Data on the economy are not precise and get revised over time. Why? The early estimates of economic data like GDP are initially released based on partial information. We only get more complete information later—sometimes much later.

Government statisticians face a tradeoff between timeliness and accuracy. If the government wants to release only GDP numbers that are very accurate, it would have to wait for a long time before releasing any data at all. For example, a fairly accurate measure of GDP for 2024 will be released by the government statistical agency (the Bureau of Economic Analysis, BEA) at the end of September 2025. That’s a long time to wait to get a comprehensive measure of how much output was produced by the U.S. economy. Instead, we got a first estimate of GDP for 2024 in January 2025, and it was revised in February and again in March. But more complete source data are available by the end of September, and the number for 2024 that will be released then will be much more accurate. For that reason, it is crucial that policymakers understand the process of producing the data.

Recently, some policymakers have focused on the revisions to the employment data for May and June 2025, and wondered if they might have been manipulated for political purposes. But the data on employment are routinely revised, sometimes by a significant amount. We need to remember that those data are monthly, and the size of the data revisions was puny, because monthly movements in the data are small. The revisions in question were less than 0.1 percent of the total number of people employed. However, it is true that the revisions were relatively large, compared with other revisions to the number of people employed at firms. Only 4 percent of all revisions from the first release to the second release since 1964 were larger in magnitude than the revision of the June value. Only 7 percent of all revisions from the first release to the third release since 1964 were larger in magnitude than the revision of the May value.

But I wonder why people focus so much on monthly data in the first place. Monthly data are very noisy—our measurement is not precise and many random things happen in a month. It would be preferable to look at quarterly data or even annual data. When you average a data measure over longer periods, you smooth out the noise and get a more precise measure. With quarterly averages, 15 percent of all revisions from the first release to the third release were larger than that in the second quarter of 2025; while 17 percent of all revisions from the first release to the second release were larger than that of the second quarter of 2025.

So, our analysis of past revisions going back to 1964 suggests that the revisions that just happened to the employment numbers were larger than normal, but far from the largest, and many revisions in the past have been even larger.

Even more importantly, the fact is that government data agencies follow strict rules for how they conduct their work. They follow well-established procedures for how to deal with the raw source data they get and then generate from those incomplete numbers the overall values of the data series being produced. If you look at when the largest revisions to GDP and employment have occurred in the past, you see that unusual events led to large revisions. The largest revisions to the GDP data occurred in the stagflation of the 1970s, the 9/11 attacks, the financial crisis in 2008, and the pandemic in 2020. Large employment revisions occurred in similar time frames. It is possible, and worth further exploration, that the cuts to government employment in 2025 by DOGE were not counted well by the BLS. If so, perhaps some modification of BLS procedures could be in order. But it is difficult for a government statistical agency to anticipate that type of shock, and modify its procedures to account for it, before it happens. In my studies of data revisions over the past 25 years, I have learned a lot about the data production process. The key thing to remember is that if data were not revised, the data would just be wrong forever. Data revisions improve the precision of the data. We are fortunate that U.S. data are among the best in the world, and always getting better, especially after they are revised.