The most contested question in the gig economy may be how much workers earn, since their hourly wages can be widely uneven. Concerns about pay have helped fuel moves in California and New York City to regulate gig-economy companies as if they were more like conventional employers.
Last week brought evidence that drivers for Uber and Lyft may be much better compensated than previously understood. But it hardly settled the debate.
A study by researchers at Cornell University found that the typical driver in Seattle made over $23 per hour after expenses during one week last fall. Previous studies for other areas had put net earnings well below $20 per hour. Another new study put the figure at less than half that.
“Given what we had read about the ride-sharing economy, we were extremely surprised by our result,” said Louis Hyman, an economic historian who was the Cornell study’s lead author.
The finding comes at an important moment for Uber and Lyft, which are facing a California lawsuit over misclassification of workers that could cost them hundreds of millions of dollars and a Seattle effort to impose a minimum wage for drivers.
But it has also raised questions from critics who have seized on the fact that Uber and Lyft sought out Mr. Hyman and provided his team with data.
Two prominent economists unaffiliated with the research team said the study had been rigorously executed, but second-guessed some of the researchers’ decisions. They also said the study reflected the limitations of research that seeks to assess the costs and benefits of prominent digital platforms — not just Uber and Lyft, but also giants like Amazon and Facebook.
In interviews, the two economists — Luigi Zingales of the University of Chicago business school and Lawrence Katz, a labor economist at Harvard — argued that because such studies were difficult to conduct without data provided by the companies, the entire body of research might be subtly skewed.
“We all make assumptions,” Mr. Zingales said. “If at the end of the day you want to please who gave you the data, you might choose a certain set of assumptions.”
The Cornell study is valuable in part because it drew on data from both Uber and Lyft, which allowed the researchers to avoid some potential pitfalls. For example, other studies may double-count a certain amount of work time because a driver who spends an hour with the Uber and Lyft apps simultaneously activated may seem to have worked for two hours rather than one.
Still, much of the result appears to have been influenced by two decisions. The first involved whether time spent waiting for a fare is work.
While other researchers have assumed that drivers are working any time their app is turned on — even if they’re not on their way to pick up a customer or don’t have a passenger in the car — the Cornell study counts such time as work only if it directly precedes a ride. If a driver turns on the ride-share app but is not dispatched on a ride before shutting it off, the authors do not count the time as work.
According to the Cornell authors, this assumption adds about $2.50 per hour to the typical driver’s earnings.
Mr. Katz, the Harvard economist, said the assumption appeared at odds with a conventional understanding of work. He cited the example of a receptionist, who is typically considered to be at work even during down time when he or she may be surfing the web.
The Cornell authors also assume that many of the costs of owning a vehicle, such as the value a car loses as it ages or financing costs, should not be considered work expenses because car owners would typically pay these costs even if they didn’t drive for Uber or Lyft.
The only costs the authors factor into their preferred calculation are so-called marginal costs — like gas and maintenance costs that accrue because of the extra miles a worker drives while on the job. This assumption results in costs that are up to about $5.50 an hour lower for full-time drivers, and a net wage that is several dollars per hour higher, than under a more conventional calculation.
But many Uber and Lyft drivers may buy a more expensive car in order to drive on these platforms. If that’s the case, the vehicle’s additional cost should be considered part of the driver’s expenses as well, according to Mr. Katz.
“I think the expense number is just way too low,” he said. He also worried that focusing on a single week might have created an unrepresentative portrait of earnings.
Analysis of wages and their impact often hinges partly on researchers’ assumptions, as with a pair of reports about the effects of Seattle’s minimum wage in 2017 and 2018 that produced somewhat divergent results.
Mr. Hyman of Cornell acknowledged that the assumptions on wait time and expenses might have overstated the results somewhat and noted that his team had provided analysis using alternative assumptions. He also said that the study had found a wide variation in earnings among drivers, and that driving might be a worse deal for full-timers than those who drive casually or part time.
Some critics on social media noted that the companies had paid Cornell $120,000 to support costs like research assistance, not an unheard-of arrangement for such work. (Mr. Hyman and his fellow professors received no money.)
Both Mr. Katz, who called the group “totally honest,” and Mr. Zingales suggested that the paper highlighted a more subtle problem: Academic work that relies on data controlled by companies tends to avoid negative findings.
Scholars typically obtain such data in two ways: They approach the company with a research question they would like to answer; one recent paper in this vein examined the wage gap between male and female Uber drivers, and another sought to put a value on the flexibility of working for Uber. Or the companies can approach scholars with a question they want answered, as with the Cornell study.
When the scholars are faculty members at an academic institution, the companies typically cede editorial control to them.
But the process still tends to skew what we know about the companies, Mr. Zingales said, because companies are unlikely to approve the release of data for a study, or approach a scholar with data, if they believe the conclusion is likely to reflect poorly on them. One such study, he has noted, recently asked whether traffic fatalities increase after Uber and Lyft start operating in a city, for which the companies did not provide detailed data.
Many scholars have an interest in maintaining a relationship with companies because it is difficult to answer key policy questions without access to their data. This can lead researchers to adopt more favorable assumptions when there is legitimate debate about how to handle a methodological question.
Mr. Katz said the problem arose with many companies, but was a growing concern with digital platforms in light of their size and relevance to the economy.
Mr. Hyman said Uber and Lyft had not influenced his results in any way. He said they had approached him after people at Uber read his 2018 book, “Temp,” a well-reviewed history of the rise of alternative work arrangements like freelancing that was relatively positive about the online gig economy.
The companies appeared to be pleased with the findings, and were quick to point out flaws in a study for the City of Seattle, also released last week, that put the typical driver’s wage at $9.73. That study calculated earnings from cruder data the city had obtained from Uber, supplemented by a survey of thousands of drivers.
“Cornell’s study is the first to provide an independent, data-driven picture of the full earnings experience of ride-share drivers,” Matt Wing, an Uber spokesman, said in a statement. A 2018 report of ride-hailing drivers prepared for the City of New York drew on detailed data it had obtained from the companies. He said the other study, by contrast, “is based on limited data and flawed assumptions.”
Mr. Wing said that Uber was willing to share more detailed data but that the city couldn’t commit to protecting its confidentiality.
Julie Wood, a spokeswoman for Lyft, which largely declined to share information with Seattle, also pointed to the merits of data used by the Cornell researchers and cited the other study’s data limitations.
Mr. Hyman seemed exasperated by the experience, particularly his dealings with what he called “the Twitter mob.”
“I would love to have dragged Uber’s name though the mud, trust me, but it’s not what the data showed,” he said. “I thought it was important for public debate and drivers most importantly to have access to the data nobody knew.”
“For me, this is not going to get me anything,” he added. “I’m tenured.”