A federal statistic that its own agency withdrew

In May 2017 the US Bureau of Labor Statistics fielded the Contingent Worker Supplement to the Current Population Survey, the household survey that produces the country's monthly unemployment figures. The supplement had not been run since 2005. Alongside its longstanding questions about temporary and contract work, it carried four new questions intended to identify what the agency called electronically mediated work: short jobs or tasks that workers find through a mobile app or a website which also arranges payment.

The results were published in June 2018 and were read as the first official federal measurement of app-mediated work. Within months the Bureau of Labor Statistics announced that the electronically mediated work estimates could not stand. On review, a substantial share of respondents recorded as electronically mediated workers had answered the questions about something else — describing ordinary employment that merely involved a computer, an app used at work, or an unrelated arrangement. The agency reissued the series with the misclassified cases removed, and the corrected share was materially lower than what had been published.

Why the original figures are not repeated here

The withdrawn numbers still circulate in secondary write-ups. They are not reproduced in this note, because the institution that produced them stated that they did not measure what they were intended to measure. A figure whose author has retracted it carries no evidential weight, and reprinting it beside the corrected series invites exactly the comparison the correction was meant to prevent.

What the episode does document is a measurement problem rather than a labour market one. The questions were new, they were asked at the end of a long supplement, and they used a description of work that respondents evidently mapped onto their own situations in ways the designers had not anticipated. The Bureau of Labor Statistics said as much in its explanation of the revision.

What the supplement was built to do

The Contingent Worker Supplement has a narrow purpose. It classifies workers by the expected duration and the contractual form of their main job during a single reference week. Its core categories — contingent workers, independent contractors, on-call workers, temporary help agency workers, workers provided by contract firms — were designed in the 1990s, long before platform intermediation existed.

Two consequences follow. The supplement counts a person's main job, so work done for a few hours beside a primary job is largely invisible to it. And it captures a single week, so activity that is seasonal or occasional falls outside the window unless it happens to coincide with the survey. Both features are documented design decisions, not defects, but they set hard limits on what the instrument can say about intermittent online work.

What the note takes from it

The correction is useful precisely because it is public. A statistical agency published a number, found the number wrong, explained the mechanism, and reissued the series. That is the process working. The reason to record it is that the retracted figure had already been repeated widely by the time the correction appeared, and corrections travel more slowly than the estimates they amend.

For anyone comparing measurements of platform work, the practical implication is narrow and specific: the household survey supplement and the administrative datasets discussed elsewhere in this index count different populations, over different windows, with different definitions of what counts as a job. Differences between them are not necessarily disagreements about the world.