Disclosure and provenance

Every value in DTG’s industry data carries information about its own origin. Three signals travel with it: a disclosure code saying where the value came from, a reliability score saying how tightly the published record constrains it, and restatement flags marking values changed when the series was restated onto current classifications. All three are available through the API and in exports.

They exist because a labor market series assembled from public sources is not uniform. Some values are published by the Bureau of Labor Statistics exactly as DTG reports them. Others are estimated because BLS withheld them, or brought forward from a more current survey, or projected into periods no survey has reached. An analyst who has to defend a number should be able to see which kind it is without asking.

Scope

These three signals apply to industry data, which is derived from the Quarterly Census of Employment and Wages.

QCEW is a census rather than a sample. It counts jobs actually reported by employers under state unemployment insurance programs, which makes it possible to say exactly which values were published, which were withheld, and how tightly the published record constrains an estimate of a withheld one. That precision is what the disclosure code, the reliability score and the restatement flags report.

Occupation data is built differently, from survey-based sources, and does not carry these signals. What it is built from and how is described on the occupation data page.

The four disclosure codes

0 — Disclosed. BLS published this value. DTG reports it as published.

1 — Estimated. BLS withheld this value to protect the identity of an employer, and DTG estimated it using its proprietary estimation process. The inputs and the constraints that process satisfies are described on the industry data page. Estimated values in published QCEW quarters carry a reliability score.

2 — Carried forward. QCEW is published roughly two quarters after the reference period ends. For quarters after the most recent QCEW release, DTG extends the series using the Current Employment Statistics and State and Metro Area Employment programs. What carries forward is the rate of change rather than the level: quarter-over-quarter movement measured in those surveys is applied to DTG’s own base, then reconciled across industry and geography so the result sums correctly at every level.

3 — Projected. Code 3 covers all projected periods. The method is described on the projections page.

Reliability scores

Suppression is a property of QCEW, not of DTG. Where BLS withholds a value, any figure reported for that industry and county is an estimate, and how close that estimate can get depends on two things: how much the surrounding published data constrains it, and how well the estimation method uses that constraint. The first varies enormously from one value to the next — in some places the published record leaves almost no room, in others a great deal. The reliability score does not introduce that variation. It measures it.

How much is withheld also varies by the size of the area. Characteristically, around 81 percent of industry values are withheld and estimated in counties with fewer than 1,000 covered workers, against around 35 percent in counties with 500,000 or more. The exact shares differ from one vintage to the next; the pattern does not.

Data Tactical Group, measured against vintage 20260913 (DTG 2026.3), 2004–2025, all NAICS levels, 3,144 counties. Shares are pooled and rounded; they drift between vintages, the pattern does not.

For values in published QCEW quarters, DTG reports a reliability score between 0 and 1.

The score reflects how tightly the published record constrains the value. Where disclosed values leave only a narrow range the unknown values can occupy, reliability is high. Where they leave a wide range, it is lower.

Consider ambulatory health care services (NAICS 621) in a county, where the sector total is disclosed at 1,000 jobs. If five of its seven component industries are also disclosed and sum to 980, the remaining two must together account for about 20 jobs. Neither is published, but neither can be far from a narrow band. If instead the disclosed components sum to 300, the remaining two share 700 jobs between them, and far more distributions are possible.

Illustration using real industry structure with illustrative figures, not a real county.

The illustration uses one industry and its components for clarity. In practice the constraint is computed across every level of the industry hierarchy and every level of geography at once, so a value is bounded by what is published above it, beside it, and around it. The process draws on every constraint the published record provides, and the range reported for a value is the range those constraints leave.

Two properties follow.

The score is computed the same way everywhere. It is not normalized to local conditions and is not a ranking of values within an area: it reports the constraint on a value directly, by the same rule in every county and at every level of aggregation.

Every value in a published QCEW quarter carries one. Disclosed values carry a score of 1. Estimated values carry a score reflecting their constraint. The score is available at every level of aggregation, not only for detailed industries.

The score is measured on employment and applies to the row. It is computed from the range the employment value could take, and the same score is shown beside that industry, area and period’s establishments and wages. The range is measured relative to the size of the value, so the calculation is not scale-neutral: scores on very small and very large values are not directly comparable.

Carried-forward and projected values do not carry a reliability score. Those values are not constrained by a published QCEW record, because none exists yet for the period.

The reliability score is not a confidence interval, a margin of error, or a standard error. Those are properties of estimates drawn from a sample. The reliability score describes how much room the published data leaves for a value to vary, not how well DTG’s estimate fills that room.

Using the score

The score reports how much the published record determines an answer. It does not tell you whether to use a number, because that depends on knowledge DTG does not have.

A high score means the surrounding published data leaves little room for the value to be much different from what DTG reports. A low score means the published record is thin there, and a reader with local knowledge — a known employer, a recent opening or closure, an industry they work in directly — should weight that knowledge more heavily against the reported figure.

The score is built for judging particular estimates within the area being studied, rather than for ranking one area against another.

The score is computed from the employment range and applies to the whole row, including the earnings and establishment counts beside it. It is not comparable between values of very different size.

Restatement flags

Within a DTG vintage, every year is expressed in a single industry classification and a single set of county definitions, and the full series is restated onto them. Values changed by that restatement carry flags: one marks a value changed by industry reclassification, the other a value changed by a change in county definitions.

Each flag is applied individually and only where the value itself changed. A detailed industry may be flagged while the sector total containing it is not, because the total was unaffected.

A flag is not a quality warning. It indicates that the value is reported under a classification different from the one in effect when the period occurred, which is what makes years within a vintage comparable to each other. It also explains why a DTG value for an earlier year may differ from what BLS published for that year at the time.

Reading the three signals together

The disclosure code says where a value came from. The reliability score says how tightly the published record pins it down. The restatement flags say whether restatement changed them. They are independent, and a value may carry any combination.

For work that must remain stable — a published report, a filing, a board document — the most durable figures are disclosed values in completed QCEW quarters. For current conditions, carried-forward quarters are the most recent read available, at the cost of being more heavily modeled and more likely to change in later releases.