Repeated "significant revisions," what exactly is wrong with the U.S. non-farm payroll data?

Why is the U.S. non-farm payroll data always revised, and what are the underlying reasons?

Ray Dalio, founder of Bridgewater Associates, publicly stated on social media platform X on August 4 that if it were up to him, he would fire the head of the U.S. Bureau of Labor Statistics, as the agency's "data estimation process is clearly outdated and prone to errors."

Analysts say Dalio's statement supporting the dismissal of the Bureau of Labor Statistics head echoes Trump's controversial decision last Friday, which was sparked by the significant revisions to the non-farm payroll data, as these revisions once again exposed the structural issues facing the U.S. employment data statistical system.

According to media reports, the fundamental reason for data revisions lies in the difficult trade-off between speed and accuracy. The Bureau of Labor Statistics needs to survey about 120,000 employers each month, but in recent months, 30%-40% of companies have failed to respond in a timely manner. When late-responding employers provide information that differs from the initial data, significant revisions occur. Additionally, the Bureau of Labor Statistics is under dual pressure from budget cuts and staff shortages, further affecting data quality.

According to a previous article by Jianwen, on August 1, the U.S. Bureau of Labor Statistics reported that the non-farm payroll data for May and June was revised down by a total of 258,000. Subsequently, Trump fired the head of the Bureau of Labor Statistics, Erika McEntarfer.

However, Dalio also emphasized that while he supports the dismissal decision, if Trump's motivation is due to "disliking weak employment data," that would be a serious issue, as manipulating data for political purposes by leaders is a typical sign of losing the rule of law and checks and balances.

Declining Survey Response Rates Drag Down Data Accuracy

Reports indicate that the Bureau of Labor Statistics tracks the number of jobs in the economy by surveying about 120,000 employers each month via phone or online, making it one of the largest monthly surveys globally.

However, in recent months, about 30% to 40% of companies have failed to respond on time, far exceeding the level of less than 20% a decade ago.

At the end of the month, the Bureau of Labor Statistics uses the available data for initial estimates, and then updates these estimates in the following two months based on delayed responses and revised seasonal adjustments. If late-responding employers happen to provide information that differs from the first batch, significant revisions will occur.

For example, the latest revisions show that the number of educators in state and local governments for May and June was about 109,000 less than previously estimated. According to Omair Sharif, founder of the analysis company Inflation Insights, delayed responses from schools are the main reason for this revision UBS economist Jonathan Pingle pointed out:

"The initial decline in response rates will only lead to larger revisions. They received a lot of additional information after the initial deadline."

The response rate for personal household surveys is also slowly declining. Ten years ago, the Bureau of Labor Statistics achieved nearly a 90% response rate in its monthly household survey, which is conducted by phone to determine the unemployment rate. In recent months, this figure has hovered below 70%.

Tightening Funds Increase Data Quality Pressure

According to media reports, the Bureau of Labor Statistics is also facing the dual pressures of budget cuts and staff shortages.

According to calculations by the Friends of the Bureau of Labor Statistics organization, the agency's funding from Congress has decreased by 22% when adjusted for inflation since 2010. This year's federal hiring freeze has further prevented the agency from replacing departing employees.

Staff shortages have forced the Bureau of Labor Statistics to stop collecting price data in Lincoln, Nebraska; Provo, Utah; and Buffalo, New York. In other parts of the country, the agency is missing about 15% of its regular price data collection on average.

Former Bureau of Labor Statistics officials stated that the tight funding situation hinders investments in modernizing data statistics. Economist Guy Berger believes:

"I do not agree with the statement that the Bureau of Labor Statistics cannot do better, but the idea that doing better is easy is equally unsupported."

Benchmark Revision Process Reveals Statistical Blind Spots

In addition to monthly revisions, the Bureau of Labor Statistics also conducts annual benchmark revisions, cross-referencing monthly data with comprehensive state-level data that covers nearly every employer in the country. This process provides direct insight into the number of new and closed businesses, which is a key blind spot in the monthly data.

Preliminary revisions in August 2024 showed that the economy created 818,000 fewer jobs from March 2023 to March 2024 than the Bureau of Labor Statistics initially estimated.

Trump at the time claimed that this data indicated early employment figures were artificially inflated, making the economy appear stronger. The final revision in March 2025 showed that the original data overestimated job creation by 598,000, rather than 818,000