Why an Economic Release Can Change After Publication

Keep three dates separate

An economic observation has a reference period, a publication date and a version. The reference period tells you when the activity occurred. The publication date tells you when a release became available. The version identifies which estimate a reader is using. A later article can discuss the same month using a different value without necessarily misquoting the original report. A useful comparison records all three rather than treating the month on a chart as the whole record.

ALFRED, maintained by the Federal Reserve Bank of St. Louis, preserves historical versions of economic data. Its documentation explains that a release date concerns availability from a source, while its own archive may add observations and revisions later. A historical database date should therefore not be treated as proof of the exact intraday moment a market participant could have seen a release. [S1]

Why payroll estimates are revised

The Bureau of Labor Statistics explains that initial Current Employment Statistics estimates are preliminary because some employers have not yet supplied payroll information. Two subsequent monthly revisions incorporate additional responses and corrections. Recalculated seasonal factors can also change seasonally adjusted estimates. An initial estimate is timely information produced with an incomplete response set, rather than a promise that its numerical value will never change. [S2]

Annual benchmarking is a different step. BLS reanchors sample estimates using a much more complete employment count based mainly on unemployment-insurance tax records. It can also make historical corrections or reconstructions. The final sample-based estimate after the monthly revision cycle is consequently not an assurance against later changes. Keep the kind of revision attached to the comparison; monthly revisions and an annual benchmark answer different measurement questions. [S2]

A comparison another reader can repeat

Use one clearly identified measure. The official BLS vintage workbook includes seasonally adjusted total nonfarm employment. In its Data sheet, reference months run across columns and release vintages run down rows. The BLS guide uses January 2021 as an example: its first estimate appeared in February 2021 and its third monthly release appeared in April 2021. Those labels identify where to compare the same reference month across releases. [S3]

For a repeatable worksheet, record the source URL, download date, reference-month column, both release rows, units, adjustment basis and both selected values. Subtract the earlier value from the later one, retaining the units. This is a procedure, not a reported calculation: no workbook values or measured revision are presented here. Keep the downloaded file with the worksheet so that a later download does not silently replace the evidence.

If comparing a monthly change instead of an employment level, label that choice explicitly. BLS provides both types in vintage form. Do not compare a level from one release with a change from another. Its Revisions sheet also distinguishes the change since the preceding publication from a comparison spanning several releases. [S3]

Match the evidence to the question

For a description of the economy using a later information set, a later estimate may be appropriate. To evaluate what a commentator could have known at an earlier decision date, preserve the available vintage and check the original release timing. These are different questions. A result calculated with revised history should not be described as a result calculated with information available at the time.

Before accepting a comparison, ask whether the reference period, release vintage, units and seasonal-adjustment basis match. Then ask whether the conclusion needs a level, a monthly change or a sequence of revisions. A revision alone does not establish the cause of a market move, the quality of a forecast or what an investor should do. Clear labels make a narrower, supportable conclusion possible.