Where the March 2026 Payroll Benchmark Moved Up—and Down
The preliminary March table ranks revisions by jobs and percent, from retail trade at -154,600 to transportation and warehousing at +135,100.
The Bureau of Labor Statistics estimated a preliminary March 2026 benchmark revision of -79,000 jobs (-0.1%) for total nonfarm employment and -178,000 jobs (-0.1%) for total private employment. Released August 28, 2026, the figures revise estimated employment levels for March 2026. They are not one-month employment changes, and they did not immediately alter the published payroll estimates, according to the BLS preliminary benchmark summary.
March 2026 benchmark revisions by industry
The modest total nonfarm revision concealed substantially larger—and offsetting—changes across industries.
The table below contains selected rows from the BLS release: the two broad employment aggregates, the listed major sectors, and four components within trade, transportation, and utilities. It is not a complete table of every detailed BLS industry.
All values are preliminary revisions to the March 2026 employment level, released August 28, 2026. Job values are expressed as signed numbers of jobs, and the sign also provides the direction. The figures come from the official BLS March 2026 benchmark table.
| Record type | Industry | Parent | Revision: jobs; percent |
|---|---|---|---|
| Summary | Total nonfarm | — | -79,000; -0.1% |
| Aggregate | Total private | Total nonfarm | -178,000; -0.1% |
| Major sector | Mining and logging | Total private | -6,000; -1.0% |
| Major sector | Construction | Total private | +62,000; +0.8% |
| Major sector | Manufacturing | Total private | -67,000; -0.5% |
| Major sector with components | Trade, transportation, and utilities | Total private | -98,000; -0.3% |
| Component | Wholesale trade | Trade, transportation, and utilities | -86,200; -1.4% |
| Component | Retail trade | Trade, transportation, and utilities | -154,600; -1.0% |
| Component | Transportation and warehousing | Trade, transportation, and utilities | +135,100; +2.0% |
| Component | Utilities | Trade, transportation, and utilities | +8,100; +1.3% |
| Major sector | Information | Total private | +87,000; +3.0% |
| Major sector | Financial activities | Total private | +85,000; +0.9% |
| Major sector | Professional and business services | Total private | -76,000; -0.3% |
| Major sector | Private education and health services | Total private | -96,000; -0.3% |
| Major sector | Leisure and hospitality | Total private | -33,000; -0.2% |
| Major sector | Other services | Total private | -36,000; -0.6% |
| Major sector | Government | Total nonfarm | +99,000; +0.4% |
This Markdown table is a static presentation. In a spreadsheet, downloadable file, database, or public data page, the job and percentage revisions should be separate numeric columns so readers can sort them independently.
The record type and parent fields are also essential. They prevent summary rows, major sectors, and components from being treated as if they were all additive peers.
The largest revisions depend on how the table is sorted
Among the industries listed, retail trade had the largest numerical decrease, at -154,600 jobs, while transportation and warehousing had the largest numerical increase, at +135,100. Sorting by percentage produces a different leader: information had the largest positive percentage revision at +3.0%, ahead of transportation and warehousing at +2.0%, as shown in the official industry table.
The rankings differ because industries have different employment bases. A revision affecting more jobs can represent a smaller percentage of a large industry, while a smaller numerical revision can be proportionally larger in an industry with fewer jobs.
A reusable version of the data should support four distinct sorts:
| Sort field | Treatment | Question answered | Example use |
|---|---|---|---|
| Signed job revision | Sort negative to positive, or the reverse | Which industries moved up or down most by job count? | Find the largest numerical increase |
| Absolute job revision | Sort by magnitude without regard to sign | Where was the largest adjustment in either direction? | Compare revision scale |
| Signed percentage revision | Sort the original percentage | Which industries had the largest relative increases or decreases? | Find the positive percentage leader |
| Absolute percentage revision | Sort percentage magnitude | Where was the largest proportional adjustment? | Flag large relative revisions |
Signed values should remain the primary stored data. An absolute-value field can be calculated for ranking, but it must not replace the original value: doing so would erase whether an employment estimate was revised up or down.
These rankings describe adjustments to previously estimated employment levels. They do not, by themselves, measure hiring momentum, layoffs, business openings or closures, industry performance, or recession risk.
Why the industry rows do not add up like a flat list
A relatively small total revision can coexist with much larger industry revisions because increases and decreases offset one another. The aggregate is the net result, not the largest movement found anywhere in the industry hierarchy.
Hierarchy creates a separate aggregation risk. Wholesale trade, retail trade, transportation and warehousing, and utilities are components of trade, transportation, and utilities. They help explain that parent sector’s revision, but they must not be added to the parent as though all five rows represented separate sectors.
The same rule applies farther up the hierarchy:
- Total nonfarm is the top-level summary.
- Total private is an immediate child of total nonfarm.
- Government is another immediate child of total nonfarm in this presentation.
- Private-industry sectors sit beneath total private.
- Components sit beneath their respective industry parents.
A safe calculation must select one mutually exclusive hierarchy level. To reconcile total nonfarm, for example, use its immediate children rather than combining total nonfarm with those children. To analyze total private by sector, use the relevant immediate children of total private and exclude total private itself. To analyze trade, transportation, and utilities by component, use its component rows and exclude the parent from the sum.
Do not use a filter such as “summaries plus major sectors.” That would mix aggregates with descendants and create double-counting. A robust data model should instead include fields such as:
record_typehierarchy_levelparent_idis_aggregate
The records can then be filtered by a specific parent and hierarchy level. Every aggregate should be excluded whenever its descendants are being summed.
Even with correct hierarchy filters, displayed values may not reconcile perfectly if the source values have been rounded or if a published extract omits relevant records. Calculations should therefore retain the original source precision where available and document the scope of the extract.
Detailed industry percentage revisions can also be larger than the total nonfarm percentage. BLS explains that statistical sampling error is generally greater at more detailed industry levels, while aggregation allows some errors and revisions to offset, as described in the official preliminary benchmark release.
What the preliminary benchmark measures—and what it does not
The Current Employment Statistics program, or CES, produces the payroll estimates commonly described as establishment-survey employment. The Quarterly Census of Employment and Wages, or QCEW, supplies more comprehensive employment counts derived primarily from state unemployment-insurance tax records.
BLS annually benchmarks the sample-based CES estimates against those broader counts. For the preliminary March 2026 benchmark, the revision equals the estimated universe employment count minus the not-seasonally-adjusted CES employment estimate for March 2026. The two counts are independently derived, and each is subject to its own potential sources of error.
That definition prevents several common misreadings:
- The total revision does not mean that the same number of jobs was lost during March.
- It is not a revision to one month’s payroll change.
- It is not a seasonally adjusted monthly employment estimate.
- Industry revisions are not direct counts of hiring or layoffs.
- A preliminary level adjustment should not be treated by itself as a recession signal or forecast.
The preliminary figures did not update the establishment-survey estimates published at the time of the August release. BLS said the final benchmark revision would be published and incorporated with the January 2027 Employment Situation release in February 2027. Until that final incorporation, the March industry figures must remain labeled preliminary rather than presented as current official payroll levels (BLS preliminary benchmark release).
A publication-ready structure for the BLS revision data
For reuse in a spreadsheet, database, downloadable file, or public data page, use one row per record and preserve these fields:
record_idrecord_typehierarchy_levelindustryparent_idparent_industryis_aggregaterevision_jobsrevision_percentdirectionreference_periodrelease_daterevision_statusunitsource_titlesource_url
The main analytical fields should have explicit data types and conventions:
| Field | Recommended type | Convention | Purpose |
|---|---|---|---|
revision_jobs |
Integer | Signed number of jobs | Preserves scale and direction |
revision_percent |
Decimal | Percentage units, not a proportion | Preserves the reported percentage |
direction |
Categorical | up, down, or no_change |
Supports labels and filters |
is_aggregate |
Boolean | true or false |
Helps prevent double-counting |
For revision_percent, define the convention in the data dictionary. A stored value of -1.0 should mean -1.0%, not the decimal proportion -0.01. Consistent units are especially important when data moves between spreadsheets, databases, charting tools, and APIs.
Store job and percentage revisions as signed numeric values rather than strings containing decorative signs. Publication formatting can add a visible plus sign to positive values, but the underlying value must remain numeric.
Derive direction from revision_jobs:
- Negative value:
down - Positive value:
up - Zero value:
no_change
Retain the original signed job and percentage fields after deriving direction. The category is useful for display and filtering, but it should not replace either source value.
Time fields need distinct meanings. Map the benchmark month to reference_period and the day on which BLS published the estimate to release_date; do not combine them into one ambiguous label. The former says when employment was measured, while the latter says when the revision estimate became available.
Every row in the current dataset should have revision_status set to preliminary. When final figures become available, preserve the preliminary snapshot rather than overwriting it. Add unambiguous comparison fields such as:
final_revision_jobsfinal_revision_percentfinal_minus_preliminary_jobsfinal_minus_preliminary_percent
The calculation direction should be documented as final minus preliminary. Separate fields are necessary because a difference in jobs and a difference in percentage points are different measures.
Provenance should travel with the data. Each reusable record should include the official source title and source URL, even if a public-facing HTML table displays the citation only once above the table. This makes a downloaded or exported row easier to audit without filling the visible page with repetitive links.
For visualization, publish separate views for job revisions and percentage revisions. A job-count chart emphasizes numerical scale, while a percentage chart emphasizes the adjustment relative to each industry’s employment base. Combining both measures into one ranking would obscure why their leaders differ.
TablePage is one option for turning a cleaned spreadsheet into an interactive public data page. Whatever publishing tool is used, the published dataset should retain its hierarchy, signed values, units, dates, revision status, and official provenance.
How unusual was the preliminary -0.1% total nonfarm revision?
By magnitude, the preliminary -0.1% revision was smaller than the 0.2% absolute average for annual total nonfarm benchmark revisions over the preceding 10 years, as reported by BLS. That comparison provides historical scale, but an average alone does not establish whether an individual revision is statistically unusual. It also describes total nonfarm employment; individual industries had materially larger positive and negative percentage revisions (BLS preliminary benchmark summary).
The central result is that the -79,000 total nonfarm revision masks much larger offsetting industry adjustments. The safest reusable dataset preserves signs and percentages, keeps aggregates separate from their descendants, records source and timing metadata, and labels every August 2026 value as preliminary. Retain that release as a dated snapshot, then compare it with the final benchmark only after BLS publishes and verifies the final figures.