MIT’s Uber study couldn’t possibly have been right. It was still important

The economics are iffy.
The economics are iffy.
Image: Reuters/Darren Whiteside
We may earn a commission from links on this page.

Some time last month, the Massachusetts Institute of Technology published a two-page brief of a forthcoming paper, “The Economics of Ride-Hailing: Driver Revenue, Expenses and Taxes.” The brief explained that through a “detailed analysis” of survey data, the authors, all affiliates of Stanford University in California, had found that the median profit from driving for Uber or Lyft was $3.37 an hour before taxes—well under minimum wage.

The brief flew under the radar until March 1, when public relations firm Berlin Rosen sent an email blast about its “bombshell findings.” The internet was soon flooded with headlines like:

Uber has conducted several of its own economic studies on driver earnings, and it didn’t take long for economists who have worked with Uber to rush to the company’s defense:

Uber CEO Dara Khosrowshahi also had sharp words for MIT’s researchers:

What happened? The MIT study was based on self-reported earnings data from a survey of about 1,100 Uber and Lyft drivers conducted in 2017 by Harry Campbell at driver blog The Rideshare Guy. The authors made a series of adjustments to the results. Most importantly, they took the response to survey question 14, “How much money do you make in the average month?” and discounted it based on the response to survey question 15, “How much of your total monthly income comes from driving?” For example, if someone answered “$1,000 to $2,000” to question 14 and “around half” to question 15, then the authors decided that person made about $1,400 a month (a statistical assumption) and half of that, around $700, from ride-hail driving.

The problem with this methodology is that it is unclear from the original survey whether drivers responding to question 14 were reporting their total monthly income from gig/on-demand work, or from everything, including any part- or full-time jobs they might have. The question itself is unclear: “How much money do you make in the average month? * Combine the income from all your on-demand activities.” It is easy to see how one person might interpret that as, “how much money do you make from on-demand work in the average month?” and another as “how much do you earn total in the average month, including but not limited to all on-demand activities?” Assuming that even some drivers interpreted the question the first way, discounting becomes unnecessary and problematic.

This is all explained in a critique of the MIT paper that Jonathan Hall, Uber’s chief economist, posted on Uber’s blog on March 2. Hall noted that the MIT findings differed “markedly” from previous studies on driver earnings. A study Uber conducted with Princeton’s Alan Krueger estimated hourly net earnings (after Uber’s fees but before expenses like gas and vehicle depreciation) at $19.04 in October 2015; a study Uber did with Stanford professors estimated gross hourly earnings (before all fees) at $21.07 from January 2015 to March 2017; and Campbell’s own analysis of his survey put net earnings at $15.68 per hour. MIT’s researchers have agreed to revisit their findings.

Read the full MIT study and it doesn’t take a degree in advanced mathematics to tell that something is off. That the conclusions are grossly out of line should signal a possible error, and the full list of survey questions, included as an appendix to the paper, reveals multiple questions that could have been confusing and elicited inconsistent responses.

“Hall’s specific criticism is valid; in retrospect the survey questions could and should have been worded more clearly,” Stephen Zoepf, lead author on the paper and executive director at Stanford’s Center for Automotive Research, said in a statement on March 5. His early reassessment finds that median profit is between $8.55 per hour and $10, depending on the calculation used and either way, up from $3.37. With the new numbers, somewhere between 41% and 54% of drivers earned less than the 2016 minimum wage. Zoepf has asked Uber “help make an open, honest, and public assessment of the range of ride-hailing driver profit after the cost of acquiring, operating and maintaining a vehicle.”

This illustrates a broader point: This study was attempted for good reason. “Estimating the economics of ride-hail driving at a large scale presents a number of challenges,” Zoepf and his co-authors write in the paper’s introduction. “An individual driver can precisely observe his or her own operational costs, but does not know whether those costs are representative of other drivers or vehicles. Researchers and policymakers are at a further disadvantage since they neither know specific revenues nor do they know the year, make and model of vehicles driven to make informed estimates of costs.”

One of the biggest problems with the so-called gig economy is how little we know about it. In Uber’s case, most of what we know about driver earnings comes from Uber, a company that famously said New York drivers earned a median income of $90,000 a year (they didn’t), and later settled with the Federal Trade Commission (FTC) for greatly exaggerating how much drivers could actually make.

Uber is constantly fiddling with how drivers are paid, most recently dropping its commission structure for a new system that pays drivers on more of a metered basis. A lot of the money drivers earn also isn’t even based on fares—it’s tied up in “boosts,” “quests,” and other gamified incentives Uber uses to nudge drivers into working particular times and places. Because of those changes, most of Uber’s in-house academic research on driver earnings is by now outdated. Papers like Krueger’s also tend to highlight the net hourly earnings of drivers, which omits very real on-the-job expenses like gas, insurance, and vehicle depreciation.

This context is crucial because it means that when Uber says drivers make X, people tend not to believe it. Drivers don’t, the FTC didn’t, and a lot of the public probably doesn’t either.

When seemingly reputable researchers from places like MIT and Stanford come along with an analysis that says Uber drivers earn less than the minimum wage, a lot of people are ready to believe them. The cost of a bad reputation is the benefit of the doubt. We saw the same thing happen last year, when half a million people deleted their Uber accounts over a “strike” the company was said to break up but actually had nothing to do with.

Yes, $3.37 an hour was a crazy number. But when people are primed to believe that driving for Uber is a crappy job, then you better bet they are going to believe a prestigious academic study that comes along telling them exactly that.

An earlier version of this post first appeared in Oversharing, a newsletter about the sharing economy. Sign up here.