Perceptions of courts under polarization

What this project found

One question runs under all of this. What actually moves how a court is seen?

The answers here are specific, and several of them are not what you would expect. Partisan feeling moves it, and moves it first. Who appointed the court's majority moves it. Political division moves it, but only once a country is among the most divided, and only after about three years. What a government does to its courts turns out NOT to move the public at all, which is a result in its own right. And nothing global is moving it, because there is no global trend to move.

Each one is a sentence, a number taken from a published table rather than typed by hand, and a link down to the section that shows how the number was got.

Ten questions, and what this project can answer

Every question below is about courts and judges, and each one is answered with what this project measured rather than with an opinion about it. This page reports about thirty analyses of thirty datasets, arranged by where the data came from. That is how the work was done and it is a poor way to read it. Below are the questions the work was meant to answer, each with the answer in a sentence, the number it rests on, and a link to the section that shows how it was got. Several of the ten answers are narrower than the question, and one of them holds for a single country.

Two words appear throughout and mean particular things. The gap is the difference between two groups of people in how much they say they trust an institution: usually supporters of the governing party minus supporters of the opposition, sometimes people on the right minus people on the left. A gap of +0.35 on a 0 to 10 scale means the first group gave answers about a third of a point higher on average. The level is how much a whole public trusts an institution, with no group comparison in it. A country can have a rising level and a widening gap at the same time, and several do.

Every number above is read from the published tables rather than typed in, so an answer cannot drift away from the evidence under it. None of these are causal claims. The surveys differ in wording and scale, the countries in each answer are not the same countries, and the sections underneath say what each one can and cannot support.

Becoming more divided does nothing. Crossing over does.

This is the sharpest thing on the page about polarization itself, and it separates two ideas that get used interchangeably.

A country can be somewhat more divided this year than last. Or it can pass into the group of countries whose politics has genuinely split into hostile camps. Those are not the same event, and only the second one reaches the courts.

Every other threshold result here compares different countries, putting Hungary next to Denmark. This follows each country through its own history, with country and year averages swept out, so nothing permanent about a place and nothing happening to the whole world in a given year can produce it.

Read the first column against the second. Ordinary increases in division do almost nothing to how a public sees its courts, at +0.222 with a t of 1.2. Crossing the line moves it by +0.809, and every year spent on the far side adds +0.086 more.

Parliament moves most, courts next, the police least. That is the ordering this project keeps finding: courts are pulled into the fight alongside everything else and somewhat less far than the elected body. The police moving too is worth knowing rather than worrying about. Division reaches all three.

Who is in this, and why

Most countries in these surveys never polarized. The median one's own division score moved about three quarters of a point across every year it was surveyed, on a scale roughly eight points wide, and thirty moved less than half a point. A country that never polarized cannot say what happens when a country polarizes, so the test runs on the 29 whose division moved by at least a full point.

That drops countries where the change being tested never happened, rather than countries where the answer came out inconvenient. Put the others back in and the estimate for courts falls to +0.419 with a t of 1.8 across 88 countries, which is what diluting a sample with cases that cannot inform it should do.

What it rests on

Where should the line between "divided" and "deeply divided" go? The table moves it and reruns the same test each time. The second column is the V-Dem score a country has to pass.

Drawn at the most divided third, the line shows nothing for courts (+0.19, t 0.4), while parliament already moves. From the most divided 30% inward, courts move by roughly the same amount wherever the line sits. The change is clear at every line in that range except the most divided 17.5% and the most divided 15% and the most divided 12.5% and the most divided 10%, where it is about the same size but less certain; a higher line leaves fewer years above it to compare. Courts need a higher line than parliament before they move. This page keeps the most divided quarter, the line it has used throughout; choosing a new one after seeing this table would fit the line to the result.

And no single country carries it. Dropping each of the 29 in turn leaves the estimate between t 2.0 and t 3.7, with 28 of 29 staying above 2.

Twenty-nine countries is small. The minimum-movement rule matters: at half a point the estimate roughly halves, and at a point and a half only fourteen countries remain to measure anything with. Division is an expert coding of a country rather than something the respondents were asked, surveys arrive at irregular intervals, and nothing here identifies a cause.

The OECD survey: courts lose the most when a country divides

What the survey is

The OECD asked 195,000 people in 43 countries how much they trust fourteen institutions, each on a scale of nought to ten. Courts are one of them. So are parliament, the government, the police, political parties, the news media and the army. Everyone answered about all fourteen in one sitting, so they can be lined up against each other fairly. There are three rounds, in 2021, 2023 and 2025.

It also asked each person whether the party they voted for is currently in government. Three quarters answered. That splits every country into election winners and election losers, and lets us ask whether the two groups see the courts differently. Everywhere else on this page that split has to be built by matching parties to cabinets. Here people were simply asked.

What "divided" means

That comes from a separate source. V-Dem is a project in which country experts score every country every year on how far its politics has split into camps that treat each other as enemies. Nobody in the survey was asked it. It is a judgment about the country, matched to the year each person was interviewed, so a country can be calm in one round and divided in the next.

A country counts as divided in a given year if it sits in the worst quarter the world has produced, across every country and year V-Dem has scored since 1975. It is the same line used everywhere else on this page.

13 of these countries are divided in at least one round: Poland, Brazil, Romania, Peru, Mexico, Slovenia, Colombia, Korea, France, Slovak Republic, Bulgaria, Ecuador and Croatia. 29 are calm in every round, and the calmest are Ireland, Norway, Luxembourg and Japan. Colombia, France, Korea and Slovak Republic change from one to the other between rounds. Canada is calm in every round, and not narrowly.

Courts change more than anything else on the list

For each institution, this measures how far apart election winners and election losers sit. Nought would mean they feel exactly the same. The table compares calm countries with divided ones.

Parliament is always fought over. Winners like it and losers do not, in Ireland as much as in Poland, and becoming divided hardly changes that. Courts are normally not like that, because both sides accept them. In divided countries that acceptance is what goes, and courts move further than any other institution in the survey.

That does not mean courts are fragile. Courts have something parliament does not, which is acceptance from both sides, and division is what takes it away. Parliament has nothing of the kind to lose.

Two checks. Taking out each divided country one at a time, the change for courts stays between +0.26 and +0.42, so no single country produces it. And four countries were surveyed once while calm and once while divided, which lets each be compared with itself:

In all four, winners and losers sat further apart about the courts in the divided year.

Would the court stand up to the government?

The 2021 round asked something sharper than trust. If a court had to rule on something that would embarrass the government, how likely is it to rule freely anyway? Election losers are more doubtful than winners, by 0.82 points. On plain trust in the courts the difference is 0.80, which is the same. On parliament it is 1.35. Losers are no more cynical about whether a court will stand up than about the courts in general.

Feeling discriminated against divides people about courts more than their vote does

The 2023 and 2025 rounds asked whether a person would describe themselves as a member of a group that is discriminated against in their country. One in five say yes. They trust the courts 0.97 points less than everyone else does.

Set that beside the difference between winners and losers, in the same countries and years:

For the government and parliament, who you voted for makes the bigger difference. For the courts and the police, the two institutions that apply the law to a person, feeling discriminated against makes the bigger difference. On the courts it is the larger of the two in 41 of 68 country-years.

Canada showed the same thing on its own, with confidence in the courts running from 22 per cent among First Nations respondents to 71 per cent among recent immigrants, about twice the distance between Liberal and Conservative voters. This is the first time the pattern could be checked across many countries at once.

Most of these countries are established democracies, so the divided group is small: 18 of 89 country-years. The rounds cover four years, which is too short to watch most countries change. A test of whether small year-to-year movements in division go with a wider gap finds nothing for courts, and finds the gap slightly narrower for the national government and international organisations; over four years that measures small wobbles rather than a country crossing into division. The "stand up to the government" question was asked in 2021 only. "Discriminated against" is how people describe themselves, and it pools very different groups. Nothing here identifies a cause.

Talking politics at home changes how a person sees the Court, and the state they live in does not.

The sections above follow countries. This one follows people. The American National Election Study interviewed the same 2,839 people in 2016, again in 2020 and again in 2024, and asked all three times how warmly they felt toward the Supreme Court. Because it is the same person each time, nothing fixed about who they are can produce what follows.

Generic division does nothing. Division about courts does.

This is the same result the cross-country work keeps returning, found again inside one country and at the level of a single person.

Sorting the states by how much their politics divided between 2018 and 2020, and then measuring how far each person's rating of the Court moved between 2020 and 2024:

The two sides pulled 15 to 17 points apart everywhere. Where the state was dividing fastest they pulled apart slightly less. Nothing about how divided a state was, or how fast it was dividing, predicts how far a person moved, and only 11% of the variation in how far people moved sits between states at all.

One thing about a place does reach a person, and it is specific to courts. Sorting the states instead by how far apart their Republicans and Democrats already sat on the Supreme Court in 2020, and measuring how people moved on a different question: whether a president would be better off not having to worry about Congress and the courts.

People living where the two parties were already furthest apart about the Court moved about four times as far toward wanting a president who does not have to answer to courts. The survey item runs the other way round, from wanting that president to refusing him, so the regression prints as -0.064 with a t of -2.2. The middle and top thirds sit together, so read this as a step rather than a slope, on three bins.

General political division leaves a person's view of judicial power alone. Division that is already about courts does not.

Talking about it moves people

Because a raw average cancels, this measures how far each person moved in their own party's direction: the change in their Court rating, multiplied by plus one for Republicans and minus one for Democrats. A positive number means the two sides pulled apart.

Against how many days a week the person discussed politics with family or friends:

Someone who talked politics nearly every day moved about twice as far as someone who barely discussed it. In regression that is +3.98 points for each standard deviation, with a t of 7.7.

This is not another name for being a strong partisan. Holding how strong a partisan the person is, where their rating started in 2020, and how much politics had already damaged their family relationships, the estimate falls only from +3.98 to +3.43, with a t of 6.5.

The habit is a standing one

The same question was asked again in 2024, which separates a settled habit from whatever the person happened to be doing at the end of the window. Both years carry independent weight, and the four kinds of person the two readings make say it plainly:

Talking in both years moves a person nearly three times as far as talking in neither. Talking in only one of the two years lands in between, whichever year it was.

The relationship runs in both directions. People who thought less of the Court in 2020 went on to talk about politics more by 2024, holding their party and their earlier habit. In units of each outcome's own spread, talking leading opinion is 0.123 and opinion leading talking is 0.057, so the heavier half of it starts with the conversation.

The place is not working through the conversation

If living somewhere divided reached people by giving them more arguments to have, the state measures multiplied by how much a person talks would do something. They do not: every one of those terms sits between t −1.0 and t +0.1. Talking carries its own weight wherever a person lives.

Where this fits

This is a mechanism for the country-level findings above. Crossing into deep division takes about three years to reach a public, and does nothing at all in the first three. A shock cannot produce a delay like that. An accumulating habit can. The people who moved furthest here are the ones who talked politics in both 2020 and 2024, and talking in only one of the two years moves a person about half as far. Division appears to reach a court through years of ordinary argument rather than through the moment a country crosses the line.

What this means for a court. A court's standing is not decided by the map. Two people in the same state ended up far apart from each other depending on how much politics was in their kitchen conversations. Geography will not tell a judge which communities are turning against the courts.

The state measures span 2018 to 2020 while the outcomes span 2020 to 2024, so the predictor comes first but the windows do not line up. A state is a weak stand-in for a person's surroundings, because people in one state can live in opposite worlds. The panel has no name generator, so the discussion item counts conversations without describing who is in them. Panel attrition is not random, four years is a long gap, and nothing here identifies a cause.

Countries like Canada almost never do this

Canadian judges asked the sharpest version of the question on this page. Among countries like ours, when they polarize, is the damage to how courts are seen worse than elsewhere? The answer is that the event is so rare among those countries that it has to be told as cases rather than estimated as a number, and that is itself the finding.

Who counts as a peer

A peer here is a long-established liberal democracy with strong courts: V-Dem polyarchy of 0.75 or better, liberal democracy of 0.70 or better, and high court independence of 1.5 or better, all three held for at least 25 years running and still holding today. That gives 19 countries.

Australia, Austria, Canada, Costa Rica, Denmark, Estonia, Finland, France, Germany, Iceland, Ireland, Lithuania, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Sweden, Switzerland

Canada has met all three for 51 years, the longest run in the data.

Of those 19, 1 has ever crossed into deep division since 1990: France, in 2023. That is too recent to say what became of French courts afterwards.

Every crossing in the world since 1990

The other way to see how rare this is: take all 130 crossings anywhere in the world since 1990 and ask how long each country had been a settled liberal democracy beforehand.

Three crossings in thirty-five years happened to a country with a quarter-century of settled liberal democracy behind it. The overwhelming majority happened to countries that had never been in that condition at all.

Note who is not in the first row. Hungary and Poland, which everyone reaches for, had about nineteen years behind them when they crossed. Canada has 51.

The United States is the case, and it just left the group

The United States met all three conditions every year from 1975 to 2024, a fifty-year run alongside Canada's. It crossed into deep division in 2016, with 41 of those years behind it. In V-Dem's 2025 coding it no longer meets two of the three: polyarchy at 0.735 and liberal democracy at 0.571, both under the floor. Its high court independence has fallen from 2.63 in 2016 to 1.55, which clears the bar by a hair.

So for a Canadian audience the American experience is not a distant cautionary tale. It is the only completed case in the peer group, and the gap there ran about +4 points for 28 readings across 43 years, held near +5 through the first three years after 2016, and reached −19 by 2024.

Does this depend on where the lines are drawn?

Moving the requirement from fifteen years to thirty-five changes the size of the group from 21 countries to 16. It does not change the answer: at every one of those settings, exactly one peer has ever crossed.

Four V-Dem thresholds are not a country. Canada's federalism, its appointment process and its two legal traditions are not in this and would matter to any lawyer. Every measure here is one project's expert coding, the 2025 figures are a single year and may move in the next release, and a country that crossed in 2023 has almost no record after it. Counting how often something has happened is not a forecast of whether it will.

France: what happened when a country like Canada divided

Why France

Canada's peers are the nineteen countries that have been settled liberal democracies with strong courts for at least twenty-five years. Since 1990 exactly one of them has crossed into deep division, and it is France, in 2023. It had been a settled democracy for forty-two years when it did.

Hungary and Poland are the examples people usually reach for, but they differ from Canada in two ways. They had about nineteen years of settled democracy behind them, not fifty. And in both, the government went after the courts. In France it did not, on the experts' own record. So France shows what division does to a country's courts when the government is not the one pushing.

Canada and France side by side

V-Dem's experts score every country's courts every year. Averaged over 2015 to 2025:

On every measure of the courts, the two countries are close to each other and far above the rest of the world. France's government has left its judiciary alone at least as consistently as Canada's. The one large difference is that France became divided. That makes France close to a controlled comparison: it shows what division does to a country like Canada when the government is not moving against the courts.

How France got there: slowly

V-Dem's experts score every country every year on how far its politics has split into camps that treat each other as enemies. France did not jump. It climbed over more than a decade.

The line that counts as deeply divided sits at +0.8. Canada, for comparison, was at −0.4 in 2025.

What the government did to the courts: almost nothing

On the same experts' codings, French high court independence stayed at 1.8 every year from 2005 to 2025, and the government's compliance with the high court stayed at 2.1. No judges removed, no court packed. Asked each year how often the government attacked the judiciary's integrity in public, the experts' answer for France falls in the band "there were no attacks on the judiciary's integrity" in every year. Their score did edge toward attacks in 2024, from 2.3 to 1.8, without leaving that band.

What the public did

The European Social Survey has asked French people about the legal system eleven times since 2003, and each time we can compare people who back the governing parties with people who back the opposition. The last column is how far apart they are.

In 2023, the first year France was over the line, the two sides were further apart about the courts than in any earlier survey, at 1.22. Before that it averaged 0.56. One earlier year came close, 2014 at 1.19, so 2023 is the high point of a jumpy series rather than a clean break.

Look at how the gap opened. Government supporters became more confident in the courts, from 5.7 to 6.2, the highest on record. Opposition supporters barely moved, 5.0 to 4.9. The gap opened because one side warmed to the courts while the other stayed where it was.

The OECD survey shows the same thing from a different angle, with the courts beside other institutions. How far apart election winners and losers sat:

Everything spiked in 2023 and came back in 2025, the courts included. And across all three rounds, overall trust in the courts held steady: 4.9, 5.3, 4.9 out of ten.

Left, centre and right: the right has lost confidence

The two surveys above compare supporters of the government with its opponents. In France the government is Macron's centre, so the opponents mix the left and the right, and a change on one side can be hidden by the other. The Eurobarometer asks French people whether they tend to trust the justice system, and splits them by where they place themselves, left, centre or right:

People on the left and in the centre trust the justice system more than they did twenty years ago. People on the right have gone the other way. Between 2014 and 2019 about 43% of them tended to trust it; in 2025 it was 30%, and in 2026 36%. The 2025 reading came in the year a French court barred Marine Le Pen, the right's leading candidate, from standing for office. These data cannot say whether that is the reason.

So the French public as a whole has not lost confidence in its courts. One part of it has.

And asked why they rate their courts' independence as they do, French people who think the courts are independent still credit the status of judges, at 80 per cent in 2024, about where it has been since 2016. Those who think they are not independent blame politicians, at 74 per cent.

What to watch

Across many countries, this project finds that crossing into deep division does little to how a public sees its courts for about three years, and then the two sides pull apart and keep pulling. France crossed in 2023. On that timetable the readings here come too early to show the effect, and the years to watch are 2026 onward.

Across many countries this project finds that governments usually move against the courts before the public splits over them. France had no such move, and its public still split in the crossing year. So the public side does not need the government side to start it. That is one country, read early, and it is the point most relevant to Canada.

France is the one live test in Canada's peer group, and it is early. So far: a long, slow slide into division; a government that has left the courts alone; a public that pulled apart in the crossing year and then partly came back; and no fall in overall trust.

One country is one country. The divided years are few, 2023 onward, and the surveys that cover them are one ESS round and two OECD rounds. The division score is an expert judgment about France, not something French respondents were asked. The ESS series is jumpy, so a single round should not be read on its own. Nothing here identifies a cause.

How much division does it take to move views of a court?

A great deal. Far more than most people assume.

In a country with ordinary political disagreement, the distance between how the two sides see their courts does not widen measurably at all. It starts widening once the country is among the most divided in the world, and then it widens sharply.

That is a usable answer rather than a general warning. It says which countries a judge should be looking at, and it says that movement within the normal range tells you nothing. The evidence is four regions, four surveys and four different ways of separating a country's two political sides, each tested on its own.

Every slope is positive, so division does move views of courts everywhere. But 4 of these 5 rows put most of that movement into the most divided group alone, where an even slope would put a quarter of it. The shaded rows are the ones where chance produces a jump that big less than one time in ten.

For a court, the practical reading is that the ground stays firm across most of the range and then gives way. There is no gradual slide to watch for and act on.

The Americas is the exception, at 30% in the last fifth, which chance produces 18% of the time. Two things are worth saying about it rather than one. Courts there do move in the same direction as everywhere else, so they are not exempt. But on the same country-years the legislature, the police and the armed forces all show the concentration at the top and courts do not, so in The Americas division appears to widen arguments about every institution at once rather than to single courts out.

Staying divided, not just being divided

A country that has been split into hostile camps for twenty years is not in the same position as one that arrived there last year, even if both score the same today. So the question is whether the number of years already spent divided does anything on its own.

Europe answers it and the other regions cannot. Comparing a country with itself, each additional year already spent among the most divided adds about +0.058 to the court gap on a 0 to 10 scale. The two European cuts are built from different variables and land on +0.060 and +0.058, which is the kind of agreement that is hard to get by accident. Ten years of it is worth more than the whole step between the fourth and the fifth group in the table above.

Africa and Asia show large duration effects when countries are compared with each other and nothing at all when a country is compared with itself. That is the sign of a variable naming a few countries rather than measuring a duration: in Asia only 5 of 15 countries ever cross the line, and one of them sits at 41 years while nothing else is near it.

Nothing here identifies a cause. The size of a gap cannot be compared between rows, because the four surveys use different wordings and scales; only the sign of the slope and the shape can be read down the column. Fifths of twenty to sixty country-years cannot separate a cliff from a steep curve. The European duration estimate rests on the handful of countries that actually crossed the line during the study, which is a real test on a small number of cases rather than a continental fact. The division score is an expert coding of a country rather than a measurement of the people interviewed.

Do governments stop obeying their courts?

Yes, and the damage starts earlier than the public's does. Everything above measures what people think about courts. This measures what happens to the court.

V-Dem codes, for every country and year since 1975, whether a government complies with judicial decisions and with its high court in particular. As a check on which end is which: the top of the 2020 distribution is Denmark, Switzerland and Australia, and the bottom is North Korea, Equatorial Guinea and Venezuela.

More divided, less obeyed

Comparing a country with itself, with country and year averages swept out, so nothing fixed about a place can produce it. A country that becomes more divided is a country whose government complies less with its own courts.

A gradient, where the public gave a cliff

This is the one place on the page where the shape is a straight line. The gap between a public's two sides is flat until a country is among the most divided and then jumps. Compliance falls from the first fifth onward, -27% of it in the last step where a straight line puts 25 per cent. A public waits for the extreme. A government does not.

The order, which is the point

Each country against its own average, by how many years it has already been divided. In the first three years the government attacks its courts and stops complying with them, and the public gap has not moved at all. The public arrives in years four to nine, after the damage.

So the sequence runs: a country divides, its government turns on its courts and starts ignoring them, and only then does its public divide about them. A divided public is downstream of a defiant government rather than its cause.

What would strengthen it

The measures above are both V-Dem readings of the same country, which is ordinary practice and carries an ordinary caveat: one team is coding both. Replacing the division score with a court gap measured by a survey removes that, at the cost of most of the sample. The signs hold and the estimates are not significant on 528 country-years. A larger independent test would settle it.

Compliance is an expert reading of a country rather than a count of defied judgments, so it carries whatever the coders thought counted. Nothing here identifies a cause, and an order in time is not one. Countries already divided when the record opens in 1975 have no crossing and contribute nothing to the timing table.

What a government actually does to a court

Ignoring a ruling is one option. There are others, and they are rarer and worse. A government can rewrite the rules the judiciary runs by, remove the judges, or fill the bench until it gives different answers.

These are extreme things and most countries in most years do none of them. So the numbers below are carried by a small number of episodes, and the episodes are worth naming before the numbers are read. Each measure is scored by experts for every country and year since 1975, and the country-years at each end name themselves.

Those are the events the scale is made of. Hong Kong in 2020 is the national security law. Malaysia in 1988 is the sacking of the Lord President. Cambodia in 1976 is the Khmer Rouge, and Turkey in 2016 sits just above it. Peru in the mid-1990s is Fujimori. At the other end, Tunisia in 2014, Colombia in 1991 and Hungary in 1990 are new constitutions building courts up.

All of it gets worse as a country splits

Each country compared with its own past, so nothing permanent about a place explains it. Everything on the list moves the same way, and removing judges moves most.

And it all happens in the same three years

Read the last column. Between 50% and 80% of each fall has already happened by the end of year three. Over the same three years the public's split about courts has not moved at all.

And it does not wear off. Each measure here is set against that country's own years of not being divided, so nothing recovers by arithmetic. Four of the five are still far below where the country started after ten years, and the government obeying its high court is at its worst there. The one that does return to normal is attacking judges in public, which makes sense once the courts have already been reformed, purged and packed.

An earlier version of this section used each country's whole record as the baseline and appeared to show all five measures recovering almost completely by year ten. They do not. A country that reaches ten consecutive divided years spends about half its record divided, so an average taken over the whole record is itself half made of divided years and forces late values toward zero. The ten-year bin is not thin: it holds 1,006 country-years across 67 countries. Anchoring on undivided years instead drops five countries that were divided throughout and 238 of 8,325 country-years.

A government does several of these things inside the same three years. It changes the rules the courts run by, removes judges, fills the bench, attacks judges in public and stops obeying rulings. The public's split about courts comes afterwards, when most of that has already happened.

Rewriting a judiciary, purging it and packing it are rare, so most country-years sit close to their own country's normal level and these averages are driven by the few where something happened. Every measure here is an expert reading of a country by one project, which means the same objection applies to all of them at once. A country already divided when the record opens in 1975 has no crossing and contributes nothing to the timing. Nothing here identifies a cause, and an order in time is not one.

When a government attacks its own court, the court usually loses

Everything else on this page measures that something happened to a court: a rule rewritten, judges removed, a bench filled, a ruling ignored. None of it says how the fight ended. For a judge that is the question, because being attacked and surviving is a different world from being attacked and falling.

Gretchen Helmke read every issue of the Latin American Weekly Report from 1985 to 2008 and coded every episode in which one branch of government tried to dissolve, replace or reduce the powers of another, across eighteen countries, recording who attacked whom and how it came out. That last part is what no other source in this project has.

A court attacked by its own executive fell in 25 of 33 cases. It is the likeliest of the four to lose. The pattern underneath is simple: when the executive attacks, it usually wins; when the legislature attacks, it usually loses. A legislature going after the courts succeeded about a third of the time, and so did a legislature going after the president.

Division lifts what the executive does, not what is done to the executive

Putting the same country-years beside this project's division measure separates two things that get run together. The rates are the share of administration-years in which each attack happened.

Read the last row against the first. Where a country is divided, the executive attacks its courts and its legislature more. Attacks on the president do not rise at all. That matters for a live disagreement: Helmke's own reading of these crises is that presidents attack because they are weak and vulnerable, and on that reading a weak president in a divided country should also be attacked more often. He is not.

Only 82 of the 474 administration-years fall in a divided country, so the right-hand column rests on very few events and is a direction rather than a rate. The counts are printed for that reason.

And it comes early

Attacks on courts concentrate in the first three years of a term and fall off afterwards. A government working through a programme should move early, while it has time and a mandate. A president lashing out from weakness should be likelier to do it late. This is the first shape.

Worth keeping straight: this clock counts years into a president's term, while the three-year window elsewhere on this page counts years since a country crossed into division. They are different clocks that happen to point the same way, and neither confirms the other.

Thirty-three executive attacks on courts and eleven legislative ones is few, so every figure here is a proportion of a small count. Eighteen Latin American countries from 1985 to 2008 is one region and one period, and presidential systems are not parliamentary ones. A crisis is coded from what one weekly publication reported, so it records reported episodes rather than every episode. Nothing here identifies a cause.

When does a government attack its own judiciary in public?

This section treats one thing as the outcome to be explained: a government publicly calling its own courts corrupt, incompetent or politically motivated.

What is measured

V-Dem's country experts answer, for every country every year: how often did the government attack the judiciary's integrity in public? That includes statements, press conferences, interviews and speeches. It covers 182 countries from 1975 to 2025. It measures attacks by the government only, not by opposition politicians, the press or members of the public.

Governments in divided countries attack their courts more

Sort every country and year into four groups by how divided the country is, from calmest to most divided, and ask how often each group lands among the worst quarter of years on record for attacks:

It rises at every step. It also holds when each country is compared only with its own past, so it is not just that some countries are always rougher than others.

Which things predict it

Each tested on its own, every country against its own history. The last column is how clear the link is; above 2 is a clear link.

The strongest link is with how deeply society is divided on the big political questions. A more corrupt or less democratic government attacks its courts more too. How strong the court is makes no difference, and neither does how many years a country has been divided.

It feeds itself

More division now goes with more attacks three years later. More attacks now also go with more division three years later, by about the same amount (+0.14 and +0.15). Neither simply causes the other; they push each other along.

When it happens

For countries that crossed into deep division, compared with their own years well before it:

Attacks begin rising before the crossing, peak in the crossing year and the three years after, stay high for several more years, and fall back after about ten. That is the same timetable this project finds for everything else a government does to its courts.

Could this be built into V-Dem's own measure?

A fair worry, since the same project codes both division and attacks. The two questions do not overlap: the division question asks whether supporters of opposing camps avoid each other at family gatherings, at work and in clubs, and it sits in a different part of V-Dem from the judiciary questions. But the same experts' overall impression of a country could still colour both.

So the test was repeated with a measure of division that no V-Dem expert touched: how spread out each public is on the left-right scale, taken from the survey answers themselves in 107 countries. It is a different measure, correlating 0.30 with V-Dem's.

Between countries, the pattern holds on the independent measure. Comparing each country only with itself it is weaker (a clarity of 1.5, against 3.6 for V-Dem's measure on the same years). That could be a shared impression among V-Dem's experts, or the difference between people spreading out on left and right and people treating the other side as enemies, or simply that surveys come only every few years. The data cannot tell these apart.

Canada

Canada's government has attacked its courts less than the world's typical government in 11 of the last 11 years.

This measures the government's attacks, not attacks by opposition politicians, the press or the public, which may be what concerns judges most. Attacks and division are both expert judgments. V-Dem's corruption index is built partly from an item on judicial corruption, so the link with corruption may partly reflect courts that are themselves corrupt, which a government could then attack. Corruption also predicts attacks in a recent study, "When do governments attack the judiciary? The explanatory power of political corruption", International Review of Law and Economics (2025). Nothing here identifies a cause.

Which democracies started attacking their courts, and what set them apart

The section above asks what predicts a government attacking its courts. This one sorts countries into kinds, using the words V-Dem's experts themselves choose from when they answer "How often did the government attack the judiciary's integrity in public?". Their five answers run from "there were no attacks" through "rare" and "more than once" to "nearly every month" and "daily or weekly". Here "often" means "more than once" or worse.

Each country's answers are averaged over 2000 to 2009 and again over 2016 to 2025. Comparing the two decades gives four kinds of country. The table covers the 85 countries that were democracies in the first decade, since that is the comparison that matters for a country like Canada.

What set the starters apart

Comparing the democracies that started attacking their courts with the ones that never did:

The starters became far more divided between the two decades. On V-Dem's scale, calm Norway and deeply divided Poland sit about five and a half points apart, so the difference here is large. The starters were also weaker to begin with: more corrupt, somewhat less democratic, and with less independent top courts, before any of the attacks began.

Did division come first?

For each starter, the year its government first attacked the courts "more than once", and how divided the country was ten years earlier and just before:

In 4 of the 7, division rose in the decade before the attacks began. In the others it did not. With 7 countries this shows that division often comes first, not that it always does.

Where familiar countries sit

Canada's government made no attacks on the courts in either decade by the experts' account. So did those of the United Kingdom, Germany, France and Australia. Israel's government attacked the courts often in both decades, and now nearly every month.

Averaging over a decade hides what happens inside it; Poland, for example, attacked its courts daily or weekly for years and then stopped after 2023. Seven starters is a small group. "Democracy" here means a score of at least 0.5 on V-Dem's electoral democracy index. Every measure is an expert judgment from V-Dem. Nothing here identifies a cause.

The question, and which kind of polarization actually moves it

The question this project exists to answer is narrower than it sounds. Not whether polarization is bad. Whether people respect courts less in polarized places and at polarized times, and which kind of polarization does it.

Because "polarized" is not one thing. A country can be deeply divided right now. It can have been divided for twenty years. It can have divided quickly. It can have crossed from one condition into another. Those are four different measurements and they do not have to give the same answer.

Run against each other on the same country-years, four of them do almost nothing. How divided a country is now, how long it has been divided, how fast it got there and how divided it has ever been all sit under a t of 1.3 for how far a public's two sides stand apart about courts.

Crossing over moves it. Following each country through its own history, an ordinary increase in division gives +0.25 on the distance between the two sides about courts, at a t of 1.4, which is nothing. Passing into the most divided group gives +0.85 at a t of 2.9, and every further year on that side adds +0.09 more. Courts are not alone in this: parliament moves further (+1.18) and the police less (+0.66).

So the answer to the question as asked is that intensity does not do it, speed does not do it, and duration only does it once a country has already crossed. The event is the crossing, and then the years accumulate.

the crossing result · what does the moving, inside one country · what people credit and blame

People credit judges when courts work, and blame politicians when they do not

Every other European measure here asks people to rate their courts. This one asks them why they rated them that way, and the answer is the clearest thing in this project about the standing of judges.

How the question works

The Eurobarometer asks everyone the same opening question: how would you rate your country's justice system for the independence of courts and judges, good or bad?

Then it asks a follow-up that depends on the answer. Both halves of the public are offered the same three explanations, pointed their way. People who said bad are asked how far each of these explains their rating: politicians interfere, business and money interfere, the status and position of judges do not sufficiently guarantee their independence. People who said good get the mirror image: politicians do not interfere, business does not interfere, the status and position of judges do guarantee their independence.

What they say

Share answering "very much" or "somewhat", 2024, averaged across countries.

The order reverses exactly. The status of judges is the first explanation among people satisfied with their courts and the last among people who are not. Political interference is the first among the dissatisfied and the last among the satisfied. The two swap places.

Why it matters

Three things follow, and the first is the one this project has been arguing without ever having the public say it.

A public already treats the standing of judges as what keeps courts independent. Asked to explain courts that work, it reaches for the position of judges before anything about politics or money.

When courts are seen to fail, judges are not blamed for it. The public points at politicians. Judges are not held responsible for the loss of their own independence.

And the blame is growing. Politicians as the explanation for courts that are not independent rose from 71% in 2016 to 77% in 2024, across nine waves. Crediting the standing of judges among the satisfied barely moved, 73% to 74%.

These are published country totals rather than respondent-level data, so nothing here can be split by a person's politics: it says what a whole public answers, not which half of it answers that way. The three reasons are offered to respondents rather than volunteered by them, and the two halves are answering about their own rating, so the bases differ by design. Nine waves, 2016 to 2024, 32 countries.

The principle holds up better than the institution

Almost everything on this page measures confidence in courts, which is a feeling about an institution. A judge is asking about something narrower and harder: whether people still accept that the law binds those in power. Two surveys ask that directly, and the answers run in a useful direction.

People disagree far more about trusting courts than about obeying them

Afrobarometer asks about court power twice over, in two directions. Whether the president must obey the laws and the courts, which is power over the powerful. And whether courts may make decisions citizens must abide by, which is power over everyone else. Beside them sits the ordinary trust question. Splitting each by whether a person is close to the governing party:

Read the standard-deviation column, because the three questions use different scales. The two sides sit about eight times further apart on trusting courts than on whether the president must obey them. A public can lose confidence in the institution and go on agreeing that the powerful are bound by it.

This also settles a live question about this project. Bartels and Kramon found in Ghana that people aligned with the president want less court power over the president and more over citizens. Had that held generally, claim one would have been an average of two things moving in opposite directions. It does not: across 43 countries the government side is more supportive of both, and the two gaps move together rather than against each other.

What division does is thin the agreement, not destroy it

The AmericasBarometer asks the question closest to a judge's own: must the authorities always respect the law, or may they sometimes act outside it to catch criminals? 148,434 people in 24 countries, and 37% say they may act outside it.

As a country divides, that share does not rise. The estimate on the level is -0.004 with a t of -0.3, which is nothing. What moves is the distance between the two sides:

Comparing each country against its own past, the gap is wider in its own divided years in 10 of 11 countries. A dividing country does not become more willing to let authorities break the law. Its two halves stop agreeing about whether they may.

For a court that is a different problem from the one usually feared, and not an easier one. A judge applying the law in such a country is applying a principle that one half of the public no longer shares with the other.

And it is not courts that are singled out

The same survey asks whether dissolving the supreme court is justified in difficult times, and whether a president closing congress is. On the 62 country-years carrying both, 20.8% would license dissolving the court and 19.5% closing congress, and the two correlate at 0.83. A public willing to license one largely licenses the other.

Whatever this is, it is an attitude to institutional limits rather than a special hostility to courts. That belongs beside anything on this page suggesting courts are picked out.

Binary items are coarse and do not append to the seven-point confidence results. Left and right do not mean the same thing in every country and about half the AmericasBarometer sample declines to place itself; the party-closeness question in Africa reaches only the part of a public that names a party. The rule-of-law item stops being asked after 2017 and the African court-power item stops after Round 7. On the American result, dropping Honduras takes the estimate from t 5.9 to t 1.2, so the count of countries above is the sturdier statement and the t is not. Nothing here identifies a cause.

How it unfolds, and to whom

The threshold above says a country has to be among the most divided before its public divides about its courts. It does not say when, or what happens in between, or whether this happens to every country. These are the three.

How long before division reaches a court?

About three years, and a court has those three years in hand.

When a country splits into hostile camps, the obvious guess is that its courts take the hit straight away. They do not. For roughly three years nothing measurable happens to how the two sides see them. Then it starts, and after that it does not stop.

The groups below are how long a country has already spent above the division line, and the number in each is the distance between how the side in power and the side out of power see the courts.

Nothing moves in the first step, the one from a country that has never divided to a country that just has. It is flat in Europe and Asia and slightly negative in Africa, while every step after it moves. The last row takes each country against its own average, so a permanently divided country cannot produce the pattern by sitting in one group.

Three years is a long time in the life of a court, and the rest of this page shows what the government is doing with them. Whether anything a court does in that window makes a difference is a question this evidence cannot answer, and it is the question worth asking next.

Only seven countries in the whole project are observed both in their first three years of division and after four, which is the strictest version of this and gives six of seven widening. The four-group table above rests on eighteen. This is a small number of countries showing the same shape in three unrelated surveys, and not a large number showing it once.

What governments do inside the delay

V-Dem codes, for every country and year, how far political leaders publicly attack the integrity of the judiciary. Every country that crosses into high division is centred on its own crossing year and measured against its own average.

Attacks rise while the public is still quiet and barely rise again once the public gap opens. That is the order the chain needs, and it rests on every country V-Dem codes rather than on the handful with repeated surveys.

And yet attacks are not the link

If attacks were what carried division through to the public, putting them into the model should have taken the duration term down with them. It does not move. Either the measure is too coarse to serve as a mechanism, or the attacks are a parallel symptom and something else fills the three years. This project does not know which, and says so rather than choosing.

Which courts does this happen to?

Twenty of the seventy-three that can be classified, and they form a recognisable group.

Averaging countries together buries the answer, because one country's split tripled while another's shrank and a third ran the other way. Sorting them instead, by a method that looks for groups without being told what to look for, produced two groups.

The pressured kind: Guinea, HU, Hong Kong, IL, Malaysia, PL, Thailand, Togo, Zimbabwe.

Hungary, Poland, Hong Kong, Israel and Thailand sit there with Zimbabwe, Cambodia, Sudan and Uganda. Their courts have almost nothing in common. What they share is the behaviour of their publics: politics in two camps, pressure on judges rising, and a wide and growing distance between the side in office and the side out of it.

A judge reading this list will recognise the first five and may be surprised by the company they keep. The method was not told which countries were democracies and which were not. It sorted on how the publics behave, and that is where they landed.

A mixture model returns classes from noise as readily as from structure. The check is the same model refitted 200 times on data with the features shuffled apart, which never matched the real fit. BIC prefers two classes over one by a wide margin and cannot choose between two and five, so two is reported as the fewest within range of the best fit. Seventy-three countries make small classes, and a country's class depends on the years it happened to be surveyed.

The United States, 1973 to 2024

No cross-national survey splits Americans by whether their side holds power, so the United States was missing from every comparison above. The General Social Survey fills it: confidence in the Supreme Court since 1973, with party identification alongside, which allows the same subtraction. 33 readings across 51 years, the longest run of any country here.

Flat for 28 readings across 43 years. Flat through the first three years after the country crossed into high division. Then three times its old size. That is the delay above, on American data, arrived at independently.

And it changes sign. Until 2020 the president's own partisans were the more confident in the Court. From 2021 they are the less, ending 19 points below. The Supreme Court's majority stopped tracking the presidency, and the public followed the Court rather than the White House.

Israel does the same with the sign reversed: there the opposition trusts the courts and the governing side does not, while attacks on the judiciary rise faster than in any of the other cases. So a public responds to a court's own majority rather than to the government's. Everywhere else the two coincide, which is why comparing government supporters with opposition supporters works as a measure at all.

A three-answer confidence item is coarser than a 0 to 10 scale, the survey draws a fresh sample each time rather than following people, and the Court's own composition changed repeatedly across these years, so what people are rating did not hold still. Classifying a year by the president rather than by Congress is a choice, and divided government is common.

What a government will say about its own courts

Everything else on this page measures what a public thinks. This measures what a state will commit to, out loud, in front of other states. Under the Universal Periodic Review every UN member is examined every few years and other governments make recommendations to it one at a time. The state answers each by supporting it or by noting it, and noting is the refusal.

Courts are the hard subject

States accept about three quarters of what they are told and about three fifths of what they are told about their judiciary. A divided state accepts 56% of the court recommendations it receives against 65% for a settled one, a drop of 10 points. On every other subject the same comparison moves 2 points. So a divided government's reluctance shows up on its courts and barely anywhere else.

Holding how much each recommendation asks for, division predicts refusal of court recommendations at -0.027 per point (t -2.2) and of everything else at -0.002 (t -0.3). The second number is the reason to believe the first.

Whose advice a divided government takes about its courts

3,225 recommendations about the judiciary, of 108,277 in all. Among the divided states only, each recommendation is scored against what an ask that demanding is supported at on average, so the comparison is between like asks.

Region does the work. A divided government accepts advice about its courts from its own regional group far above the going rate and refuses it from outside. Being similarly divided is worth about a third of that, and the other state merely being divided is worth almost nothing. So this is not two polarized states recognising each other. It is a government taking correction from its neighbours and not from strangers.

Before difficulty is held, advice from a divided state looks more welcome, at +0.024 with a t of 3.4. Holding it, the same term falls to +0.008 (t +1.3). Most of what looked like one government listening to another turns out to be divided governments making easier asks.

Courts as a political asset

Refusing advice is one signature and a weak one. Three others are visible in this record, and the first two work together.

A divided government is told about its own judiciary more often and raises other governments' judiciaries less often, at the same time. Per point of division that is +0.0049 on what it receives and -0.0056 on what it makes. A state that becomes a target and stops throwing stones is behaving as though it holds something worth protecting.

Advice about courts is weighed by who offers it

Measuring how far apart two states sit on V-Dem's liberal democracy scale, and holding both region and how much the recommendation asks for, a court recommendation from a politically distant state is refused at -0.392 per point of distance. The same distance costs -0.198 on every other subject. Advice about a judiciary is weighed by its source about twice as heavily as advice about anything else.

The government pulls back before the public does

The fall lands in the same three years where attacks on the judiciary rise and the public has not moved yet. In that window a government is attacking its courts at home and refusing correction about them abroad. The partial recovery afterwards is not something this project can explain, and these counts are small.

The regional effect is not gentler wording

Same-region acceptance is higher inside every category of ask, and the difference is largest on the most demanding one. The recommendations are the same length either way. So who is allowed to criticise a judiciary is itself a political question, and a state's own bloc has standing that outsiders do not.

None of this identifies a cause, and there are three further reasons to read it narrowly. Supporting a recommendation is cheap and costs nothing to ignore afterwards, so these numbers record what a government was willing to say in a diplomatic setting. States choose whom to advise and about what, so a refusal rate is partly a fact about who bothered to ask. And the judiciary recommendations are found by their wording, which will miss some and catch some that are really about something else.

Is it a slope or a cliff?

A cliff. Countries divide about their courts only once their politics is among the most divided in the world. Below that the gap barely moves.

This matters for anyone deciding whether to worry. If the relationship were a slope, every increase in polarization would cost a court something, and the warning sign would be the direction of travel. It behaves instead like a threshold, so the warning sign is a country arriving in the top group rather than moving within the rest of the range.

The bottom four groups sit between +0.26 and +0.37, a range of 0.11 points across three quarters of the distribution. The last step alone is +0.21, which is 60% of all the movement between groups. A straight line would put 25 per cent there. Canada, measured a different way entirely, points the same way: taking the Canadian provinces and asking where in the distribution polarization matters, the coefficient rises from +0.115 at the 25th percentile of the gap to +0.221 at the 90th. That rests on 37 province-years and proves nothing by itself, and it is an independent case pointing the same way.

Parliament does the same thing on the same country-rounds, which says the shape belongs to political division rather than to something about the court question. Courts simply sit lower throughout, which is the other finding on this page.

The same exercise on a different outcome gives a different shape, which is worth holding onto. How much a whole public trusts its courts, rather than how far apart its two sides sit, changes as a gradient: 24% of the movement falls in the last step, against 25 per cent for a straight line. So polarization appears to shave trust off everywhere by degrees, and to pull the two sides apart only at the extreme.

Five groups of about fifty country-rounds cannot separate a cliff from a steep curve, and the most divided fifth is not a random fifth of Europe. If those countries are also the poorest, or the most recently democratic, the step could belong to that instead. Nothing here identifies a cause, and the cut into fifths is a choice: the same data as a continuous quantile regression is in the section on where in the distribution the effect sits.

Asia, where one finding held and the other broke

Two claims on this page were built on Europe and Africa. The Asian Barometer is the first time either has been tried anywhere else. The threshold held. The comparison with parliament did not.

This survey settles one problem the others create. It asks the interviewer to record whether the person voted for the winning or the losing camp, in every country, the same way, so the government side and the opposition side are identified in the fieldwork rather than reconstructed afterwards from party databases. What follows rests on 56 country-waves across 15 countries, fieldwork from 2006 to 2023.

Start with the crosstab

That is 51,390 people answering the court question, before any model. Winners trust courts more than losers do in 49 of the 56 country-waves, and in 2 the two sides are the same to within half a per cent of the scale.

The comparison with parliament breaks

Everywhere else this project has looked, the court gap is about half the parliament gap: 0.50 in Europe, 0.68 in Africa. In Asia the two are the same size. The court gap averages +0.078 and the parliament gap +0.078, a ratio of 1.00. Asian publics divide about their courts as much as about their parliaments.

That is a real break and it is stated here rather than softened. What survives of the original claim is the direction, which holds everywhere, and not the size.

The threshold holds, and only for courts

Sorting the 56 country-waves into fifths by how divided their politics is gives the same shape Europe gave: flat, flat, flat, flat, jump.

A straight line would put 25 per cent of the movement in the last step. Courts put 66% there. Dealing the same 56 numbers out at random against the division scores 2,000 times produces a last step that large in 0.6% of tries. Parliament, the police and the military show nothing of the kind, so the threshold belongs to courts rather than to political division wearing everything down at once.

Four other ways of asking, all of which found nothing

The threshold is the only shape that appears. Every way of treating division as a quantity that pushes steadily came back empty.

A fifth way looked like the strongest result here and is the one to distrust. How many years a country had already been divided predicted the court gap at +0.0020 per year, with a t of 4.9. Only four of the 15 countries ever cross that threshold, so the variable is close to a label for Malaysia, which sits at 41 years while nothing else is near it. Cambodia has the widest court gap of the lot on zero years. Comparing a country with itself the coefficient becomes -0.0033, t -0.5. It is a country and not a duration.

A quantile regression says the same thing from a different direction. Division does nothing to the middle of the spread of court gaps (+0.0016 at the median) and a good deal to its wide end (+0.0418 at the 90th percentile). So what division changes is how wide the widest cases get.

Fifteen countries and 56 country-waves cannot separate a cliff from a steep curve, and eleven country-waves per fifth is thin. A winning camp in a dominant-party system is not in the same relationship to power as one in a competitive system, and about half of respondents did not vote or were not sure. Eighteen of these country-waves are matched to a division score at a published fieldwork midpoint because Waves 2 and 3 record no interview year. The division score is an expert coding of a country rather than a measurement of the people interviewed. Nothing here is a panel, so no attitude can be shown to come before another.

South Asia, where courts stand second

5 countries this project had not covered: Bangladesh, India, Nepal, Pakistan and Sri Lanka. The same institutional trust questions, put to 16,150 people across 2005 and 2013.

Read this section as a smaller claim than the one above it. Everywhere else on this page the comparison between institutions is a difference between two political sides. These files cannot produce one: the item recording whether a person voted for the winning or the losing camp is absent, the 2005 which-party question is answered only in India, and the 2013 survey carries no party identification at all, because variables a country team collects with its own money stay with that team. So what follows is how much of a public trusts each institution, which is a different quantity from how far apart two sides sit, and it neither confirms nor breaks anything said above about gaps.

Where courts stand

Courts are the second most trusted of the 8, at 73% in 2013, behind the army at 81%. They sit above parliament at 61% and above the national government at 62%. Over the eight years courts moved +1% while the national government moved -7%.

One country went the other way

Bangladesh is the case worth looking at. Trust in courts there fell 25% points, from 85% to 61%, while the other 4 countries all rose, Pakistan furthest at +16%. This project cannot say why from these two numbers, and a single country moving hard against its neighbours is the kind of case that is worth a closer look rather than an explanation.

Two readings eight years apart are not a trend. The 2005 and 2013 fieldwork used different questionnaires under different names, so part of any change is the rewording rather than the public. These are levels and not gaps, and nothing in this section speaks to polarization, because the surveys carry no measure of which political side a respondent is on.

Do government supporters trust courts more than everyone else?

In most places, yes, and by less than they trust parliament more. This table sets that out.

How to read a cell. Each one covers a country in one round of the survey. It shows the average trust in the legal system among people who voted for a party sitting in cabinet on the day they were interviewed, minus the average among people who voted for a party that was not. Both averages are on a scale of 0 to 10. A positive number means the government side trusted courts more. The smaller figure underneath is the same subtraction for trust in parliament, from the same people, so the two can be compared directly.

What the table shows

The government side trusted courts more in 209 of the 267 country-rounds, and trusted parliament more in 246. The court gap widened over two decades, from about 0.33 points in 2002 to 0.77 in 2023, and the parliament gap widened over the same years from 0.54 to 1.03. So courts appear to divide people along the same political line as parliament, and about half as far. The averages hide a lot: countries differ widely in every round, and a handful run the other way entirely.

In 14 country-rounds the two numbers disagree, because the government side trusted parliament more by at least half a point while trusting courts less: Spain 2020, Spain 2024, Britain 2002, Britain 2004, Britain 2021, Israel 2015, Israel 2016, Israel 2024, Italy 2003, Italy 2024, Lithuania 2011, Netherlands 2010, Slovenia 2006, Slovenia 2020. These are the cases where courts are not simply following parliament. Israel is the largest of them, and it rests on smaller groups on each side than most other cells, so it should be read with that in mind. Its round 11 interviews ran from December 2023 to October 2024, so every one of them was taken after the judicial overhaul fight and after 7 October 2023, which is the context that number carries.

Does the same division show up between left and right?

It does, and it is a different cut of the same people. Here the comparison is between respondents who place themselves at 6 to 10 on a left-right scale and those who place themselves at 0 to 4, so nobody has to be matched to a party. A positive number means the right trusted the institution more.

Two reasons to look at this version. It covers countries that ParlGov does not, and it is the one to trust where a government changed while interviewing was still going on, because a person's own left-right placement does not depend on knowing who was in office that week.

Does the European pattern hold outside Europe?

Yes, on 1981 to 2023 of the Integrated Values Surveys. Courts divide the right from the left in most countries surveyed, and less than parliament does.

The European Social Survey reaches about 30 European countries from 2002. The Integrated Values Surveys, the World Values Survey trend file appended to its European sister survey the European Values Study, reach 107 countries from 1981. Both halves are built on the same common dictionary, so the courts and left-right items carry the same names and the two files append. Each cell is again the right minus the left, with the parliament gap underneath, and countries run down the page by wave.

Two differences from the European figures above matter for reading the numbers. The Integrated Values Surveys ask about confidence in four steps, from a great deal to none at all, rather than trust on an eleven-point scale; the values here are reversed so that higher means more confidence and then stretched onto a 0 to 10 range so a gap can be set beside the European one. That stretch is arithmetic, not a claim that the two questions measure the same thing. And left and right come from a 1 to 10 self-placement with no midpoint, so the left is 1 to 4, the right is 7 to 10, and the two middle positions are dropped.

What the wider coverage adds

The right is more confident in courts in 283 of the 376 country-years, and the court gap is smaller in absolute terms than the parliament gap in 229 of them. Averaged across country-years the court gap is 0.59 points against 0.73 for parliament, on the rescaled range.

The growth over time does not carry across. Within the 61 countries surveyed at least three times, the absolute court gap is wider at the last observation than at the first in 22 of them, so the doubling visible in the European series since 2002 does not appear as a global trend. The country mix changes from wave to wave, which is why this comparison is made within countries rather than across the pooled averages.

10 country-years run the opposite way, with the right more confident in parliament by at least half a point and less confident in courts: Armenia 2021, Australia 2005, Bulgaria 1997, India 2012, Italy 2005, Italy 2009, Lithuania 1997, Pakistan 2012, Poland 2008, Poland 2017.

Which division splits people most about courts: party, or something else?

Belonging to an ethnic minority divides trust in courts about as much as being on the right or the left, and more than supporting the government or the opposition. That is the strongest evidence here for a division about courts that does not follow party lines.

Round 5 of the European Social Survey, fielded in 2010, is the only round that carried a module on the police and the courts. It asked the same people the standard trust questions and a set of questions about their own experience, which makes it possible to take one country in one year and compute the gap in trust in the legal system under five different divisions, then set them side by side. Every cell below is a difference of weighted means of the same 0 to 10 item among the same respondents, so the five columns are directly comparable. The small figures give the size of the two groups.

What the comparison shows

Averaged across countries, the division between people who belong to an ethnic minority and everyone else is about the same size as the division between right and left. It is larger than the division between government and opposition supporters. Having been stopped by the police, or having been a victim of crime, divides trust in courts less than either.

0.54 ethnic minority against everyone else, across 20 countries

0.52 right against left, across 27 countries

0.37 government supporters against opposition supporters, across 25 countries

0.32 crime victims against non-victims, across 27 countries

0.22 stopped by police against not stopped, across 27 countries

In 13 of the 18 countries where both can be measured, the minority division is wider than the government one. This project asks whether courts divide people on a line other than the party line, and this is the clearest place where such a line appears.

Three limits belong with that reading. The minority samples are small, from 59 to 440 respondents, and twenty countries have enough minority respondents to measure at all. The division runs in opposite directions in different countries: minorities report more trust than the majority in Bulgaria, Switzerland, Germany, Spain, France, Britain, Greece, Ireland, Israel, Russia, and less in Czechia, Estonia, Lithuania, Netherlands, Ukraine. And the module ran once, in 2010, so it compares countries rather than tracking change.

What minorities say about courts

The same module asked what people think courts actually do. Respondents who belong to an ethnic minority were on average 6.5 percentage points more likely than others to say that, of two people falsely accused of an identical crime, the one from a different ethnic group is more likely to be found guilty. On the other items they answered much like everyone else: agreement that courts protect the rich and powerful, agreement that court decisions are politically influenced, how often judges take bribes, and how often courts decide fairly all differ by under two percentage points or a tenth of a point.

So the complaint minorities report is a specific one about ethnic bias in verdicts. It does not appear as a general suspicion of the courts, because on every other item they answer much as everyone else does.

Does the division among citizens track what was happening to the courts?

Weakly, and in the direction you would expect. Where high courts are more independent the two sides differ less about the legal system, and where politics is more divided they differ more. Both links are modest, and neither shows that one caused the other.

The surveys measure how far apart two groups of citizens are. V-Dem measures, for the same country and the same year, how independent the courts were and how polarized the politics was. Attaching one to the other asks whether the division among citizens tracks the condition of the institution. Every European Social Survey country-round matched a V-Dem country-year, using the year in which the interviews were actually done rather than the round's label. The round number is a poor guide to the year: 120 of the 267 country-rounds were fielded in a different year from the one the round label implies, and thirteen are two years out.

The first column correlates each indicator with the government-minus-opposition gap in trust in courts across all country-rounds. The second removes each country's own average first, so it compares a country with itself over time rather than richer countries with poorer ones. The third does the same for the difference between the court gap and the parliament gap.

What the correlations suggest

Where high courts are more independent, government and opposition supporters differ less about the legal system, and the relationship survives removing each country's own average (-0.23). Where politics is more polarized into camps, they differ more (+0.21). Both associations are modest and neither is an effect estimate.

Government attacks on the judiciary carry the largest correlation in the table. Within a country over time, they track the difference between the court gap and the parliament gap at +0.41. The V-Dem indicator runs in the direction its name does not suggest: a high score means the judiciary is left alone, and the score falls as a government attacks it. Poland's score falls from about +2.3 before 2015 to about -3.6 by 2020, while Denmark stays near +1.8 throughout. So in the years when a government is attacking the courts, the party division over parliament widens faster than the division over courts.

Poland is the clearest case in the file. Under the Law and Justice governments the gap between government and opposition supporters over parliament reached about two and a half points, while the gap over the legal system stayed below one. Attacking the courts did not, on this measure, make courts the institution the two sides disagreed about most.

Does it matter how much partisans dislike each other?

This project cannot answer that yet, and the honest summary is that its two sources disagree. The American panel shows a link and the fifteen elections outside the United States do not, and fifteen elections are too few to settle it.

The distinction matters. Everything above measures how far apart two groups' average trust sits, which is a gap between groups. Affective polarization is a different quantity: how much warmer people feel toward their own side than toward the other side. Every other number on this page uses V-Dem's expert coding of how far a society has split into hostile camps, which is not the same thing. Two sources here measure affective polarization directly and ask the same respondents about courts.

The United States, the same people three times

The American National Election Studies panel rates both parties and the Supreme Court on the same 0 to 100 scale, and follows the same respondents from 2016 to 2024. Affective polarization is the warmth a partisan reports toward their own party minus the other one, and the last two columns are the correlation, within each party, between a person's affective polarization and their rating of the Court.

In 2016 Democrats and Republicans rated the Court 60.3 and 58.8, within two points of each other, and a person's partisan hostility told you nothing about their view of it. By 2024 the ratings are 33.8 and 65.1, and the same correlation runs -0.29 among Democrats and +0.34 among Republicans. Over eight years the Court was drawn into the affective structure of partisanship, having sat outside it.

The panel adds a caution. Among the same people from 2020 to 2024 the Democratic rating falls about twenty points while their affective polarization barely changes, and the two changes are uncorrelated. The fall was collective rather than a by-product of individuals growing more hostile.

Which comes first inside a person

Everything else on this page compares countries or compares groups. The panel interviews the same Americans in 2016, 2020 and 2024, which allows the question this project asks of countries to be asked inside a person. Does growing to dislike the other party come before souring on the Supreme Court, or does the Court rating move first and the partisan feeling follow? Each row below regresses the later value on its own earlier value and on the other attitude's earlier value, so what is estimated is a change rather than a level, and both are standardised so the two directions can be compared with each other.

The answer is that partisan feeling comes first, and the pooled row hides it. Between 2020 and 2024 a Republican's dislike of Democrats in 2020 predicts a warmer rating of the Court four years later at +0.357, and a Democrat's dislike of Republicans predicts a colder one at -0.178. The two sides move in opposite directions, so together they average to almost nothing. In the other direction the Court rating predicts almost no change in partisan feeling for either side, at -0.010 among Republicans and -0.049 among Democrats. Between 2016 and 2020 neither direction reaches much at all, which fits the finding above that the Court sat outside the partisan structure in 2016.

Two things that ordering does not depend on. The first is where a person lives. Respondents sit in states, and states differ in how sorted their politics is, so the same model was run with a random intercept for the state and then with the person's partisan feeling multiplied by the share of their state at the ends of the party identification scale. Neither adds anything: the state term reaches t 0.9 and the cross-level term +0.017 at t +0.7, while the individual coefficient does not move. Before anything is added, only 1.8% of the variation in the 2024 Court rating sits between states rather than between people inside them. The pattern is about who someone is and not about the political weather around them. The second is falling out with people. The panel has no question about who a person discusses politics with, so their network cannot be described, and it does ask in both waves how much political differences have hurt their family relationships. That measure correlates with partisan feeling at only +0.03, so it is a different thing rather than a second reading of the same one. Run through the same cross-lag it gives -0.076 one way and -0.072 the other, which are the same size. Damaged relationships and views of the Court travel together, and nothing here says which came first.

Fifteen other elections, where it does not repeat

The Comparative Study of Electoral Systems asks respondents to rate every significant party from 0 to 10 and asks the same people about trust in the judiciary. Affective polarization here is the spread of a person's party ratings. Within electorates the two are close to unrelated, and splitting by ideological side, as the American case requires, the signs oppose in 8 of 15 studies while almost every correlation sits under 0.2.