What if the number making headlines today is not the number we should remember tomorrow?
Earlier this year, we looked at the revisions the Bureau of Labor Statistics (BLS) has been making over the years when it comes to the monthly jobs numbers. As a quick refresher, the BLS typically releases the initial estimate of the number of jobs created three weeks after the end of the previous month. The BLS will then make two rounds of revisions once more data and information is collected and available for analysis. The first number is the one that always tends to get attention in the press, even though it is the least complete estimate by definition. Historically, revisions tend to be split between the initial numbers being an overestimate and an underestimate. In the past, the year with the most lopsided results was 2014, when the initial estimates were revised upward in 11 out of the 12 months. Twelve years later, we have seen the opposite pattern: all 12 months of 2025 were subsequently revised downward. Even though all initial estimates were rosier than they should have been, the numbers were not to the liking of the President, who fired the former BLS Commissioner, Erika McEntarfer, in early August of 2025. Since then, three months have been revised up, while seven (7) have been revised down. In the first half of 2026, it has been an even 3-3 split. According to their own analysis, the BLS has pointed to two challenges they are dealing with: one is non-response bias, and the other is an outdated view of how many people are dying and being born each year. While the former feels very related, the latter doesn’t feel as naturally connected. This has been an issue for a while, and the changes to the ARIMA model being used (the way of analyzing linear data) go back to the early post-COVID era. We will see if the first half of the year points to a more typical pattern in the revisions.
Key Takeaways:
🔢🔢🔢 It’s the numerator, stupid. In March of 2025, one of the bigger revisions was made to the overall number of jobs. Nearly a million jobs (~ 898,000) were removed from the total number of jobs in the United States. This is not to say that the BLS said 898k jobs were lost, but that the total number of jobs that were filled or able to be filled was being overestimated. This is a revision of 0.6%, which is 3x the typical revisions in the past.
🚫👥👉👤It’s not us…it’s you. BLS identified response and nonresponse error as the main reason why the numbers were not hitting the mark. Businesses who did participate in the initial studies tended to overestimate the number of jobs they were going to be creating. They also found that companies not participating in the initial surveys tended to be companies that were cutting jobs. While not mentioned in the analysis, it feels safe to posit that the political climate of 2025 likely had an effect on the data collection.
🔴⬆️ Nothing is perfect. Even if the problems above were perfectly solved, not all businesses can immediately be captured by the survey. New businesses do not immediately enter the CES sampling frame; instead, BLS relies on unemployment-insurance records to measure the effect of new business. While this is a long-standing practice, there is a significant lag between a business beginning operations and becoming directly observable through the monthly survey. Over the past few years, there has been a notable increase in new small businesses, and this could have an outsized effect on the results.
