When two economists revised their own widely cited estimate

In 2015 Lawrence Katz of Harvard and Alan Krueger of Princeton commissioned a version of the government's Contingent Worker Supplement through the RAND American Life Panel, an online survey panel. The federal supplement had last been fielded in 2005 and there was no official reading of the intervening decade. Their working paper, circulated in 2016, reported that the share of US workers in alternative work arrangements — independent contractors, on-call workers, temporary help agency workers and workers provided by contract firms — had risen from about 10.7 per cent in 2005 to about 15.8 per cent in late 2015.

The figure was cited heavily. A rise of roughly five percentage points in ten years, arriving as ride-hailing and task platforms entered public argument, was read as evidence that the employment relationship itself was shifting. The authors were careful about what their categories contained — the great majority of alternative arrangements had nothing to do with apps — but that caution travelled less well than the number.

The reassessment

When the Bureau of Labor Statistics fielded its own supplement in May 2017, it found little change from 2005. Katz and Krueger responded by re-examining their result rather than defending it. In a 2019 paper in RSF: The Russell Sage Foundation Journal of the Social Sciences, they concluded that the increase they had measured was considerably smaller than their earlier estimate implied, and that the alternative work share had risen only modestly over the period.

Two explanations carried most of the weight in their account. The first was the instrument: an online panel and an in-person household survey recruit and question people differently, and differences in mode can shift how respondents classify their own work. The second was timing. Their survey ran in late 2015, closer to the aftermath of the recession, when involuntary contract and temporary work was more prevalent than it would be a few years later.

What the revision does not say

The reassessment is not a finding that platform work does not exist or has not grown. Alternative work arrangements are a much broader category than app-mediated work, and a stable total can conceal movement between its components. The paper addresses the aggregate share and the comparison between two surveys; it does not measure earnings, hours or the number of people who take on tasks through an app beside another job.

It also does not settle which instrument was closer to correct. It documents that two credible measurements disagreed, identifies plausible mechanisms for the disagreement, and lowers the confidence that should be placed in the first result. Krueger died in 2019, shortly after the reassessment appeared.

Why it sits in this index

The sequence is unusually clean: a well-designed survey produced a striking number, a second measurement disagreed, and the original authors published a correction of their own work rather than leaving it to others. The original estimate nonetheless remains in circulation, often without the revision attached.

Read alongside the agency correction recorded elsewhere in this index, the case makes a narrow point about reading any single headline share of the workforce. Survey mode, reference period and category definition each move the number, and a difference between two studies is frequently a difference between two instruments rather than a change in the labour market.