What is labor market data?

What makes labor market data good

The distinguishing property is not recency and it is not granularity. It is attribution. A defensible figure states four things: where it came from, the geography it applies to, the time period it covers, and — if it was derived rather than published — the method used to derive it.

That sounds obvious. It is also the thing most commonly missing. A number in a slide deck that reads "12% projected growth" has none of the four, and cannot be checked, defended, or reproduced next year.

Four questions, four different kinds of data

Most labor market questions reduce to one of four, and each is answered by a different kind of data. Reaching for the wrong one is the most common analytical mistake in this field.

1. How many people work in this industry here?

Answered by employer records that cover nearly every establishment, published by industry and ownership sector down to county level. When you want to know the actual size of an industry in a specific place, this is the answer.

What it cannot do: it is not occupational. It knows a hospital employs 2,400 people; it does not know how many are nurses. It also runs roughly two quarters behind.

2. What does this job pay here?

Answered by a large annual occupational wage survey, published as a full distribution — 10th through 90th percentile — by occupation and area.

What it cannot do: small cells are suppressed to protect respondents, so some county-level occupations have no published wage. It is annual, so it lags fast-moving compensation. And a median alone is misleading — a new graduate enters near the 10th percentile, not the middle.

3. Where is this heading?

Answered by official employment projections — a ten-year national outlook, plus long- and short-term occupational projections produced by each state.

What it cannot do: annual openings — growth plus replacement demand — are published nationally only. Any local openings figure you see anywhere is modeled. That is not a criticism; it is unavoidable. The question is whether the figure tells you it is modeled, and how.

4. Who is being trained for it?

Answered by postsecondary completions reported by every institution, by program, award level and student demographics, and linked to occupations through a program-to-occupation crosswalk.

What it cannot do: the crosswalk is many-to-many, so naive supply–demand arithmetic inflates demand badly. Completions also lag by about a year, and graduates move between regions.

A word on scraped job postings

Several vendors build primarily on scraped online job postings. Postings are genuinely useful for reading demand signals month to month, and for seeing the skills and tools employers name — which official statistics do not capture.

But a posting database is a sample of what employers advertise online, and no vendor publishes its sampling frame. Duplicate postings, staffing-agency listings, evergreen requisitions and roles never posted publicly all distort it in directions that cannot be quantified from outside. Postings are a leading indicator; they are not a measurement of employment.

The practical rule: use postings for direction and detail, and official statistics for magnitude and for anything you will have to defend.

Published, derived, and modeled

Three categories, and a good source keeps them distinct.

• Published — stated directly by the statistical program. Industry employment for a county. A median wage for an occupation in a metro.
• Derived — arithmetic on published figures, with no assumptions added. A location quotient is derived: it is two published shares divided.
• Modeled — an assumption has been introduced to produce a figure nobody publishes. Local annual openings. A county-level industry forecast. An occupation's employment where the published cell was suppressed.

All three are legitimate. The problem is presenting the third as though it were the first. When you evaluate any labor market data provider, the single most useful question is: which of your numbers are modeled, and can you show me the method? A provider that cannot answer that clearly is one you should not cite.

Five pitfalls worth knowing

1. National rates for local decisions. An occupation growing 12% nationally can be flat in your region. Always resolve to the geography where people will actually look for work.
2. Medians without distributions. The median tells you about the middle of the market; entry wages sit near the 10th percentile. Two programs can look identical on medians and be very different for a student.
3. Growth without replacement demand. Most openings come from people leaving, not from growth. Occupations projected to shrink still hire, sometimes heavily.
4. Double-counting through the crosswalk. Charging an occupation's full openings to every program that feeds it inflates total demand several-fold. The openings must be allocated.
5. Ignoring suppression. If a provider always has a number at county level, ask what it does where the published cell was suppressed. Silently imputing it is a choice, and it should be disclosed.

A short evaluation checklist

When someone hands you a labor market figure — from us or anyone else — five questions settle whether it is usable:

1. Where does this figure come from, and which release?
2. What geography and what time period does it describe?
3. Is it published, derived, or modeled — and if modeled, by what method?
4. What happens where the underlying data was suppressed?
5. Could I reproduce it, given the method?

Any provider worth using can answer all five without a follow-up call.

---

Further reading: how our evidence base is kept current · how each modeled figure is computed · glossary of terms