Competing Criteria: Rethinking Congressional Redistricting and Representation

The Process of Congressional Redistricting

Congressional redistricting requires every state with multiple congressional districts to redraw its legislative boundaries at least once per decade. Theoretically, there are an infinite number of ways to create single-member congressional district boundaries within any state. Therefore, the people and institutions who redistrict turn to specific criteria to provide guidance about where the new district lines will go.

Across the United States, different institutions have the responsibility of redistricting. Historically, state legislatures have overseen redistricting for congressional districts in the vast majority of states. In recent years, independent redistricting commissions have become more common, sometimes passed by popular ballot measures. ¹ As of 2022, eight states required independent commissions for congressional redistricting (AZ, CA, CO, ID, MI, MT, NY, and WA) and three used politician commissions (HI, NJ, and VA). In the 2010–2012 redistricting cycle, only four states used independent commissions. In the other 33 states with multiple congressional districts, state legislatures are responsible for redistricting, although some have advisory (IA, ME, and UT) or backup commissions (CT, IN, and OH). ²

Although states designate redistricting institutions to redraw their congressional districts, these institutions are not always who ultimately make the final congressional map. In every redistricting cycle since the 1960s, state or federal courts have become redistricting institutions, drawing remedial plans for legislative districts that are ultimately used in elections. For example, a large majority of states had their congressional plans challenged in either state or federal court (or both) in 2000 (37 states) and 2010 (42 states ³ ). In some instances, these challenges were rejected. In others, the initial redistricting institutions were required to make new plans, which were then approved by the courts. In some, however, the courts drew or imposed redistricting plans themselves, becoming the redistricting institution for that state because its map was used in the election. ⁴

When an institution begins the process of redistricting, it uses redistricting criteria to determine where to draw the districts lines. Redistricting criteria can be any guiding principles that focus on specific goals and allow for the systematic drawing of district boundaries. Common redistricting criteria include population equality among the districts, partisan advantage (or fairness) in the statewide plan, competitiveness of elections, protection of incumbents, promotion (or dilution) of minority voting power, compactness of districts, preservation of political subdivisions (towns, counties, or cities), contiguity, preservation of previous districts, protection of communities of interest, and other geographic considerations. Some of these criteria are based on tradition (such as contiguity of districts) or political incentives (such as partisan advantage) while others are based on federal law (such as population equality among districts) or state law (such as protection of political subdivisions). ⁵

As Bruce Cain observed in 1984’s Reapportionment Puzzle, the choice of criteria for a new redistricting plan is also essentially a zero-sum process. Criteria necessarily conflict with other potential criteria, and therefore choosing certain criteria precludes the option of using other criteria (Butler and Cain 1992; Cain 1984). For example, if a redistricting institution prioritizes compactness of districts as the predominant criterion, it will necessarily subordinate other potential criteria, such as the goal of maintaining county boundaries. Some criteria directly conflict with one another, such as partisan advantage and electoral competitiveness. Others may conflict more indirectly, such as population equality and partisan advantage. Overall, following any one criterion makes it more difficult to follow another, and increasingly difficult to follow a third, fourth, or fifth.

Although redistricters may be able to favor multiple criteria (such as legally required population equality and minority representation, plus partisan advantage), they cannot favor all criteria with equal weight. For example, redistricting to maximize partisan advantage would be much easier with unequal district populations. Redistricting institutions are forced to make decisions that elevate certain goals while demoting others. There will be winning and losing criteria. There is no “nonpartisan, noncontroversial reapportionment process” ⁶ (Butler and Cain 1992; Cain 1984).

Redistricting and Representation Scholarship

Conceptualizing redistricting as choices among competing criteria would not matter significantly for representation if all redistricting institutions favored the same criteria. However, as many researchers have shown, institutions make different decisions from one another on specific redistricting criteria.

The clearest finding is that single-party control of legislative redistricting leads to less competitive districts, with a map that favors their own party (Campagna and Grofman 1990; Erikson 1972; Gelman and King 1994b; McDonald 2004; Stephanopoulos 2017). All other institutional arrangements—commissions, courts, and bipartisan legislatures—are less partisan and more competitive than partisan-controlled legislatures on average in nearly every category (Carson and Crespin 2004; Carson, Crespin, and Williamson 2014; Cottrill and Peretti 2013; Lindgren and Southwell 2013; Nelson 2023). Additionally, independent commissions are likely to draw more compact districts, with more retained cores of previous districts and fewer split political subdivisions (Edwards et al. 2017; Grainger 2010). Maps made by divided or bipartisan control may lead to incumbent protecting gerrymanders or court fights (McDonald 2004) as well as lower partisan favoritism (Stephanopoulos 2017).

Further, research has found that the process of redistricting impacts constituent representation. Much of this research focuses on specific forms of representation or political behavior. For example, redistricting can have significant impacts on voter knowledge and participation, especially when constituents are moved to a new district with a different incumbent (Hayes and McKee 2012; McKee 2008). Redrawing legislative districts can affect legislators’ agendas, roll call votes, policy responsiveness, and the relationship with constituents (Bowen 2014; Crespin 2010; Gelman and King 1994a; Hayes, Hibbing, and Sulkin 2010; Yoshinaka and Murphy 2011). The new composition of a given congressional district can have a substantial impact on how the congressional representative deals with constituents, campaigns, and behaves in congress (Adler and Lapinski 1997; Bertelli and Carson 2011; Bullock 1995; Gamble 2007).

Research demonstrates that many of the representational effects of redistricting have an outsized impact on specific constituent groups. Researchers have shown that redistricting and district composition can have a significant impact on voting power and representation for Black voters and other historically underrepresented minority groups in the U.S. Congress (Canon 2022; Cameron, Epstein, and O’Halloran 1996; Lublin 1999; Sharpe and Garand 2001; See Footnote 19). Traditional criteria, specifically compactness, can have substantial effects on minority voting power (Barabas and Jerit 2004). The foreign-born population has also been historically underrepresented in the federal legislature (Gaynor and Gimpel 2021). Because partisan control over the redistricting process typically leads to plans biased toward their co-partisans, it also has a significant impact on partisan constituents (Abramowitz 1983; Cain 1985; Campagna and Grofman 1990; McGhee 2020; Swain, Borrelli, and Reed 1998). Traditional criteria can limit the abuse of political or partisan criteria such as incumbency protection (Forgette and Platt 2005), lower the costs of elections (Bullock 2021), and impact responsiveness and electoral prospects for legislators (Bowen 2014; Niemi, Powell and Bicknell 1986; Winburn and Wagner 2010).

While extensive, the existing redistricting scholarship falls short in some important areas. Much of the redistricting scholarship is narrow in its focus. Because of the interdependent decision making that defines the redistricting process, as explained by Cain, this research yields only limited representational conclusions. For example, many studies look at only a select number of redistricting criteria, such as compactness (Edwards et al. 2017) or the Efficiency Gap (Stephanopoulos 2017). While this delivers an in-depth analysis and helpful insights about the criterion of focus, and which institutions may correlate strongly with that criterion, it does not deliver answers as to what criteria are used by the redistricting institutions that do not have a strong relationship with the criterion of focus. As Cain’s account points out, the totality of choices about criteria is what is critical in the redistricting process, not a single choice on one criterion.

Relatedly, many studies also only focus on certain redistricting institutions or group together institutions that have substantial differences in their function. For example, the combination of federal and state courts together as one “courts” category happens frequently (e.g., Carson, Crespin, and Williamson (2014)). However, the reality of state and federal courts is that they face completely different incentives and constraints with redistricting cases, including the cases on their dockets and the law that is applicable. A federal court, for example, is much more likely to be confronted with a racial gerrymandering case arising out of the Voting Rights Act, whereas a state court may view a challenge to a plan based on a state constitutional provision. Further, researchers may group “commissions” as one category without considering the differences between independent, partisan, bipartisan, or back-up commissions, which may strongly influence the process.

Additionally, Cain’s description of the redistricting process in The Reapportionment Puzzle is personal and qualitative. It is based on his own observations as a technical expert for the California redistricting process. It does not provide an empirical analysis of these competing criteria, nor does it present a research design like in this paper. Other scholars have subsequently shown the existence of some of these tradeoffs empirically. For example, with compactness and fairness (Chen and Rodden 2013), minority representation and partisan fairness (Shotts 2001) or with competitiveness and minority representation (McDonald 2006). However, this paper is the first attempt to adapt Cain’s process-based framework into a more comprehensive quantitative study.

Despite these gaps, this trove of strong political science research on redistricting and representation presents a variety of measurements, research designs, and findings that are influential on this paper. ⁷

A Theory of Redistricting and Representation

Connecting these two sets of ideas, a few conclusions are likely: (1) Redistricting can be conceptualized as the zero-sum process of elite choices among competing redistricting criteria representing political goals to guide line-drawing from infinite possibilities; (2) redistricting institutions favor different redistricting criteria from one another when they make maps; and (3) the redistricting process impacts representation.

This paper builds on this existing scholarship to argue that a research design that follows Cain’s description of the redistricting process, using the full array of institutions and criteria, allows for a better understanding of the representational effects of redistricting. This framework also has important implications for group representation. The elite choices among conflicting criteria represent the substantive goals redistricting institutions value when they draw a line and group individuals in districts. Therefore, because institutions use different criteria, and because these criteria matter for representation, understanding which criteria are correlated or not correlated with each institution also explains which representational qualities may be experienced by each constituency.

Why should each redistricting criterion be conceptualized as representing constituents based on specific group identities? Redistricting criteria requires the categorization of people in a geographic space by specific group qualities or characteristics. The chosen criterion defines the preeminence of some group characteristics as opposed to some other potential group characteristics in the new redistricting map. For example, when a redistricting institution promotes the criterion of partisan advantage over compactness, it is attaching the representational value of partisanship as the preeminent identity of the constituents in the map at the expense of any other group identities, such as residents of town or county. These elite decisions about criteria assert a form of group representation in the new redistricting plan. The redistricting criteria determine whether constituents are grouped and divided by party, by race, by county, by geographic closeness, by previous voting history, or other group identity characteristics.

Put simply, redistricting requires institutions to use specific criteria to draw districts, these criteria necessarily conflict with one another, and they represent specific political goals that require grouping constituents based on distinct group identities, elevating the representation of certain group identities, while subordinating others. Therefore, understanding which criteria are correlated or not correlated with which institutions has implications for group representation in congress.

Partisan Criteria

The redistricting process presents an opportunity to enhance party power in the congressional delegation. One way to do this is to use partisan criteria in the map-making process to enhance the number of seats won by one party at the expense of the other party. The exact means of partisan gerrymandering may vary, ¹³ but all techniques are focused on creating a partisan advantage or bias toward one party. To measure partisan criteria, I use three common partisan gerrymandering metrics as dependent variables—the Efficiency Gap, the Partisan Bias Test, and Mean-Median score—as well as Democratic seat share. Each of these variables uses the partisan outcomes in a plan to approximate the criteria that may have been used by the institution. Each of the gerrymandering metrics measures aspects of the relationship for how votes are translated into seats and whether one party is benefiting more than the other. These data were collected from official House of Representatives election returns ¹⁴ and the PlanScore website. ¹⁵

The descriptive data on Democratic seat share (Figure 1) show institutional variation in the average percentage of Democratic seats won in the plans created over six redistricting cycles. Unsurprisingly, partisan legislatures have the most dramatic differences, with Democratic legislature-made plans featuring an average seat share of about 67 percent and Republican legislature-made plans with 28 percent Democratic seats on average. All other institutions fall in between, with state courts and political commissions both featuring average seat shares above 55 percent for Democrats and only divided legislatures below 50 percent.OPEN IN VIEWER

The partisan gerrymandering variables tell a similar story. The first metric, the Efficiency Gap, measures how many votes are cast above the threshold necessary to win a district in a statewide plan (Stephanopoulos and McGhee 2015). It attempts to quantify the effects of cracking and packing on the electoral outcomes of a statewide plan. Second, the Partisan Bias test measures the difference between a party’s seat share in a redistricting plan if the two parties had a tied vote share of 50 percent in a theoretical election. For example, if the hypothetical 50–50 election led to Democrats receiving 53 percent of the legislative seats, the plan would have a 3 percent Democratic bias (Grofman and King 2007). Third, the Mean-Median Difference test subtracts a party’s median vote share from its mean vote share. If the difference is low, then the redistricting plan has a more normal distribution of districts (McDonald and Best 2015). Each of these three metrics helps measure an institution’s use of partisan advantage as a criterion. In this analysis, all metrics are coded positive for Democratic advantage and negative for Republican advantage.

The descriptive data for these three variables highlights the means for each redistricting institution (Figure 2). The data show clearly that historically plans made by the partisan legislatures had the most dramatic biases toward co-partisans. Plans made by Democratic legislatures provided partisan advantage for Democrats in all three metrics, while plans made by Republican legislatures measure as biased toward Republicans. The results for the other institutions are more muddled or muted. Political commissions show bias toward Democrats in two of the three metrics. For other institutions, biases exist, but they are not as strong or clear across all three metrics. Only state and federal court-made plans lean consistently toward one party or the other, but without large averages.OPEN IN VIEWER

These raw means do not separate the institutional variance from other aspects of the plans such as the political culture of the state or the political climate in which the plans were made. To isolate the relationship between institutional variation and the criteria in this large data set, I use a series of multivariate regressions (Table 1, Figure 3) as described in the research design.

OPEN IN VIEWER

Table 1.

Partisan Criteria Models.

	Partisan Criteria
	Model 1:	Model 2:	Model 3:	Model 4:
	Democratic Seats	Efficiency Gap	Partisan Bias Test	Mean-Median Test
Republican legislatures (n = 67)	−0.663* (0.377)	−0.019 (0.023)	−0.058** (0.026)	−0.035*** (0.009)
Democratic legislatures (n = 84)	0.943*** (0.294)	0.050** (0.023)	0.059** (0.024)	0.022*** (0.008)
Political commissions (n = 29)	0.141 (0.329)	−0.012 (0.017)	0.033 (0.022)	0.013 (0.011)
Independent commissions (n = 8)	2.605 (1.667)	0.044 (0.035)	0.023 (0.030)	0.009 (0.007)
Federal courts (n = 28)	0.591 (0.409)	0.046* (0.026)	0.016 (0.022)	0.002 (0.008)
State courts (n = 19)	0.252 (0.435)	0.037 (0.032)	0.043 (0.314)	0.018* (0.011)
Divided legislatures (n = 24)	ref	ref	ref	ref
Number of plan districts	0.216 (0.161)	−0.002 (0.002)	0.001 (0.002)	0.000 (0.001)
Change in plan seats	−0.619* (0.322)	−0.003 (0.006)	−0.005 (0.008)	−0.005 (0.003)
VRA section 5 coverage	0.226 (0.601)	0.030 (0.035)	0.001 (0.038)	0.012 (0.021)
Democratic vote share	7.577*** (1.423)	0.219*** (0.081)	0.245** (0.114)	0.077* (0.045)
1972	ref	ref	ref	ref
1982	−0.021 (0.288)	−0.017 (0.014)	0.008 (0.017)	0.009 (0.009)
1992	0.053 (0.344)	−0.020 (0.020)	−0.033 (0.021)	−0.025** (0.010)
2002	−0.424 (0.321)	−0.017 (0.016)	−0.008 (0.026)	−0.008 (0.015)
2012	−0.209 (0.371)	−0.049* (0.025)	−0.080** (0.030)	−0.038*** (0.011)
2022	−0.059 (0.368)	−0.022 (0.032)	−0.054 (0.026)	−0.033** (0.012)
Constant	−0.740 (1.845)	−0.087 (0.065)	−0.151* (0.078)	−0.028 (0.036)
R2	.7972	0.3601	0.5774	0.5663
n	256	145	126	126

OPEN IN VIEWER

Figure 3.

Partisan gerrymandering metric estimates—Models 2, 3, and 4.

All four partisan criteria models yield similar results. ¹⁶ There are strong, statistically significant relationships between partisan-controlled legislatures as map makers and plans that advantage their co-partisans. For example, when a Democratic legislature creates a redistricting map, it is estimated to have an Efficiency Gap favoring Democrats 5 percent more than plans made by divided legislatures, the reference category in this model.

The other redistricting institutions’ estimates on these metrics are inconsistent, close to zero, and lack statistical significance. The only other statistically significant relationship—Federal Courts in Model 2—is less notable with the lack of measured bias through the other metrics. This is also important. In the context of this paper, and a research design highlighting the competitiveness of criteria, the lack of correlations is important. These models also highlight the predictive quality of vote share on seat share as well as the Republican bias in the redistricting cycle in 2012.

Viewing these partisan gerrymandering models together (Figure 3), it is likely that much of the partisan bias observed in the descriptive data can be attributable to the institutional variation for the two partisan legislatures whereas for the other institutions, it may be attributable to other factors, such as state political culture.

Political Criteria

Partisan advantage and bias are not the only electoral goals that may be achieved during redistricting. To account for nonpartisan political criteria that institutions may pursue during congressional redistricting, I focus on competitiveness and incumbency protection.

Competitiveness has been one of the most discussed criteria in the elusive search for a neutral redistricting criterion (Forgette, Garner and Winkle 2009). Advocates for competitiveness argue the criterion could give voters more choices, promote better representation, and prevent the ills of redistricting, such as partisan gerrymandering and incumbency protection (Hirsch 2003). Others argue that competitive elections hurt representation by harming critical connections between constituents and representatives while maximizing the unhappy voters in a district (Brunell 2010; Buchler 2005).

I also measure incumbent winners for a criterion that is explicitly political but does not necessarily provide partisan advantage or competitive elections. This section includes four dependent variables to measure the two criteria: Competitive seats won, incumbent seats won, competitive seat share per plan, and incumbent winner percentage per plan.

The descriptive data on these two political criteria metrics (Figure 4) highlight the high rates of incumbent reelection and low percentages of competitive seats across institutions in congressional plans over the past six decades. Divided legislatures have the highest levels of incumbent winners when they draw redistricting plans (92 percent), in line with previous research (McDonald 2004). They also have the highest levels of competitive elections. Political commissions have the second highest level of incumbency winners per plan on average (84 percent).

OLS regression models for the dependent variables of competitive or incumbent seats won (Table 2—Models 5 and 6) did not yield any evidence of a strong relationship between the redistricting institutions and criteria. In the seventh regression model, the dependent variable was competitive seat share—the percentage of seats in a plan where the winner receives less than 60 percent of the vote ¹⁷ (Carson, Crespin and Williamson 2014). Competitive districts in Congress can be rare—there were only 46 competitive districts out of 435 in 2022. This model shows a negative relationship between Democratic legislature-made plans and competitive districts per plan, predicting a nearly 50 percent decrease when Democrats make a plan compared to divided legislatures. The model does not highlight strong connections to other institutions; however, it does show interesting effects for the reapportionment control variable as well as three of the redistricting cycles.

OPEN IN VIEWER

Table 2.

Political Criteria Models.

	Political Criteria
	Model 5:	Model 6:	Model 7:	Model 8:
	Competitive Seats (OLS)	Incumbent Seats Won (OLS)	Competitive Seat Share (GLM)	Incumbent Seat Share (GLM)
Republican legislatures (n = 67)	−0.201 (0.598)	−0.590 (0.681)	−0.375 (0.345)	−0.923** (0.424)
Democratic legislatures (n = 84)	−0.247 (0.629)	0.333 (0.987)	−0.564* (0.315)	−0.830* (0.442)
Political commissions (n = 29)	−0.558 (0.599)	−1.259 (1.168)	−0.436 (0.341)	−0.648 (0.468)
Independent commissions (n = 8)	1.933 (1.940)	−3.584 (2.601)	−0.014 (0.364)	−1.359*** (0.457)
Federal courts (n = 28)	0.475 (0.824)	−1.851 (1.431)	−0.489 (0.337)	−0.806* (0.439)
State courts (n = 19)	−0.327 (0.503)	−0.039 (1.245)	0.279 (0.343)	−1.187*** (0.461)
Divided legislatures (n = 24)	ref	ref	ref	ref
Number of plan districts	0.250*** (0.067)	0.462 (0.376)	−0.021*** (0.006)	0.005 (0.007)
Change in plan seats	0.675 (0.425)	−1.834* (0.934)	0.089** (0.036)	−0.147** (0.007)
VRA section 5 coverage	−0.266 (0.742)	1.003 (1.436)	−0.315 (0.195)	−0.167 (0.154)
Democratic vote share	−0.579 (1.191)	−3.504 (2.931)	0.0250 (0.972)	−0.549 (0.691)
1972	ref	ref	ref	ref
1982	0.663 (0.462)	0.389 (0.483)	0.511** (0.246)	−0.109 (0.261)
1992	1.481*** (0.458)	4.37*** (0.992)	0.499** (0.231)	−0.278 (0.218)
2002	−0.692** (0.342)	7.859*** (1.358)	−0.261 0.239	0.299 (0.274)
2012	1.018** (0.397)	0.066 (0.454)	0.355 (0.270)	0.007 (0.320)
2022	−1.850*** (0.604)	7.35*** (1.566)	−1.33*** (0.383)	0.032 (0.299)
Constant	0.753 (1.257)	−1.850 (4.647)	−0.246 (0.649)	2.587*** (0.611)
R2	0.6287	0.5325	-	-
n	256	256	256	256

The most notable finding of this political criteria analysis is in Model 8, where the dependent variable was the seat share of incumbent winners in a plan in the election directly following the implementation of a new plan—1992, 2002, 2012, etc. The model highlights statistically significant estimates for several institutions, all of which are negative compared to the reference category of divided legislatures. These findings highlight the impact divided legislatures have toward incumbency protection in their plans, supporting previous findings (McDonald 2004). There is no evidence that partisan legislatures and courts prioritize this as a criterion. State courts in particular correlate with a strong, negative relationship to incumbent success when they draw new districts. Political commissions do not have a notable negative relationship to incumbency protection. This datapoint together with the descriptive means points to a relationship between the political commissions and incumbent winners as an outcome.

Minority Voting Criteria

To measure minority voting power as redistricting criteria, this project focuses on majority–minority districts in three dependent variables. One variable measures the total number of majority-minority districts for Black and Hispanic constituents in a plan. The second looks at majority–minority districts as a percentage of the total districts in a plan. And the third variable is an original measurement: the Majority-Minority Proportionality Test. ¹⁸ This novel measurement highlights the proportionality of the plan's electoral representation to the statewide population. All three variables use population data from the U.S. Census and the American Community Survey to quantify a form of racial and ethnic minority voting power.

These dependent variables as measurements for minority representation and political power are far from perfect. They neglect many racial and ethnic minority groups in the U.S., most notably Native American and Asian populations. They are simple measurements and do not account for contemporary conceptions of influence or opportunity districts, coalition districts, districts based on voting age population only, or communities of interest. These metrics, and therefore these models, lack the detail and nuance necessary for a full assessment of a topic as important to American representation as minority voting power. However, for the purposes of this analysis, these metrics approximate a source of comparison among the redistricting institutions about their emphasis on descriptive racial and ethnic representation. ¹⁹

The descriptive means (Figure 5) for two of the minority voting criteria variables—percentage of majority–minority districts per plan and the Majority-Minority Proportionality Test—show clear institutional variation. Independent commissions and the federal courts create the highest percentage of majority–minority districts per plan on average—11 percent and 16 percent, respectively. Additionally, they each produce the highest ratio of majority–minority districts to population, where “1” would represent perfect proportionality. While the independent commission means are high, the number of plans made by independent commissions is quite small—only 8, and many of these in the last two redistricting cycles. Federal courts have a longer history and are more likely to make a map when a case comes to the courts for violating the Voting Rights Act dealing directly with racial gerrymandering and minority voting rights. Based only on the descriptive data, the state courts, political commissions, and divided legislatures have created the least minority representation on average, while both partisan legislatures score about the same on these two metrics.OPEN IN VIEWER

OPEN IN VIEWER

Figure 5.

Average minority voting criteria per plan.

To better assess the relationship between these institutions and minority voting criteria, I use three regression models—Models 9, 10, and 11 (Table 3). Model 9 uses the total number of majority–minority districts in a plan as the dependent variable in an OLS regression. Its most prominent finding and effect size is unsurprisingly that a state’s Black and Hispanic population is a strong indicator for how many Black and Hispanic majority–minority districts a plan will have. Plan size and reapportionment variables also had significant relationships.

OPEN IN VIEWER

Table 3.

Minority Voting Critreria Models

	Minority Voting Criteria
	Model 9: Maj-Min Districts (OLS)	Model 10: Maj-Min % (GLM)	Model 11: Maj-Min Dist-Pop Proportion (OLS)
Republican legislatures (n = 67)	0.388* (0.217)	0.912** (0.379)	−0.004 (0.026)
Democratic legislatures (n = 84)	0.139 (0.201)	0.826* (0.498)	0.007 (0.024)
Political commissions (n = 29)	0.307 (0.197)	0.371 (0.557)	−0.029 (0.024)
Independent commissions (n = 8)	1.504 (1.151)	0.900** (0.433)	0.176** (0.067)
Federal courts (n = 28)	0.381 (0.277)	0.875* (0.512)	0.101* (0.586)
State courts (n = 19)	−0.075 (0.199)	−0.585 (0.702)	−0.002 (0.028)
Divided legislatures (n = 24)	ref	ref	ref
Number of plan districts	0.313*** (0.068)	0.004 (0.009)	0.015** (0.006)
Change in plan seats	−0.437** (0.182)	−0.181** (0.072)	−0.023** (0.011)
VRA section 5 coverage	0.749 (0.517)	0.234 (0.322)	0.056 (0.050)
Democratic vote share	−0.630 (0.684)	−0.229 (0.883)	−0.127 (0.101)
State Black + Hispanic population	5.399* (2.789)	9.024*** (1.465)	-
1982	ref	ref	ref
1992	0.362** (0.138)	0.602*** (0.205)	0.092** (0.036)
2002	0.271 (0.201)	0.389 (0.244)	0.070** (0.032)
2012	0.097 (0.257)	0.160 (0.375)	0.060 (0.036)
2022	0.072 (0.427)	0.050 (0.547)	0.0455 (0.036)
Constant	−3.378*** (1.111)	−5.889*** (0.579)	0.0544 (0.092)
R2	0.7268	-	0.3073
n	214	214	214

Model 10 uses the percentage of majority–minority districts in a plan as the dependent variable in a GLM with a binomial family and logit link due to the bounded nature of the measure. It shows statistically significant and positive relationships between partisan legislatures, independent commissions, and federal courts in relation to the reference category. The similar effect sizes help highlight the institutional influence on this metric relative to other institutions, but again, state Black and Hispanic population is the key predictor variable.

Model 11 uses the novel Majority-Minority Proportionality Test as the dependent variable. The results of Model 11 (Figure 6) highlight the strong and statistically significant relationships between federal courts and independent commissions on proportional majority–minority districting relative to the other redistricting institutions. Despite the small n, independent commissions have a strong predicted relationship on the majority–minority districts even when accounting for the controls, year, and state fixed effects.OPEN IN VIEWER

For the federal courts, the cause of this relationship has some clear endogenous possibilities because the federal judiciary is often the site of Voting Rights Act enforcement. However, in terms of correlation of criteria, this is still notable as a relationship for the federal courts when they adopt redistricting plans. Control variables for reapportionment (negative) and plan size (positive) are also statistically significant estimates.

Taken together, these models, and the descriptive data, show that independent commissions and federal courts have the strongest relationships with minority representation as criteria. The relatively large and positive effects of independent commissions on the proportionality score are especially notable because of the demographics of the states that use independent commissions such as California and Arizona, and the role of the federal courts in guaranteeing voting rights.

Traditional Criteria

The fourth set of redistricting criteria is Traditional Criteria, including compactness and splitting of political subdivisions.

Compactness has long been the preeminent “traditional criteria” for considering “gerrymandering” based on the theory that the most simple and compact districts are the least manipulated. Compactness is a quantification of how geographically close the boundary of a district is to its center in a redistricting plan. Compactness has attracted more than 100 different ways to be calculated (McDonald 2019). For this project, I use the approach and replication data from Barry Edwards et al.’s 2017 Journal of Politics article on redistricting institutions and traditional criteria (Edwards et al. 2017). I build on the scholars’ district-level data, which measured compactness using the Polsby–Popper, Reock, and Convex-Hull measurements. ²⁰ I aggregated this district data up to the statewide level creating new statistics for the plan-wide compactness means, which I use as dependent variables. An average compactness score for a plan is far less accurate than a district-level analysis, but it provides a measure of compactness to view in relation to the other institutions. In line with previous research (Edwards et al. 2017), all three compactness measures should be viewed together to make conclusions. I also include the 2022 Polsby–Popper and Reock data from the Princeton Gerrymandering Project in my data set and analyses. ²¹

The descriptive data (Figure 7) of the compactness means shows some variation in average district compactness, with state courts and divided legislatures creating plans with more compact average districts than partisan legislatures or political commissions.OPEN IN VIEWER

OPEN IN VIEWER

Figure 8.

Estimates of compactness measures—Models 12, 13, and 14.

The regression table (Table 4) for Models 12, 13, and 14 displays the results from the three compactness dependent variables. The three compactness models should be read together to get a better understanding of each institution’s accurate relationship with this criterion (Edwards et al. 2017). The consistent results across the three models (Figure 8) show two important conclusions: First, Democratic-led legislatures have lower average compactness scores for their plans than the other institutions. Second, independent commissions have significant and positive effects on the average compactness scores in their plans. As noted previously, these models help understand the relative impact of redistricting institutions’ emphasis on this criterion but are not as accurate as district-based analysis on measuring absolute compactness. These analyses also highlight the importance of considering politically appointed and independent commissions separately—the relationships are distinct. The control variables in all three compactness models also highlight interesting results about less compact districts in the 1990s and 2000s as well as the statistically significant and substantially small relationship between plan size and average compactness.

OPEN IN VIEWER

Table 4.

Traditional Criteria Models.

	Traditional Criteria Metrics
	Compactness			Political Subdivisions
	Model 12: Mean Polsby–Popper (GLM)	Model 13: Mean Reock (GLM)	Model 14: Mean Convex-Hull Ratio (GLM)	Model 15: County Split (OLS)	Model 16: City Split (OLS)
Republican legislatures (n = 67)	−0.154 (0.104)	−0.012 (0.071)	−0.122 (0.082)	6.76* (3.62)	−12.37 (8.49)
Democratic legislatures (n = 84)	−0.365*** (0.096)	−0.10* (0.053)	−0.286*** (0.079)	10.29** (4.14)	−10.99 (11.93)
Political commissions (n = 29)	−0.360*** (0.130)	−0.003 (0.049)	−0.112 (0.078)	3.16 (4.68)	−0.957 (8.97)
Independent commissions (n = 8)	0.165 (0.120)	0.172** (0.075)	0.327*** (0.127)	5.20 (4.88)	−13.81 (10.79)
Federal courts (n = 28)	0.032 (0.112)	0.076 (0.051)	−0.018 (0.089)	5.67 (3.72)	−29.06** (12.41)
State courts (n = 19)	0.063 (0.142)	0.076 (0.055)	0.067 (0.106)	−5.06 (5.20)	−13.69 (9.43)
Divided legislatures (n = 24)	ref	ref	ref	ref	ref
Number of plan districts	−0.009*** (0.003)	−0.005*** (0.002)	−0.012*** (0.003)	0.632 (1.58)	−7.70 (5.35)
Change in plan seats	0.006 (0.028)	0.011 (0.007)	0.025** (0.012)	1.07 (1.19)	11.57*** (3.48)
VRA section 5 coverage	−0.281*** (0.090)	−0.009 (0.052)	−0.054 (0.071)	11.58 (7.94)	Collinear
Democratic vote share	−0.428 (0.335)	−0.117 (0.124)	−0.309 (0.208)	−11.64 (10.31)	−22.736 (27.64)
Total # of counties	-	-	-	0.156*** (0.041)	-
Total # of cities	-	-	-	-	0.020 (0.041)
1972	ref	ref	ref	-	-
1982	−0.006 (0.065)	−0.043 (0.033)	−0.052 (0.034)	-	-
1992	−0.067 (0.077)	−0.096** (0.039)	−0.207*** (0.055)	ref	ref
2002	−0.162** (0.077)	−0.148*** (0.041)	−0.284 (0.0521)	1.24 (2.40)	−3.81 (4.33)
2012	−0.199** (0.079)	−0.153*** (0.046)	−0.340*** (0.064)	0.817 (2.090)	−7.303 (9.052)
2022	−0.096 (0.107)	−0.078 (0.062)	-	−14.05*** (3.99)	-
Constant	−0.459** (0.194)	−0.238*** (0.076)	1.588*** (0.139)	6.019 (16.655)	149.1*** (38.94)
R2	-	-	-	0.4983	0.4284
n	252	252	252	168	129

Another important traditional redistricting criterion is the protection of political subdivisions—whether town, city, or county boundaries are split by district lines. I use the total number of counties and cities split by districts as separate dependent variables in the analysis. This data is also aggregated from replication data (Edwards et al. 2017) and collected for counties from the Princeton Gerrymandering Project for 2022. These analyses include the unique control variables of the number of total statewide counties or cities, respectively.

The descriptive data (Figure 9) shows wide variation among institutions for how county lines are split per plan and less variance in city lines being divided by congressional districts. Independent commissions, federal courts, democratic legislatures, and political commissions are among the highest county line splitters per plan on average. These initial data are intriguing because protection of political subdivisions is criteria that fit well into Cain’s framework of competing criteria. If a different criterion is prioritized, such as compactness or partisan advantage, protecting county or city boundaries may be criteria that need to be neglected.OPEN IN VIEWER

I use OLS regression models to better assess the relationship that institutional variation has with these criteria. Table 4 and Figure 10 display the results for Models 15 and 16.OPEN IN VIEWER

For Model 15, where the number of county boundaries split by district lines is the dependent variable, partisan legislatures are shown to split more county lines when they are map makers. Democratic-controlled legislatures are estimated to split more county boundaries when they draw a plan as opposed to other institutions, twice the estimated impact of Republican-made plans. The number of counties in a plan is a statistically significant indicator for how many county boundaries will be split by districts. The 2022 redistricting cycle had negative effects, showing fewer county lines split compared to other years.

For city boundaries (Model 16), the federal courts were estimated to split fewer municipal boundaries than other redistricting commissions. Other institutional effects lack statistical significance and clear emphasis. The change in number of seats in a redistricting plan was a significant indicator for how many city boundaries would be split by district lines.

There is one other important redistricting criterion that is not emphasized in this analysis: the equal population of districts.

Since the 1960s’ Reapportionment Revolution and establishment of One Person, One Vote for congressional districts (Wesberry v Sanders 1964), equal population among federal legislative districts has been a required criterion for redistricting institutions. As a result, contemporary congressional redistricting plans have had almost zero population variance among districts.

Although the population equality data may not be an interesting criterion to model in this analysis, it is a critical aspect of the larger picture of redistricting and representation. Because all redistricting institutions must fulfill this requirement, it constrains the other choices they can make. If their true goal is competitiveness or partisan advantage, they first must fulfill the population equality criterion. This subordinates their other criteria goals and only serves to further subordinate other potential criteria. This constraint does apply to all institutions however and still allows for the relative comparison of criteria.

Legislatures and Political Commissions Facilitate Partisan Group Representation

The data show that both Democratic- and Republican-controlled legislatures make congressional redistricting maps that correlate highly with partisan advantage. Political commissions also had Democratic bias with seat share and two partisan gerrymandering metrics in the descriptive data. Additionally, divided legislatures and political commissions correlate with incumbency protection as a key criterion.

To achieve a partisan majority in a state delegation or protect an incumbent member of congress, the redistricters must classify constituents and potential voters based on the group identity of partisanship (or likely partisanship). Common gerrymandering techniques like cracking or packing require conceptualizing voters by partisan group identity. However, using partisan or political criteria as the predominant redistricting criterion among all of the competing criteria also means that constituents are being categorized and grouped primarily through their partisan identity at the expense of other identities that could be considered for districting or representation, such as their town/city/county residence, community membership, race, sex, socioeconomic status, ethnicity, citizenship, age, community of interest, religion, occupation, or others.

Federal Courts and Independent Commissions Facilitate Racial Group Representation

Minority voting criteria, as measured by the three majority–minority metrics used in the models, are key for two redistricting institutions: federal courts and independent commissions. An emphasis on majority–minority districts would require the consideration and grouping of constituents based on race and ethnicity as opposed to other potential group identities, such as geographic or partisan. It may also require the consideration of voting age and citizenship for majority–minority districting.

This paper argues that the institutions that correlate most with minority voting criteria facilitate a form of racial representation in their congressional districts above other forms of representation like party or geography by emphasizing race and ethnicity as the salient group identities. The measurements used in this analysis emphasize a specific understanding of racial representation focused on majority–minority districting for Black and Hispanic Americans. This conclusion could be improved with additional measurements of this criterion.

For the federal courts, the emphasis on race as key group identity in redistricting seems clear based on these results. Because the federal courts do not have strong or positive relationships with all other criteria, they must consider constituent race as a key basis for the district lines they draw. As mentioned previously, this is likely related to the process by which federal courts end up drawing redistricting maps. Endogenously and unavoidably, federal courts will often draw redistricting plans because of constitutional or Voting Rights Act litigation related to race.

State Courts and Independent Commissions Facilitate Geographic Group Representation

Traditional redistricting criteria are most correlated with independent commissions and state courts. Traditional redistricting criteria—especially the forms measured in this data set—promote a grouping of constituents based on geographic identity, such as city or county residence or spatial closeness.

The results make it likely that state courts emphasize traditional criteria and geographic identity most when drawing congressional district lines, over other group identities, like race or partisanship. These findings deserve additional research and highlight the importance of studying state and federal courts as separate institutions in redistricting. These results may point to the importance of state constitutional provisions and statutes on traditional criteria, or the importance of nonpartisanship in judicial legitimacy, or both, or neither. These findings are ripe for further exploration in scholarship related to place-based identity.

Independent Commissions Have One of the Most Complete Profiles from this Analysis

Independent commissions have positive relationships with several key criteria, including compactness, the protection of city boundaries, and minority representation, while clearly disfavoring incumbents and having no clear effect on partisan advantage.

Based on these results, it is likely constituents are grouped by geographic and racial identities as opposed to partisan identities in districts made by independent commissions. These results also show that there are good reasons for separating independent and political commissions in analyses.

The questions in this paper are not just academic—they are at the heart of some of the highest-level political and legal disputes related to redistricting. The 2024 Supreme Court case Alexander v. South Carolina State Conference of the NAACP provides a concrete example of the theory that this paper puts forward as well as the political, legal, and representational stakes.

The central question in Alexander was whether constituents in South Carolina were being grouped by the state legislature based on their racial identities or their partisan identities when congressional boundaries were redistricted. The Republican-controlled legislature argued that the plan they drew after the 2020 census emphasized partisan advantage, cracking Democratic voters and helping Republicans throughout the rest of the state. They argued that the gerrymandering was permissible, and beyond the scope of federal court justiciability, ²² because it is partisan gerrymandering—grouping individuals based on their partisan identities.

The South Carolina chapter of the National Association for the Advancement of Colored People (NAACP) sued the state, arguing that residents were actually being districted by race, not party. By using race as the predominant factor, the plaintiffs alleged, the state violated the 14th Amendment’s Equal Protection clause. A federal district court agreed. It ruled that race was the predominant factor in South Carolina’s congressional districting—that the legislature did group constituents based on their racial identity and it was therefore a violation of the law.

In May 2024, a 6-3 Supreme Court overturned the lower court decision, arguing that it had not sufficiently proven that the state legislature was categorizing constituents by their racial identity as opposed to their partisan identity. “To make that showing, a plaintiff must prove that the State ‘subordinated’ race-neutral districting criteria such as compactness, contiguity, and core preservation to ‘racial considerations’, ²³ ” wrote Justice Samuel Alito in his majority opinion.

Alexander emphasizes the utility and the stakes of this paper’s topic: how people are grouped by redistricting institutions, and by what criteria, matters for congressional representation and democratic rights. In the context of race and party, it also matters for legal remedies. There are distinct legal avenues for constituents to challenge redistricting plans based on how they are being grouped—racial gerrymanders can be challenged in federal court, but partisan gerrymanders cannot.

Alexander underscores the relevance of this paper’s argument by conceptualizing the process of redistricting as competing criteria that can be prioritized or subordinated by redistricting institutions with important implications for representation. Additionally, the case highlights some of the limitations of this project and areas for further study.

This paper advocates for a more comprehensive research design for redistricting and representation. However, this design is not fully comprehensive of all potential criteria that institutions may use. Future approaches could be even more comprehensive and interactive, with stronger measurements. Similarly, as Alexander underscores, racial gerrymandering would benefit from better measurement. Alexander directly confronts questions like, how should racial gerrymandering be measured? What measurements can be used to differentiate correlated group representations such as race and party?

Overall, this project is designed to provide a data set and theoretical basis for further research on redistricting and representation. There are many implications and applications for this paper’s findings that lead to related research questions, including: What drives the institutional variation in correlated criteria? How do constituents experience the differences of the favored redistricting criteria in policy and responsiveness? Are the institution-imposed group identities congruent with the constituent preferences? To what extent do these criteria impact the group identity formation and the voting behavior of constituents?