---
title: "Occupation data — how occupation figures are built | Data Tactical Group"
route: "/docs/methodology/occupation"
updated: "2026-09-12"
---

# Occupation data

What occupation figures count, and how they are constructed.

Updated 2026-09-12

DTG reports employment and earnings by occupation for every county, state and the nation, quarterly and annually, from 2004 through the projection horizon — the same areas and periods as industry data.

## What is reported

**Employment**, in jobs, on the same basis as industry data.

**Average earnings**, as a per-period figure: a quarterly query returns average quarterly earnings, an annual query returns average annual earnings. Earnings are reported as averages.

Every occupation figure is reported for one area and one ownership — federal government, state government, local government, or private — because the construction is carried out separately for each ownership.

Occupations are classified using the Standard Occupational Classification system, at five levels from the all-occupations total through major groups, minor groups, broad occupations and detailed occupations.

Occupations sum through that hierarchy, with one exception worth knowing: not every broad occupation splits into detailed occupations. Summing detailed occupations alone therefore recovers less than the total. Read each level against its own parent rather than assembling totals from the deepest level.

## Sources

Three inputs, described more fully on the [data sources](/docs/methodology/sources) page.

**DTG industry data** supplies the employment totals that occupation figures must agree with. It is the fixed quantity in the construction.

**The BLS Employment Projections national industry-occupation matrix** supplies the structure: which occupations staff which industries, and in what proportions, nationally.

**OEWS** supplies the area dimension: how many people work in each occupation in a given area, and what they earn. These figures are reported at the national, state, and substate levels.

## How the data is built

The national matrix says how industries are staffed. DTG industry data says how much of each industry is in each area. OEWS says how many people work in each occupation there. Those three do not agree on their own, and the construction is the process of making them agree.

Staffing patterns are constructed separately for each ownership. Everything that follows happens within an ownership, and the figures for one ownership are never mixed with another’s.

### Staffing patterns

The construction runs from the top down, and each level inherits a pattern already fitted at the level above it.

The EP national matrix is applied to DTG’s national industry employment, producing a first estimate of national occupation employment by industry. That estimate is fitted against two sets of totals at once: each industry’s occupations must sum to that industry’s DTG employment, and each occupation’s total is fitted to national OEWS employment for that occupation. The fitting alternates between the two until both hold.

The fitted national pattern is then applied to each state’s industry employment and fitted to that state’s OEWS figures in the same way. Each state’s pattern is applied to the substate regions within it — the metropolitan and nonmetropolitan areas OEWS reports for — and fitted to each region’s OEWS figures.

Once the substate regions are built, state patterns are rebuilt as the sum of the substate regions within them, and the national pattern is rebuilt from the states, so the levels agree with each other rather than being computed independently.

Two rules govern the fitting. **DTG industry employment never moves.** OEWS figures are three-year survey averages rather than counts, and they are treated accordingly: the fitting retains them in spirit, but allows them to move within reason where the industry structure implies it, and does not match them exactly. Where OEWS suppresses an occupation total, that total is estimated from the published totals above it in the occupation hierarchy before fitting begins.

Occupation codes are expressed in the EP vocabulary. Where one OEWS occupation maps to several EP occupations, its employment is divided evenly among them.

Each public [industry data page](/data) lists the occupations that staff its industry, with their estimated jobs, from this pattern.

### Geography

Staffing patterns are built at three levels: the nation, each state, and the substate staffing regions OEWS reports for.

**There is no county-level staffing pattern.** A county takes the pattern of the substate region containing it, applied to that county’s own industry employment. Two counties in the same metropolitan area will share a staffing pattern, though the final occupation figures will vary based on how much their industry composition differs. Each state’s unallocated area takes the state’s pattern.

### Earnings

Earnings are not regionalized the same way employment is. Employment is regionalized as a *share* of an industry; earnings are regionalized as a *ratio* to the industry’s own per-worker earnings in that area.

The practical consequence: a DTG occupation wage is always DTG’s industry pay for that area multiplied by an OEWS-derived ratio. It is never an OEWS wage reported directly. An occupation’s pay premium or discount relative to its industry comes from OEWS; the level it is applied to comes from DTG industry data.

Only employment is reconciled to industry data. OEWS earnings are hourly wages annualized at 2,080 hours, and not every job is paid for forty hours a week, fifty-two weeks a year. Occupation earnings totals are therefore not reconciled to industry earnings totals, and should not be expected to sum to them.

## Staffing patterns over time

OEWS is annual. DTG uses OEWS figures from 2015 onward. Each year’s pattern is a three-year average centered on that year, which smooths the survey’s year-to-year movement.

Years outside that span take the nearest available average. Every year from 2004 through 2016 uses the same earliest three-year average. Every year from the most recent OEWS year forward — including all projected years — uses the same latest three-year average.

This is a step function, not an interpolation between endpoints. Within each constant-pattern span, occupation figures still change from year to year, because the industry employment the pattern is applied to changes. What does not change is the pattern itself.

Two consequences follow, and both affect how the early years should be read.

**Occupational change before 2017 reflects industry change.** With the staffing pattern held constant, an occupation’s employment moves only as its industries move. Apparent growth in an occupation across those years is growth in the industries that employ it, not a change in how those industries staff themselves. For analysis of occupational mix — whether an industry is staffed differently than it used to be — work within the period DTG’s OEWS use covers.

**Early-year values can change between vintages.** The staffing patterns that 2004 through 2016 depend on are re-derived from their sources with every release, and those sources are revisable. This follows from the same principle as restatement: each vintage is built in full, from current inputs, rather than carrying earlier results forward.

## Projections

Occupation data is projected across the full ten-year horizon, for every area industry projections cover.

A projected occupation figure is DTG’s projected industry employment with the most recent staffing pattern applied to it. There is no separate projection model for occupations.

**Staffing patterns are held at recent history rather than projected forward.** The BLS Employment Projections program publishes a view of how staffing patterns themselves will shift over the coming decade — the effects of automation, technology adoption and changing skill requirements within industries. DTG does not incorporate that view. A DTG occupation projection answers what happens to occupations if industry employment evolves as projected and industries continue to staff themselves as they do now. It is not a forecast of occupational change within industries.

That distinction matters most for occupations concentrated in industries where staffing is expected to change, and least for occupations whose fortunes track their industries’ employment directly.

## Provenance

Occupation values carry **no disclosure code, no reliability score, and no classification flag**.

The reason is the nature of the sources. QCEW is a census, so for industry data it is possible to say exactly which values BLS published, which it withheld, and how tightly the published record constrains an estimate of a withheld one. Occupation data has no equivalent published record to be measured against: it is constructed from a survey and a national matrix, and every value is the result of that construction. There is no meaningful line between published and estimated to report.

Whether an occupation value falls in a surveyed period, a period carried forward from more recent surveys, or a projected period is determined by the period itself. The boundaries between those spans are recorded with each vintage and reported alongside the data.

Within a vintage, all years use a single occupation classification, and the full series is expressed in it. Unlike industry data, values changed by that restatement are not flagged.

Still stuck? A person can help.

[Get support](/support)