Special Report
How the Monthly Jobs Numbers Are Built
The monthly employment report is the most consequential recurring economic release in most countries. It moves markets, shapes policy debate, and is treated in coverage as a single reading of the labour market. It is in fact two separate surveys with different populations, different definitions and different error properties, published in one document on the same morning.
Survey one: households
A sample of tens of thousands of households is interviewed each month about the activity of every member of working age during a specific reference week. From those answers each person is classified into one of three states: employed, unemployed, or not in the labour force.
The definitions are strict and are the source of most public argument. Employed means having done any paid work in the reference week, including part-time and self-employment — one hour counts. Unemployed means having no work, being available for it, and having actively looked within a set recent period. Not in the labour force covers everyone else: retired people, students, carers, and anyone who wanted work but had not looked recently.
The unemployment rate is the unemployed divided by the labour force — the employed plus the unemployed — and not by the population. That denominator is what produces the most-cited oddity in all of labour statistics: a person who gives up looking leaves the numerator and the denominator, and the unemployment rate falls. Statistical agencies publish broader measures alongside the headline that add discouraged workers and involuntary part-timers precisely because of this, and the Bureau of Labor Statistics releases all of them in the same document every month.
Survey two: employers
Separately, several hundred thousand worksites report how many people were on the payroll for the pay period covering the twelfth of the month. This produces the "jobs added" headline.
The differences from the household survey are consequential:
- It counts positions, not people. Someone with two jobs appears twice.
- It excludes the self-employed, unincorporated business owners, and unpaid family workers entirely.
- It is a much larger sample, so its estimate of the monthly change is statistically far more precise — which is why the payroll figure and not the household employment figure is the market-moving number.
- It cannot directly observe firms that were born or died since the sample frame was drawn, and uses a statistical model to estimate their net contribution. That model is accurate over a cycle and is least accurate at turning points, which is exactly when it matters most.
Seasonal adjustment
Employment has enormous, entirely predictable seasonal swings: retail hiring before the holidays, construction in spring, education staff over the summer. Unadjusted, these dwarf everything else and would make month-to-month comparison useless. Seasonal adjustment estimates the normal pattern from past years and removes it, so what remains is the movement beyond what the calendar predicts.
The procedure is well established, transparent, and occasionally a genuine source of noise. When the normal pattern itself changes — a shift in holiday hiring practice, or an unusual year that distorts the estimated pattern for several years afterwards — the adjustment can add error rather than remove it. Agencies publish both adjusted and unadjusted series, and looking at the same month a year earlier in the unadjusted data is a useful cross-check when an adjusted figure looks strange.
Why the number changes twice after publication
The first estimate is published while a substantial share of survey responses is still outstanding. Two subsequent monthly releases incorporate the late returns, and an annual benchmarking exercise realigns the whole series to near-complete administrative records from unemployment insurance filings.
Both stages matter more than the coverage suggests. Monthly revisions are frequently larger in magnitude than the change being reported, and the annual benchmark revision can shift a year's cumulative total appreciably. There is no way around this: the choice is between a timely estimate that will be revised and an accurate figure that arrives too late to be useful, and every statistical agency has chosen timeliness. The consequence is that the first print deserves less confidence than it receives.
What an announced layoff tells you
Almost nothing about aggregate employment, and this is the single most common misreading in the field. In a large economy, millions of jobs are created and millions destroyed every month through routine business formation, expansion, contraction and closure. A net monthly change in the hundreds of thousands is the small residual of two vastly larger flows.
So an announced reduction of even tens of thousands of positions at a single employer is, against the gross flows, statistically minor — and it may be spread over several years, may be achieved through attrition, and may consist partly of unfilled vacancies rather than people. It is a real event for the people involved and a legitimate story. It is not evidence about the direction of the national labour market, and it is the concrete-versus-diffuse asymmetry described in The Balance Sheet at its purest.
A reader's checklist for jobs day
- Which survey? Payroll change comes from employers; the unemployment rate comes from households.
- What happened to participation? A falling rate alongside falling participation is a different event from a falling rate with rising participation.
- What were the revisions? They appear in the same release, usually in the second paragraph of the official summary and rarely in the coverage.
- Is the change bigger than the error margin? For a single month, often not.
- What did wages do in real terms? Nominal earnings growth below inflation is a fall in purchasing power.
For the definitions in full, the Bureau of Labor Statistics publishes both a short official glossary of the definitions used above and a Handbook of Methods covering the sampling and estimation procedures in full. For the broader labour-market context, see work and wages.