Census A/B test finds question wording with definition best matches CPS standard for ‘active’ job search; employed workers account for many active searchers

Matthew Pesner, an economist in the Census Bureau’s Labor Force Statistics Branch, presented experimental findings on measuring job search in modernized SIPP. Using a May 2025 A/B test on the Household Trends and Outlook Pulse Survey (HTOPS), researchers randomized respondents into several question formats and compared classifications against a CPS-style two-step approach. Pesner said the ACS/SIPP consolidated question with added instructions produced active-search classifications closest to the CPS-style benchmark for both the non-employed and the employed.

Pesner emphasized caveats: the HTOPS sample is experimental, relatively small, skewed older, and lacked nonresponse follow-up, so findings are preliminary. He said the Census will continue validation work and adopt the ACS/SIPP wording with the additional definition in further testing.

On prevalence, Pesner presented descriptive results showing that roughly 30% of employed respondents did something to look for work in the prior month, while active search methods (actions that can directly lead to an offer, such as submitting a resume) occurred in about 17% of employed respondents. Among prime-age adults (25–54) the not-employed looking rate was near 50%, and employed people made up a majority of active searchers. Pesner also reported method-specific rates: looking at job ads was the most common passive method (~24% of employed), and submitting resumes was the most common active method (~12%).

Pesner highlighted industry heterogeneity: Public Administration showed an active-search rate near 8%, while Professional and Scientific Services showed a rate near 23%; for those two estimates the confidence intervals did not overlap in the test data. He noted that SIPP’s larger future samples could reduce sampling variability and enable breakdowns by industry, occupation and work arrangement.

An audience question about whether wording changes affected item nonresponse drew a response from Pesner that nonresponse impacts are experiment-specific and the bureau had not yet cleared results for that metric from this A/B experiment.

What happens next: Census plans further validation in larger, more representative tests and to incorporate the improved wording into future SIPP testing and development.