Download Uber's Taxi Leasing Experiment: An Analysis of Driver Participation and Earnings and more Lecture notes Linear Algebra in PDF only on Docsity!
American Economic Journal: Applied Economics 2021, 13(3): 1– https://doi.org/10.1257/app.
1
Uber versus Taxi: A Driver’s Eye View†
By Joshua D. Angrist, Sydnee Caldwell, and Jonathan V. Hall*
Rideshare drivers pay a proportion of their fares to a ride-hailing platform operator , a commission-based compensation model used by many service providers. To Uber drivers , this commission is known as the Uber fee. By contrast , traditional taxi drivers in most US cit- ies make a fixed payment independent of their earnings , usually a weekly or daily medallion lease , keeping every fare dollar net of lease costs and other expenses. We assess these compensation mod- els using an experiment that offered random samples of Boston Uber drivers opportunities to lease a virtual taxi medallion that eliminates the Uber fee. Some drivers were offered a negative fee. Drivers’ labor supply response to our offers reveals a large intertemporal substitu- tion elasticity , on the order of 1.2 , and higher for those who accept lease contracts. At the same time , our virtual lease program was undersubscribed: many drivers who would have benefited from buy- ing an inexpensive lease chose to sit out. We use these results to com- pute the average compensation required to make drivers indifferent between rideshare and taxi-style compensation contracts. The results suggest that rideshare drivers gain considerably from the opportu- nity to drive without leasing. ( JEL J22, J31, L84, L92)
Other Driver issues [the New York Taxi and Limousine Commission] identified include the perceived inflexibility of leases currently offered by lessors as well as the stress associated with starting shifts “in the red” having paid a set lease price at the beginning of shifts. —New York Taxi and Limousine Commission Resolution ( 2015 )
T
raditional taxi drivers in most large American cities must own or lease one of a limited number of medallions granting them the right to drive. Until recently,
- Angrist: MIT, IZA and NBER (email: [email protected]); Caldwell: UC Berkeley (email: scaldwell@berkeley. edu); Hall: Uber Technologies, Inc. (email: [email protected]). Alexandre Mas was coeditor for this article. Special thanks go to Emily Oehlsen for indispensable project management, to the staff of Uber’s Boston city team for host- ing and assisting with our study, and to Phoebe Cai, Anran Li, Gina Li, and Yulu Tang for research assistance. Thanks also go to seminar participants at the Bank of Italy, Chicago, Harvard, The IDC (Israel), MIT, NBER Labor Studies, Nova (Lisbon), Princeton, the Stanford Graduate School of Business, Tinbergen, Uber, U Tokyo, and Uppsala, and to Daron Acemoglu, David Card, Liran Einav, Amy Finkelstein, Henry Farber, Ed Glaeser, Nathan Hendren, Guido Imbens, Michael Ostrovsky, Amanda Pallais, and Frank Schilbach for helpful discussions and comments. The views expressed here are those of the authors and do not necessarily reflect those of Uber Technologies, Inc. or MIT. Angrist and Caldwell’s work on this project was carried out under a data use agreement executed between MIT and Uber. This study is registered in the AEA RCT Registry as trial no. AEARCTR-0001656. Caldwell thanks the National Science Foundation for support under NSF Doctoral Dissertation Research in Economics Award No. 1729822. This is a revised version of NBER Working Paper 23891. The project was approved by MIT’s Committee on the Use of Human Subjects† (COUHES) protocol #1605555831. Go to https://doi.org/10.1257/app.20190655 to visit the article page for additional materials and author disclosure statement(s) or to comment in the online discussion forum.
2 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2021
limited supply had turned medallions into valuable assets, typically held by inves- tors or fleet owners and worth hundreds of thousands of dollars. Most big city taxi drivers therefore lease their medallions by the shift, day, or week. Taxi drivers can drive as much or as little as they want, but they are on the hook for the lease. The rise of rideshare platforms, including Uber, means that many workers now have the opportunity to add to their earnings by driving private vehicles, no medallion lease required. By the summer of 2016, Uber had almost 20,000 active drivers in Boston, a figure that can be compared with Boston’s long-fixed 1,825 taxi medallions. In addition to reducing entry barriers and perhaps taxi fares, an important feature of the rideshare model is a proportional compensation scheme, with few fixed costs. 1 In return for a percentage of their earnings known to drivers as a fee or commission, rideshare drivers can set a work schedule without having to worry about covering a lease. Drivers who work long hours are still better off leasing because they keep every dollar earned on a relatively high farebox. But drivers with low hours should prefer to work on a rideshare platform. This paper uses a series of randomized experiments to compare the value of the proportional compensation scheme offered by rideshare companies with traditional taxi compensation (Angrist, Caldwell, and Hall 2021). The latter can be seen as an exemplar of work arrangements whereby workers buy the firm in the sense that they keep every dollar earned after expenses. Our experiments offered random samples of Boston Uber drivers the opportunity to buy a virtual lease that eliminated or reduced the Uber fee. Some lease-paying drivers were offered a negative fee, capturing a pos- sibly higher taxi wage. We use drivers’ labor supply behavior and lease choices to estimate the parameters that determine the value of a rideshare compensation contract. The first key parame- ter of interest is the labor supply response to temporarily higher wage rates, or inter- temporal substitution elasticity (ISE). A large ISE tends to make medallion-type compensation contracts more attractive because elastic drivers collect additional surplus by driving longer hours when their hourly wage goes up. Drivers’ response to experimental wage changes reveals an ISE for the wage effect on Uber hours, on the order of 1.2 overall, and around 1.8 for drivers who lease. These estimates are broadly consistent with experimental estimates reported for Swiss bicycle messen- gers by Fehr and Goette ( 2007 ).^2 Our estimated ISEs are remarkably stable across groups with varying levels of experience and work intensity. They’re also broadly in line with the Mas and Pallais ( 2017 ) experimental estimates of compensated elastic- ities for part-time workers who work flexible hours. The second key parameter in our framework quantifies the extent to which attrac- tive leasing arrangements were undersubscribed. Many drivers to whom we offered a lease indeed took it. But many drivers who would have benefitted from leasing failed to take advantage of the opportunity to do so. We refer to this behavior as “lease aversion” and use a model of context-specific loss aversion to explain it. Specifically, we compute a behavioral lease-aversion coefficient that rationalizes the
(^1) Some cities, including New York and (until recently) Houston, impose additional licensure and training costs on ride-hailing drivers. 2 See Farber (2005, 2015 ) for more on taxi driver supply elasticities.
4 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2021
A. Budget Sets
We capitalize “Taxi” when referring to the lease-based compensation schemes offered to Uber drivers in our experiment. This is cast against a simplified but real- istic characterization of the “Rideshare” contract facing Uber drivers. Fares are cast in terms of average hourly earnings, w , taken to be the same for Rideshare and Taxi drivers. This is unrestrictive because differences in wages can be modeled as part of the Rideshare fee, or reflected in a negative fee for Taxi drivers. Drivers drive for h hours, so their weekly farebox is wh. Their compensation schemes are as follows:
- Rideshare drivers earn y 0 = w ( 1 − t 0 ) h , where t 0 is the Rideshare fee.
- Taxi drivers earn y 1 = w ( 1 − t 1 ) h − L , where L is a Taxi lease price and t 1 ≤ 0 reflects a possibly higher Taxi wage.
Drivers can choose not to work and earn nothing, but leases must be purchased in advance. The quantity t 0 − t 1 is the difference in tax rates imposed under the two contracts. Our experiment ran for one week at a time. Many drivers indeed lease weekly, so it is natural to think of L as a weekly lease, with drivers choosing Rideshare and Taxi week by week. Alternately, we can imagine Taxi as permanently displac- ing Rideshare or vice versa, in which case the relevant decision-making horizon might be longer, with L scaled accordingly. After laying out the basic framework, we briefly consider the contrast between Taxi and Rideshare in a life-cycle framework where the opportunity to choose between contracts may be transitory and future wages are uncertain. Figure 1 sketches the Rideshare and Taxi budget sets when w = 20, L = 100, t 1 = 0 , and t 0 = 0.25 , so the difference in tax rates in this example is just the Rideshare fee (these are realistic values for wages and fees in Boston, but real-world medallion lease costs are much higher). In general, the budget lines cross where the farebox solves
wh = _ t L 0 −^ t 1
≡ B ,
a quantity we call the Taxi breakeven. This is $400 in the figure, attained by driv- ers who drive at least 20 hours. Drivers who collect more than $400 in fares come out ahead under Taxi, while drivers with a lower farebox take home more under Rideshare. Note that the indifference curves sketched in this figure reflect increas- ing utility as the curves shift northwest. A driver with indifference curve u 0 prefers Rideshare, while a driver with indifference curve u 1 prefers Taxi. Figure 1 compares a pair of drivers with fareboxes above and below breakeven. Drivers with hours above breakeven clearly benefit from Taxi. But some drivers with a below-breakeven farebox under Rideshare may respond to the higher Taxi wage
with similar schemes. Neither rideshare platform requires drivers to make advance payments analogous to medal- lion leasing, though rideshare upstarts such as Fasten have tried such schemes.
VOL. 13 NO. 3 (^) ANGRIST ET AL.: UBER VERSUS TAXI 5
by driving longer hours, thereby clearing breakeven. This scenario is sketched in Figure 2. As in Ashenfelter’s ( 1983 ) analysis of welfare program participation, we compute the theoretical take-up threshold by expanding an excess expenditure func- tion that approximates the cash transfer required to attain a reference utility level. The expenditure function for a generic labor supply problem is
e ( p , w , u – ) ≡ min x , l px + wl subject to u ( x , l ) = u – ,
giving the minimum spent on consumption ( x ) at price p and leisure ( l ) at price w in the effort to reach utility u –. Excess expenditure is spending minus the value of drivers’ time endowment, T , that is,
s ( w , u – ) ≡ e ( p , w , u – ) − wT.
Using the fact that expenditure is minimized by compensated demand functions, xc and lc , we can write
s ( w , u – ) = p xc^ + w lc^ − wT = p xc^ − w hc.
The cash needed to reach a given utility level is the difference between consumption spending and driver earnings when these quantities are chosen optimally.
− 200
200
400
600
800
5 − L
u 0
u 1
0 10 15 25 30 35 40
300 = ( 1 − t 0 ) B = B − L
wh ( 1 − t 0 )
− L + wh
Hours
Earnings
20
L = 100 w = 20 t 0 = 0.25 B = 400
Figure 1. Rideshare and Taxi Budget Lines
Notes: This figure contrasts the Rideshare and Taxi budget sets. The red line shows the Rideshare budget for a driver who collects $20 in fares each hour and pays a 25 percent fee. The blue line shows the corresponding Taxi budget set given a lease price of $100. The two lines cross at the breakeven point with $300 of after-tax earnings. The two black lines depict indifference curves for a driver who prefers the proportional fee ( u 0 ) and for a driver who pre- fers to lease ( u 1 ).
VOL. 13 NO. 3 (^) ANGRIST ET AL.: UBER VERSUS TAXI 7
where h 0 is Rideshare labor supply. 4 It is useful to rewrite the Taxi participation rule in terms of the Taxi breakeven,
( 3 ) w h 0 Rideshare ⏟ farebox
_ L t 0 (
1 + _δ 2 _^ t^0 1 − t 0 )
− 1
adjusted breakeven
where δ is the substitution elasticity evaluated at the after-fee Rideshare wage,
δ ≡ ∂ h _ c ∂ w
w _( 1 − t (^0) ) h 0 =^
_∂ hc ∂ w
_^ w^0 h 0
and w 0 = ( 1 − t 0 ) w. This shows that a positive substitution elasticity reduces the participation threshold by the proportional amount
___________^1 1 + 0.5δ _ 1 − t^0 t 0
Eligible drivers with a Rideshare farebox that clears breakeven should always prefer Taxi. But some with a farebox below breakeven should also accept a Taxi contract. With a unit-elastic compensated response and a fee of 25 percent, for example, we expect the participation threshold to be reduced relative to breakeven by about 14.5 percent.
B. Compensating for Taxi-Type Compensation
To model driver choices between work arrangements, we derive the payment required to make up for loss of the opportunity to drive under a proportional fee-based contract. This is compensating variation (CV), where the baseline condition is the Rideshare budget line with an interior solution and the alternative is the Taxi budget set. Positive CV means payment is required for imposition of Taxi, while negative values arise for drivers who prefer Taxi. Although CV is tied to the specifics of the compensation scheme on offer, the results of our experimental Taxi-Rideshare comparisons can be used to extrapolate compensation values to other markets where workers might choose between paying a proportional tax on their earnings and pay- ing a fixed up-front fee. Formally, CV is the difference in cash required to reach a reference utility level given the Taxi and Rideshare budget lines:
f ( w , u 0 ; 0, L ) − f ( w , u 0 ; t 0 , (^0) ),
where u 0 is the Rideshare utility level. Using Shephard’s lemma as in equation ( 2 ), the CV required as compensation for Taxi can be shown to be
( 4 ) CV = (^) { L − t 0 w h (^0) } − t 0 w h 0 _^ δ^ t^0 (^2) ( 1 − t 0 )
(^4) We omit the superscript indicating that this is the level of work determined by the compensated supply function.
8 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2021
Rideshare drivers for whom CV is negative take the Taxi scheme when offered, pro- ducing the participation rule described by ( 2 ). A Leontief (δ = 0 ) driver should be paid the difference between his or her lease costs and Rideshare fees. Elastic labor supply favors Taxi, reducing CV. Even so, the principal determinant of CV for most drivers is likely to be L − t 0 w h 0 , the differ- ence between lease costs and Rideshare fees. This difference is largest for Uber and Lyft’s many low-hours drivers. Recall also that in the absence of substantial income effects on the demand for leisure, CV approximates the difference in driver surplus yielded by the two compensation schemes (this in turn equals the corresponding equivalent variation). The left panel of Figure 3 illustrates the CV calculation generated by a move from the Rideshare to Taxi budget lines. A Rideshare driver working at point A drives 10 hours and is on indifference curve u 0. Faced with a Taxi budget line, this driver drives 13 hours, but is worse off on u 1. It seems natural to compensate this driver by an amount equal to the excess of his lease over what he used to pay in Rideshare fees. But a payment of L − t 0 w h 0 puts non-Leontief drivers above point C on u 0 , as indicated by the blue line extending from point A with a slope equal to the Taxi wage. Payments equal to lease costs minus ex ante Rideshare fees overcompensate for Taxi because the Taxi scheme increases wages, yielding additional driver sur- plus. The term w h 0 (δ t 0 /( 2 ( 1 − t 0 ))) in equation ( 4 ) captures this surplus, a term denoted by σ in Figure 3. The surplus generated by higher Taxi wages is the product of the proportional Taxi wage advantage, t 0 /( 1 − t 0 ) , the substitution elasticity ( δ ), and driver fees, t 0 w h 0. This product approximates the area under the driver’s supply curve between his net-of-fee Rideshare and Taxi wages.
Choosing Not to Drive .—The compensation formula above presumes Rideshare drivers accept the Taxi budget line as a condition for compensation. But we might instead allow former Rideshare drivers to refuse Taxi, taking some of their com- pensation in the form of increased leisure. In this scenario, drivers are made whole by imagined unemployment insurance (UI) in an amount that takes them to the u 0 ordinate, a scenario illustrated in the right panel of Figure 3. To compute the compensation needed in this case, we assume the marginal util- ity of leisure is zero at h = 0 , so drivers with a wage of zero drive zero hours. Expanding the excess expenditure function for Rideshare utility with a wage of zero around Rideshare expenditure with a fee of t 0 , we have
( 5 ) s (0, u 0 ) = s ( w 0 , u 0 ) + (− h 0 )(− w ( 1 − t 0 )) − 1 _ 2 ∂ h
_ c ∂ w
w^2 ( 1 − t (^0) )^2.
By definition of u 0 , Rideshare drivers with no unearned income and no lease to cover have consumption equal to their Rideshare earnings, so s ( w 0 , u 0 ) = 0. The compensation required for the replacement of Rideshare work opportunities with UI is therefore
( 6 ) UI = (^) ( 1 − t (^0) ) w h 0 − 1 _ 2 (∂ h _ c ∂ w
_^ w (^1 −^ t^0 )
h 0 )([^1 −^ t^0 ] w h^0 )
( 7 ) = (^) ( 1 − t (^0) ) w h (^0) [ 1 − _δ 2 ].
10 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2021
where r is the reciprocal of the marginal utility of wealth, vs ( x , l ) is period- s utility, and wages and prices in period s are time-varying. The profit function imagines consumers valuing their utility at price r ; profit is then the monetary value of utility plus earnings, net of expenditure on inputs in the form of consumption. Consider a driver making a life-cycle plan in the face of known wages and prices, choosing between Rideshare and Taxi at time (week) s. This driver prefers Taxi if the Taxi contract is profitable for that week. That is, Taxi beats Rideshare in week s when
π s ( r , ws ) − π s ( r , ws [ 1 − t 0 ]) > L.
This comparison presumes the utility price is unchanged by Taxi, either because the Taxi opportunity and parameters are known at the time plans are made, or because the Taxi option is short-lived. We assume goods prices are constant, so ps is left in the background.^5 Expanding π s ( r , ws ) around the value of Rideshare profits, π s ( r , ws [ 1 − t 0 ]) , the life-cycle participation rule for Taxi at week s is approximated by
∂_______________π s ( r , ws [ 1 − t 0 ])
∂ w
ws t 0 + 1 _ 2
________________^ ∂^2 π s ( r ,^ ws [^1 −^ t^0 ])
∂ w^2 (
ws t (^0) )^2 > L.
Applying a life-cycle version of Shephard’s lemma, this can be written
( 9 ) ws hs 0 Rideshare earnings ⏟
_ L t (^0) (
1 + δ^ _ f 2
_^ t^0 (^1 −^ t 0 ))
− 1
life-cycle breakeven
where δ f^ ≡ (∂ hs^ f^ ( r , ws )/∂ ws )( ws ( 1 − t 0 )/ h ) and hs 0 ≡ hs^ f^ ( r , ws [ 1 − t 0 ]) is Frisch
labor supply for Rideshare drivers in period s. The earlier Taxi participation rule therefore stands—but with the Hicks substitution elasticity replaced by the possibly larger ISE, δ f. The revision to CV in a life-cycle framework parallels that for participation. Specifically, CV is the sum of the difference in within-period profits under the two compensation schemes:
CV = [π s ( r , ws ) − L ] − π s ( r , ws [ 1 − t 0 ]).
Using the expansion yielding equation ( 8 ), this becomes
( 10 ) CV = (^) { L − t 0 ws hs (^0) } − t 0 ws hs 0 _^ δ^ f^ t^0 (^2) ( 1 − t 0 )
This is the same as ( 4 ), with the ISE δ f^ again replacing the substitution elastic- ity, δ. Since the ISE (weakly) exceeds the Hicks substitution elasticity, a life-cycle
(^5) Our streamlined notation also ignores the the fact that wage and price variables determining profits in a future period s are discounted back to the decision-making date; see Browning, Deaton, and Irish ( 1985 ) for details.
VOL. 13 NO. 3 (^) ANGRIST ET AL.: UBER VERSUS TAXI 11
perspective tends to favor Taxi. Because our experimental design offers temporary wage changes, we interpret the experiment as identifying δ f. In practice, drivers considering a weekly lease must do so without knowing next week’s wage or farebox. Suppose that a Rideshare driver who doesn’t know next week’s wages is offered the opportunity to buy a one-week lease. Although marginal utility of lifetime wealth presumably changes little as a result of wage surprises, some idea of w is required to make a wise near-term choice. Assuming drivers know how they will respond to wages, a predicted wage implies a predicted farebox. The econometric framework outlined in Section IV therefore embeds farebox prediction in an empirical model for Taxi participation.
D. Outside Options
The drivers in our experiment can typically drive as many hours for Uber as they like at the implicit market wage, but many Rideshare drivers work at another job (Hall and Krueger 2018). We model alternative employment as characterized by declining earnings opportunities. For alternative jobs with institutional limits on hours, such as shift work or salaried office work, the decline is likely to be precip- itous. On other sorts of jobs, including alternative ride-hailing platforms, any pay advantage over Uber may taper smoothly. We might imagine, for example, that Lyft takes lower fees than Uber, but offers its drivers less steady trip demand. This mar- ket structure is captured by assuming that drivers earn e ( a ) for a hours worked on an alternative job, where e ( a ) is increasing but concave. 6 The excess expenditure function for a driver who holds an alternative job is
sa ( p , w , u – ) = min x , h , d px − wh − e ( a ) subject to u ( x , T − h − a ) = u – ,
where the a superscript indicates that this is excess expenditure for someone who works an alternative job. As always, excess expenditure is minimized by the com- pensated demand functions xc , hc , ac , so
sa ( p , w , u – ) = p xc^ − w hc^ − e ( ac ).
Writing f a ( w , u – , L , t ) for the cash required to reach utility u –^ in this scenario yields the relevant excess expenditure functions:
- Rideshare: f a ( w , u – ; t 0 , 0 ) = p xc^ − w ( 1 − t ) hc^ − e ( ac^ ) = sa ( w ( 1 − t 0 ), u – )= sa ( w 0 , u – ) ;
- Taxi: f a ( w , u – ; 0, L ) = ( p xc^ + L ) − whc^ − e ( ac^ ) = sa ( w , u – ) + L ,
(^6) This setup is inspired by the Gronau ( 1977 ) model of home production, where workers get utility from a single consumption good and from leisure, and can produce the consumption good under diminishing returns at home or buy it with money earned on a job paying constant wages.
VOL. 13 NO. 3 (^) ANGRIST ET AL.: UBER VERSUS TAXI 13
differ from other drivers, but an analysis of drivers grouped by hours driven shows little systematic variation in the behavioral parameters of interest to us. Roughly 45 percent of Boston drivers were eligible for inclusion in the exper- iment. Although the cap on hours per week reduces average hours in the eligible sample relative to the city average, drivers in the eligible sample are otherwise sim- ilar to the pool of active Boston drivers (that is, the group who took at least four trips in the previous month). For example, 14 percent of both the active and eligible samples are female and both groups had used the Uber platform for an average of 14 months. These comparisons appears in the first two columns of Table 1. A total of 1,600 eligible drivers were selected for inclusion in the experiment. The experimental design randomized compensation schemes within strata defined by average hours driven in July 2016, driver fee class (commission rate), and vehicle model year. The low-hours stratum includes drivers who averaged 5–15 hours per week, while the high-hours group averaged 15–25 hours per week. The 20 percent fee class includes veteran drivers who signed up before September 2015, while other drivers paid a commission rate of 25 percent. Because Lyft requires its drivers to use cars no older than 2004, our strata distinguish between drivers with cars from model year 2003 or older and drivers with newer Lyft-eligible cars. Table 1 also shows
Table 1—Boston Uber Drivers
All Boston drivers
Eligible drivers
Experimental drivers
Strata-adjusted difference ( 1 ) ( 2 ) ( 3 ) ( 4 )
Female 0.14 0.14 0.14 0. (0.35) (0.34) (0.35) (0.01) Age 40.90 41.58 41.80 0. (12.13) (12.20) (12.29) (0.36) Average hours/week in the month before selection 14.42 13.13 14.51 0. (14.39) (5.69) (5.81) (0.08) Average hourly earnings in the month before selection 15.39 17.59 17.40 −0. (8.64) (6.19) (6.05) (0.17) Average weekly Farebox in the month before selection
372.06 310.91 342.82 −0. (447.51) (192.04) (198.12) (3.93) Months since sign-up 13.89 14.26 11.14 −0. (9.43) (9.25) (8.67) (0.15) Vehicle solutions participant 0.08 0.08 0.08 0. (0.26) (0.27) (0.28) (0.01) Car model year 2003 or older 0.03 0.03 0.12 --- (0.17) (0.17) (0.33) --- Car model year 2011 or newer 0.64 0.64 0.56 −0. (0.48) (0.48) (0.50) (0.01) Commission 22.34 22.24 23.21 0. (2.50) (2.49) (2.40) (0.01) Observations 19,316 8,685 1,600 8,
Notes: Columns 1–2 compare Boston drivers to the subset of drivers eligible for the experiment. Eligible drivers are those with valid vehicle year information who made at least 4 trips during the past 30 days and drove an aver- age of between 5 and 25 hours/week in July 2016. Column 3 shows means for drivers in the experimental sample. Treatment was randomly assigned within strata defined by hours (high/low), commission ( 20 /25 percent commis- sion), and car age (older/newer than 2003). Column 4 shows strata-adjusted differences between the experimental sample and the rest of the eligible pool. Average hourly earnings include surge pay but are net of Uber fees. Vehicle solutions drivers lease a car through an Uber-sponsored leasing program.
14 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2021
the proportion of drivers with cars newer than 2010, since Lyft’s most important promotion requires drivers operate newer vehicles. Drivers were randomly sampled and randomly assigned to the first or second opt-in week within these three strata. As can be seen in column 4 of Table 1, which reports strata-adjusted differences in means, the experimental sample has characteristics similar to those of drivers in the rest of the eligible sample.
B. Opt-In Weeks
The 1,600 drivers in the experimental sample were offered the opportunity to drive for one opt-in week with no Uber fee. Half of the drivers (Wave 1) were offered fee-free driving in the first opt-in week. While the first wave was driving fee free, drivers in the second half sample (Wave 2) were offered the chance to opt in to fee-free driving the following week. This split-sample design was meant to bal- ance wealth effects induced by the higher fee-free wage. Online Appendix Table A shows that driver characteristics are similar in the two waves. Drivers in both waves were offered fee-free driving by email, text message, and in-app notification on Monday morning of the relevant offer week; they had until midnight the following Saturday to opt-in. Sampled drivers received up to three emailed reminders to opt-in by the deadline. Drivers who opted in paid no Uber fee on all trips taken in the subsequent week. This was reflected in their immediate in-app trip receipts and weekly pay statements (participating drivers saw a fee of zero in receipts and statements). Fee-free driving increased a driver’s total payout by 25 percent in the 20 percent fee class ( 0.25 = ( 1 /0.8) − 1 ) and by 33 percent in the 25 percent fee class ( 0.33 = ( 1 /0.75) − 1 ). Take-up rates for fee-free driving are shown in Table 2. Roughly 64 percent (1,031/1,600) accepted the fee-free offer. Although fee-free driving should be attractive to all drivers, many appear to ignore Uber messaging beyond the offer of trips. This likely reflects the fact that (during our experimental period) Uber drivers received many electronic messages each week. 7 The struggle for driver attention is reflected in a decline in take-up from Wave 1 (71 percent) to Wave 2 (58 per- cent), after we stopped the opt-in reminders midweek. Messaging was reduced in view of higher-than-expected take-up and a consequent risk of running over budget. Discussions with Uber’s Boston team suggest our take-up rates compare favorably with take-up rates for other no-lose driver promotions requiring an opt in. Incomplete take-up may also reflect the fact that drivers who opted in consented for their data to be used in academic research and to receive further Earnings Accelerator offers. Table 3 shows that drivers who opted in drove and earned more than other drivers during the period preceding opt-in weeks. In the pooled sample including both high and low hours drivers, those who opted in had a pre-experimental farebox roughly $51 higher than the farebox of drivers who opted out. Those who opted in also drove four more hours that week. On the other hand, these gaps are much smaller when averaged over the month of July. This is consistent with the idea that inattention drove
(^7) In view of this, Uber moved later to cap the number of promotion-related messages sent to drivers.
16 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2021
fee-free driving and 20 percent were offered negative fee driving in the form of a 12.5 percent wage increase ( t 1 = − 0.125 ). Lease prices in the first Taxi week ranged from $45 to $165. The treatments in week 2 were less generous—the nega- tive fee treatment was replaced with a half-fee treatment—but also less expensive, with leases priced between $15 and $60. Each treatment during week 2 was offered to 30 percent of drivers within strata. Figure 4 summarizes experimental staging and design parameters. As with solicitation for fee-free driving in opt-in week, treated Taxi drivers were offered Taxi contracts via email, text messages, and in-app notifications. These offers were sent one week in advance and highlighted the breakeven amount. For example, drivers in the 25 percent fee class who were offered a half-fee treatment for $35 were told, “As long as your weekly total fares + surge exceed $280, you’ll come out ahead.” Email and text messages included links to a simple table con- trasting the former and revised fee calculation for a sample trip. Emails and text messages also included links to a calculator that showed net earnings with and without treatment for any driver-selected value of fares plus surge pay. Figure A in the online Appendix shows a message delivered in the Taxi promotions; online Appendix Figure A2 shows the calculator. Drivers who opted-in to Taxi’s virtual lease had lease payments deducted from their pay for the opt-in week. This deduction appeared as a negative entry on
Table 3—Who Opts In?
Pooled High hours Low hours Opt-out mean Difference
Opt-out mean Difference
Opt-out mean Difference ( 1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ( 6 )
Female 0.13 0.03 0.12 0.01 0.14 0. [0.33] (0.02) [0.32] (0.02) [0.35] (0.03) Age 42.75 −1.46 44.81 −3.06 40.89 −0. [12.61] (0.65) [12.68] (0.95) [12.27] (0.89) Commission 23.11 0.16 22.97 0.33 23.24 0. [2.43] (0.13) [2.46] (0.18) [2.39] (0.17) Vehicle solutions 0.06 0.03 0.07 0.04 0.05 0. [0.24] (0.01) [0.26] (0.02) [0.23] (0.02) Vehicle year 2010.40 −1.76 2010.56 −3.50 2010.26 0. [4.45] (1.96) [4.39] (3.82) [4.51] (0.33) Months since signup 11.60 −0.71 12.53 −1.61 10.75 0. [9.03] (0.46) [9.19] (0.67) [8.81] (0.63) Average hours/week the 14.16 0.53 19.67 −0.18 9.16 0. month before selection [6.01] (0.31) [3.01] (0.22) [2.84] (0.21) Average hourly earnings the 16.19 1.88 17.46 2.30 15.03 1. month before selection [5.56] (0.30) [4.80] (0.38) [5.95] (0.45) Average weekly Farebox the 310.06 50.85 447.65 57.13 184.93 24. month before selection [180.52] (9.90) [145.18] (11.48) [100.90] (7.54)
Observations 569 1,600 271 800 298 800
Notes: This table compares the characteristics of drivers who opted in to fee-free driving with those of drivers who were offered fee-free driving but did not participate. Standard deviations appear in brackets. Columns 2, 4, and 6 report the strata-adjusted difference between drivers who opted in and drivers who did not opt in. Standard errors are in parentheses. Average hourly earnings include surge pay but are net of the Uber fee.
VOL. 13 NO. 3 (^) ANGRIST ET AL.: UBER VERSUS TAXI 17
otherwise standard weekly pay statements on the line that typically would show pay- ments earned through Uber promotions. 8 These deductions were labeled “Earnings Accelerator buy-in.” During opt-in week, participating drivers’ trip receipts reflected the reduced fee (see Figure 5 for sample trip receipts and Figure 6 for a participating driver’s weekly pay statement). Earnings Accelerator lease amounts are well below the price of a traditional taxi medallion lease: before the advent of ride-hailing, Boston medallion leases (includ- ing vehicle) ran around $700/week and over $100/day. Our virtual medallions were priced from $50–$165/week. These amounts were calibrated to appeal to drivers with weekly earnings in particular ranges, as explained below. As a measure of the empirical relevance of our design, it is noteworthy that in 2016 a Boston ride-hailing upstart (Fasten) offered its drivers the option to pay $80/week or $15/day to drive fee free. 9
(^8) A few drivers who earned less than needed to cover their lease carried a negative balance into the following pay period. 9 In 2010, the Boston medallion lease cost for a single driver was capped at $700/week, $139/day, and $77/12-hour shift (BPD Circular Date 12-30-09 “2010 Standard Shift Rental Agreement”). Newer cars leased for an additional $170/week. Drivers could split a weekly lease for no more than $800. Before the advent of ride-hailing, short supply meant medallions typically leased at the cap. Side payments to Boston fleet owners also appear to have been common (See the 2013 Boston Globe stories linked under http://www. bostonglobe.com/metro/specials/taxi). Data on medallion prices is spotty; a CommonWealth Magazine arti- cle (http://commonwealthmagazine.org/transportation/ taxi-medallion-owners-under-water-and-drowning/)
Table 4—Earnings Accelerator Taxi Parameters and Take-up
Strata Treatment Offers and Opt Ins
Hours Group Fee
Number in group Lease New fee Breakeven Offer rate Opt-in rate ( 1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ( 6 ) ( 7 ) ( 8 )
Week 1 High 0.20 180 $110 0 $550 0.4 0. $165 −0.125 $508 0.2 0. High 0.25 349 $110 0 $440 0.4 0. $165 −0.125 $440 0.2 0. Low 0.2 177 $45 0 $225 0.4 0. $75 −0.125 $231 0.2 0. Low 0.25 325 $45 0 $180 0.4 0. $75 −0.125 $200 0.2 0. Week 2 High 0.20 180 $60 0 $300 0.3 0. $25 0.10 $250 0.3 0. High 0.25 349 $55 0 $220 0.3 0. $35 0.125 $280 0.3 0. Low 0.2 177 $40 0 $200 0.3 0. $15 0.10 $150 0.3 0. Low 0.25 324 $35 0 $140 0.3 0. $15 0.125 $120 0.3 0.
Notes: This table describes the Taxi phase of the experiment in which drivers could purchase a virtual lease giving them a week of fee-free driving. During each of two Taxi weeks, drivers within each stratum (listed in columns 1–3) were randomly assigned to one of two lease treatments (60 percent) or to the control group (40 percent). Columns 4 and 5 describe these treatments. Column 6 reports the breakeven associated with each treatment. Opt-in rates in column 8 are reported as a proportion of drivers offered. Lease prices were chosen so as to be attractive to roughly 60 percent of the drivers in each stratum.
VOL. 13 NO. 3 (^) ANGRIST ET AL.: UBER VERSUS TAXI 19
a Taxi contract. Participation estimates distinguish extensive from intensive mar- gin effects, identify possible changes in average hourly compensation, and reveal anticipatory or post-treatment labor supply changes that might signal confound- ing wealth effects. The participation analysis yields three important findings: (i) Earnings Accelerator participation had no effect on the extensive margin (that is, effects on whether drivers drive at all); (ii) participation boosted hours driven and driver earnings considerably during treatment weeks, with no corresponding change in average hourly earnings; (iii) we see no evidence of anticipatory or post-experiment effects in the treated group. The analysis sample for 2SLS estimation of participation effects stacks data for two pairs of weeks: the first pair contains data on 1,600 drivers from the first two waves; the second pair includes observations from the two Taxi weeks for the 1,031 drivers who opted in to fee-free driving and agreed to receive Taxi offers later. The endogenous variable in this setup, Dit , indicates fee-free driving in week t or purchase of a Taxi contract during the Taxi opt-in weeks, to be used in week t. The instrument, Zit , indicates offers of fee-free driving or a Taxi contract in
Figure 5. Earnings Accelerator Trip Receipts
Notes: This figure shows two trip receipts. The left is for a trip taken while the Earnings Accelerator was active; this driver paid no fee. The right is for a trip taken when the Earnings Accelerator was inactive; this driver paid an Uber fee. The right shows how additional Uber promotions (in this case, Boost) are reflected on trip receipts.
20 AMERICAN ECONOMIC JOURNAL: APPLIED ECONOMICS JULY 2021
week t. For example, Zi 1 is switched on for the 800 drivers offered fee-free driving in Wave 1 and for the 619 drivers offered a Taxi lease during the first week of the Taxi trial. For a set of weekly labor supply outcomes denoted by Yit , the 2SLS setup used to compute participation effects can be written
( 13 ) Yit = α Dit + β Xit + η it ,
( 14 ) Dit = γ Zit + λ Xit + υ it ,
where Xit includes dummies indicating the strata used for random assignment, driver gender, the number of months a driver had worked on the Uber platform, one lag of log earnings, and indicators for whether a driver used Uber’s vehicle leasing pro- gram and whether a driver had a car from model year 2003 or older. Because drivers appear in the sample more than once, standard errors in this setup are clustered by driver.
Josh Angrist to Emily, Casey, Jonathan, Sydnee 9/6/ nice - shows zero fees all down the line! -------- Forwarded Message -------- Subject: Date: Your earnings for the period ending Sep 05, 2016Tue, 06 Sep 2016 16:05:17 +0000 (UTC) From: To:
Payment Summary Mon, Aug 29 - Mon, Sep 05
$51. Total Payout
Current Rating
Hours Online
6 Trips
Day Trips Fares 1 Surge Uber Fees Payments Aug 29 0 $0.00 $0.00 $0.00 $0. Aug 30 0 $0.00 $0.00 $0.00 $0. Aug 31 0 $0.00 $0.00 $0.00 $0. Sep 01 0 $0.00 $0.00 $0.00 $0. Sep 02 0 $0.00 $0.00 $0.00 $0. Sep 03 1 $20.30 $0.00 $0.00 $20. Sep 04 5 $27.09 $4.25 $0.00 $31. Total Payout $51. Fees that don't impact your payments are not shown here. Refer to your detailed payment statement to see more details.
[email protected] [email protected]
Figure 6. Weekly Pay Statements
Notes: This figure shows a weekly pay statement for a driver participating in the Earnings Accelerator. The column at the right lists net pay for each day.
