This is the third and final in a series of three articles from the end of 2019 that we have dug out of the archives in support of the Barcelona taxi drivers’ current conflict with Uber (see HERE)
Over his long career, Hubert Horan has worked on many of the most critical issues in the field of transportation economics, including the impact of regulation and mergers on industry efficiency, and the impacts of consolidation and other industry structural changes on consumer welfare. Mr. Horan graduated from Wesleyan University in 1976 with a B.A. degree and Honors in Economics. In 1980 he graduated from Yale University’s School of Management with an MPPM (MBA) degree.
Cross-posted from ProMarket of the Stigler Center
Uber’s employees co-authored academic papers with brand name scholars that were then used to back the company’s PR and lobbying strategy. Published in respected journals, those articles are based on proprietary data and non-replicable analysis. Moreover, they all don’t discuss the subsidies that make it possible for Uber to pursue market dominance despite its endless losses.
Part one of this series documented Uber’s uncompetitive economics, its inability to earn sustainable profits in competitive markets, and its explicit pursuit of quasi-monopoly industry dominance. Part two described Uber’s manufacture and promulgation of PR narrative claims, which were almost entirely wrong. Uber used propaganda-based techniques that had proven successful in partisan political settings to create the widespread belief that a company that has lost over $20 billion and had been openly pursuing quasi-monopoly industry dominance was highly innovative and successful and created huge welfare benefits for consumers and cities.
This first part of this post outlines the major problems found throughout Uber’s “academic research” program, which should be seen as an integral part of Uber’s overall narrative program. The second part of the post presents reviews of four prominently publicized papers that illustrate that this program was producing research that formally met traditional academic standards but whose results are highly problematic. A number of other Uber-sponsored and supported journal articles that I have reviewed are just as dubious as the ones discussed here.
Unless links to other sources are provided, more detailed explanations and source documentation of points about Uber’s economics and narrative programs in this series can be found in my 2017 Transportation Law Journal paper and my recent American Affairs Journal article.
The core findings of all four papers directly support important Uber PR claims: that Uber’s growth was driven by major productivity advantages; that the regulations Uber evaded significantly reduced traditional taxi productivity; that Uber’s drivers have higher earnings and greater job satisfaction than traditional taxi drivers; that Uber has created billions in annual consumer welfare benefits; and that any regulatory limits on Uber’s operating practices would significantly reduce driver welfare. However, the core findings of all four papers are highly problematic.
|“None of the papers actually analyzed their nominal subjects (comparative taxi operating productivity, the labor market for taxi drivers, changes in consumer welfare since Uber’s entry, or factors affecting driver welfare).”
Uber’s “Academic Research” to Support Its PR narrative
Uber’s “academic research” program was established by David Plouffe and Jonathan Hall in 2014 to allow Uber supporters to assert that major narrative claims were backed by rigorous, independent academic research. In each case, Uber engaged the participation of well known, often brand name economists, who usually were open public supporters of Uber’s agenda. The articles were published in journals considered prestigious.
Employing academics to consult on corporate PR and lobbying projects isn’t necessarily a problem. If Uber had clear evidence of productivity advantages, or higher driver pay, or large consumer welfare benefits, it could publish that evidence under its corporate name and invite independent outsiders to review the evidence and vouch for its legitimacy.
The problem is that Uber supported academic research that formally followed the same procedures and standards as independent or university-sponsored research but whose outcome would almost certainly prove useful to its strategy. Uber benefited tremendously not only from journal articles, but also from simplified versions of the articles’ findings that were then widely publicized by pro-Uber columnists and think tanks in non-academic channels such as newspapers and internet blogs.
This allowed Uber to transform narrowly stated journal claims into much broader, tweetable claims (“academic research proves Uber produces big benefits for drivers”) aimed at the same mainstream press and policymakers who Uber’s overall PR narratives had been targeted at.
Editors of prestigious economic journals were unable to critically evaluate articles about detailed aspects of Uber’s business model. Responses to the original articles would be impossible since other economists would not have access to Uber’s proprietary data or its research funding.
Just as Uber knew the mainstream press was largely pliant and uncritical about its overall PR narratives, it knew that none of the reporters or policymakers reading the simplified claims would have the time or ability to scrutinize the claim, or to determine whether it was actually supported by the original paper, or to evaluate the claimed independence of the original analysis.
Luigi Zingales’s October 9th ProMarket post pointed out that Uber’s policy of only allowing selected academics access to its proprietary data raised concerns about how academics could “become an unintended instrument in the PR efforts of powerful firms.”
As the discussion of the four papers below indicates, there was nothing unintended about Uber’s efforts to utilize academics in support of its PR objectives. Any academic journal claims put forward by Uber and their external academic consultants should be presumed dubious.
There are countless examples in other fields of academic research actually sponsored or ghostwritten by private companies. Big corporations such as Monsanto used to cooperate with scientists to support research projects that could be useful for their strategies, as internal e-mails revealed. In the case of Uber, companies’ employees directly co-authored academic papers.
Cramer and Krueger’s Claim
The 2015 Cramer/Krueger paper Disruptive Change in the Taxi Business: The Case of Uber, published initially by the National Bureau of Economic Research and then by the American Economic Review in 2016, provided an analytic background to two major Uber PR objectives. It gave credibility to the claim that Uber had a huge productivity advantage (38 percent overall; 66 percent in some cities) over traditional taxis thanks to its cutting-edge technological innovations and its evasion of traditional regulations. It also distracted the public from the huge subsidies that were the primary driver of observable taxi vs Uber competitive dynamics.
Alan Krueger acknowledged having worked as an Uber consultant in 2014-2015 when he was also writing the initial draft of one of this paper.
The paper significantly overstated its Uber vs taxi utilization advantage by using incompatible measures of total work hours—total shift hours for taxis, time with the app on for Uber. Taxi drivers have a huge incentive to drive very long shifts in order to cover daily vehicle lease and fuel costs, while Uber drivers have incentives to drive short shifts during demand peaks, and have incentives not to turn their app on unless prepared to immediately accept ride requests. The paper concealed this by not presenting any of the actual data (hours/miles with passengers, total hours/miles worked) used in the utilization comparison.
Cramer and Krueger seemed to have little grasp of the peaking and empty backhaul problems that actually drive utilization rates, and they made no effort to present any analysis of historic taxi operating issues. As the Chen, Chevalier, et.al. paper (discussed below) documents, most Uber drivers only drive short shifts during demand peaks, and have incentives not to turn their app on unless prepared to immediately accept ride requests. Thus, the Uber data largely reflects seat occupancy during peak demand periods, while the taxi data measures seat occupancy across an entire day. The paper did not present any of the actual data (hours/miles with passengers, total hours/miles worked) used in the utilization comparison.
Cramer and Krueger presented Uber’s superior productivity as driven by “technology” and “scale” effects and regulatory evasion, but they provided no evidence showing that any of these factors increased productivity. They can’t point to anything Uber’s “technology” did to smooth demand peaks or stimulate new demand closer to each drop-off point.
The paper notes that radio dispatch technology is not very new, but provides no evidence that Uber’s slightly faster transmission of demand requests could possibly explain the large claimed utilization differences. Their “scale effects” claim is similarly unsubstantiated. Market share did not shift to Uber because its size had dramatically reduced unit costs, or because passengers strongly prefer larger companies, it shifted because Uber’s massive subsidies gave customers much more capacity and much lower prices than could be economically justified
Cramer and Krueger’s only example of how Uber’s regulatory avoidance might have improved utilization (jurisdictional limits, such as those preventing Manhattan drivers from picking up return fares at Newark Airport) was trivial, and their discussion of the evils of occupational licensing had nothing to do with the utilization rates purportedly at issue.
Normally, superior productivity would create a significant overall cost advantage and would drive superior financial results, but Cramer and Krueger avoid any mention of Uber’s staggering losses. Uber may have had a better ratio of revenue miles to total miles during this period, but this would have been explained by Uber’s massive subsidies, not by operational efficiencies.
Traditional taxi supply was limited to what actual fare revenue could cover, while passengers increasingly flocked to Uber vehicles since (in the period studied), they only had to pay 41 percent of the cost of their trips on average, and peak-period passengers received much larger subsidies. Cramer and Krueger improperly suggest that their purported utilization advantage explains why Uber could be charging less traditional taxis. They not only ignored that sustainable prices require sustainable profits but made no attempt to determine whether their claimed seat occupancy advantage gave Uber an overall competitive advantage.
Stripped of its unsubstantiated claims, the Cramer/Krueger paper considered car service utilization rates in five cities but failed to present any of the actual data, made comparisons based on inconsistent utilization measures, and failed to demonstrate any link between their data and overall competitiveness or financial results. It seems highly improbable that American Economic Review would have accepted a paper using inconsistent measures to compare bus utilization rates in five cities.
Hall and Krueger’s “Uber Drivers Earn More Than Taxi Drivers and Have Flexibility Benefits” Claim
The 2015 Hall/Krueger paper An Analysis of the Labor Market for Uber’s Driver-Partners in the United States was written when one of the authors was an Uber’s employee and the other, Krueger, an Uber consultant, raising the question whether it was commissioned for research purposes or as part of a PR campaign. What is certain is that Uber used its findings as a part of its strategy to address the PR crisis that developed after earlier Uber claims that its drivers earned on average $90,000 a year in New York and $75,000 in San Francisco had been exposed in the press as being entirely fabricated.
Krueger, who had been a White House colleague of then Uber-PR chief David Plouffe, actively helped promote the paper’s findings. Simplified versions of the Hall/Krueger conclusions were popularized through a variety of mainstream publications, thereby countering emerging criticisms of Uber’s business model. This normalized pro-Uber talking points, such as a Congressional testimony claiming that “A survey of Uber drivers showed that the vast majority are happy working for the company. They greatly value the flexibility in terms of when and how much to work. . . . They also seem happy with the pay.”
The paper seriously overstates Uber drivers’ earnings by using gross revenue instead of true take-home pay. The paper briefly mentioned that Uber drivers must bear the full vehicle expenses and financial risks while the taxi drivers they are being compared to do not and failed to analyze the impact of these additional expenses and risks on true driver earnings.
While claiming to be a labor market analysis, the paper can’t explain labor market dynamics, since it only looks at data from a single firm at a single (highly unrepresentative) point in time. There is no data on driver wages, working hours or conditions or turnover rates, no analysis of how Uber’s market entry changed the demand for drivers, or how the increased demand changed wages or working conditions.
The paper’s claims about overall Uber driver job satisfaction and the value of driver flexibility are based on surveys fraught with serious methodological problems (very low response rate, loaded and deliberately misleading questions, sample bias, etc.) that later journal articles have documented in detail.
The authors repeatedly highlight the rapid growth of Uber driver numbers but failed to tell readers about Uber’s huge subsidies and losses, which explains that growth and invalidates the paper’s claim that they reflect a superior working experience. Comparing the driver economics of the viable (but marginal) traditional taxi industry with a new entrant that has lost billions is a meaningless exercise.
The paper serves to distract readers from Uber’s awful economics by repeating the Uber narrative claim that “modern technology, like the Uber app, provides many advantages and lower prices for consumers compared with the traditional taxi cab dispatch system.”
Hall/Levitt et al.’s “Uber Increased Consumer Welfare” Claim
The 2017 Cohen/Hahn/Hall/Levitt/Metcalfe paper Using Big Data to Estimate Consumer Surplus: The Case of Uber claimed Uber annually creates billions in consumer welfare benefits. It also allowed Uber supporters to trumpet over-simplified versions of that claim in non-academic mainstream media channels. Steven Levitt, through his Freakanomics media franchise, played a major role in this process. “It’s not that we have to wonder whether demand curves exist or not, because we know they exist because we define them and they’re there. But I wanted to touch one; I wanted to hold a demand curve, and I had never had a chance to do that until I took Uber and it suddenly occurred to me that here was a chance to hold a demand curve in my own hand”—this 2016 comment by Levitt explains very well why so many economists were bamboozled by Uber and its virtually infinite amount of new data to do research with.
Other outlets that uncritically presented simplified versions of the paper’s claim included the New York Times, Wall Street Journal, Forbes, Reuters, and Bloomberg. The claims consistently went beyond anything supported by data and analysis in the paper. To cite one example, economist and Bloomberg columnist Tyler Cowen claimed the $6.8 billion represented the “social value” of Uber and claimed that this evidence of huge consumer benefits justified his view that current industry competition was “a fight between progress and protection.”
The idea that a company that lost $2 billion with a negative 149 percent profit margin in 2015 (and has subsequently lost an additional $18 billion) and was openly pursuing a global monopoly can be said to have created $6.8 billion in sustainable consumer welfare gains in ridiculous. The paper withheld any mention of Uber’s losses, made no attempt to explain what Uber had done to create these gains, and failed to tell readers that they depended on massive, unsustainable subsidies.
The paper did not present a serious welfare analysis, as it failed to present any before-and-after evidence showing the impact of Uber’s entry on industry prices or service levels. The methodology, which had never been used in any previous academic analysis, claimed that one can derive the medium/long term welfare impacts of industry competitive changes from a small sample of one company’s data about extremely short-term consumer purchase decisions. That sample was limited to frequent Uber users in four of the wealthiest cities in America who had already decided to order an Uber vehicle. Can one measure the impact of United Airlines on overall consumer welfare by extracting a sample of consumer reactions from its loyal frequent flyers to prices shown at its website at the point they urgently need to book a flight?
The paper was not measuring consumer welfare, it was measuring the very short-term price elasticity of a highly unrepresentative sample of customers. The willingness to pay 20 percent more when you need a car right now does not mean an across-the-board 20 percent price increase would leave demand unchanged. The study used very short-term elasticity estimates of wealthy, loyal customers in order to inflate their “consumer surplus” estimate which in turn inflated their $6.8 billion “consumer welfare” claim.
No one in the profession pointed out that the consumer welfare impact of Uber’s market entry couldn’t be calculated from very short-term, Uber-only price elasticity measures. Most of the summaries published by Uber supporters ignored these (rather significant) caveats and claimed that a demand curve based on inappropriate data was actually a great accomplishment.
Chen/Chevalier et al.’s “Uber Drivers Think Their Schedule Flexibility Is Worth As Much As 40 Percent Higher Cash Wages” Claim
The 2017 Chen/Chevalier/Rossi/Oehlsen paper The Value of Flexible Work: Evidence from Uber Drivers proved useful to Uber in countering growing criticisms of driver exploitation, and legal challenges to Uber’s claims that its drivers were totally independent entrepreneurs who were not entitled to the same labor law protections employees are entitled to.
When the article was first published, Uber spread these claims through various mainstream articles. The paper’s central claims—Uber drivers think their scheduling flexibility is as valuable as 40 percent higher pay, and that drivers would withdraw two-thirds of current supply if this flexibility was lost—helped Uber claim that opponents of its labor practices were trying to hurt drivers and reduce major economic welfare benefits that Uber created. Uber’s need for this type of “evidence” on driver benefits recently increased given the threat of California’s AB5 legislation, although there is nothing in that law that would limit the existing flexibility of Uber’s drivers.
While purporting to analyze the “value of flexibility” and its “impact on driver welfare,” the paper does not present any data about Uber drivers’ compensation over time, or any data about the alternative wage/flexibility packages available in the market. As with other Uber-financed studies, it uses gross driver revenues instead of the true driver take-home pay Uber could have developed. Supply elasticities are based on inappropriately short-term data from a single point in time. The study fails to present any data on driver turnover rates or Uber’s huge losses and subsidies. The paper doesn’t consider Uber’s massive, unilateral reductions in driver compensation (roughly 40 percent since 2015) or that drivers now earn much less than traditional taxi drivers did prior to Uber’s market entry.
The paper claims that the alleged driver welfare benefits were created by “technology” and “[o]ur expectation is that technology will enable the growth of more Uber-style work arrangements.” There is nothing in the paper substantiating any link to technological innovation or explaining why (after ten years of Uber operations) no other large company has profitably adopted Uber’s work practices.
The paper’s claim that the Uber operating model “impos[es] no constraints on labor supply flexibility” badly misrepresents actual driver flexibility. The paper does not mention that drivers locked-in to expensive vehicle obligations have very little “flexibility” to quit once they realize that Uber take-home pay is much less than promised, or if they discover other jobs with a better combination of pay and conditions. The paper also fails to mention that drivers’ ability to decline work was seriously constrained by Uber’s practice of unilaterally terminating drivers who failed to respond to given percentages of ride requests.
In addition to overstating the flexibility in the Uber model, the paper’s conclusions are based on a strawman comparison with a fictitious alternative with the same wage uncertainty and financial risks but no flexibility to decline work. The real-world alternatives are low-wage jobs with more rigid schedule requirements but complete certainty about hourly wages and the total work available, and no requirement to provide capital assets.
The paper misrepresents a system designed for Uber’s benefit as a system that significantly enhances driver welfare. Because Uber demand has major (but short) evening peaks, Uber’s systems are designed to attract part-time drivers during these hours. There is nothing wrong with a system that tries to tie driver supply more tightly to Uber’s revenue maximization opportunities, but this doesn’t increase driver “flexibility.” Drivers can’t flexibly achieve similar earnings whenever they want or by driving as long as they want; they can only make money if they operate short shifts during the hours of greatest value to Uber. The study acknowledges that the “driver surpluses” that drive its headline benefit finding disappear once drivers want to work full time.
The ability to decline work on a given day undoubtedly has some value, but the paper implausibly claims the vast majority of its estimated driver welfare benefits are due to the ability of drivers already working to decline trips in real-time. These trip refusals are not because drivers exercise the “flexibility” to have coffee with a friend, but because the offered trips would have been uneconomical due to the low likelihood of return/onward rides. Thus, the paper’s claim that drivers gain the equivalent of 40 percent higher pay from this “flexibility” is largely driven by situations where drivers would actually lose money.
The Four Papers’ Problematic Conclusions
None of the previous papers actually analyzed their nominal subjects (comparative taxi operating productivity, the labor market for taxi drivers, changes in consumer welfare since Uber’s entry, or factors affecting driver welfare). None of the papers provide readers with any of the relevant pricing, service, utilization or wage/working conditions data.
The four papers failed to mention anything about Uber’s massive losses and subsidies, which invalidate all of the stated conclusions since data about a company that is billions away from breakeven cannot reflect sustainable productivity breakthroughs or permanent welfare enhancements.
Each paper makes broad claims that would require analysis of industry/market data over time, but the papers only look at very-short term Uber-only data from a single time period. Most of the Uber data used is inappropriate, including the use of gross driver revenue instead of data on true driver take-home pay, and the use of extremely short-term data to measure longer-term supply or price elasticities.
All of the authors claim that the alleged improvements are all due to the superior economics of Uber’s business model (e.g. innovative technology, scale effects) but the papers provide no objective evidence substantiating any of these claims. Several make pronouncements supportive of Uber claims totally unrelated to any of the data and analysis.
It is unclear why the journals that published these papers would not have identified many of these questions as part of their normal review process. It is especially unclear why those editors accepted papers presenting powerful conclusions about Uber’s efficiency and welfare benefits, without requiring the authors to disclose Uber’s losses and explain the industry’s volatile competitive situation.