A new Commerce Department policy announced last week bans the use of noise infusion, a statistical technique that allows agencies like the Census Bureau to publish detailed data about individuals and businesses while protecting their privacy. The likely result, according to Nathan Goldschlag, Research Director at the Economic Innovation Group and former Census Bureau principal economist, won't be more precise data — it'll be less data overall, weakening the public's ability to understand local economies and make informed decisions about jobs, business investments, and policy.

Noise infusion works by slightly altering granular data so that individual identities remain hidden while keeping statistics useful. The County Business Patterns, for example, uses multiplicative noise infusion — multiplying each business's employment by a random number to increase or decrease it slightly. In a county with 20 breweries, this allows the Census Bureau to publish total brewery employment without revealing any single brewery's exact headcount. For statistics covering large numbers of people or businesses, the noise cancels out and the figures end up very close to actual values. For smaller counties with five or ten businesses in an industry, the difference may be larger, but Goldschlag argues it's still better than nothing. Without noise infusion, agencies might only release state-level data, which is far less useful to local policymakers, businesses deciding where to invest, or workers choosing which skills to acquire and where to move.

The report explains that federal statistical agencies are required by law to protect the privacy of survey respondents, meaning they must ensure no one can back out a specific individual or business from published statistics. Agencies have three main tools to manage this: coarse aggregations (releasing less detail), cell suppression (removing certain statistics entirely), and noise infusion (fuzzing the data). According to the report, noise infusion has been used for decades in datasets including the County Business Patterns, Business Dynamics Statistics, and Quarterly Workforce Indicators. The new Commerce policy states that "coarsening shall be the preferred category of disclosure avoidance methods" and permits suppression only as a last resort, effectively forcing agencies toward releasing less granular data or no data at all for certain statistics.

Goldschlag agrees with the goal of making data more precise but argues this policy achieves the opposite. Statistical agencies already tend toward limiting data release due to privacy concerns — the Census Bureau's 2022 proposal to restrict Current Population Survey data was walked back only after user pushback. Without noise infusion, agencies face a stark choice: either publish data for a county with just a handful of businesses (risking privacy violations) or roll that data up to the state level, eliminating the local detail that makes it valuable. The report cites a 2005 study noting that for Quarterly Workforce Indicators, which track employment, hires, separations, and earnings by industry, geography, age, sex, race, and education, "only the application of state-of-the-art protection methods allows the Census Bureau to publish these statistics." Workers rely on this granular information to decide where to move and which jobs to seek. Businesses use it to determine investments. Policymakers design policies around it.

The timing makes the policy particularly troubling, Goldschlag writes, because there's growing bipartisan agreement that the country needs significant investments in its statistical system to measure AI's economic impact. He spent nearly 17 years at the Census Bureau before becoming Research Director at the Economic Innovation Group in 2025, and he observed firsthand the agencies' tendency to err toward caution and suppress data rather than release it. If the administration wants to rebalance toward producing more and higher-quality data, he argues, the solution isn't banning a disclosure method — it's changing the laws governing privacy protections or promoting leadership focused on data volume and quality. Otherwise, the likely outcome is that Americans lose access to the local economic information they need to understand their communities and make smart decisions about their futures.