What the evidence supports

established

Political division reaches courts only once a country crosses into the most divided group

Following each country through its own history, with country and year averages swept out, being more divided gives +0.251 on how far the two sides sit apart about courts, at t 1.4. Crossing into the most divided group gives +0.855 at t 2.9, and each further year there adds +0.085. The line can sit anywhere from the 70th to the 90th percentile of world division and give the same answer, and dropping any one of the 29 countries leaves it standing.

against it Twenty-nine countries is small, the minimum-movement rule matters, and division is an expert coding of a country rather than anything respondents were asked.

the field McCoy and Somer (ANNALS 2019) theorise that polarization harms democracy by changing kind rather than degree. This measures that change of kind and separates it from degree inside the same countries, which they do not do. It SHARPENS their account.

supported

It takes about three years to reach the public

Binning years-divided as none, one to three, four to nine and ten or more gives the same shape in Europe, Africa and Asia separately and within countries: nothing in the first three years, then the gap opens. Within countries the first step is −0.024 and everything after it adds +0.238.

against it Only seven countries are observed both at nought to three years and at four or more years of their own division, and surveys arrive at irregular intervals.

the field The delay is the project's own. Nothing in McCoy and Somer or Svolik predicts a lag, and claim nine's finding that talking politics is a standing habit rather than a mood is the only mechanism offered for it.

supported

A government works through the courts before the public moves

Five V-Dem measures, reform, purge, packing, attacks and compliance, all fall inside the same window, and between 54 and 80 per cent of each fall has already happened by year three while the public gap has not moved. Measured against each country's own UNDIVIDED years, four of the five are still far below baseline after ten years.

against it All five are expert codings by one project, so a single objection hits all of them. Reform, purge and packing are rare, so these averages rest on the few country-years where something happened. An earlier version used each country's whole record as the baseline and wrongly appeared to show full recovery by year ten.

the field This is Bermeo's executive aggrandizement (Journal of Democracy 2016) given a timetable, so it CONFIRMS her. Helmke (Institutions on the Edge, 2017) reads the same behaviour as presidential WEAKNESS. Tested on her own ICLA data, 474 administration-years and 33 attacks on courts: division runs as a clean gradient across thirds (4.4, 7.5, 9.0 per cent) and raises the rate for strong and weak presidents alike, while presidential strength is not a gradient at all. Both clear significance alone and both survive together at z 1.8, so both roads are open. Her own data also shows division lifting what the executive DOES while leaving what is done TO the executive flat, and attacks landing early in a term rather than late, which are two marks against weakness as the whole story.

supported

Once a government stops obeying its courts, it does not start again

Compliance with the high court falls −0.261 (t −4.3) per point of division and compliance with the judiciary −0.259 (t −5.0), across 171 countries, and it does not recover in any bin. A claim that democracies recover better was withdrawn on 18 September 2026 once it was traced to a demeaning artifact.

against it V-Dem compliance may partly record courts avoiding confrontation rather than governments defying orders, which this cannot separate.

the field Staton and Vanberg (AJPS 2008) model compliance as depending on the public imposing costs for visible defiance. If a polarized public will not, the pressure that would restore compliance is gone, which gives the irreversibility a mechanism. The irreversibility itself is new.

suggestive

This happens to a particular kind of country, and there are two kinds

A latent profile analysis on country features returns two classes, with 20 countries in the pressured group. The United States belongs to it at 92 per cent.

against it Class counts depend on the parsimony rule, and a shuffle test on scrambled features is the only check run.

the field UNDER-ANCHORED. No literature was found establishing that this statistical grouping corresponds to any typology in the comparative backsliding literature. This is the weakest of the nine on prior work.

established

Party feeling moves first, and views of the court follow

Following the same Americans from 2020 to 2024, partisan feeling in 2020 predicts the 2024 Supreme Court rating (+0.357 for Republicans, −0.178 for Democrats) while the 2020 Court rating predicts almost no change in partisan feeling.

against it Four years is a long lag, panel attrition is not random, and a cross-lagged coefficient is a prediction rather than an effect.

the field This is the Gibson and Nelson versus Bartels and Johnston dispute, settled by events in Levendusky et al. (Science Advances 2024) and Davis and Hitt (AJPS 2026). The within-person direction test is the project's contribution, not the divergence.

supported

What decides which side trusts a court is who holds the court

The United States and Israel are the two cases where the court and the government came apart, and in both the public took its cue from the court rather than the government. In the United States the gap changed sign from 2021; in Israel it runs the other way round from the start.

against it Two cases, chosen because they are the ones where the two came apart, so this interprets rather than tests.

the field Consistent with the legitimacy literature's positivity and diffuse-support tradition, and with Böhringer and Boucher (2024), who find the winner-loser gap on courts smaller than on elected bodies across 34 countries.

supported

Living in a polarizing place does not change a person. Arguing about politics at home does

Following 2,839 Americans from 2016 to 2024, no state measure predicts how far a person moved (every t at or below 1.4, only 11 per cent of the variation between states), while days a week talking politics with family or friends gives +6.6, +7.6 and +13.3 across bands, and +3.28 per standard deviation at t 6.3, surviving partisan strength, the 2020 baseline and family strain at +2.72.

against it The state windows and outcome windows do not line up, the panel has no name generator, and the discussion item counts conversations without describing who is in them.

the field Mutz (Hearing the Other Side, 2006) finds cross-cutting exposure increases awareness of the other side's rationale. That is about WHO a person talks to; this measures HOW MUCH. The two are not the same claim and this does not test hers.

suggestive

Before polarization, talking to the other side pulled people toward the other side's court

In the 1987 General Social Survey, the only American survey ever to ask about discussion partners' party and confidence in the Supreme Court together, Democrats with Republican partners had MORE confidence in a Republican-appointed Court (+0.221, t 3.1) and Republicans with Democratic partners had LESS (−0.201, t −2.5). Congress is flat for both, so this is not general trust.

against it One cross-section, 1,327 people, nothing here is a change or a cause, and an alter's party is reported by the respondent.

the field This CONFIRMS Mutz on an outcome she did not use. It should be presented as her mechanism measured on courts rather than as new. Whether it still holds needs the unreleased ANES-GSS 2024 Joint Study.

established

Courts divide a public less than elected institutions do

Across 30 European countries the average left-right gap in trust is 0.67 points for the legal system, 0.81 for parliament and 1.60 for satisfaction with government. The same ordering appears in Latin America and in the values surveys. This is also the one claim here that was already published, by Böhringer and Boucher in 2024 across 34 countries, and our data reproduces it independently.

against it It says nothing about change over time and nothing about causes.

established

Courts are the last institution a public divides about, and usually it never does

Of 19 Latin American countries over 30 years, the public ends up divided along political lines about the government in 15, the presidency in 12, and the courts in 3. In 21 European countries over 23 years the legal system divides a public in 4, against 14 for satisfaction with government. Ranking institutions by the year they divide places courts later than six of seven others in Latin America and seven of ten in Europe.

against it Onset dating needs a long series, so this rests on the two sources that have one.

established

A public divided about its courts stays divided

Of the states a country can be in, the one where the public is split about its courts is the hardest to leave, in both regions. Spells last about twice as long as any other kind of division. Shuffling each country's own states 2,000 times, keeping its exact mix and destroying only the order, never reproduces the observed persistence in either region.

against it Smoothing does part of the work: without it, courts are the most persistent of the DIVIDED states but not of all states. And it is not about courts alone. In Europe there is not one country-year in which a public is split about its courts and united about its politicians, parties, parliament and government.

supported

In the United States, and almost nowhere else, the Supreme Court was absorbed into party conflict

Between 2018 and 2024 the partisan gap in saying you have hardly any confidence in the Court moved 30 points. The next largest movement in the same direction among eleven other institutions asked of the same Americans in the same interview is organised religion, at 2. Congress and the executive branch move the opposite way.

against it A biennial series cannot attribute a change to an event, and the 2021 and 2022 rounds differ in fielding after the survey's pandemic redesign.

supported

Canada does not follow the American pattern

The Liberal-Conservative gap in confidence in courts moved 6 points between 2019 and 2025, against 30 in the United States, in the opposite direction, and less than Elections Canada at 7. A cluster of federally appointed bodies drifts together rather than the judiciary being singled out.

against it The Liberals governed throughout, so the party gap is also a government-opposition gap.

supported

At the extreme of polarization, governments stop accepting outside advice about their courts

In the Universal Periodic Review, recommendations about a state's own courts are accepted about 6 points more readily than others. Above the 80th percentile of polarization that advantage disappears, at -0.08 with a t of -5.33. The same test on seven other subjects finds nothing comparable, including women's rights and children's rights with over 13,000 recommendations each.

against it It is a threshold rather than a gradient: across four fifths of the range nothing distinguishes courts. Cross-sectional, and no causal claim is available.

suggestive

Where a public is more ideologically spread, it trusts political institutions less, courts among them

Across 19 Latin American countries with up to 24 waves each, the correlation between the spread of left-right placement and the level of trust runs −0.29 for government, −0.28 for parties, −0.23 for parliament and −0.17 for courts, all with intervals excluding zero. Church, military and police show nothing.

against it The measure shares respondents with the outcome. A direct test of that contamination has not been run.

not supported

Polarization erodes public trust in courts

Across 71 countries in the Gallup World Poll the within-country correlation is −0.04. Across 32 countries on the expert measure the mean correlation with the spread of court attitudes is +0.06, interval −0.10 to +0.21, and courts sit third behind hospitals and schools. Confidence in courts ROSE in most of the world over the period: the median change across 61 countries is +7 points and only 14 fell.

against it The United States ranks second of 61 on the size of its fall, so the general null and the American case are both true.

not supported

At the extreme of polarization, publics divide further about their courts

The same threshold test that works on governments finds nothing in the survey data. Across 5,711 country-year-institution observations in 50 countries the change in the court gap above the 80th percentile is +0.04 with a t of 0.5. What does widen sharply is division over satisfaction with government, satisfaction with democracy and the presidency; what narrows is division over the United Nations, the European Parliament, the church and the army.

against it This is the clearest disagreement between the two halves of the project, and it is informative: extreme polarization changes what governments will agree to about courts without changing what their publics think of them.

not supported

Polarization presses harder on institutions that check a government than on those it uses

With both groups widened so the comparison is fair, the difference between them is −0.004 with a t of −0.7 on one polarization measure and −0.005 with a t of −0.9 on a second, independent one.

against it An earlier version of this page claimed the opposite, on a narrower set of institutions.