November 18, 2022
The United States is rapidly becoming an unrepresentative democracy. In an ideal democracy, every vote counts and all votes count equally. In the U.S. some serious deviations from that ideal were already written into the Constitution as part of the compromises needed to create a union in the first place. The Electoral College in Presidential elections and the awarding of equal representation in the U.S. Senate to all States, independent of their population, both give disproportionate weight to voters in sparsely populated states. It would take an enormous groundswell among U.S. citizens to push for Constitutional amendments to change those basic elements of the U.S. electoral system. Certainly, the smaller-population states are not going to willingly give up power to reform the Electoral College or the Senate makeup.
The more serious problem for the moment – and one that voters could change if they wanted to – is that the vast majority of members elected to the U.S. House of Representatives, as well as in State houses around the country, are effectively chosen by a small, and wholly unrepresentative, fraction of the general voting public. The problem arises from several features of district drawing, primary elections, and vote counting that are not at all Constitutionally mandated:
- The combination of geographical self-sorting of voters by their partisan preferences (e.g., rural Republicans vs. urban Democrats) with aggressive partisan gerrymandering by State legislatures and Governors have made an increasingly large percentage of U.S. voting districts “safe,” in the sense that one party will predictably win that district, pretty much independent of whom they nominate and of year-to-year shifts in the overall partisan leanings of the electorate at large. Figure I.1 illustrates the steady decline in competitive districts over the past several decades.
- The winner of safe seats is therefore effectively determined in primary elections that were not even foreseen in the U.S. Constitution. The choice of primary election protocols is determined by individual States or parties, and relatively few of them to date permit fair representation of all voters in a district. In particular, 10 states feature closed primaries in which registered independent or unaffiliated voters are prohibited from participating in the selection of either major party candidate.
- Primary election turnout is nearly always quite low and is dominated by the strongest partisan activists in a district, who tend to be considerably more extreme in their views and candidate preferences than the population at large.
- Voting in the general elections is usually fair, but the voters at that point can only express a single preference among candidates that have survived the primary elections held by the two major parties. General election voters usually just vote their party prejudices, no matter how extreme, or even unlikeable, their candidate, rather than honestly weighing qualifications or character of the candidates.
- Senate elections are not directly affected by gerrymandering since the elections are statewide. However, since serving House members frequently run for Senate in subsequent elections, the pool of Senate candidates is indirectly affected by gerrymandering. And Senate elections and political calculations of sitting Senators are strongly affected by the skewed turnout in primary elections: their re-election concerns are often dominated by the fear of being outflanked by a more extreme candidate in the party’s next primary election.
The extent of the deviation from representative democracy is starkly quantified in an analysis by Unite America of 2020 voting in all Congressional districts. They conclude: “Of 234.9 million eligible voters, only 23.4 million cast ballots (10%) in primaries that effectively decided the outcome of 361 Congressional districts (83%).” So, some votes, often from voters with more extreme views, are “more equal” than others. We will describe additional aspects of the Unite America analysis in Section II of this post.
Why is this system problematic, when there are many safe Democratic districts as well as Republican districts, so that the majority in the U.S. House does occasionally shift back and forth between the two parties? This skewed voting leads ineluctably to increasing partisan rancor and Congressional dysfunction. Representatives elected from safe districts do what they can to stave off defeat in the next primary election by an even more extreme candidate. Part of that calculation leads to increasing buy-in to party- or President-determined positions and talking points, no matter how outlandish the policy or behavior they may be defending – few Representatives want to stand out as bucking their own party (see Cheney, Liz). Political compromise is disfavored by the primary system and low voter turnouts in primaries.
In place of honest attempts to find compromise solutions to real problems, we get political games of “chicken,” in which minority parties threaten government shutdowns or U.S. debt default as a method of extortion to get their pet policies enacted. The skewed primaries furthermore lead to candidates accepting absurd policy positions, such as the nearly universal buy-in of 2022 Republican candidates to Donald Trump’s Big Lie of a stolen 2020 election, despite the complete absence of supporting evidence and the obvious threat this stance poses to the future of American democracy. Extreme Congress members elected by an unrepresentative minority of eligible voters then demonize their opposition, promote lies, and help to stoke partisan tribalism that divides the country further.
Much of the effort in the House of Representatives is wasted on passing strongly partisan bills that have little chance of garnering support in the usually more balanced Senate, or to partisan Congressional investigations of the opposing party’s issues and politicians. (We do not consider the ongoing January 6 Select Committee in the U.S. House to represent such a partisan investigation.) Oversight hearings in the House often devolve into partisan cat-fights. Policy preferences in House bills alternate radically depending on the party in narrow power, since there is still competitive balance nationwide, with both parties selecting their voters carefully. Congressional control is effectively determined in the dwindling number of competitive districts. Most moderate lawmakers come from the relatively small number of competitive districts.
One profound consequence of this dysfunction is that laws favored by large majorities of the general voting public are unable to garner bipartisan support in Congress because they offend the small minority of voters who dominate partisan primaries or the big donors who help candidates navigate those primaries. Examples include universal background checks for firearm purchases, favored by about 90% of the public in polls, and allowances for abortion exemptions in the case of rape or incest, favored by more than 70% even in the most conservative states. Congressional dysfunction leads inexorably to Presidents taking over more power from the Legislature. That trend establishes the conditions under which a suitably inclined President (see Trump, Donald J.) can attempt a non-democratic or fascist takeover of the country’s governance.
It becomes clearer with each new Congress that the U.S. electoral system is in serious need of reforms to make the democracy more representative of citizens at large. Without reforms, the U.S. political system is in a positive feedback state that drives it toward increasing polarization and decreasing effectiveness. Parties that dominate individual States or Congress, and politicians in general, will not normally support such reforms, because they make their own re-elections dicier. Redistricting reform acts have been introduced in both houses of Congress in 2021 – requiring transparent processes open to citizen input and comment and, in the Senate version, outlining guidelines for independent redistricting commissions and for a district map that does not markedly favor one party over another – but have not progressed in either chamber. Thus, the push has to come from the voting public — a large fraction of whom now view Congress with disdain — mostly through citizen-generated statewide referendums posing particular reforms.
In this post, we will deal with evidence to date regarding the effectiveness of several specific proposed types of reform to redistricting, to primary election processes, to increasing voter turnout, and to the evaluation of votes in general elections. Evidence is provided, sometimes with sophisticated statistical analyses, by experience in different countries or U.S. States that have adopted different procedures. The goal is to find reforms consistent with the U.S. Constitution that increase the fraction of the voting public that influences elected representatives and the competitiveness of districts, so that candidates have to establish appeal to a representative sample of the population, rather than just to extreme partisans. As we will show in Section II, more competitive elections also tend to increase voter turnout. The same reforms can also, in principle, open elections more effectively to third-party candidates if voters are dissatisfied with both Republican and Democratic candidates.
In Part I of this post we present additional evidence defining the problem with current electoral systems (Section II) and we evaluate proposals for redistricting reform (Section III). In Part II we describe alternatives to the closed partisan primary elections still used in most States (Section IV) and the advantages of ranked-choice voting (Section V), where voters’ second or third choices among multiple general election candidates from different parties can influence the election outcome – for example, by disfavoring candidates who elicit strongly negative perceptions from a sizable fraction of the electorate — without the need for delayed runoff elections. In Section VI we summarize our findings and recommend paths forward to make American democracy more responsive to the entire electorate.
II. some data defining the problem
Different analyses of U.S. voting results use somewhat different definitions of “competitive” House districts, but all agree that the percentage of competitive districts is declining rapidly. In the analysis from FiveThirtyEight shown in Fig. II.1, competitive districts are defined as those with a Cook Political Report Partisan Voting Index value indicating a partisan lean of 5 percentage points or less toward either major party. The fraction of all U.S. House districts that fall within this category has been cut in half over the past two decades. The fraction of districts that prefer different parties in their Presidential and U.S. House elections has dwindled from a peak in the mid-1970s to almost zero. The vast majority of Congressional districts are predictable and largely immune to candidate deficiencies. For example, “In 2020, FairVote, a leading nonpartisan reform organization, accurately predicted the winner of 357 congressional seats with 99.7% accuracy, relying solely on prior voting patterns of the district (rather than polling or other inputs).”
In the analysis of 2020 voting results by Unite America, competitive districts are taken as those in which the winner of that district in the previous Presidential election won by a vote margin no greater than 8% (i.e., 54% to 46% or tighter). By this measure, the fraction of competitive districts during the 2020 election was only 17%. The winning candidates for Congress in the 83% of districts considered safe were selected in 2020 by an average of 10% of nationwide eligible voters who cast votes in the primary elections selecting the sure-to-win candidate from the strongly preferred party. In the most extreme example, just 0.4% of eligible voters in Virginia’s 5th Congressional district chose the district’s Republican Representative. They did this by participating not in a primary election, but rather in a closed convention organized by the Republican Party, in which “only 2,500 delegates participated by casting ballots from their cars in a church parking lot.” The sitting Republican Congressman from that district, Denver Riggleman, lost his re-election bid in that convention because he had officiated a same-sex wedding of members of his Congressional staff.
The 10% who vote on average in primary elections in safe districts tend to have much more strongly partisan (i.e., left-leaning or right-leaning) self-reported views than the general voting public. This is illustrated, for example, in Fig. II.2, representing self-reported policy preferences among voters in the 3rd district in Colorado from a survey commissioned by Unite America after the 2020 election. In the primary election in that safe Republican district, the gun-toting QAnon-friendly Lauren Boebert defeated the five-term incumbent Scott Tipton. 60% of the Republican primary voters identified as “very conservative,” compared to 35% of Republican voters who only participated in the general election and 25% of all general election voters in that district. 37% of the Republican primary voters favored leaders who refuse to compromise, compared to 19% of general election voters. The “ideology index” in Fig. II.2 takes account of responses on a suite of analogous questions. There is an analogous tilt toward very liberal voters in safe Democratic districts, such as New York’s 14th district, where Alexandria Ocasio-Cortez defeated the incumbent Democratic Representative Joseph Crowley in a 2018 primary election by attracting 4.3% of all eligible voters in the district.
As the granularity of voting preference data and the sophistication of district drawing software improve, the percentage of competitive districts dwindles with each census. As we will show in Section III, redistricting following the 2020 U.S. census has reduced the fraction of competitive Congressional districts further from 17% to 14%. This continues a trend seen over recent censuses. The politicians who generally draw these districts are not interested in serving the general public or improving the country; they are interested primarily in doing what they can to guarantee their party’s continuing dominance. As the country loses competitive districts it also loses the more moderate Representatives (see the “missing middle” in Fig. II.3), who might be effective in brokering compromises in Congress. In addition, the safer the district, the lower (on average) the voter turnout even in the general election (see Fig. II.4), effectively ensuring that the district will remain safe in the future. Safe districts and closed primary elections thus enhance not only voter disenfranchisement, but also voter disengagement from the political process.
The root cause of all these problems and of ongoing Congressional dysfunction is not the U.S. Constitution or the overall political polarization in the country or what is commonly understood by the term “election fraud.” It is rather allowing politicians to rig their own games in the drawing of Congressional districts, the organization of primary elections, and even the method of voting in winner-take-all elections. We treat each of these rigging attempts in turn in the remaining sections and we suggest reforms that voters themselves can insist on.
III. redistricting reform
A bit of history:
Here is the entirety of what the original U.S. Constitution says about “redistricting:”
“The House of Representatives shall be composed of Members chosen every second Year by the People of the several States, and the Electors in each State shall have the Qualifications requisite for Electors of the most numerous Branch of the State Legislature.… Representatives and direct Taxes shall be apportioned among the several States which may be included within this Union, according to their respective Numbers, which shall be determined by adding to the whole Number of free Persons, including those bound to Service for a Term of Years, and excluding Indians not taxed, three fifths of all other Persons. The actual Enumeration shall be made within three Years after the first Meeting of the Congress of the United States, and within every subsequent Term of ten Years, in such Manner as they shall by Law direct. The Number of Representatives shall not exceed one for every thirty Thousand, but each State shall have at Least one Representative…”
There is absolutely no mention in the Constitution of Congressional districts or of the processes to be used for the election (as opposed to the apportionment) of House members. The Constitution commands a census every ten years to determine the number of Representatives from each state, and the population in each state is now understood to include all inhabitants; the infamous “three-fifths” compromise, which allowed southern States to gain some benefit in apportionment from slaves who were denied the right to vote themselves, was repealed by the 14th Amendment to the Constitution. But the Constitution does not require elections within districts. Thus, it would have been completely consistent with the Constitution for states to have adopted the proportional representation approach used in many other countries, where the election is statewide and the number of Representatives from each party is apportioned to represent the percentage of popular votes that Party received in the statewide election (perhaps with some threshold on percentage of the total vote needed to earn at least one seat). With redistricting as practiced in the U.S., representation deviates substantially from proportionality, as we will describe below.
The Founders did consider representation in the U.S. House by single-member districts, but chose not to include that in the Constitution because they considered it too prescriptive. As pointed out by the FairVote website: “James Madison wrote in Federalist Paper Number 56, ‘divide the largest state into ten or twelve districts and it will be found that there will be no peculiar interests…which will not be within the knowledge of the Representative of the district.’ Nevertheless, most of the original thirteen states used multi-member districts in the first congressional elections; none of course used proportional systems, which had yet to be developed… In 1842, six states were electing representatives at-large and twenty-two states were electing representatives by single-member district. Three states had only one representative… This arrangement changed with an apportionment act in 1842… This act set the House membership at 223 members and contained a requirement for single-member districts. It stated that representatives ‘should be elected by districts composed of contiguous territory equal in number to the number of representatives to which said state may be entitled, no one district electing more than one representative.’ Thus single-member districts were officially instituted by Congress.”
The constitutionality of the 1842 apportionment act was questionable and not all states adhered to it at first. However, the argument that single-member districts lead to better representations of individual community interests carries some weight, although that weight is diminished in the highly gerrymandered, egregiously extended districts that State legislatures sometimes draw. The FairVote site goes on to say: “Apportionment acts continued to be passed after each national census. These acts increased the number of representatives as the population of the country increased and more states were added to the Union. However, the districting provisions of the 1842 act were modified inconsistently, resulting in several important changes in congressional districting policy.“
It was, in fact, not until 1967 that a uniform approach to Congressional districts was established by Congress. “In 1967 Congress passed a law…which prohibited at-large and other multi-member elections by states with more than one House seat. Only two states, Hawaii and New Mexico, were affected by this legislation: all other states by this time were using elections by districts. This law was passed largely because of two factors. The first concern was that, in the wake of the 1965 Voting Rights Act, southern states might resort to winner-take-all at-large elections to dilute the voting strength of newly-enfranchised blacks in the South. The second concern was that the courts might order at-large elections in states which were having difficulties with redistricting. Such elections could have threatened the position of incumbents whose district seats were considered safe for re-election.”
However, the 1967 law did not mandate, as some previous Apportionment Acts had, that the districts be “contiguous, compact, and equally populated.” Thus, it left State legislatures to their own devices in constructing districts to optimize the chances that the Party controlling the State legislature at the time of the census will also dominate the State’s membership in the U.S. House. States had long since figured out how to use single-member district drawing to political advantage. In 1812 the Massachusetts legislature drew the tortuous state senate district pictured in Fig. III.1 to favor the ruling party, and the district drawing was approved by Governor Elbridge Gerry. Although the district is pictured in the cartoon as a dragon, others commented that it had the shape of a salamander – hence, the origin of the term Gerry-mander.
The 1965 Voting Rights Act was intended to prohibit State policies that enacted various forms of racial discrimination in voting. Two sections of the Act dealt specifically with the drawing of electoral districts. Section 2 prohibited jurisdictions from drawing electoral districts that are aimed at diluting the votes of protected racial minorities. Section 5 required pre-clearance by the federal government (specifically by the U.S. Attorney General or the U.S. District Court for the District of Columbia) of any changes, including in redistricting, that affected voting in specific states and localities that had long engaged in egregious racial discrimination practices in voting prior to the passage of the Voting Rights Act.
However, the U.S. Supreme Court has weighed in on the constitutionality of several aspects of redistricting and on provisions of the Voting Rights Act in particular. In Reynolds v Sims (1964), the Supreme Court held that “the Equal Protection Clause [of the 14th Amendment to the Constitution] requires that the seats in both houses of a bicameral state legislature must be apportioned on a population basis,” assuring that maps achieve “substantial equality of population among the various districts,” with deviations from strict equality allowed if they are “based on legitimate considerations incident to the effectuation of a rational state policy.” In Karcher v. Daggett (1983), the same “substantial equality of population” principle was applied to Congressional districts.
In Shaw v. Reno (1993) the Supreme Court held that “Legislative and congressional districts will be struck down by courts for violating the Equal Protection Clause if they cannot be explained on grounds other than race. While not dispositive, ‘bizarrely shaped’ districts are strongly indicative of racial intent.” And in Miller v. Johnson (1995), the Court explicitly held that “A district becomes an unconstitutional racial gerrymander if race was the ‘predominant’ factor in the drawing of its lines.” In Alabama Legislative Black Caucus v. Alabama (2015) the Court clarified that racial gerrymandering claims should proceed district by district, and not be judged on the basis of an entire state plan.
However, conservative members of the current Supreme Court seem intent to unravel, little by little, restrictions on voting laws passed by state legislatures. In Shelby County v. Holder (2013), the Court decided effectively that “Section 5 of the  Voting Rights Act no longer applies to any jurisdictions in the United States. As a result, redistricting plans and other changes in voting laws, such as voter identification requirements, need not be approved [by the federal government] before they take effect.” The questionable reasoning used in Chief Justice John Roberts’ majority opinion argued that the coverage formula that led to disparate treatment of different states was “based on 40-year old facts having no logical relationship to the present day” because “the country has changed.” Section 2 of the Voting Rights Act is the subject of a current Supreme Court case, Merrill v. Milligan, brought by the state of Alabama – no longer required to get pre-clearance after Shelby County v. Holder — whose 2022 redistricting plan was judged by a lower court to include racially motivated gerrymandering. We await the Court’s decision on this important case.
Of course, gerrymandering can be used strictly to gain partisan advantage without an obvious racial motivation. In June 2019 the Supreme Court ruled that federal courts have no role in policing gerrymandering for partisan advantage – the “one person, one vote” guideline be damned (in the language of George Orwell’s Animal Farm, some votes are “more equal” than others). This ruling rendered invalid some previous Supreme Court attempts to provide guidelines that might be used to judge excessive partisan advantage, which proved too difficult to apply consistently. That issue is therefore left to State courts and the political process, including the possibility of Congressional legislation to add back restrictions on districts that had been introduced in various earlier Apportionment Acts, following early censuses. Now the jurisdiction of State courts is being questioned further in the upcoming Moore v. Harper case. In that case, the North Carolina legislature argues that its State Supreme Court, which had struck down as a violation of the State Constitution the highly gerrymandered 2022 Congressional district map the legislature passed, should have no jurisdiction over the State legislature in matters pertaining to federal elections. The North Carolina legislature has been trying to adopt district maps with a strong partisan favoring of the Republican Party for a decade now, as we will examine further below.
Before going on to describe gerrymandering as it is practiced as an art form today, we note with the FairVote site that single-member Congressional districts are not written in stone. “Provision for simply giving states the option to have multi-member districts requires only repeal of the single-member district requirement passed by Congress in 1967… When combined with proportional representation (such as cumulative voting or preference voting), multi-member districts would provide states with a viable method to reduce problems inherent in single-member districts.” This would be the most direct reform of redistricting abuse and would bring the U.S. House closest to the intent of the Equal Protection clause of the 14th Amendment, but it requires Congressional action, rather than State actions, and an action that would threaten the re-election of many House members. In addition, multi-member districts have problems of their own, and might well make Representatives even less responsive to the needs of all their constituents.
Current redistricting practice:
The drawing of districts for both the U.S. House and State legislatures leaves states a good deal of leeway in tilting the playing field to produce unproportional representation. The Wikipedia article on gerrymandering uses Fig. III.2 to indicate how different choices of district boundaries give rise to different numbers of representatives from each of two parties, given the same underlying popular vote within the state.
Partisan redistricting today is aided by granular statistics on citizen voting preferences and sophisticated district drawing software. Examples of gerrymandered districts are shown in Figs. III.3 and III.4. The first of these shows the Texas 24th Congressional district prior to the state’s 2022 redistricting, compared to its subsequent subdivision into the new 24th and 32nd districts. According to the 2020 census, Texas gained two seats in Congress, largely due to increases in Hispanic population. But the Republican-controlled legislature used that flexibility to add two more safe Republican districts, while piling Democratic voters as much as possible into a few districts like the 32nd. The 24th district used to be competitive, but in the new district drawing is now safely Republican. The redistricting, of course, does not tend to increase the influence of Hispanic voters despite the increased apportionment of seats coming mainly from increases in Hispanic population.
Of course, Democrats use similar tricks in states where they control the legislature and the Governorship. Figure III.4 illustrates gerrymandered districts in Maryland and Pennsylvania prior to the 2022 redistricting. These are hardly compact though they satisfy the letter — certainly not the spirit — of any law requiring contiguousness by the use of extremely narrow, barely visible, isthmuses of land containing few, if any, voters, joining basically disjoint areas! Bizarrely shaped, elongated districts are often a signal of egregious gerrymandering, but seriously partisan gerrymandering can also be accomplished without such offensive district shapes.
The aim of partisan gerrymandering is two-fold: (a) to create strong deviations from proportional representation, and (b) to make those deviations as resistant as possible to overall shifts in statewide party preference among the electorate. The recognition that nationally coordinated creative redistricting could actually change control of the U.S. House led the Republican Party to focus on taking over State legislatures following the election of Barack Obama in the 2008 Presidential election. They devised Operation REDMAP: “With $30 million in funds from corporations like Wal-Mart, tobacco giants Altria and Reynolds, and from [Republican strategist Karl] Rove’s SuperPAC, American Crossroads, the GOP mounted a targeted drive to capture control of as many state legislatures as possible in the 2010 election in order to dominate congressional redistricting after the 2010 census.”
Republican gains in House seats in the 2012-2016 elections were the “prize” for the victors. Their gains are well illustrated by the North Carolina Congressional district maps in 2012 vs. 2010 (Fig. III.5). “Before REDMAP, Democrats in North Carolina carried the popular vote for Congress and won a 7-6 majority in the state’s House delegation in the 2010 election, under the redistricting plan drafted by Democrats in 2001. Two years later, in 2012, Democratic House candidates again won a statewide popular vote majority for the House. But a new gerrymandered map had been drawn by the Republican-dominated legislature, and Republicans emerged with 9 seats to 4 for the Democrats. A miraculous switch of three seats with little change in population.”
The national impact of Operation REDMAP in generating disproportionate representation in the U.S. House is illustrated in Fig. III.6 from an analysis by the Center for American Progress. Nearly half of the states show a partisan advantage in seat assignment, relative to proportional representation, of 15% or more. On average, for each of the three elections from 2012 to 2016, 59 House seats would have changed hands nationwide to the opposing party if the percentage of seats won matched the percentage of votes cast. Republicans saw a net gain of 19 additional seats each election because of districts biased in their favor. The net gain in percentage of seats in each state largely reflected the political control of the State legislatures after the 2010 census. Of course, in states with only a single or a few Representatives and winner-take-all elections, proportional representation is not possible. A map of the number of excess seats gained in each state would look somewhat different from Fig. III.6; for example, California’s modest 11% disproportionality led to an excess of 6 Democratic House seats, while Texas’ 7% disproportionality led to an excess of 3 Republican House seats.
In response to Republican gains from Operation REDMAP, Eric Holder and Barack Obama launched the National Democratic Redistricting Committee (NDRC) in 2018 to try to wrest back control of some state legislatures prior to redistricting after the 2020 census and to file lawsuits to overturn especially egregious Republican redistricting maps. The NDRC efforts have helped to partially counterbalance the Republican advantage in disproportionate House representation prior to the 2022 midterm election. However, an arms race between the two parties to see who can out-gerrymander the other is likely to come at the expense of the voting public. The basic aim of partisan gerrymandering is to reduce competitive districts and the relevance of general election votes.
The latter point is illustrated nicely by a more detailed analysis of North Carolina gerrymandering impacts, carried out by a group at Duke University, under the leadership of Jonathan Mattingly, a professor in the Mathematics department. Figure III.7 shows the impact of the State legislature’s 2022 drawn boundaries of its 13 Congressional districts when combined with the Democratic and Republican party vote shares recorded in the 2012 and 2016 Presidential elections, precinct-by-precinct across the state’s 3000 or so precincts. The graphs plot the implied Democratic vote share for the two elections in each of the 13 districts, ordered from most Republican to most Democratic district. The gold and purple curves represent the vote shares that would have resulted from sorting all those precincts according to the State legislature’s proposed district map. But, as we indicated above, the State Supreme Court has struck down the legislature’s map and replaced it with ones reflected by the green curves in each frame. These are similar to the yellow curves, which reflect the median vote shares among a large number of alternative district maps simulated according to Mattingly’s algorithm, which will be described in more detail later in this section.
The State legislature’s district map was gerrymandered specifically to introduce the big gap seen in each frame of Fig. III.7 between the leftmost 10 and rightmost 3 districts. That goal was achieved by “cracking” (moving a fraction of Democratic voters out of what would otherwise be the most competitive 4 or 5 districts) and “packing” (moving those displaced voters into districts that were already overwhelmingly Democratic, to make them even more overwhelmingly Democratic). This map leads to highly disproportional representation in the U.S. House. But perhaps most concerning is that it would have produced precisely the same seat apportionment – 10 Republican seats and 3 Democratic seats — in both elections, despite the fact that in 2012, when Barack Obama was re-elected, the vote share in each district shifted about 5 percentage points more Democratic than in 2016. With this gerrymandered map, North Carolina’s representation in the U.S. House would be largely unresponsive to the wishes of voters!
Jonathan Mattingly makes a persuasive argument that in a healthy democracy responsiveness to votes cast should be a fundamental requirement on any district map, more so than proportionality (i.e., 40% of the vote share should yield 40% of the apportioned seats) or symmetry (i.e., if Republicans get 6 seats when they have 51% of the vote share, then Democrats should get 6 seats when they get 51% of the vote share). Neither proportionality nor symmetry can be guaranteed when there is so much geographic concentration of partisan leanings. But responsiveness can be guaranteed by striking down gerrymandered maps with large cracking-packing gaps in figures like Fig. III.7. We will try to quantify a test for such cracking-packing gaps below in suggesting a set of nonpartisan criteria for judging state district maps.
Independent redistricting commissions:
As citizens begin to understand that they are the ultimate losers from gerrymandering, more and more states have begun to adopt or consider reforms to the way they draw new Congressional and state legislature district maps. Figure III.8 illustrates the Congressional redistricting methods in play in every U.S. state following the 2020 census.
Six states have only a single member of Congress, so no need for redistricting: Vermont, Delaware, North Dakota, South Dakota, Wyoming, and Alaska. Ten multi-member states have by now adopted some form of redistricting commission for Congressional districts, rather than relying on the state legislature or Governor for decisions that are bound to be politically influenced: New Jersey, Virginia, Michigan, Colorado, Arizona, California, Washington, Idaho, Montana, and Hawaii. Utah and Ohio have special arrangements that allow both legislature and independent commission input to the drawing of electoral maps.
The independent redistricting commission in Arizona was established in a voter referendum in 2000 but was subsequently challenged in court by Arizona’s Republican-dominated state legislature. The Republicans argued that under the U.S. constitution, it was only state legislatures that were empowered to draw district maps – an odd claim, since the Constitution did not even prescribe the existence of Congressional districts. The case ended up in the Supreme Court in 2015 (Arizona State Legislature v. Arizona Independent Redistricting Commission), where the 5-4 majority decision was written by Ruth Bader Ginsburg. The decision ruled in favor of the state’s voters, claiming that “the animating principle of our Constitution is that the people themselves are the originating source of all the powers of government.”
Among the states that currently use independent commissions, no two states use identical systems. The sizes of independent commissions vary from 3 (in Arkansas, strictly for state legislative districts) to 16 (newly adopted for Virginia following a state constitutional amendment passed by referendum in the 2020 election). Most of the commissions make decisions by majority vote, but super-majorities are required in California, Idaho, and in Missouri for legislative districts only. Roughly half of the states with independent commissions require bipartisan membership, with members of the two major parties chosen in equal numbers (within one) by political leaders in the state, but other states (Alaska, Arkansas, Ohio, Virginia) do not require any specific partisan balance. Several states (Arizona, California, Colorado, Michigan), but not all, require at least one commission member who is independent of both major parties. Some states have restrictions that rule out commission membership for present or near future elected office-holders, while other states specifically include particular elected officials.
In addition to the states that already have some form of non-legislature-based redistricting, citizen movements demanding or even suing for redistricting reform have sprung up in recent years in Illinois, Indiana, Maryland, Minnesota, Oregon, Pennsylvania, South Dakota, Wisconsin and North Carolina. The lawsuits were based on citizen claims that they were being denied Equal Protection guaranteed by the 14th Amendment under partisan gerrymandering. But the Supreme Court ultimately decided in 2019 that federal Courts have no role in restricting partisan gerrymandering. So the ball is in the voters’ court.
The appointment of a bipartisan independent redistricting commission is probably the easiest reform for voters to understand and approve in referendums. But it is unlikely to be the most effective reform. Eli Rosenbaum considered the question of effectiveness already back in 2005 in the Washington Post, for the three states then operating with bipartisan commissions. Rosenbaum concluded that bipartisan districting commissions did not truly address the problems: “as the current commissions demonstrate, equal representation on the panel for both parties tends to favor the status quo. The commission is indeed bipartisan, not nonpartisan, and each party’s delegates on the panel are closely connected to their state parties and politicians. To avoid gridlock and approve a plan, commissioners must draw a map that is pleasing to both sides, and of course nobody on either side really wants a competitive district. Political scientists even have a name for this type of redistricting scheme: bipartisan gerrymandering.”
While there are more redistricting commissions now than there were in 2005, they differ so much in details that it is difficult to draw firm conclusions about the efficacy of redistricting commissions in generating fairer, and especially more competitive, districts. We have the data from the redistricting in place for the 2022 midterm elections, which can be subdivided by method of drawing maps, as in Fig. III.9 from the Brennan Center for Justice at the New York University School of Law. Overall, the latest redistricting has reduced the fraction of competitive districts (won by no more than 8 percentage points in the 2020 Presidential election) from 17% to 14%. Redistricting by Republican-controlled state governments was slightly worse than the average, going from 16% to 12%. Those controlled by Democratic state governments were even worse, going from 12% to 6%, as the Democrats tried to “catch up” to the advantage the Republicans gained from their REDMAP initiative, while controlling the drawing of many fewer districts. The redistricting commissions did only slightly better than average, going from 17% to 15%, which is not an encouraging trend. The best performance in creating competitive districts come from maps mandated by State courts, increasing competitive districts from 18% to 21%.
Proportion of Competitive Districts, by Map Drawer
On the other hand, there is some evidence that commission-drawn 2022 state district maps are among the fairest ones for the 2022 election. This assessment comes from non-partisan “report cards” provided for various state district maps, on the basis of several criteria, by the Princeton Gerrymandering Project.This group uses an algorithmic approach that “generates around one million potential districting plans for each state, providing us with a baseline of what’s possible to draw in a state given its political landscape and redistricting rules.” Grades are then assigned to each state’s final district map based on several metrics. The fairness grade assesses the map’s partisan fairness and competitiveness. The competitiveness grade is based on where the state’s map falls, in terms of number of competitive districts, with respect to the distribution among those million simulated plans. Partisan fairness is judged based on the ratio of major party seat shares to statewide vote shares for the state’s map (applied to precinct vote results from the previous election), compared to the distribution among simulated maps and also to a theoretical model for that ratio.
To date, the Princeton Gerrymandering Project has completed grading for 15 states based on their Congressional district maps for the 2022 election (the grading takes significant time since a million potential district maps must be simulated for each state). Among those 15 only 4 states received a grade of A for fairness: Arizona, Colorado, Michigan, and Virginia. These are the only four states using independent redistricting commissions whose maps have yet been graded. In contrast, seven of the 15 states receive F grades for fairness: Florida, Illinois, Maryland, Ohio, Oregon, Texas, and Wisconsin. These are all states that continue to use partisan redistricting by state legislatures, and they include both Republican-controlled and Democratic-controlled legislatures. Both parties prefer safe districts that favor the re-election of incumbents! Texas is worthy of special mention here, because the Republican-controlled state legislature and Governor managed to produce a map that reduced the number of competitive districts by 10, shifting 10 (including 2 new districts) to safe red and 2 to safe blue.
To give a flavor of the Princeton Project grading, we display their results for Wisconsin in Fig. III.10 and for Arizona in Fig. III.11. In each figure, the districts themselves are color-coded to represent the estimated partisan vote share within each district, ranging from deep red (Republican) to deep blue (Democratic). Both of these states are presently considered “purple” states, which might vote Republican or Democratic in a Presidential election. But Wisconsin uses the traditional method of legislatively drawn maps with approval by the Governor, while Arizona uses an independent redistricting commission. In addition to the fairness grades, each state also receives a grade based on geographic features of the map, namely, the average compactness of the drawn districts and the number of counties that are split by district boundaries, in both cases in comparison to the ensemble of simulated district maps. We will describe compactness measures and simulation approaches in more detail below, when we consider nonpartisan algorithmic approaches to redistricting.
The data to date thus suggest that independent redistricting commissions provide some benefit over strictly partisan redistricting by state legislatures and Governors, although they still tend to favor safe districts over competitive districts because that is the preference of both parties. The Brennan Center has proposed guidelines to be adopted by all states for independent commissions. Their guidelines are perhaps most similar to the actual composition and procedures of the present California Citizens Redistricting Commission written into the state Constitution as Article XXI. The Brennan Center proposes a 15-member committee, with 5 members registered to each of the two major parties and 5 not affiliated with either, with no recent federal, state, or local officers or lobbyists allowed in the pool. The commissioners would be chosen randomly from an applicant pool — after winnowing via initial interviews to judge candidate commissioner skills and willingness to compromise, and review by the majority and minority leaders of each chamber of the state legislature — by an independent agency for the first 9 members (3 from each group), and then by bipartisan vote of those 9 members for the remaining 6. The membership as a whole should reflect the state’s geographic and demographic diversity to the extent possible and include an array of skills needed to carry out the commission’s work.
According to the Brennan Center proposal, meetings of the commission would be open and transparent, with publicly accessible agendas and results, and encouraged citizen input. With input from various sources, including citizens, the commission should construct districts that are roughly equally populated, geographically contiguous, respecting the integrity of communities of interest and the geographical boundaries of precincts to the extent practicable, and not unduly favoring any political party when viewed as a statewide map. The commission should publicize beforehand metrics by which the candidate maps will be judged and include with each proposed map a written evaluation of how well the map meets those metrics. The written evaluation should also include data on population, geography, prior election results, and any other characteristics that have influenced the choices. Proposed maps should be available for at least 14 days for legislature review and public comment, before a final map is chosen and forwarded to the Secretary of State. We will discuss suggestions for appropriate nonpartisan metrics below.
Nonpartisan algorithmic redistricting:
Independent redistricting commissions, even when bipartisan, may tend to minimize competitive districts and thereby to reduce responsiveness of Congressional (and state legislature) elections to representative voter preferences. This concern has led a number of groups to consider ways of redistricting by understandable algorithms according to truly nonpartisan criteria.
One of the most straightforward, strictly geometric, proposals is known as the shortest splitline algorithm. It leads naturally to compact, contiguous, equally populated districts with straight-line boundaries. Here is the algorithm:
“1. Start with the boundary outline [and population distribution] of the state.
2. Let [the number of districts assigned to the state] N=A+B where A and B are as nearly equal whole numbers as possible. (For example, 7=4+3. More precisely, A = ⌈N/2⌉, B=⌊N/2⌋.)
3. Among all possible dividing lines that split the state into two parts with population ratio A:B, choose the shortest. (Notes: since the Earth is round, when we say “line” we more precisely mean “great circle.” If there is an exact length-tie for “shortest” then break that tie by using the line closest to North-South orientation, and if it’s still a tie, then use the Westernmost of the tied dividing lines. “Length” means distance between the two furthest-apart points on the line, that both lie within the district being split.)
4. We now have two hemi-states, each to contain a specified number (namely A and B) of districts. Handle [each of] them recursively via the same splitting procedure [of step 3].”
The splitline algorithm is illustrated in Fig. III.12 by a side-by-side comparison of its output with the 2007 Alabama Congressional districts drawn by the Republican-dominated state legislature.
Such a strictly geometric approach to redistricting has the great advantage of completely removing partisan bias from the drawing, but the disadvantage that it may not respect precinct boundaries, geographical dividing features, communities of common interest, or the demands of the Voting Rights Act. It could be used as a starting point, with tweaks to accommodate these shortcomings. But it is becoming clear that the best chance to remove political self-preservation as the primary factor motivating redistricting, while adhering to desired nonpartisan constraints, is to adopt algorithmic approaches that employ computer simulations (which are currently used in any case to generate highly gerrymandered district maps for state legislatures with partisan control).
The development of such simulation approaches is now a burgeoning field in applied mathematics. The basic approach in these simulations has been described in a talk by Prof. Jonathan Mattingly of the Mathematics Department at Duke University. In addition to Mattingly’s group at Duke and the Princeton Gerrymandering Project, similar approaches are under development by the Metric Geometry and Gerrymandering Group (led by mathematician Moon Duchin) at Tufts University and a group at Harvard led by political scientist Kosuke Imai. The approaches used by these different groups differ in details, but they achieve largely consistent results. The steps in such simulations are described below.
- Begin by choosing a seed districting map that respects all precinct boundaries and has as close as possible to equal population among districts, with reasonably compact districts. The details of how this seed is chosen don’t matter too much, as simulations will generate as many as a million alternative district maps by deviating in some specified steps from the seed map. For example, one could start from a seed map drawn initially from the shortest splitline algorithm discussed above. But then to accommodate actual precinct boundaries, move any precincts split by those shortest splitlines into one of the initial districts that share that precinct.
- The next step is to make small changes to that initial seed map to generate a nearby map. The probability of that nearby map adhering better or worse than the seed to a set of chosen criteria can be judged by a set of predetermined metrics — such as equality of population, compactness, county and municipal preservation, and adherence to the Voting Rights Act — each of which contribute to an overall map score. For example, one could modify the seed map by moving one precinct at a time into a neighboring district and, for each such modification, reevaluate the resulting map score. One can decide to keep one of the newly generated maps or revert to the seed map, based on their relative probabilities of meeting the criteria.
- When one has allowed for all single-precinct moves, one has a final map generated from the initial seed. But now make larger, random alterations to that final map to generate a new seed map and repeat the process. Keep doing this until you’ve generated all the simulated maps you want in your ensemble. All simulated maps should respect precinct boundaries and generate nearly equally populated districts. One example showing how such simulated maps “evolve” in computer simulations is shown in the video at https://www.youtube.com/watch?v=Lq-Y7crQo44. In the case of Mattingly’s group, the alternative maps are generated using mathematical techniques adapted from the field of statistical mechanics.
- Using all the candidate district maps reached in your simulation, generate a histogram showing the percentage of maps in the ensemble that produce each possible number of Democratic or Republican Congressional seats, when each map is used in conjunction with the precinct-by-precinct vote map (as distinct from population map) from the preceding election. Examples of such histograms drawn from the North Carolina 2012 and 2016 Congressional elections are shown in Fig. III.13.
The most probable number of Congressional seats apportioned to Democrats in these simulated district maps of Fig. III.13 was 7 and 5, respectively, in 2012 and 2016, with the change between the two years reflecting the overall shift in voting preferences between the two years. And there are many possible district maps that would produce this most probable result. In contrast, the district maps generated by the North Carolina legislature in each of those years would have yielded 4 and 3 Democratic seats, respectively. As can be seen from the figure, the legislature’s maps were extreme outliers in the small tail of the probability distribution, even though they did not feature districts of egregiously distorted shape (and other outliers could be drawn with no egregiously-shaped districts at all). As such, they were rejected by state courts, which judged the maps to be drawn with the specific intent to make Black voters count as little as possible in determining the state’s apportionment of Congressional seats. The maps drawn by judges are labeled by green boxes in Fig. III.13.
Such simulations allow one to clearly identify outlier district maps. However, when these algorithmic results were presented to the U.S. Supreme Court in 2019 in support of cases objecting to legislatively drawn maps in North Carolina and Maryland (the cases are Rucho v. Common Cause and Lamone v. Benisek), the Justices decided by a 5-to-4 vote that the math was too complicated and the issue too political for them or any federal court to weigh in on, so that the Supreme Court has nothing applicable to say about partisan gerrymandering. The decision did, however, leave open the possibility for state courts to intercede in redistricting, but it remains to see whether this will happen in cases other than those that violate the Voting Rights Act, assuming that Section 2 of that Act survives the ongoing Supreme Court review.
A proposal for redistricting reform:
In many ways, the simplest redistricting reform, because it would affect all states, would be a law passed by Congress and signed by the President mandating that all Congressional districts be drawn, starting with the 2030 census and for all subsequent censuses, to be “compact, contiguous, equally populated to the extent possible, and drawn in such a way as to be consistent with the Equal Protection Clause of the 14th Amendment and the 1965 Voting Rights Act.” In order to avoid having to get federal courts involved in interpreting such guidelines, the act could furthermore specify allowed ranges for nonpartisan metrics guiding the requirements. For example:
- The population in any district should fall within ±5% of the average district population for the state.
- There are several possible metrics for compactness in use. For example, the Princeton Gerrymandering Project (see Figs. III.10 and III.11) uses a Reock value defined as the ratio of the district area to the area of the minimum circle that completely circumscribes the district. This is a value that must fall within the range from 0 to 1. The Reock value is 0.64 for a square district and 0.255 for a rectangular district with a long side/short side ratio of 3:1. To meet the requirement for compact districts, it would be reasonable to require a Reock value larger than 0.20 for every district, as this would eliminate the most egregiously elongated or distorted districts, as well as districts where contiguousness is maintained (as in Fig. III.4) by sparsely populated isthmuses joining disjoint areas. Alternatively, one can measure compactness by the ratio of the perimeter squared to the area of the district. The minimum possible value of this ratio is 4π or about 12.6 for a circular district. The value for a square is 16.0 and for a 3:1 rectangle is 21.3. The compactness requirement could be met by requiring the perimeter-squared-to-area ratio be below 23 for every district.
- In interpreting consistency with the Equal Protection Clause, we follow Jonathan Mattingly’s proposal to require a reasonable level of responsiveness in seat apportionment when statewide voter preferences shift from election to election. Responsiveness is a nonpartisan requirement much less stringent than demanding proportional or symmetric representation. As seen in Fig. III.7, responsiveness is compromised by large “cracking-packing” gaps in the vote share by district. These gaps are designed specifically to maintain the resultant seat apportionment regardless of significant statewide shifts in voters’ party preferences. Hence, rather than relying on the large computer simulations used by Mattingly’s and other groups, we suggest the following simpler approach:
- For any proposed district map, plot the projected vote share for the two major parties by district, using the precinct-by-precinct votes for House Representatives recorded in the preceding election. This should look similar to Fig. III.7, with the districts arranged in order from most Republican to most Democratic (or vice-versa).
- In that plot, measure the magnitude of the largest gap in vote share between adjacent points in the vicinity of 50% vote share. Take the ratio of that largest vote share gap to the average of the two nearest-neighbor vote share gaps. This can be done only for states with at least 4 Congressional districts. That ratio should always be greater than or equal to 1.0, but a much larger value is evidence of a gerrymandered map intended to reduce responsiveness to voters.
- On the basis of our own analysis (see Fig. III.14) of district maps drawn for the 2022 midterm election, we suggest that any map where that ratio of largest-to-nearest neighbor vote share gaps is greater than 4.0 should be suspected of engineering too much partisan advantage and too little voter responsiveness.
- Deviations from such numerical guidelines should be allowed only when they can be justified by the need to meet other mandated requirements or to allow for unusual geographic features, such as islands or long peninsulas, within the state’s boundaries.
For the analysis producing Fig. III.14, we have used the district-by-district vote share tables generated to date by the Princeton Gerrymandering Project for final district maps for the 2022 midterm election from 32 states with 4 or more Congressional districts apiece. Those tables can be generated without requiring any simulated district maps. Figure III.14 plots the number of states whose 2022 district maps have a largest district vote share gap in the vicinity of 50% vote share, divided by the average of nearest neighbor vote share gaps, between 1 and 2, 2 and 3, and so forth. 15 of the 32 states have values of that ratio below 3.0, indicative of acceptable, though not always ideal, responsiveness to statewide voter preferences. However, another 15 states have ratios larger than 4, and all the way up to 45 (!), indicating egregious discontinuities in district vote share guaranteed by gerrymandering with significant-to-extreme cracking and packing.
The most offending states in Fig. III.14, listed in ascending order of ratio value, together with the fairness grades assigned so far by the Princeton Project, are: Florida (F), Ohio (F), Illinois (F), Massachusetts (not yet graded), Virginia (A), California (not yet graded), Oregon (F), Mississippi (not yet graded), South Carolina (not yet graded), Wisconsin (F), Tennessee (not yet graded), Maryland (F), Iowa (not yet graded), Nevada (not yet graded), and Alabama (not yet graded). Not all states have yet received a grade because the grading requires the full simulation of a million potential districts for each state. Note that among these, Mississippi, Iowa and Nevada each has only 4 districts, so defining a largest vote share gap may be influenced by “end effects” in these cases. Also note that the above list includes both red and blue states; both parties prefer the incumbent “tenure” enhanced by cracking-and-packing gerrymandering.
The fact that both parties are happy to engage in gerrymandering makes it somewhat unlikely that Congress would, in fact, pass legislation of the sort we’re proposing here. But voters in each state can work to pass referendums that require either independent redistricting commissions or other citizen commissions to evaluate district maps according to these or similar nonpartisan metrics, and to report the results to the public. Statewide referendums can apply to both Congressional and state legislative district maps. In states where such commissions are not given the authority to demand alternative districting, such quantitative metrics are nonetheless useful in appealing to state courts to strike down offending maps. Of course, there is a problem with state-by-state decisions on redistricting reform: as long as any states choose to draw districts for partisan advantage, other states are dissuaded from adopting a nonpartisan technique, for fear that one party will lose too many seats by choosing a neutral path. State referendum efforts thus require some national coordination.
Redistricting reform is necessary but not sufficient:
Redistricting reform is a necessary, but not sufficient, change to address the currently unrepresentative aspects of U.S. elections. At best, redistricting reform will increase the fraction of competitive districts nationwide to perhaps 25-30%, but many districts will remain safe as a result of the geographical self-sorting of voters by their partisan preferences. And extreme candidates will still be able to survive in such districts even if they are out of step with the preferences of a majority of general election voters in the district. Reform to the types of primary elections used in a majority of states and to the way general election votes are counted – topics we deal with in Part II of this post – represent additional important improvements to bring the country closer to the Equal Protection goal of the 14th Amendment.