What tax filings show that household surveys miss

Household surveys ask people to describe their own work. Tax records observe what was reported to the revenue authority. The two instruments answer different questions, and a 2019 study by Brett Collins, Andrew Garin, Emilie Jackson, Dmitri Koustas and Mark Payne used the second to look at a question the first has struggled with: how far platform work has displaced conventional employment in the United States.

The authors, working with Internal Revenue Service statistics of income, examined roughly two decades of individual tax returns and the information returns that accompany them — the 1099 forms that firms and platforms file when they pay a non-employee. That combination allowed them to identify filers with income from what they term the online platform economy and to follow the same filers across years.

What the records described

Two patterns dominate their account. Participation in platform work grew sharply from the early 2010s, and the growth was concentrated in transportation. At the same time, the income involved was typically modest and frequently intermittent: many filers reported platform income in one year and not the next, and for most it sat beside other earnings rather than replacing them.

That combination — many participants, small and unstable amounts each — is what allows an apparently large phenomenon to leave a faint trace in aggregate employment statistics. The authors also found that the growth in self-employment recorded on tax returns had not displaced traditional employment at the scale the public argument often assumed.

The limits the authors state

Administrative data removes the recall and self-classification problems of surveys, and introduces its own. Tax records observe reported income, so activity that is not reported is not there. Information reporting thresholds matter: a platform is only obliged to file a return above certain amounts and transaction counts, and those thresholds have changed over the period studied, which affects comparability across years.

The unit is also different. A tax return covers a filer and a year. It does not record hours worked, so it cannot produce an hourly rate, and it does not capture the unpaid time spent waiting or searching for work that other studies treat as central to measuring platform pay. Costs borne by the worker — a vehicle, fuel, equipment — enter only through whatever deductions were claimed.

Reading it beside the surveys

Placed next to the other entries in this index, the study illustrates why apparently competing measurements often are not in conflict. The household supplement counts a main job in a single reference week. The revised survey comparison estimates a share of the workforce in broad contractual categories. The tax records count filers with reported income of a particular kind across a full year.

Work that is occasional, secondary and small is close to invisible to the first, ambiguous to the second, and visible to the third. A responsible reading names the instrument before quoting the number, which is the practice this publication tries to keep.