The Narrow Middle

How far apart the parties are

Voteview publishes DW-NOMINATE scores for every member of Congress, estimated from roll-call votes. The first dimension, which Voteview says is often interpreted as economic liberalism-conservatism, is used here. The conventional polarization measure is the Republican mean minus the Democratic mean in each chamber and Congress (Exhibit 17). The House gap was 0.52 in the 80th Congress (1947), 0.57 in the 95th (1977) and 0.91 in the 119th; the Senate gap was 0.48, 0.59 and 0.92 (Voteview). Senate independents (party code 328) are in neither mean.

The window starts near a low. In the same Voteview file, which runs from the 34th Congress, the House gap peaked at 0.84 in 1895 and the Senate gap at 0.81 in 1899, then fell to between 0.47 and 0.57 in the 77th through 80th (1941–1949). The 119th is above every earlier value except the House’s 34th, which had one Republican-coded member (Voteview). Comparison across eras assumes that one dimension describes the votes in all of them.

The common reading is that the gap widened because Republicans moved right further than Democrats moved left. The party means support it in both chambers on every base shown (Exhibit 18). From the 95th, the House Republican mean rose 0.26 and the Democratic mean fell 0.08, and the Senate means moved 0.29 and 0.05; from the 80th, the House means moved 0.24 and 0.15 and the Senate means 0.26 and 0.18 (Voteview). The Republican mean moved 3.2 (House) and 6.0 (Senate) times as far as the Democratic from the 95th, and 1.6 and 1.5 times as far from the 80th (Voteview). The 95th is the main base because the gap stayed between 0.48 and 0.60 in both chambers from the 80th through the 95th, and every later Congress is above its 95th value (Voteview).

How the means move depends on the scaling. Voteview describes NOMINATE scores as assuming a static position across a career, so a party mean changes only when different people serve; turnover accounts for all of the change by construction. Its per-Congress scores (the nokken_poole_dim1 field) let a continuing member move. On that scaling, continuing members account for 30% of the rise in the House gap between the 95th and 119th Congresses and turnover for 70%; in the Senate, continuing members account for −24%, because on net they moved slightly closer together, and turnover for 124% (Voteview). The file cannot say which scaling is right; under both, turnover accounts for most of the rise.

What the gap measures is disputed. A 2024 article in the Quarterly Journal of Political Science describes the dispute. It notes that the content of the votes can produce roll-call estimates that indicate polarization without any change in preferences, so it is difficult to tell whether rising disagreement is ideological without accounting for what is voted on.

The article tests the question with candidate surveys. Project Vote Smart’s National Political Awareness Test asked the same questions in each election from 1996 to 2008, which holds the agenda constant (working paper, 2022). In that sample, the replacement of relatively moderate legislators with more extreme ones, particularly among Republicans, explains virtually all of the recent growth in polarization (published abstract). The March 2022 working-paper version of the same study draws two conclusions that point in different directions. First, the growth in roll-call polarization “is not exclusively an artifact of a changing legislative agenda”: with the agenda fixed, a gap still appears, from the average Republican becoming substantially more conservative with less evidence that the average Democrat became more liberal. Second, the surveys show no systematic movement by members during their tenure, and it concludes that DW-NOMINATE “may overstate the degree to which the parties are ideologically polarized.” Exhibits 17 and 18 fix one score per career and cannot show within-member movement, so the second conclusion does not apply to them directly; the agenda concern applies to both.

The surveys are a convenience sample: 698 House candidates completed the survey at least twice, and those who served were about 28% of the unique legislators in the 105th through 111th Congresses; the sample’s DW-NOMINATE trend matches the House as a whole (working paper, 2022). The window ends in 2008 and covers the House only. The data cannot settle whether the pattern held afterward, whether the Senate behaves like the House, or whether the roll-call gap measures distance in preferences or the degree to which each party votes as a bloc. They do show that the Republican mean moved further than the Democratic mean in both chambers on every base shown.

The middle between them

A simple test of the middle is whether any Democrat scores to the right of the most liberal Republican in the same chamber. In the 80th Congress 72 of 194 House Democrats (37%) and 23 of 47 Senate Democrats (49%) did (Voteview). Of those, 64 House members (89%) and 18 senators (78%) represented one of the 11 former Confederate states (Voteview). The House share fell to 9% by the 96th Congress (1979) and below 1% in the 107th (2001), and reached zero in the 108th (2003). In the Senate it was zero from the 97th through 105th Congresses, one Democrat in each of the 106th through 108th, and zero since the 109th (2005) (Voteview). Exhibit 19 shows both series.

The overlap was always thin. Taken over every Democrat-Republican pair, the share in which the Democrat scores to the right of the Republican never exceeded 4% in any Congress since 1947; the peak is 3.95%, in the Senate in the 85th Congress (Voteview). Counted pair by pair, the crossings in the 80th Congress were 0.8% of House and 2.6% of Senate Democrat-Republican pairs (Voteview). The overlap sat in the tails of each party, and one member at the edge can set the threshold. In the 94th through 98th Houses a single Democrat scored 0.884, which put every Republican to the left of that member (Voteview).

The measure cannot separate shared preferences within a party from party unity in voting, the conflation described above. The absence of overlap since 2005 is a fact about how members vote. It does not say why the middle thinned.

How thin the majorities are

Exhibit 20 shows the majority’s margin at the start of each Congress. The House series is the majority party’s seats minus all other seats in the House Historian’s initial election results, which exclude Delegates. The Senate series is the majority caucus’s seats, counting independents that a publisher states caucused with the majority, minus all other seats, as printed on Senate.gov, which gives the division immediately following the election unless it notes otherwise. The Method lists seven Senate rows where the printed division is not the opening one or a stated opening division replaces it.

The House margin averaged 80 seats in the 80th through 103rd Congresses (1947–1995) and 29 in the 104th through 119th (1995–2027) (House Historian). The initial election results for the 119th were 220 Republicans and 215 Democrats, a margin of 5, the smallest of the 40 Congresses; the next smallest are the 83rd and 107th, at 7 (House Historian). CRS’s opening-day count excluding Delegates gives 219 to 215 with one vacancy, a margin of 4, on a different basis that exists only for the 119th. The Clerk’s Official List of Members of October 1, 2026 lists 218 Republicans, 214 Democrats, 1 independent and 2 vacancies, so the Republican side can lose one member and still prevail 217 to 216 if the independent votes with the Democrats.

The Senate margin has been smaller. It averaged 14.5 seats in the 80th through 103rd Congresses and 6.1 in the 104th through 119th, and was 10 seats or fewer in 15 of the 16 Congresses since 1995, against 11 of 24 before; the exception is the 111th, at 18 (Senate.gov). It is zero in four Senates: the 83rd, 84th, 107th (the first of four periods, a 50–50 tie) and 117th (the division after the January 20, 2021 swearing-in) (Senate.gov). On October 17, 2026 Senate.gov listed 53 Republicans, 45 Democrats and 2 independents who caucus with the Democrats (CRS R48535), a margin of 6.

With a separated middle, a majority with margin M prevails on a party-line vote as long as fewer than M/2 of its own members vote with the other side: with a margin of 3, one defection still wins and two lose. These data describe margins and separation, not how members vote or what Congress passes. The November 2026 election will set the margins for the 120th Congress; this study offers no forecast.

Crossover seats are the district-level counterpart of the overlap: seats one party holds in a district the other party’s presidential nominee carried. Two counts are shown (Exhibit 21), and they are different measures. Cook Political Report counted 16 members who won in 2024 in seats the opposing nominee carried, 13 Democrats and 3 Republicans, matching the 2020 count, which Cook describes as the record low. Sabato’s Crystal Ball reported on October 8, 2026 that on the maps in place for 2026 the crossover seats the Democrats are defending rose from 13 to 22 and the Republican total from 3 to 8, and attributes the change to redistricting. The study sums them to 30. The 8 include open seats and one independent-held seat, and seats defended after redistricting are not the same quantity as members who won, so 16 to 30 is not a like-for-like rise. The counts do not show why voters split their tickets.

What the public thinks

Gallup has asked whether Americans approve of the way Congress is handling its job since April 1974. Its table holds 400 readings through September 1–17, 2026 (Exhibit 22). The latest shows 16% approving, 79% disapproving and 4% with no opinion (Gallup).

Approval was 10% on April 1–15, 2026, with 86% disapproving (Gallup). Gallup said at the time that 10% was barely above the all-time low of 9% and that 86% tied the record for disapproval. The 9% reading came on November 7–10, 2013 and remains the lowest in the series; the 10% ties two readings from 2012 (Gallup). Disapproval at 86% appears in five readings: December 2011, February 2012, November 2013, November 2015 and April 2026 (Gallup). The series high for approval is 84% on October 11–14, 2001 (Gallup).

Gallup states that approval has been mostly underwater since 1974, averaging 28% approval and 65% disapproval. Since January 2010 the table holds 202 readings. None exceeds 36%, and 151 are 20% or lower (Gallup). Approval has not reached 40% in any reading since February 2005 (Gallup).

The series measures opinion of the institution. It does not separate opinion by party or identify whether respondents are reacting to the gap between the parties, to the thinness of the margins, to output or to something else. Approval and the party gap in Exhibit 17 both moved over the same decades, as did many other things, and the data here cannot say whether one affects the other.

Options

Five changes are considered here to alter who is nominated and elected. Seven keyword queries on the OpenAlex index (journal articles; titles, abstracts and keywords), as the index stood on October 20, 2026, with the first 30 results of each screened, gave 172 distinct records in all. A study is cited only where its abstract was read and tests a question named here, and each finding is stated as the abstract gives it. Most of the evidence comes from states: those that adopted primaries and commissions, and state legislatures for multi-member districts and chamber size.

Open or top-two primaries. A top-two primary sends the two leading candidates to the general election regardless of party. A 2020 study of top-two introductions and withdrawals from 2003 to 2018 finds lower legislator extremity under top-two, with some evidence of lower extremity under open primaries. Four studies qualify it: a 2010 study finds little evidence that primaries, turnout or competition are associated with congressional roll-call polarization; a 2013 study finds that primary openness has little, if any, effect on extremity; a 2017 study of California and Washington finds an inconsistent effect, partly coinciding with a redistricting commission and a term-limit change; and a 2016 experiment finds that California voters in 2012 failed to distinguish moderate from extreme candidates.

Ranked-choice voting. CRS reported on October 8, 2026 that Maine and Alaska are the only states using it for statewide and federal general elections (LSB10837). It records proponents’ claim that it reduces partisanship by encouraging a broader coalition and opponents’ claims that it burdens the right to vote and lowers turnout. A 2023 assessment of Alaska’s 2022 elections, held under a top-four primary and a ranked-choice general election together, reports more choice, more accommodative campaigning and generally more moderate outcomes than likely under the old rules, and a 2025 study of the same elections finds winners more likely to be centrists than before despite more extremists standing. A 2024 study finds that bipartisan cosponsorship rose after Maine adopted it in its swing House district but not its safe one, and calls the direction of causation open. The limits of that evidence are two states, one or two cycles, and comparison with a counterfactual or an earlier period; the 2023 assessment also reports critics’ concern that weaker parties could worsen polarization.

Independent redistricting. Commissions had primary responsibility for congressional lines in 11 of the 44 states apportioned more than one seat after the 2020 census (CRS IN11053). CRS records proponents’ belief that commissions can prevent partisan gerrymandering and may create more competitive districts, and others’ view that the effect is overstated. On competitiveness, studies from 2013, 2014 and 2023 find commission-drawn districts more competitive, the last finding commissions 2.25 times as likely as legislatures to have competitive elections, and a 2016 study finds modest gains in competitiveness and modest declines in polarization after California moved to a commission. A 2017 study comparing alternative plans with enacted maps finds that commissions give incumbents virtually the same insulation as legislatures. On polarization, a 2009 study finds very little evidence that districting causes it, a 2012 study finds small effects of partisan redistricting in state legislatures, and a 2014 study finds no evidence that commissions differ from legislatures in effects on districts or members. A 2007 study finds districts changed by redistricting became more polarized, an effect it calls relatively modest, and a 2010 study of California’s legislature associates legislative redistricting with more polarization.

A larger House. The House has had 435 seats since the 63rd Congress (1913), apart from a temporary 437 after Alaska and Hawaii became states (CRS IN11547), and CRS puts the possible size at about 50 to about 11,000 members. A larger House gives less populous districts: average apportionment population per Representative was 761,169 in 2020 against 210,328 in 1910 (Exhibit 23; Census Bureau). A 2023 study finds larger legislatures, relative to population, associated with better descriptive representation of women in cross-national data and of women and several minorities in state legislatures. Against: CRS lists chamber and office space, operating cost and possible shifts of seats and Electoral College votes among states; a 2014 study finds large state legislatures have highly partisan collaboration networks of low density; and a 2022 simulation of House sizes finds that more districts exacerbate polarization, with a diminishing marginal effect. No study reviewed estimates the effect of House size on margins.

Multi-member districts. Federal law has required single-member districts since Public Law 90-196 (1967), and House members have been elected only from them since the 92nd Congress (CRS IF12567). Bills to modify or end the mandate, including the Fair Representation Act (H.R. 3863, 117th Congress), had not advanced past introduction as of that report. For: CRS records the argument that multi-member districts, with other reforms, could give minority groups representation where they lack a majority in any single district, and a 1996 study finds that the parties in Illinois’s multi-member House were more ideologically diverse than those in its single-member Senate, a difference that disappeared once the House adopted single-member districts. Against: a 2004 study finds more ideological extremism in Arizona’s multi-member House than in its single-member Senate, and CRS records supporters’ view that single-member districts place citizens of different backgrounds in one political community and advance minority representation. CRS also records that in 1967 some civil rights supporters saw the mandate as preventing the dilution of minority votes through at-large districts and general ticket elections.

Where effects on polarization have been estimated, they are small, inconsistent or disputed. The effects measured are mostly on competitiveness or on who wins, and none of the abstracts reviewed tests an effect on House margins.