The Survey That Changed Each Year It Was Taken
A commissioned count of American freelancers grew, shrank, and grew again. The methodology moved with it — and the funders' names were on every edition.
Who Commissioned the Count
The Freelancing in America annual report ran from 2014 to 2019 and became, in that period, the most-cited source for the size of the US freelance workforce. It was commissioned jointly by Upwork — the platform formed from the 2013–14 merger of Elance and oDesk — and the Freelancers Union, the membership organisation founded by Sara Horowitz. Edelman Intelligence conducted the fieldwork. That combination of funders matters before a single figure is read: a labour platform whose revenue depends on freelancer volume, and an advocacy organisation whose influence tracks with the size of the constituency it claims to represent, were jointly paying for a count of freelancers.
The Bureau of Labor Statistics, which runs the Contingent Worker Supplement to the Current Population Survey on an irregular basis, produced nothing between 2005 and 2017. Into that vacuum stepped privately commissioned research, and Freelancing in America filled it loudly.
What the Numbers Did
The 2014 edition put the freelance workforce at 53 million Americans — roughly a third of the employed population at the time. The figure circulated in press releases, congressional testimony, and policy papers. By 2016, the count had shifted downward in relative terms even as absolute numbers fluctuated, partly because the methodology was adjusted to clarify what types of work qualified. By 2019, the report claimed 57 million freelancers, a figure covering anyone who had done any freelance work in the past twelve months, a definition broad enough to include people for whom freelancing was occasional and incidental rather than a primary source of income.
That definitional latitude is the technical core of the problem. The Bureau of Labor Statistics distinguishes between contingent workers — those without an implicit or explicit long-term employment contract — and self-employed own-account workers. The Freelancing in America methodology drew on neither framework consistently. Online panel surveys, which the Edelman fieldwork used, introduce self-selection bias: respondents who identify as freelancers are more likely to complete a survey about freelancing. The Freelancers Union and Upwork disclosed their methodological notes, but those notes were appended to reports whose headlines were already written.
The McKinsey Global Institute's 2016 independent analysis, which estimated 54 to 68 million independent workers in the United States using a different instrument, suggested the Upwork–Freelancers Union figures were broadly in the right range, though McKinsey's category definitions and data sources also differed substantially from BLS standards. Ranges that wide, across serious institutions using different instruments, indicate that the underlying population genuinely resisted clean measurement — not that one number was simply wrong.
Why the Funder Is Part of the Methodology
None of this makes the Freelancing in America series useless. It tracked sentiment, working conditions, reasons for freelancing, and income security over five years, producing longitudinal data that the BLS supplements, taken only occasionally, could not supply. But the series consistently chose the broadest plausible definitions, and those choices were not neutral. A wider definition produced a larger number, and a larger number served both funders: Upwork could cite a 57-million-strong workforce as a market; the Freelancers Union could cite it as a constituency deserving legislative attention.
Reading any commissioned survey requires identifying who wanted what answer. The Freelancers Union made its policy agenda explicit and its funding transparent, which is more than many survey commissioners do. Upwork disclosed its role on the reports' cover pages. Transparency about the funder is necessary but not sufficient: the prior question is whether the methodology would have been designed the same way had the funders had no interest in the outcome.
The BLS Contingent Worker Supplement returned in 2017 and counted a substantially smaller population under its narrower definitions — approximately 5.9 million contingent workers under its broadest measure, about 3.8 percent of total employment. That gap between 5.9 million and 57 million is not a disagreement about facts; it is a disagreement about what question is being asked. The methodological choices made before fieldwork begins are already an argument. The funders named on page one are already telling you which argument they preferred.
Elsewhere in The Count