Controversy Over BLS Data Collection Methods
After Trump announced the dismissal of McEntarfer, the White House website published an article accusing her of dereliction of duty and data inaccuracies during her tenure. The article cited excerpts from multiple media reports, claiming that the BLS had consistently overestimated employment growth before "quietly making significant downward revisions." For example, the agency revised down the annual benchmark wage growth data as of March 2024 by 818,000 jobs, marking the second-largest annual revision in history. The White House also criticized the agency for frequent technical errors and sensitive data leaks under McEntarfer’s leadership.
In response, several former statistical agency officials defended McEntarfer and the BLS. The "Friends of BLS" organization, co-founded by William Beach (BLS Commissioner during Trump’s first term) and Erica L. Groshen (Commissioner under the Obama administration), issued an open letter accusing Trump of trying to shift blame for bad news and stating that the U.S. employment statistics process is "designed to be decentralized to prevent interference."
Beach further explained that the 818,000-job revision was due to the BLS obtaining actual employment data each March through the Quarterly Census of Employment and Wages (QCEW). However, since QCEW is based on unemployment insurance records, it only provides a complete statistical snapshot for March. In routine surveys, the BLS uses the "birth/death model" for estimation, which tends to produce deviations during economic turning points—underestimating job growth during recoveries and overestimating it during downturns.
Gary Evans, Head of Research Solutions at BCA Research, commented this week: "Every time the jobs data comes out, I’ve said it’s terrible—probably the most watched but worst-quality data out there. The numbers are often heavily revised, and the birth/death model doesn’t work well. Wage data is typically revised sharply downward before recessions and upward in early expansions. Trump’s criticism about poor data quality has some merit, but it’s probably hard to do much better."
Regarding whether the birth/death model has become outdated, she told First Financial that traditional statistical models (e.g., small business survival curves, startup rate predictions) assume a relatively stable economic environment. However, high-frequency shocks in recent years—such as the pandemic, supply chain crises, and the AI revolution—have disrupted this stability.
Notably, in June, the BLS announced plans to scale back CPI data collection. By late July, the agency revealed further details, stating the actual reduction far exceeded expectations. Data shows that in June, the agency suspended about 19% of traditional data collection.
Meanwhile, when direct price data collection is impossible, statistical agencies often resort to "imputation" techniques—using similar local or broader regional data to estimate missing values. However, BLS data indicates that the agency has increasingly relied on cross-regional rather than local data for imputing missing CPI survey values. From June 2024 to June 2025, the share of "local imputation" (using same-area data) dropped from 92% to 65% due to declining price submissions from local businesses.
Omair Sharif, founder of Inflation Insights, noted: "My main takeaway is that their data collection problems are far worse than we thought." He argued that relying on estimation methods inevitably widens error margins.
