Why we never compare countries — and what we compare instead
The three rules behind every number on this site: compare each kind of work only against its own market, publish nothing below the floor, and separate what we measure from why it moved.
Every number on SHIFT is a comparison, and the comparison always happens inside one country. We count the job advertisements matching a kind of work in a given market, and then ask a single question: did that count move faster or slower than the whole basket of work in that same market, over the same two years?
It is deliberately a modest question, and it is the only one the data can answer honestly. Advertisement density, population, and how much hiring happens online at all differ enormously between countries. A country with ten times fewer advertisements is not a country with ten times less hiring — it is a country where less of the hiring is advertised online, somewhere our source can see it. Any figure comparing those raw counts across borders would be measuring the internet, not the labour market.
So we never compare absolute levels between countries. Dividing by the same country's own market is what cancels the problem: if a country's coverage is thin, the numerator and the denominator shrink together and the ratio does not move. That is what makes one market's reading legible beside another's — not the counts, the ratios.
The floor
A market appears for a kind of work only once we counted at least 500 advertisements for it in the base window. The number is not a matter of taste: below it, the sampling error is larger than the movement we would be reporting, and what we published would be noise wearing the clothes of a finding. This is why plumbers are visible in a handful of markets and software developers in almost all, and why coverage grows on its own — a market enters the moment the data qualifies.
Direction and pace, not cause
SHIFT measures the movement of demand. It does not, by itself, prove why demand moved. A fall could reflect automation, a change in job titles, a seasonal pattern, or a shift in how employers advertise — the index cannot tell these apart, and we don't pretend it can. Where we offer a reading of what might be driving a change, we label it clearly as interpretation, kept separate from the measurement.
The index measures the shape of change. The explanation is always a separate — and more careful — question.
Three rules, then: compare each kind of work only against the market it sits in, publish nothing below the floor, and keep what we measure separate from why it moved. They make the numbers less dramatic than a headline might like — and far more trustworthy. That trade is the whole point.
