Revisions and Vintages: The Number You Remember Is Not the Number in the Record

Most economic figures are published before the data are in, then corrected for months or years. The first version is the one everyone reads.

In one sentence

A vintage is a figure as it stood on a given date; most statistics have many vintages, and the one in the headline is the earliest and least complete.

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

The analogy

The early edition. A newspaper printed at ten at night carries the story as it was known at ten; the morning edition corrects the score, the count, the spelling of the name. Both are the paper. The difference is that most people read the early edition, remember it, and never see the correction, which lands on an inside page a month later. The jobs report's first print is the early edition. The figure in the almanac's history is the morning edition, sometimes the one printed a year later.

The almanac keeps both. The "as first reported" companions on four charts are the early editions, rebuilt from the St. Louis Fed's archive of every vintage; the main line is the current one; the gap between them is the revision. Each number revises differently:

Jobs added as first reported (lighter line) against the current figure, with the three-month average. The first-print line sits above the current one through 2008 and again in 2024–25: in slowdowns, the early edition runs high. Full chart and context →

Where the analogy breaks

A newspaper's early edition is wrong at random. Economic first prints are wrong with a direction. Businesses that are closing don't return surveys, and the model that fills in unsurveyed firms assumes the recent trend continues, so first prints run high when the economy is turning down and low when it is turning up. The chart shows it: the first-print line sits above the current one through 2008 and 2024–25. And the morning edition is final. An economic figure never is; the benchmark resets the level a year later and the annual GDP revisions go on for years, so "the latest" is the best available description, not the truth.

What people argue about

Whether revisions mean the numbers are rigged. A run of downward revisions in 2024–25 became a political charge in August 2025, when the head of the Bureau of Labor Statistics was dismissed after a report that cut the prior two months by 258,000. The argument on the data: the same downward pattern appears in 2008 under a different administration and commissioner, the methods are published before the data, and the household survey, run separately, told the same story. The argument against the bureau: the pattern is persistent enough to be a bias, and a model that fills in unsurveyed firms should be fixed rather than defended. The chart shows the pattern. Intent isn't in it.

Whether to wait for better data. Survey response rates have fallen for twenty years: the establishment survey's first-print response rate is now under half, down from well over half a decade ago, which makes the early edition thinner. One side says publish later with more responses; the other says markets, the Fed and the public want the number on the first Friday and would read a leak instead. The statistical agencies' budgets are the quieter version of this argument.

Whether to read first prints at all. Some economists argue the three-month average and the unemployment rate (which barely revises) are the only monthly figures worth reacting to. Others say a first print with a known bias is still information if you know the bias, which is the case for keeping the vintage line on the chart.

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