How these figures are published, and what they leave out

One household, three incomes

The OECD publishes, for every member country, what a specific person actually pays. The person is a single adult with no children, no income other than employment, and no reliefs beyond those that apply automatically. The incomes are 67%, 100% and 167% of that country's own average wage, so the comparison is between people at the same position in their own wage distribution rather than between people on the same number of euros.

That choice is what makes the series comparable and it is also its main limitation. Everything published describes that one household. A couple, a parent, somebody with rental income, somebody with a pension contribution or a student loan, and anybody at a salary between or beyond those three points, is outside it. There is no way to interpolate honestly: the tax systems concerned have thresholds, tapers and cliffs between the published points, and a straight line drawn between two of them would be a figure this site invented sitting beside figures it did not.

An average rate is not a band rate

The single most misread number in this subject is the headline rate. A country with a top band of 45% does not take 45% from somebody who reaches that band. It takes 45% of the part of their income above the threshold, and lower rates on everything below it, and the result across the whole salary is the average rate. That is the figure these pages lead with, because it is the one that answers what a year actually costs.

The marginal rate answers a different question: what is taken from the next unit earned. It decides whether a rise, a bonus or an extra shift is worth taking, and it is routinely far above the average rate at the same income. The published marginal figure is a net personal rate, so it includes the withdrawal of any cash benefit as income rises, which is why it can exceed the top statutory band in countries where benefits taper.

Both numbers appear on every reference page and the gap between them is stated for each. Quoting one when you meant the other is the commonest error made with a table of this kind, and it is an error that changes a decision rather than a description.

Why employer contributions are shown but excluded

Every take-home figure on this site excludes what your employer pays, because it never appears in your gross pay and including it would change how a country ranks for a reason that has nothing to do with what lands in your account. Two countries with identical employee deductions can look very different once employer contributions are added, and the difference is invisible on a payslip.

The reference pages nevertheless print the employer rate, labelled, because a reference page reports what its source publishes. Leaving a published figure out because it does not fit our own convention would be editing the source, and the whole claim these pages make is that they do not do that. Where the employer side is larger than the employee side, the cost of employing somebody is a good deal further above their gross pay than their payslip suggests, and the page says so with that country's own two numbers.

Why a jurisdiction gets a reference page instead of a calculator

Every jurisdiction this site computes has to reproduce the published rates from its own statutory schedule, for the same person, at all three incomes, within half a percentage point. That is not a style rule and it is not cautious for its own sake. It caught a live page showing an income tax rate of 13.70% where the published figure for the same person was 19.22%, because the allowance withdraws with income and the schedule we read did not record the withdrawal. The page had been up for a day and nothing about it looked wrong.

Half a percentage point is the tolerance because a miss of that size on an ordinary salary is real money and a reader has no way to see it. A calculator that is wrong by two points still produces a number that looks exactly like a right one, which is the failure worth designing against: not the obvious error, the plausible one.

When a jurisdiction misses, the ingest does not give up quietly. It builds every reading the published schedule allows, computes each one with the real engine, and keeps whichever reproduces the published outcome. A jurisdiction is refused only when no reading survives that, and the page for it states the miss of the closest one in percentage points along with how that reading was assembled. Those numbers are a work list rather than an apology: they say exactly how far a page of ours would have been from the truth, which is more than a site publishing every country can tell you about any of them.

A gap under a point is usually something the compiled dataset does not carry and the national authority publishes plainly, so it is the kind that a proper national ingest closes. A wide gap usually means something structural about how that country taxes income is missing, and closing it means reading the authority directly rather than adjusting a number until it fits.

What a reference page does not tell you

• What you would pay. One household type, three incomes. A different household is a different answer and these pages cannot produce it.
• Anything at your own salary. Three points are published and nothing between them is.
• Regional variation. Where income is taxed below central government, the series uses a representative rate rather than your municipality, canton or province. In the countries where that layer is most of the income tax, this matters more than any other caveat on this list.
• Anything about being foreign. Residence rules, treaty relief, remittance treatment and the interaction with the system you are leaving are not in the series. For a US citizen, the obligations that follow you regardless of where you live are on the guides.
• What the taxes buy. A ranking compares what is taken and says nothing about what is provided. Two countries a place apart can feel entirely different to live in, and no figure on any of these pages speaks to that.

Why the schedule and the outcome disagree

Every reference page carries two things that look like they should match and do not: a statutory band table, and an average rate far below the lowest band in it. Both are correct, and the distance between them is the most useful thing on the page.

Four things account for most of the gap. An allowance or a zero-rate first band means the early part of every salary is untaxed, so the average is dragged down by income that never reached the ladder at all. A credit comes off the tax rather than the income, so it reduces the average rate without touching any band. Contributions that are deductible from the income tax base shrink what the ladder runs on, which is a different operation again from either of the first two. And where income is taxed below central government as well, the central table understates rather than overstates, which is the one direction that surprises people.

Those four also explain why a compiled schedule is hard to compute from. The published data carries the numbers but not always what each one means: the same field is a percentage of the tax due in one country, a contribution on gross in a second and a levy on taxable income in a third, and the feed does not say which. Rather than pick a reading, the ingest builds every reading the data allows, computes each with the real engine, and keeps only one that reproduces the published outcome at all three incomes. A reading that survives that is a finding. A reading asserted would be a guess with a citation attached, which is harder to catch than an obvious error and does more damage.

Where a jurisdiction has no band table at all on its page, the reason is upstream rather than ours. Two published ladders in this dataset step down as income rises, which a progressive schedule does not do, so a rate is wrong where it was published. Others publish a count of thresholds that cannot pair with their count of rates, and pairing them by position would invent a schedule rather than report one. Repairing somebody else's arithmetic and presenting the result as a published fact is not something this site will do, so those pages say plainly that no schedule is shown and why. The outcome rates are unaffected: they come from a separate series the OECD computes itself.

How far apart the jurisdictions actually are

Across the 38 jurisdictions this site holds the series for, the combined rate taken from a single person on the average wage runs from 0.0% in Colombia to 39.5% in Belgium, a spread of 39.5 percentage points. Every reference page states where its jurisdiction sits in that range and names the ones immediately above and below it.

Why a reference page is worth having at all

The obvious alternative to a reference page is no page, and for a while that is what this site had. A jurisdiction failed the tolerance, its rules were deleted so that no stale answer could survive the refusal, and the reader searching for it found nothing here. That is defensible and it is not the best available answer, because the reader's question does have a published answer. Refusing to compute a number is not the same as refusing to tell somebody a number that already exists.

The other alternative is what most pages on these subjects do, which is to state a top band rate and leave the reader to assume it applies to them. That produces a page that is never exactly wrong and is useless for the question actually being asked, because nobody pays their top band rate on their whole income anywhere. A page reporting the average rate, the marginal rate and the schedule together, each labelled, is a smaller claim and a far more useful one.

The cost of the approach is that a reference page cannot answer a question about your own salary, and it says so rather than approximating. That limit is real and it is preferable to the alternative, which is an interpolation this site invented sitting in the same table as figures it did not.

How to use one of these pages next to a calculator page

The two page types answer the same question to different standards and it is worth knowing which you are looking at. A calculator page takes a salary you choose and computes the deductions band by band, with the rule behind every rate, because that jurisdiction reproduces the published outcome closely enough for the arithmetic to be trusted. A reference page reports three published points and refuses to compute a fourth.

Comparing across the two is legitimate at the average wage, where both state a figure for the same household, and stops being legitimate anywhere else, because only one of them can produce a figure anywhere else. If you are weighing a move between a country with a calculator and one with a reference page, the honest comparison is the combined rate at the average wage on both, and the honest thing to say about your own salary is that half the comparison is missing.

Which jurisdictions have a reference page, and why those

The rule store holds this series for every jurisdiction the OECD publishes it for, and generating a page for each would cost nothing. The pages are limited to 10 anyway, because English-language search demand was measured for those and not for the rest. A page built on unmeasured demand is a page built on an assumption, and the marginal page being free makes that easier to do rather than safer.

The jurisdictions with no page still have their figures in the published ruleset, with the same quote and hash on every row.

How this figure is produced, and how it is checked

Nothing on this page is typed in. Every rate, threshold and allowance is a rule in a store, and each rule carries the document it came from, the date it was read, an exact quote and a hash of that quote as it stood on the day. The calculation reads rules through a lens that records every one it touches, so the citations underneath a number are derived from the arithmetic that produced it. A page cannot cite a source the calculation did not use, and cannot use one it does not show.

A missing rule is a build failure rather than a fallback. That sounds severe until you consider the alternative: the failure worth preventing here is not a wrong number, it is a plausible one. A calculator that quietly substitutes last year's allowance produces a figure no reader can tell apart from the right one, and every check that matters is designed around making that impossible rather than unlikely.

Four gates run before anything is published. Golden fixtures check the arithmetic against worked examples at several incomes. An independent second reader compares every rule against a separately published compilation of the same law and records each disagreement rather than resolving it silently. A reconciliation gate computes tax for a single person at 67%, 100% and 167% of the average wage and requires the result to reproduce the average rates the OECD publishes for exactly that person. And a page gate checks that what is rendered matches what was computed, because a template can drift from the engine behind it without either being wrong on its own.

Ruleset sha256:7c8a1de9fc6d87cc. The full method is on methodology, every source on sources, every change on the changelog, and anything found wrong on corrections.