-
Notifications
You must be signed in to change notification settings - Fork 3
Readings: State Case Studies
Clelland, J., et al. (2021). Colorado in Context: Congressional Redistricting and Competing Fairness Criteria in Colorado. Journal of Computational Social Science.
Summary:
This article analyzes the 2011/2012 Congressional district plans in Colorado using ensemble methods to generate a large sample of plans, which can serve as a baseline from which to evaluate the relationship between partisan outcomes, the number of counties which are split, and number of competitive districts in a plan.
Using GerryChain to apply ensemble methods, the authors randomly generated hundreds of thousands of viable Colorado districting plans and then use 2018 election data to determine the likelihood of different election outcomes with these unbiased maps.
Legislative History:
In 2018, Colorado’s State Legislature adopted a new framework for redistricting and put control of the process in the hands of an independent commission. This new framework requires that maps must: *have contiguous districts *keep district populations as equal as possible *preserve communities of interest and political subdivisions *minimize the number of divisions when a city, county or town is divided *have districts as compact as reasonably possible *maximize the number of politically competitive districts
And the maps must not: *endeavor to protect incumbent Legislators or declared candidates for the Legislature *be designed to benefit any political party *be designed to or result in denying any citizen the right to vote based on race or language, or dilute any minority group’s electoral influence
This new legal framework does not define compactness, and competitiveness is defined as “having a reasonable potential for the party affiliation of the district’s representative to change at least once between federal decennial censuses.”
Landmark Court Cases:
Wesbury v. Sanders (1964) – upheld the precedent established in Baker v. Carr (1962) that federal courts have jurisdiction in cases involving state-drawn legislative plans, particularly if they are alleged to violate the “one person, one vote” principle
Reynolds v. Sims (1964) – extended the federal judicial jurisdiction and “one person, one vote” principle to state legislative districts
Karcher v. Daggett (1983) – Established that it challengers to legislative plans can show that even small deviations in population between districts could have been reduced, then the state must prove the population deviations were necessary to satisfy some other legitimate goal (e.g. preserving political boundaries or compactness)
Evenwel v. Abbott (2016) – established that total population (as opposed to voting-age population or citizen residents) is a valid metric for complying with one person, one vote, but did not exclude other metrics.
Rucho v. Common Cause (2019) – effectively limited the Supreme Court’s jurisdiction over partisan gerrymandering cases by establishing that “Federal judges have no license to reallocate political power between the two major political parties, with no plausible grant of authority in the Constitution, and no legal standards to limit and direct their decisions.” However, allows that even though partisan gerrymandering does not violate the Equal Protection Clause, US states and Congress can pass laws to restrict partisan gerrymandering.
The Legislative Council for the State of Colorado has determined that the order of precedence most consistent with the new redistricting amendments is as follows:
- Population equality as required by the federal constitution
- Compliance with the federal Voting Rights Act
- Preserving whole communities of interest and political subdivisions
- Compactness
- Maximizing the number of politically competitive districts
Political Landscape of Colorado:
Widely considered a “purple” state since gubernatorial and presidential elections going back to 1980 have often been fairly evenly split among Republican and Democratic vote shares. The rural areas of the state (Western Slope and Eastern Plains) tend to be more conservative, while urban areas broadly trend liberal, there is notable variation to this trend. One of the larger cities in the state, Colorado Springs, is heavily conservative; while, Costilla county is consistently one of the most Democratic-voting counties in the state despite being in a rural area of Southern Colorado.
Quotes:
“The fundamental goal of ensemble analysis is to model the political geography of a region (in this case, the state of Colorado) in order to better understand what might be expected for a “typical” districting plan for the state.” (p. 8).
“Plans may be evaluated with regard to a variety of measures: partisan balance of election results, geographic compactness of districts, competitiveness of district elections, preservation of communities of interest, racial/ethnic population within districts, etc. The key idea is to create a large number of randomly generated, valid plans that satisfy all relevant legal constraints—an “ensemble” of plans—and to use real voting data to compute the measures of interest for each plan in the ensemble. The result is a statistical range of possible outcomes for each measure, to which any proposed plan may be compared. If a proposed plan appears to be an extreme outlier compared to the ensemble, this may suggest that the plan was deliberately designed to achieve some specific goal, such as partisan gerrymandering.” (p. 8-9)
Locating the Representational Baseline: Republicans in Massachusetts
Moon Duchin, Taissa Gladkova, Eugene Henninger-Voss, Ben Klingensmith, Heather Newman, & Hannah Wheelen
MA Republicans have not won a seat in the US house of Representatives (n = 9-10 seats) since 1994, even though GOP candidates receive 30-40% two-way vote shares in statewide elections.
Authors purport this is NOT a product of gerrymandering, but instead a structural feature of voting patterns.
- "uniformity can block desired representational outcomes for a group in the numerical minority"
numerical uniformity
- vote shares across building-block units are extrememly consistent
issue: it is numerically impossible to build R-favoring collection of towns/precincts with enough population to be a Congressional district (even without requiring contiguity, compactness, other spatial constraints) -> low variance in GOP town/precinct voting results
Naive Bounds on Gerrymandering:
If Party X has a vote share of 0 <= V <= 1, and has just over 50% of votes in districts it wins, and 0% in losing districts (most efficient), then seat share will be 2V
If Party X is inefficient and V < 0.5, can lose all districts. If V > 0.5, it's opponent will have 1-V share, and if opponent is efficient, O will have 2(1-V) = 2 - 2V seat share, leaving 1 - (2-2V) = 2V -1 minimum seat share for party X
bounds: 2V -1 to 2V if V > 0.5
Low variance in voting patterns prevent minority party from capturing seats
Even when numerically feasible, not spatially possible to create districts favoring Republicans (contiguous) because GOP favored areas are scattered
The clustering patterns in MA are uniform; only when high-variance in voting patterns are present, are there spatial clusters
- study of votes, not voters (data that is available)
-
efficiency gap
- EG - measure of parity of wasted votesEG = 2V - S - 1/2
, whereV
is vote share andS
is seat share - rule of thumb - 8% EG could flag gerrymandering, but a score below 11% was impossible to achieve regardless of how redistricting plans are drawn in MA
Takeaway:
It is only legitimate to compare partisan outcomes against the state space of actual possibilities to determine presence of gerrymandering. Range of possible representational outcomes under valid redistricting is not always in keeping with the range that classic modleing appraoches predict from vote share alone
Assessing Significance in a Markov Chain without mixing (2017) Maria Chikina, Alan Frieze, and Wesley Pegden
Goal: statistical test to detect that a state in reversible Markov chain is an outlier (not originating from) a stationary distribution without any additional info on mixing rates [alternative to bootstrapping, which might not be possible in state space]
Claim: observing presented state is an outlier significant at
Assumption: Markov chain is reversible
Key Terms:
- reversible - distribution of reversible sequence of RVs are equal
- reducible - for a pair of states is inaccessible from state 0 via legal transitions
DeFord, D., & Duchin, M. (2019). Redistricting reform in Virginia: Districting criteria in context. Virginia Policy Review, 12(2), 120-146.
Context: SJ306 (2019) is a constitutional amendment that establishes a bipartisan redistricting commission. Virginia is site of much reform activity from civil rights advocacy groups.
- Challenges include: SJ306 language is weak on how can participate in the commission; electoral data quality (precinct shapefiles are not maintained, local officials can only change precincts)
Districting Criteria: "Preserve opportunities to elect candidates of choice for substantial racial or ethnic minority communities... Form districts to be compact and contiguous" (p. 5)
Methods:
Recombination walk: (a.) Begin with a plan; (b.) Consider possible new plan by merging two adjacent district; (c.) Re-splitting them with spanning tree; (d.) Plan will be rejected if doesn't align to criteria/rules, plan will be accepted with some probability based on priorities set by random walk.
Four types of random walk:
- Neutral ensemble: Congressional district population is allowed to vary 2% above and below ideal populations and Senate districts 5% above and below ideal populations.
- Split ensemble: Require steps do not increase locality splits until threshold is passed, then impose exponential weighting prioritizing fewer splits.
- Population ensemble: Congressional district population has tighter (0.5%) or looser (5%) above and below ideal populations, and Senate districts population has tighter (2%) or looser (10%) above and below above and below ideal populations.
- BVAP ensemble: Order districts by Black Voting Age Population (BVAP) as share of total population. BVAP% is number of Census-identified Black residents at least 18 years of age, divided by the total number of residents at least 18 years old.
50,000 ReCom steps, discard the first 2000, ensembles include 48,000 plans
Findings:
- Population Deviation: Little measurable effects of tighter or looser limits
- VRA: Statistical behavior of the enacted plans comports well with the ensemble designed to hit BVAP targets, with no cost to partisan fairness
- Preserving Localities: High priority to preserving localities impact partisan scope, but eliminates most extreme partisan outcomes for both Republicans and Democrats.