What the First Print Got Wrong

Seventy years of jobs reports, each one compared with the number it became. The first print is usually too low, except when it matters most.

The record

The almanac keeps every monthly jobs number as it was first announced, rebuilt from the St. Louis Fed's archive of vintages, back to mid-1955. Comparing those 856 first prints with the figures in the record today is the cleanest test there is of how much the number you read on jobs Friday should be trusted. The answer has a shape.

Over the whole record, the first print has been revised up more often than down. In months outside recessions, the current figure is on average about 22,000 higher than the first announcement; in recession months it is about 16,000 lower. Late responders and the model that fills in unsurveyed businesses both lean toward the recent trend, so the early edition undercounts a growing economy and overcounts a shrinking one. The revisions page explains the mechanism; this is the record.

DecadeMonthsAverage revision (current minus first print)Average size of revision, either directionShare revised down
1950s (from mid-1955)55+5811736%
1960s120+238845%
1970s120+499932%
1980s120-109352%
1990s120+499328%
2000s120-147450%
2010s120+184943%
2020s (to Sept. 2026)81-1910963%

Two things stand out. The 2010s were the most accurate decade in the series: the average revision in either direction was about 49,000, half what it had been in the 1960s. And the 2020s reversed the long-run sign: for the first decade in the record, most months have been revised down, and in 2024 and 2025 the average revision was about 94,000 a month downward, with the economy not in recession. Whether that is the birth-death model lagging a slowdown in new businesses, falling survey response, or the immigration drop making the population estimates stale, is argued below.

Every month since 1955: jobs added as first reported against the figure in the record today. The pandemic months are clipped. Where the first-print line sits above the current line, the announcement was too high; through most of the record it sits below. Full chart and context →

New articles and charts arrive in one short email on Monday mornings. Get the weekly digest

Around the turns

The sign matters most at turning points, which is where the first print is least reliable. The table shows the average revision in the year before each recession began, during it, and in the year after it ended.

RecessionAverage revision, 12 months before it beganDuring12 months after it ended
Sep 1957 – Apr 1958+36+16+77
May 1960 – Feb 1961+50-31+4
Jan 1970 – Nov 1970+32+52+80
Dec 1973 – Mar 1975+52+70+26
Feb 1980 – Jul 1980-25+75+7
Aug 1981 – Nov 1982+7-32+42
Aug 1990 – Mar 1991+10-5+2
Apr 2001 – Nov 2001-11-19-43
Jan 2008 – Jun 2009-8-125-16
Mar 2020 – Apr 2020+0-333+59

The pattern: going into a downturn the first prints run close to right or a little high, during it they run high (by 125,000 a month on average in 2008–09, by 333,000 in the two pandemic months), and coming out they run low. That is the early edition lagging the turn in both directions. The largest single misses in the record are all at turns: March 2020 was first reported as a loss of 701,000 and now reads a loss of 1.4 million; November 2021 was first reported as a gain of 210,000 and now reads 658,000. A reader in either month, planning on the first print, was off by about half a million jobs.

The other three

The other first-print lines on the almanac are far better behaved, which is itself useful to know.

GDP revises most. Across the 47 quarters since the archive begins in late 2014, the advance estimate has moved by about 0.9 of a point on average in either direction, and in two quarters the sign flipped: the second quarter of 2022 from minus 0.9% to plus 0.6%, and the first quarter of 2025 from minus 0.3% to plus 0.1%. Revisions have run upward about two-thirds of the time.

The unemployment rate barely moves: the average revision is seven-hundredths of a point and seven in ten months are revised by a tenth or less. It comes from a separate survey with no late responders and no model, which is the reason economists who distrust the first jobs print lean on it.

Inflation moves even less, about six-hundredths of a point on average, almost entirely from the annual recalculation of seasonal factors each February. The index itself is never revised.

What people argue about

Why the 2020s flipped. One camp: the birth-death model, which assumes new businesses open at the recent rate, overstated job creation as that rate fell, and a model that has leaned the same way for two years is a bias to fix. Another: response rates have dropped below half and the first print is simply built on less, so the error is noise with a bad streak. A third: population growth slowed sharply when immigration fell in 2025, and the benchmark against tax records caught up with a change the survey weights hadn't. The Bureau of Labor Statistics has said it is reviewing the model. The data show the direction; they don't show the cause.

What to do with the early edition. One side reads the record as a case for the three-month average and the unemployment rate, and for ignoring the Friday headline. The other says a first print with a known tilt is still information: in an expansion, assume the truth is a little higher; at a turn, assume it is worse than it looks. Both readings follow from the tables above, and the almanac shows the first-print line so a reader can apply either.

See also

Sources

The weekly digest

One email on Monday mornings: the week's new articles and charts, the numbers that moved most, and what comes out in the week ahead. No ads, and no tracking of opens or clicks.

Past issuesHow addresses are kept