Projections

The Quarterly Census of Employment and Wages is published roughly two quarters after the reference period ends. Anyone working from QCEW alone is therefore always looking at conditions that are at least six months old, and has nothing at all for periods still ahead.

DTG extends the series in two stages. Quarters after the most recent QCEW release are brought current using monthly survey data. Periods beyond the reach of those surveys are projected, ten full years forward.

Both stages are labeled. Survey-extended quarters carry disclosure code 2 and projected periods carry code 3, described on the disclosure and provenance page. Neither carries a reliability score, because neither is constrained by a published QCEW record.

Bringing the series current

Two relevant Bureau of Labor Statistics programs publish monthly. The Current Employment Statistics program covers the nation; the State and Metro Area Employment, Hours, and Earnings program covers states and metropolitan areas. DTG uses the unadjusted series from both.

Neither program publishes at the industry and geographic detail DTG reports. CES publishes industry data at varying levels of NAICS detail, so a mapping is built from the lowest level each program publishes down to 6-digit NAICS. CES does not publish county data at all, so every county is mapped to the metropolitan or nonmetropolitan area covering it in SAE.

Monthly figures are combined into quarterly averages. Because survey counts do not match QCEW counts, what is taken from them is the rate of change, not the level: quarter-over-quarter movement is measured as a percentage by industry and region, and applied to DTG’s own data starting from the most recent QCEW quarter.

Because the two programs publish at different industry and geographic granularity, the resulting values do not naturally sum. Each extended quarter is reconciled — first across geography, from the nation down to states and counties, then across industry, from the all-industry total down to detailed industries — and then fitted so that every industry level in every area is consistent with every other. The same summing guarantees that hold for published quarters hold here.

Ten-year projections

Beyond the reach of CES and SAE, DTG projects ten full years forward. Three inputs are used: DTG industry data, the survey-extended quarters above, and the BLS Employment Projections program’s national industry projections.

The national outlook

The Employment Projections program publishes employment by industry and occupation for exactly two points in time: a base year, roughly nine months after it ends, and a projection year ten years later. It publishes for the nation only, using custom industry aggregations, and its values are subject to the same disclosure limits as other published data. DTG unsuppresses EP using a similar process that applies the same methodology it uses for QCEW, described on the industry data page.

Two published points do not describe a path. The interim years have to be inferred, and how they are inferred matters: analysis of past EP projection cycles shows that projected change tends to arrive earlier in the period rather than being spread evenly across it. A straight line between the two points would therefore understate movement in the early years. DTG interpolates the interim years using a dampened logarithmic curve, and extends the same curve to cover the years beyond EP’s published horizon.

Projecting each year

Projections are made one year at a time. The first projected year is the year containing the quarter after the last quarter covered by CES; where that leaves a partial year, the remaining quarters are projected so the year is complete.

Employment is projected for every industry at every level, for the nation, every state and every county. Each industry in each area is projected individually, using the Holt-Winters method, also known as triple exponential smoothing, which accounts for both year-over-year trend and seasonal movement within the year. Because DTG data is quarterly and unadjusted, seasonality is not a nuisance to be removed but a pattern to be carried forward.

Where an industry reports no employment in the most recent quarter, it is treated as no longer present in that area, and no projection is made.

Establishments and earnings

At the national and state levels, establishment counts are projected the same way employment is. Earnings are projected as average earnings per worker rather than as total wages, and the projected per-worker figure is applied to projected employment.

Because establishments and employment are projected independently, two constraints keep them from diverging implausibly. Where employment is projected to grow while establishments are projected to fall, or the reverse, establishment counts are held at their level in the same quarter of the previous year. And because establishment counts do not in practice move as fast as employment, establishment change is limited to a fraction of the rate of employment change.

At the county level, only employment is projected directly. Establishments and earnings are derived from the previous year’s establishments-per-worker and earnings-per-worker ratios, applied to projected employment, and then reconciled against the projected state and national totals. County earnings therefore carry the trend of the levels above them rather than a forecast of their own.

Adjustment to the national outlook

Once preliminary national projections are made for a year, they are adjusted to the EP outlook. The adjustment is made on percentage change from the previous year, not on the level EP implies for that year. Adjusting to levels would override a trend already visible in the recent data; adjusting to change preserves it while still following the national outlook.

EP’s custom aggregations do not align with NAICS levels, so after adjusting the industries EP represents, the parents and children of those industries are adjusted in turn, repeatedly, until every industry level for the nation is consistent.

One exception: the first projected year is not adjusted to EP when it contains at least one quarter already covered by CES. A live survey signal takes precedence over a national outlook published against an older base year.

Reconciliation

Each projected year is then fitted so that all industry levels and all geographic levels agree with one another, exactly as the survey-extended quarters are.

What projections are, and are not

A DTG projection carries forward the trend and seasonality in an area’s own history, disciplined by the national outlook BLS publishes. It is not a forecast of events. Plant closures, openings, policy changes and recessions are not anticipated, because nothing in the inputs anticipates them.

Projections are most useful near the present and least useful far from it. A projected quarter shortly after the last surveyed one is a short step from observed data. The further a period sits from the last surveyed quarter, the more of its value comes from the shape of the projection and the less from observation.