Despite the voluble testimony of pundits and bar companions, the world of work is not one of Uber drivers and temp workers. In fact, the share of US employment accounted for by contingent and “alternative” arrangements is lower now than it was in 2005 and 1995.

That testimony is derived from several original sources. For example, a much-ballyhooed 2014 study commissioned by the Freelancers Union — which is not a materially disinterested party — reported that a third of workers are freelancers. The claim of a 2016 paper by Lawrence Katz and Alan Krueger that “all of the net employment growth in the US economy from 2005 to 2015 appears to have occurred in alternative work arrangements” was widely quoted and quickly became folk wisdom. That paper was based on an online survey conducted by the RAND Corporation. The survey was small — fewer than 4,000 respondents — and its sample wasn’t very representative of the overall population, a flaw the authors corrected through vigorous statistical handiwork.

Data released yesterday morning by the Bureau of Labor Statistics should put an end to this chatter. According to a special edition of their Current Population Survey, a monthly poll of 60,000 households conducted jointly with the Census Bureau, just 3.8% of workers were classed as contingent in May 2017, meaning they don’t expect their job to last a year. That’s down from 4.1% in 2005 and 4.9% in 1995. (Reports from the years before 2017 are here.) Tighter definitions show smaller shares, but also down from earlier years. In 2017, 96.2% of workers were non-contingent, compared with 95.1% twenty-two years earlier.