The rise of the government statistical apparatus in the 20th century was heavily based on survey data, including surveys of households and businesses. But the willingness to respond to such surveys has been in decline. David S. Johnson, Maggie Meinhardt, and John Sabelhaus provide an overview of the problem in “Why did people stop responding to federal economic surveys? What can be done?” (Brookings Institution, April 14, 2026).
For a sense of the challenge, consider the falling survey response rates from households over time for key federal surveys. The CPS or Current Population Survey collects an array of data on households and income, along with employment and unemployment, poverty, and health insurance. Ten years ago the response rate was almost 90%; now it’s fallen almost to 60%. The CPI-Housing Survey is a monthly survey of 50,000 rental households for what they pay in rent–and the cost of shelter determines about 30% of the Consumer Price Index. The CE or Consumer Expenditure Survey collects detailed data on households and their detailed spending habits, sometimes through surveys and, for a certain group, through diaries.
Response rates are falling for business surveys as well. The CPI-C&S is a “Commodities and Services” survey that gathers price information from retail stores, service providers, and websites as an input to calculating the inflation rate. The CES or Current Employment Statistics survey collects data from about 700,000 workplaces on number of employees, hours worked, and earnings. The ECI or Employment Cost Index is based on data from the National Compensation Survey of businesses that collects data on wages, salaries, and benefits. The IPP or International Price Program survey collects data on prices of imports and exports.
The authors discuss the standard reasons for declining response rates: a proliferation of surveys that makes people and firms less willing to bother with any of them, concerns over privacy of results, and more. But what’s to be done?
One answer is to rely more heavily on “administrative” data: for example, instead of asking people in a survey to report what they earn, get the data from Social Security or income tax records. Instead of asking people in a survey what government benefits they receive, get the data from the government program making the payments. The good news about administrative data is that it’s often much more accurate than survey results. The bad news is that governmetn programs were not set up to be sources of data for social scientists. Their record-keeping isn’t organized in that way. Moreover, there are concerns about preserving privacy. There has been a considerable shift in this direction in economic research, although it’s still a work in progress.
As an additional step, one can imagine doing more to scrape the web for data on prices, job openings, and the like, although steps along these lines are already underway as well. I suppose one can also just ask an AI tool to survey the web and offer an estimate, but that approach is not yet ready for prime time.
But at the end of the day, some survey data isn’t easily replaceable. For example, the definition of “unemployment” is that a person is both out of work and also actively looking for a job. (If you aren’t actively looking, you are counted as “out of the labor force” rather than unemployed.) It’s not clear how to find out if people are looking for work unless you ask. Similarly, it is not always easy to get data on certain kinds of future expectations, like whether firms are planning to hire in the near future, without a survey tool.
The declining survey response rate is unlikely to reverse itself. But one can imagine a future where the surveys can be scaled back, when other approaches are available to gather the data, and can also become shorter so that they focus on the questions that can only be answered with survey data. The authors also note: Finally, although respondents to large-scale economic surveys like the CPS and CES participate voluntarily, other private and government surveys often use incentives—small gift cards or entries into a drawing for a larger prize—to compensate participants for their time.” Maybe the future of government data surveys will involve handing out lottery tickets, and publicizing the winners?
