Historically, the U.S. has been a leader in providing reliable and timely statistical information to support business strategy and policymaking. The value of information provided publicly and free of charge to businesses, households, and governments is immense. Yet underinvestment over the past 15 years is a key reason why the U.S. lost its position on the cutting-edge of public statistical services worldwide.

Since the beginning of the second Trump administration, this underinvestment has accelerated, and the administration has made intentional efforts to degrade the effectiveness and independence of the federal statistical agencies (FSAs). This accumulation of threats to the effectiveness of the FSAs will rapidly degrade the value of the key public good they provide, unless policy changes course sharply.

This blog post provides just one example of how cumulative underinvestment has blocked the ability of a key FSA to respond to developments, making its data less reliable over time. The Bureau of Labor Statistics collects a range of necessary data tracking the performance of the U.S. labor market. This BLS data are a key input into high-stakes decisions across the U.S. economy—including for both public and private actors. For example, the Federal Reserve relies on BLS data about unemployment rates, payroll job growth, wage growth, and price indexes to set monetary policy. The more volatile the BLS data are from month to month, the worse the information that guides Federal Reserve decisions.

Private industry also relies heavily on these statistics. A 2018 survey conducted by the National Association for Business Economists found that 95% of businesses responded “yes” to the question: “Are government data important for analyses and forecasting that drive business decisions?” Employment and unemployment data produced by the BLS were rated as the most important data source for informing business decisions.

Yet over the past 15 years, the BLS has gradually lost personnel and funding, which has been undermining their mandate of producing timely, accurate statistics on wages, prices, and the labor market. More recently, the Trump administration’s choices to freeze BLS hiring has further strained Census field staff charged with collecting household survey data. Worst of all, the Trump administration took the unprecedented step of firing the commissioner of the BLS simply because the agency accurately reported data that the administration happened to find politically inconvenient.

Even without further blatant political pressure on the BLS’s independence, the agency will encounter growing difficulty in doing its job effectively in coming years. One of their most important efforts is the fielding of the Current Population Survey (CPS), a survey of thousands of households across the U.S. taken every month, which provides detailed employment and wage information. The CPS is the source data for the monthly estimate of the nation’s unemployment rate, for example. This is in turn a key criterion for assessing whether the economy is heading into recession. In recent years—after the COVID-19 pandemic—the response rates for the CPS have sharply declined. These declines, if not countered with greater investment in response rates, may make it harder for economists and policymakers to make timely sense of changes in labor market, particularly for populations that already have small sample sizes, such as rural areas or detailed demographic groups.

The rest of this blog post highlights the problem of falling response rates, demonstrates that they have made some labor market measures more volatile month to month, and shows that these falling response rates have occurred over the same period as the retrenchment in resources for the BLS.

Nonresponse reduces sample size in the Current Population Survey

The Current Population Survey asks questions about employment and other labor market characteristics to 60,0000 households or about 110,000 individuals every month. Between 2005–2016, the Current Population Survey household survey was able to steadily receive responses from around 107,000 people, ages 16 and older. However, as noted by others and shown in Figure A, the number of households responding to the survey has declined since the mid-2010s and then fell precipitously after the COVID-19 pandemic. In the first few months of 2026, just over 75,000 individuals, ages 16 and older, had responded to the monthly CPS.

The decline in response rate has likely occurred for a few reasons. The Bureau of Labor Statistics notes that the rate of refusals had been increasing as early as the 1990s, likely as the world became more connected with computers and the internet, leading to less reliance on in-person interactions to conduct business. Social trust has also gone down over the past few decades, and the share of adults who agree that “most people can be trusted” has decreased by more than 15% since 1984.

More recently, the COVID-19 pandemic, coupled with concerns for privacy and distrust in the government, may be the reason that the rate of decline grew in recent years. The COVID-19 pandemic forced many workers to transition to remote work, and concerns about contagion limited overall social interactions, making response collection increasingly difficult. Additionally, concerns about privacy or retribution from the state felt by groups like immigrants may make some people more reluctant to answer questions for fear of deportation. 

Finally, distrust in the federal government, fueled by recent overtly political activity, could be behind some of the reduction in response rates. For example, when the Bureau of Labor Statistics published two consecutive months of large negative revisions to the number of payroll jobs in mid-2025, the Trump administration leveled charges—which were baseless and never backed up by any evidence—that the BLS had manipulated the data for political purposes and fired then Commissioner Erika McEntarfer. People are less likely to trust government if they think publicized information and facts are politically motivated. 

Smaller sample sizes are linked to less precision in key labor-market estimates

If the size of sampled households is large enough, declining participation does not have to significantly affect the reliability of statistics produced from the survey. However, if declines in participation reduce usable sample sizes too much, this can lead to estimates with less precision, which can reduce researchers’ ability to parse a signal from statistical noise in a timely manner, especially for economically vulnerable groups.

For example, because the unemployment rate for Black workers is volatile, it can be difficult to accurately diagnose labor market softness for this group. If the sample size is too small to generate statistical precision in each month, researchers will require increasingly more months of data to be able to diagnose labor market softness, which could jeopardize the timeliness of proper policy responses to support the labor market.

Every month, the Bureau of Labor Statistics publishes statistical significance summary tables, identifying whether changes in labor force indicators are statistically significant at the 90% level. BLS publishes these statistical significance tests for dozens of indicators across several demographic groups, including for Black workers. We collected these tables over time and documented the margin of error needed in order to claim a 1-month change in unemployment was statistically significant, shown in Figure B.

While the margin of error that is needed to claim a change is statistically significant varies with the level of unemployment rate, the reduction in precision from lower response rates is evident when we hold the unemployment rate constant. The two red lines in Figure B identify the effect size needed to claim statistical significance for a change from a starting unemployment rate of 7.3%. In November 2017, when the sample size of the labor force was 63,346, a 0.66 percentage point change in unemployment would have been considered a statistically significant change. In April 2026, when sample size of the labor force decreased to 45,416 respondents, a 0.84 percentage point change in unemployment is required to claim statistical significance.

If the declines in survey participation are not random across the U.S. population, estimates may also be biased, which runs the risk of conveying inaccurate information about the state of the economy. For example, if nonresponse is more likely to occur among unemployed respondents compared with employed respondents, the statistics derived from these samples may suggest labor market softness when there is none. These concerns are already materializing: The Census reported that nonresponse had biased income statistics from the CPS Annual Social and Economic Supplement upward by 2%–3% since 2020.

Researchers and field staff at Census and the BLS are aware of potential concerns of bias in their estimates and do their best to weight estimates using population counts from administrative data and other sources so that these issues don’t happen. However, if sample size declines continue on this trajectory, the BLS will need to create new methodologies and sampling strategies, all of which will require funding.

Steady throttling of BLS funding makes all decision-makers—public and private—less well informed

The declining precision of estimates in the Black unemployment rate is just one of the key indicators affected by a BLS that lacks resources to respond effectively to growing data collection challenges. Achieving a larger sample size for key surveys requires a well-functioning and well-funded BLS with personnel who can take on the challenges of administering surveys in the 21st century. Yet this is the exact opposite of what is happening. Figure C shows that from 2005 to the present, the staffing at the BLS went from roughly 2,500 employees to just over 2,150, a drop of about 15%.

Funding has followed a similar trajectory. Since its high-water mark in 2010, the BLS budget has declined from $810 million to $636 million in inflation-adjusted terms, a decrease of 20%. These cuts don’t hurt just the estimates generated by the Current Population Survey. In the past couple of years, the BLS has been forced to reduce data collection for the Consumer Price Index and to discontinue certain Producer Price Indexes in an effort to cut costs. At a time when affordability and price changes are top of mind for U.S households and businesses, depriving public and private decision-makers of accurate and timely information about prices makes little sense.

Increased funding would allow the BLS to maintain all their current functions and implement new procedures to address declining sample sizes. In 2023, BLS began to modernize the collection process of the CPS to improve response rates by allowing online self-completion of the survey and other collection process improvements for certain data products. This BLS initiative is happening in parallel to similar initiatives in several other countries undertaking modernization efforts. The United Kingdom, the Netherlands, Australia, and Canada have all received funding to launch similar modernization efforts for their own household surveys to address declining response rates. However, the BLS requests for increased funding for the modernization efforts have not been fully granted.

The decision to steadily defund the BLS is especially striking when weighed against the large economic benefits provided by the agency and other federal statistical agencies. The BLS provides up-to-date precise estimates of economic indicators that policymakers and business leaders alike rely on. Previous research finds that increased economic uncertainty can have negative effects on the economy, proving the important role that the BLS plays. Moreover, some economists have estimated in 2025 that the BLS generates economics benefits of about $25 for every $1 spent on the agency’s budgets. The 2025 FY BLS budget was approximately $636 million, meaning the BLS currently generates about $15.9 billion in economic benefit. Across all agencies, in FY 2022, the combined budget request for statistical agencies was $7.1 billion or 0.3% GDP, yet the benefits have been measured to be around $770 billion.

Conclusion

At a time when more information on the economic and social well-being of people and communities is needed, not less, funding the BLS should be a top priority. Addressing nonresponse will require substantial effort and creativity to counteract declining levels of social trust and anti-government sentiment. It will, for example, require public campaigns to convey that information provided to the BLS is confidential and safe, and changes in methodology to render the correct statistical adjustments, such that the statistics generated are unbiased. 

Rather than tackle these challenges head on however, the Trump administration put forward a proposal that would reduce the number of statistics about rural and less populous substate areas that could be published without running the risk of disclosing personally identifiable information. These proposals are a lazy solution to the real but solvable problem of making public data widely available and fully confidential. They would provide less information on the economic and social well-being of citizens, likely leading to delays in accurately diagnosing economic and social problems.

When agencies like the BLS are underfunded and understaffed, they aren’t able to conduct the critical functions of their agency or serve the public to the degree their mission entails. Funding for these organizations shouldn’t be up for debate, given how strong of an economic benefit they deliver.