How we get a figure, and why we stopped adding them up

The short version

Every country is described four times, once for each reason someone leaves: school before eighteen, a degree, a trade, or a move for good. Each description is four or five ingredients, and each ingredient is a figure with a source attached — a fee in dollars, a rent in dollars a month, a band from a written rubric.

We do not combine them. There is no total, no position, and no default order: a list page opens alphabetically and you choose what to compare on. That is the change this page exists to explain, and it costs something real — one number is easier to act on than five. What it buys is that nothing on this site is true only because of a weighting you never agreed to.

What has not changed is the discipline underneath. We still do not write the ingredients by hand: the figures go in with their sources and lib/scoring.ts does the reading. It means we cannot nudge a country up by deciding it deserves to be up — which was always the point of the separation, and is the part worth keeping.

Right now that comes to 52 countries — 10 of them being worked through at the moment and the rest waiting their turn, which each of their pages says — 208 country-and-reason pages and 988 individual figures. Of those, 113 are printed in the source we cite, 821 are numbers we made from what a source does say — a basket, a formula, a band from a written rubric — and 54 are marked unknown because we could not find them. Every page tells you which is which, and the balance is the honest shape of the thing: almost everything here is our arithmetic on somebody else’s evidence, and the fees are the part you can look up.

Why the same country gets four different pages

A country that is excellent for a degree can be poor for a family with a nine-year-old, because the two decisions do not rest on the same things. So the ingredients change with the reason.

This table used to carry a weight against each cell — how much that ingredient counted toward the total. The totals are gone and so are the weights. What is left is the question the table was always answering underneath: which ingredients does this reason actually turn on?

Read the dashes as hard statements. A relocation page carries no tuition and no education, because someone retiring to Portugal is not buying schooling; a school page carries no residency, because the question there is whether a parent can be present, not whether anyone can settle for good. “Outcome” changes meaning with the column: for a family it is whether a parent can stay, for a graduate whether they can work, for a trainee whether the trade will have them.

Costs and fees run the other way from the rest — lower is better — so where they are put on a 0–100 scale, for the width of a bar or the order of a “cheapest first” chip, they are inverted against a ceiling set per reason. Eight thousand dollars a year is cheap for an international school and expensive for a German public university, and one scale for both would be meaningless. That scaling happens within an ingredient and never across two of them; combining across was the step that has gone.

Where each ingredient comes from

Only safety is marked verified, and only one ingredient deserves it. The Global Peace Index measures the thing we want to measure, for every country, on one scale, published by someone else. Nothing else here has that. Calling the rest verified would be a claim we could not defend.

What we refuse to use, and why

Section 14-3

A ranking is only as free as its sources. Four of the obvious ones are closed to us, and the reasons are worth stating, because they shape what this site can be.

QS World University Rankings are published under CC BY-NC-ND. The NC clause forbids commercial use; the ND clause forbids derivative works. Turning QS positions into figures of our own is both of those at once — precisely the two things the licence rules out. Every ranking you have seen quoted on a comparison site was either licensed or used without permission. We are not able to do the first and not willing to do the second.

The WHO Global Health Observatory is CC BY-NC-SA, which is worse for us than QS. NC closes it the same way. SA would then require us to publish our own work under the same licence, so a single health indicator would set the terms for the whole site.

Numbeo charges for commercial use and forbids scraping. It is the cost-of-living source that most comparison sites quietly rest on.

OECD indicators are open enough, and they fail us for a different reason: they cover member countries. A site that exists to surface the places nobody names would be able to score Germany and not Malaysia, Portugal and not Georgia. The bias would run in exactly the direction we were built to correct.

What is left — the World Bank, the Leiden Ranking, the Global Peace Index, and government and school documents read one at a time — is slower and narrower. It is also why we can publish this page.

Education: a rubric, not a formula

Section 9-1

There is no single published number for how good a country’s education is. Safety has one; education does not, and the ones that look like it are either licensed away from us or measure something else. So instead of a formula we use a written rubric, and we publish it here so that the same standard is visibly applied to every country.

The rule we hold ourselves to is that the band comes first and the number second. You pick which description a country matches, then choose a score inside that band — never the reverse. Every education figure in the data names its band in the source text, so you can check the judgement against the rule.

For a degree, the question is: how far does this qualification travel?

One trap is worth naming because it caught us more than once. A foreign university with a campus in Tashkent or Jakarta usually awards its own degree, not the host country’s. Westminster International University in Tashkent says plainly that its graduates receive University of Westminster degrees. That is a British degree taken in Uzbekistan, and it does not lift Uzbekistan’s education figure. The mirror case counts, though: the American University of Armenia is an Armenian institution that holds United States accreditation, so its degree is Armenia’s and it does count.

For a trade, the question is different: does the industry know the name?

Section 9-2

The difference between the two matters, and getting it wrong produces wrong figures. A degree question is about institutions: is the qualification recognised, will a regulator accept it. A trade question is about reputation: does the industry know the name. So mutual recognition across the EU raises a degree figure and does nothing for a trade one. Polish vocational qualifications are recognised across Europe and the industry has never heard of the schools, which puts them in “locally useful only”.

Schooling before eighteen has no rubric yet

Section 9-3

We have not written one. The early-study education figures are judgement, and they are the weakest thing in this dataset. See what we know is weak, where this is the first entry.

Research impact is quoted, and kept out of the figures

Section 9-4

Degree pages carry a research-impact percentile from the CWTS Leiden Ranking Open Edition, which is CC0 and free to use. It never enters the band. It measures how often a university’s papers are cited, which is a fact about its professors, not about what a student is taught, and we label it that way every time it appears.

It also carries a sampling artefact that is easy to misread. Leiden only ranks institutions above a publication threshold, so the number of universities a country contributes varies enormously, and we report the median of them. Indonesia has 129 ranked universities and its median lands at the 6th percentile; Rwanda has one and its median lands at the 39th. That is a statement about how deep each sample goes, not about which country teaches better. It is exactly why the figure sits beside the band instead of inside it.

Outcome and residency: we measure the structure, not your odds

Sections 10-1 to 10-4

“Your chance of getting a US work visa” is not one number. It is completely different for an Indian-born computer science graduate and a Brazilian-born art graduate, and any single figure would be false for both. So we do not score your odds. We score the structure of the rules, which is the same for everyone.

Six questions decide where a country lands:

• Does a route exist at all? If not, bottom band.
• Is it automatic, or is it somebody’s discretion?
• Does it depend on an employer, or on conditions outside your control?
• Is there a lottery or a quota?
• Does the answer change with your country of birth? That itself is a structural weakness, and it lowers the band rather than being treated as noise. That the United States green card queue turns on where you were born is true for everybody.
• Is the period defined, and is it long enough to be worth anything?

Each purpose then asks its own question of those rules, and the subject of the sentence changes each time.

The vocational column is judged on its own rather than derived from the university one, because the two genuinely disagree. A German apprenticeship is a paid employment contract from the first day, which scores higher than the eighteen-month job-search permit a German graduate gets. Deriving one from the other would make that impossible to see.

Two rules place a country inside its band. A rule flagged as under review sits at the bottom of whatever band its current terms earn — the band is set by the rules as they stand, the position by how much we trust them to stand still. And where a route covers only part of a case, we judge the typical case: a parent visa that runs out when the child turns eleven covers half a school career, and is scored as covering half.

Healthcare: can you get care, predictably?

Section 13-2

This column is not a ranking of medicine. It asks four questions in order: can a foreign resident get into the system at all; what does it cost and is that cost predictable; how long is the wait; and does it hold up when something serious happens.

This is why the United States sits in the fourth band despite having some of the best hospitals in the world. That is not a bug in the rubric; it is the rubric. The question is not how good the medicine is, it is whether you can predictably reach it.

The fourth question is the one that does the most work. A country where everyday care is cheap and fast but serious illness means a flight to Bangkok is a different proposition from one where serious illness is treated at home, and the first three questions cannot tell them apart. Malaysia is a country people fly to for treatment; Cambodia is one people fly out of. Both are inexpensive.

Health pages also quote the World Bank’s out-of-pocket share of health spending, and always with a warning attached. The measure counts only what insured people pay — not premiums, and not the uninsured — which is why the United States comes out at 10.9%, ahead of most countries on this site. We print the caveat next to the number every time, because a number like that without one teaches the wrong thing.

Living cost: the direction is evidence, the size is our judgement

Section 12

The licensed cost-of-living indices are closed to us, so we build the figure from a US reference budget for each purpose and move it by the World Bank’s household consumption price level, which is CC BY 4.0 and covers the countries the commercial indices skip. There is exactly one published series behind it and no parameter of our own invention.

Every cost figure on this site says, in its own source text, that it is a judgement directed by evidence rather than computed from it. That wording is deliberate. The direction is well founded; the size is ours; and we would rather say so than let a number look more precise than it is.

Why we use the latest year rather than the benchmark survey

Section 12-1-1

We first used the 2021 benchmark. Adding Japan showed that was wrong: on the 2021 benchmark Japan came out at 0.995 of the United States, almost identical, because that was the yen of 2021. The yen has since fallen by more than a third. The benchmark had not moved and the figure had become false.

We did not carve out an exception for Japan. A threshold written to exclude one inconvenient country is a number reverse-engineered from the answer you wanted, and the discrepancy was systematic rather than Japanese. So the whole dataset moved to the annually updated series, on a conceptual argument rather than a freshness one: someone arriving with foreign money meets this year’s prices at this year’s exchange rate, and the annual series is what measures that.

One vintage is used for everyone, currently 2024, because these are comparisons between countries. Where a country’s series has stopped — Argentina’s ends in 2021 — that country’s cost is left unknown rather than dragging every other country back to an older year.

The remaining error is described in what we know is weak, and it is not small.

Tuition: where we look, and in what order

Section 12-4

A fee cannot be banded — the amount is the meaning — and there is no global dataset of them. So instead of a formula we discipline the source: we ask the government first, then the institution, then whoever collected it, and we record which kind of document the figure came from.

That is an order to search in, not a grade. These five used to be numbered tiers, and the number was printed at the front of every fee we published — which meant a judgement of ours was being read as part of the source. There is no best kind: an aggregator quoting a ministry correctly beats a university page nobody has touched since 2019, and the only way to know is to open both. So we open both, and the source line says what we read.

We read aggregators with suspicion because they have been wrong to our faces. A summary put Poland tenth on the peace index; the published table said twenty-second. An aggregator put an international school in Tbilisi at $16,480; the US State Department’s fact sheet, carrying the school’s own figures, said $24,900. A summary put secondary fees at a Bangkok school at 600,000–750,000 baht; the school’s own PDF said 840,000. When thirty-three degree fees were re-read from their own sources, thirty-two changed, and the badly wrong ones had that kind of document in common.

What the figure prices is fixed per purpose, so that countries are comparable: an international school’s annual secondary tuition, day rate, in the country’s main city; the published sticker price of a bachelor’s degree for a non-national, before any scholarship; the annual fee at a specialist institution in the field.

Which schools we price, and which we leave out

Section 12-4-1

One country does not have one fee. Inside a single city, international school tuition runs three to five times from bottom to top, and the curriculum does not tell you where you are standing — Kuala Lumpur’s cheapest and dearest are both teaching British, American or IB programmes. So if the school is not chosen by a rule, the column stops measuring what schools cost and starts measuring which school somebody happened to pick.

The rule: the middle of the city. Where a city has four or more international schools we name the most expensive ones and leave them out, take two or three from the middle, and compare them all at the same school year, because fees climb with every year group. Where it has two or three we use all of them and say the city has no tiers to pick from. Boarding is excluded, and so are private schools teaching the national curriculum, because a family moving for an international education is not buying those.

Leaving out the top is not an attempt to make a country look cheap. It is a refusal to write the assumption that an international school is whatever the priciest school in town charges — an assumption we have never tested. The schools we excluded are named in the source line with their fees, so the reader can put them back.

The same problem has a subtler form inside a table of figures. Spain’s six university fees are three regional decrees, and Valencia publishes three bands where Catalonia publishes one; a plain middle of the rows would give Valencia three votes for being more explicit about its own pricing. So the middle is taken over the bodies that set the prices, not over the rows. And the middle is taken in the published currency and then converted, rather than converted and then averaged — Malaysia showed us the difference is real the first time a country had an even number of them.

The biggest hole: what vocational training costs

Section 12-5

Twenty-nine of the countries here have no published fee for vocational or specialist training, and that is the single largest gap in the dataset by a factor of four. It is not for want of asking. Each one was taken down the same path — the ministry, then the colleges, then whoever collected it — and each one ended without a number.

The pattern is not random and it is worth saying out loud, because it decides what the gap means. The countries that publish a vocational fee are the ones selling the training to foreigners as a product: a culinary academy printing its local and international rates side by side, a maritime university heading a table “cost of programmes for foreigners”. Where that market does not exist, vocational training is a public service for citizens and the price is something you ask about. Uzbekistan is the clean case: its ministry publishes admission arrangements for its own citizens and no rate for anybody else, and the institutions that do print fees are foreign university campuses rather than trade schools.

Two repairs were considered and refused. Assuming vocational training costs less than a degree would be a number with nothing behind it — German apprentices are paid a wage and Swiss hotel schools cost more than universities, so even the direction is unreliable. Quietly reweighting the page to hide the hole would be worse than guessing, because it would stop the reader seeing that anything is missing. So the count on the card names the missing ingredient instead, and this list is kept where it can be counted.

It shrinks two ways. One is finding a fee. The other is going back, failing again, and recording the month — a gap with a date and an account of what was tried is a gap somebody has been to, and it stops being a debt and starts being a finding.

How fresh a rule is

Section 11

Immigration rules are the only ingredient that goes wrong while sitting still. Georgia began requiring work permits in March 2026, and every guide written before that became incorrect without a word of it changing.

So every policy figure — outcomes, residency, healthcare — carries the month we last checked its source, and one of three ratings. Stable means no change known to us; note that this says what we know, not that we re-confirmed it today, which is what the date is for. Recently changed means it moved in roughly the last eighteen months, so older guidance elsewhere is already wrong. Under review means a change is announced or under discussion, or the terms move often enough that you should not plan around the current ones. The last two appear as a badge next to the figure.

A rating other than stable has to rest on a change we actually recorded on that country’s page. We do not apply it on a hunch.

What we do when we don’t know

Sections 14-5, 14-7 and 14-2

Unknown is a designed exit, not a failure. But it has a condition: unknown means we looked and could not find it, never that we did not look. To mark something unknown we have to have gone down the source ladder in order, and to write into the data what we tried and why it failed. Those accounts are on the pages — open any ingredient marked unknown and you can read the attempt.

A gap used to be filled. It had to be: a weighted average cannot have a hole in it, so a missing ingredient took the dataset’s median for that ingredient and reason, and the substitution was charged against the page’s coverage. That machinery was defensible and it was also the clearest sign the total was the wrong shape — we were inventing a figure for a country in order to be able to add it up.

Nothing is added up now, so nothing has to be invented. A missing ingredient is missing. Every card says how many ingredients it has out of how many the reason asks for, and names the ones it does not — “Tuition unknown” rather than “80%”. A count is something you can check against the page in front of you, which was never true of a share of a weighting.

No country is dropped for being thinly known. There used to be a floor: below 40% of the evidence a country left the ranking, because a position it had not earned would have been a claim we could not support. Nothing is claimed by appearing in an alphabetical list, so every country is in it, carrying its own count.

As it stands, 49 of 208 country-and-reason pages are missing at least one ingredient, 54 ingredients in total. The largest single hole is vocational tuition: 30 of 52 countries publish no fee a foreigner can check.

Why we leave our own mistakes on the page

Re-reading the fees from their own sources changed most of them, and several by a lot: a South African degree went from about $3,100 a year to $8,350, a Spanish one from about $7,500 to $1,800, an international school year in Portugal from $9,500 to $23,050. Those were our figures, not anyone else’s, and every one of them was wrong in a way that was checkable at the time.

So the country pages carry them. Each says what we had, what it is now, why the old one was wrong, and the month we read the source that fixed it — printed on the page the figure lives on rather than folded into a changelog nobody opens. A reference work that corrects quietly is asking to be trusted; one that prints its corrections is giving you something to check.

It also says something we could not have said on our own authority. The survey our old Spanish figure came from put the Universitat de Barcelona at around €6,600 a year when the university’s own regulations charge about a sixth of that. We are not claiming the aggregators are unreliable; we are showing one instance we walked into and can prove, which is a different and smaller claim, and the only one we have.

What a filter hides, and why ours refuses to

A filter is where a gap does the most damage, because a filter’s job is to make countries disappear and a reader cannot tell one kind of disappearance from another. Ask for vocational training under $5,000 a year and thirty of our fifty-two countries have no fee on file at all. Drop them the way most filters would, and what comes back is Britain at $34,000 and Canada at $12,700, and the page has used our own ignorance to prove that only the expensive countries exist.

It would be a lie in a particular direction, too. A country publishes a price for vocational training when it sells that training to foreigners; where the training is a public service for its own citizens, the fee is something you get by asking. So the countries missing from that list lean cheap, and deleting them silently is not neutral.

So every filter here answers in three counts rather than two: how many match, how many do not, and how many we could not check. The third is printed the same size as the other two, it names the ingredient that is missing, and it will list the countries if you ask. A country only appears under “do not match” when something we could actually check failed — if the rent is five times your budget, an unknown fee does not turn that into a maybe, but if nothing checkable failed then the honest answer is that we do not know.

Money is filtered on the range of what we verified rather than on the middle of it. A Spanish degree is $1,150 a year in Catalonia and $8,350 for engineering in Madrid; asking for under $2,000 finds Spain, because Catalonia is real. A reader who wants no surprises anywhere can tick a box and ask that the whole country fit, which is a different question and gets a different answer. Neither is our judgement about Spain.

A clean licence does not make every row right

Section 14-2

Before a price level is used it is put beside its neighbours. Bulgaria’s 2024 row comes out at 0.216, next to Romania at 0.447 and Slovenia at 0.625, which is plainly wrong — the underlying figure looks like a unit error. The publisher is reputable and the licence is open and the row is still wrong. The same check caught a Global Peace Index score we had recorded as 2.507 when the published table said 2.535.

Tell us what we got wrong

83% of the figures here are numbers we made rather than numbers we found. The fastest way for one of them to become a figure printed in a source is somebody who has actually been.

hello@pathoff.com

What is most useful to us, roughly in order:

• A fee you were quoted, with the document or the email it came in.
• A rule that changed after the date we last checked — that date is on every policy figure.
• Anything marked unknown that you happen to know, especially a vocational fee.
• A number that is simply wrong. Tell us what it should be and how you know, and we will either change it or write down why we did not.

We read everything and we do not have a form, because a form would mean collecting personal data on a site that currently collects none.

Licence and attribution

Our own work on this site — the rubric judgements, the written assessments, the selection of countries and ingredients, the weights, and the prose — is published under CC BY 4.0. Use it, build on it, sell things made with it; credit Pathoff and link back.

Third-party figures carry their own terms, which we cannot extend to you. One of them needs naming rather than burying: the safety figure is a restatement of a Global Peace Index score, and we pass on no rights in it. The index is published by the Institute for Economics and Peace, we cite its rank and score with attribution on every page, and we have not been able to establish what licence the current edition carries — the report we read does not state one in a form we could extract. Until we have asked them and can say, treat the safety column as IEP’s and go to the source. Everything else on the page is ours to give.