Market Data
Why Manufacturing Data Gets Revised: How to Read a Number That Will Change
The first release of almost every manufacturing statistic is a well-educated estimate that will change. Understanding why, and by roughly how much, is the difference between reacting to noise and reading a trend.
The headline number for almost any manufacturing series, industrial production, orders, employment, is a preliminary estimate, and it will be revised as more complete data arrives. This is not a flaw or a cover-up; it is the honest consequence of publishing quickly. Statistical agencies face a genuine trade-off between speed and completeness: release fast on partial survey responses, or release slow on complete ones. They choose fast and then correct, which means every early print carries a knowable amount of "will change." Reading the data well starts with treating the first number as a draft, not a verdict.
Where the revisions come from
The three main sources of revision
- Late responses. The first estimate is built on the survey forms returned by the deadline; the rest arrive afterward and shift the number.
- Better source data. Preliminary figures lean on samples and models; later, more complete administrative records replace the estimates.
- Updated seasonal factors. The adjustment that strips out predictable seasonal swings is itself re-estimated as new years of data accumulate, which nudges historical values.
The practical rule that follows: the most recent one or two data points in any series are the least settled, and the ones most likely to move. Industrial production currently reads 98.70 index (2017=100) (Jun 2026), up about 1.1% from a year ago, and the honest caveat is that the latest print is a preliminary figure carrying more revision risk than the values behind it. A single early data point that seems to break a trend is more often revision noise than a real turn.
The first print is the agency's best guess under a deadline. Treating it as final is how you end up reacting to a number that was never really there.
How this site handles it
Rather than pretend the latest number is settled, the approach here is to make revision risk explicit: the trust stack tracks vintages (what the data said as of a given date), flags which recent points are preliminary, and states uncertainty instead of burying it. A number you can trust is a number that tells you how much it might still move. That is why the disciplined read of any fresh release is to weight the trend over the tip, watch how the last few points revise, and reserve strong conclusions for the values that have had time to settle. Honest data is not data that never changes; it is data that is upfront about how much it still might.
The series that gets revised most
- 1972: 39.54 (Archive begins 1972; selected years shown)
- 1974: 37.78
- 1977: 44.58
- 1984: 50.74
- 1991: 58.39
- 1998: 87.28
- 2005: 101.00
- 2007: 106.61
- 2012: 99.24
- 2019: 98.96
- 2026 (latest): 98.70
Over the 54-year record industrial production has moved decisively rather than oscillated: from 39.54 at the close of 1972 to 98.70 today, up 150%, and now the highest since April 2022. A change of that size across a span this long is a level shift, not a cycle, and planning that assumes a return to the 1972 figure is planning against the whole record.
Read how vintages, revisions, and uncertainty are tracked across every series on the methodology page. See the methodology
Published 2026-08-05.