Labour market

Reading unemployment rates across countries

Why Spain's 10.4% and Japan's 2.5% cannot be compared at face value: definitions, survey design and labour-force participation behind the ILO numbers.

An unemployment rate is a fraction with a definition inside it, and the definitions differ enough across countries that headline numbers summarize rather than compare. Spain's 2025 ILO rate of 10.4% and Japan's 2.5% are both real, but saying one economy is "four times worse" than the other would mistake a survey output for a welfare verdict. This note explains what the rate measures, what drives the differences, and how to read the published series sanely.

The figures come from the ILO-modelled unemployment estimates — World Bank-sourced series as republished on the Unemployment desk — for 2025. The ILO modelled series harmonises definitions across economies, which is precisely why it is comparable in a way raw national survey releases are not; the price of harmonisation is that each figure is a modelled estimate, not the raw national read.

What the unemployment rate actually measures

The ILO definition counts a person as unemployed when they are not in paid work, are available for work, and have actively looked for work in the recent reference period. Three consequences follow. People without work who are not seeking and available are classed as outside the labour force, not unemployed. People working even one hour in the reference period are employed, however precariously. And the rate is a ratio — unemployed divided by labour force — so the denominator, not just the numerator, drives it. A country where many people give up searching entirely sees the rate fall, not rise, because thousands leave the labour force.

The 2025 published figures

Why lower is not always better

A rate near zero is suspicious, not triumphant. Japan's 2.45% partly reflects a tight labour market, but also an institutional structure recognising short-hours and older-age work, and long-standing cultural patterns of labour-force participation. Very low rates can also coincide with under-employment — workers with fewer hours or lower skill use than they want — which the headline rate does not capture. So a strong headline rate is "good" only in combination with employment rates, hours, participation and wage growth elsewhere in the picture.

Similarly, high rates carry structural stories rather than only cyclical ones. Spain and Greece have shown double-digit or high-single-digit rates in some recent years; part of that reflects difficult transitions for young workers and long spells of unemployment, which affect the labour force denominator and the rate's interpretation for years.

Participation drives the denominator

The same unemployment rate can accompany very different labour markets when participation differs. If a large share of the working-age population stays outside the labour force (full-time study, early retirement, home duties by choice or constraint), the denominator shrinks and the rate can shrink with it. Comparisons that ignore participation can mistakenly call an economy "healthy" for the wrong reason. The way analysts handle this is to read the unemployment rate alongside participation and employment-to-population measures on the same desk pages.

Modelled versus national releases

The series Younivi carries on the desk are the ILO modelled estimates, built to be produced under a common definitional frame. Each economy also has its own national statistical release — for New Zealand, the household labour force survey run by Stats NZ — which is the more precise figure for that country, updated more often, and which can differ from the modelled number. If precision about one economy is the goal, use the national release through the ILO's citation; if comparing across many economies, use the modelled series — but remember it is a modelled output, revised as the underlying data improve.

Denominator changes that mimic improvement

Because the rate is a ratio, events in the denominator can move it without anything happening among the jobless. Two examples worth naming. Ageing populations shrink the denominator over time: as a larger share of the working-age population retires and standard unemployment questions are asked of ever-chosen narrower groups, a rate can drift lower with no change in hiring. And education expansions do the same: keeping more young people in full-time study removes them from the labour force, mechanically improving the youth rate and slightly improving the headline one. A falling rate is thus good news only when participation is stable or rising.

The same logic warns against country comparisons during one-off shocks. After a recession in which discouraged workers stop searching, the rate can fall while underemployment and non-participation rise — a pattern statisticians call a discouraged-worker effect. The ILO's harmonised series cannot repair that on its own; it reports the labour-force frame, not desire.

Timing and revision

Unemployment data of the modelled kind are typically annual or near-annual in the published series, with revisions in later vintages as national sources update. A figure you remember for a prior year may change, and Younivi's copy will change with it at the next rebuild. The producing agency — the ILO, via its data programme — remains the authority on any number, as noted on the desk and our methodology page.

Where this fits in the wider picture

Cross-country unemployment comparisons are legitimate only because someone harmonised the definitions first. National statistical agencies measure unemployment in ways that suit their own labour markets and legal traditions: the threshold for what counts as actively searching, the treatment of people working a small number of hours, the age at which someone enters the labour force, and the handling of people who have stopped looking. Left as published, those national figures describe different things, and ranking them against each other would compare measurement conventions rather than labour markets.

The ILO modelled estimates exist to solve that problem. They impose a shared definition on the underlying national data, filling gaps and adjusting for known differences, so that a rate for one country is built on the same conceptual basis as another. That is what makes a comparison meaningful, and it is also the source of the estimates' limits: modelling smooths and imputes, so each figure is an estimate rather than a direct count, and it is revised as better national data arrive.

The practical consequence is that the harmonised series should be used for exactly what it was built for - cross-country comparison and long-run direction - while the national release remains the precise figure for any single economy. A reader who keeps those two uses apart will avoid most of the errors that headline ranking invites, including the common mistake of reading a fall in measured unemployment as an unambiguous improvement when participation has also fallen.

How to read this on Younivi

The Unemployment desk's home page lists the 2025 figures for the economies in our sample with charts and the World Bank/ILO attribution; the youth page carries the 15–24 series separately (see our youth note). Companion material sits on the wages and GDP desks. Figures are republished official data, may be revised, and none of this is advice.