Uberralenti? Ride-hailing and traffic jams in Austin (Search of the Month)

Matthew Tarduno, The congestion costs of Uber and Lyft, Journal of Urban Economics, 2021, vol. 122

Abstract:

• In Austin, the cessation of Uber and Lyft operations in May 2016 appears to have slightly improved traffic flow (the time required to travel 1 mile decreased).
• The effect seems to be concentrated mainly at midday, outside peak hours.
• The societal cost of the slowdowns caused by Uber and Lyft appears to be close to the value of their services to consumers.

The emergence of ride-hailing companies is one of the major innovations in urban transportation in recent years. However, does the public really benefit from these innovations? There are two possible scenarios. In the first, rideshare trips replace trips that would otherwise have been made in private vehicles. In the second scenario, these trips would otherwise not have been made or would have been made in a different way, for example by public transport. While in the first case the number of vehicles on the road remains stable, the second scenario leads to an increase in traffic. In doing so, ride-hailing companies exacerbate congestion problems and generate negative externalities (slowdowns or traffic jams). In the latter case—especially if the nuisance is significant—the government may have an interest in taxing or limiting the activity of these companies.

Matthew Tarduno, in his article « The congestion cost of Uber and Lyft, » seeks to measure the impact of these two companies on road traffic in Austin and the well-being of drivers. His work illustrates the difficulties economists face when trying to estimate the impact of a new service (in this case, ride-hailing activities).

Technical and econometric difficulties

In order to measure the impact of a ride-hailing company on road traffic, the first challenge is to measure that traffic. To do this, Matthew Tarduno uses Bluetooth sensors installed by the city of Austin. These sensors detect Bluetooth devices (including smartphones and vehicle Bluetooth systems) and measure the speed at which these devices are moving. Thanks to these sensors, the author is able to measure the time it takes to travel one mile (1.6 km) on 79 segments (a road or section of road) at every hour of the day between the end of March and August 2015, and then between March and August 2016.

This time frame allowed Matthew Tarduno to overcome a second difficulty. In order to measure the impact of ride-hailing companies, it is necessary to exploit a variation in their activity. One possibility would be to study a city before and after the arrival of one of these companies. However, this strategy raises the following question: why would a ride-hailing company decide to start operating in this city at this particular time? If the reason is somehow related to road traffic, then such a strategy will lead to biased results. This would be the case, for example, if ride-hailing companies decided to « enter » only cities where traffic flows smoothly, allowing them to increase the number of rides and thus their revenue. There would then be a negative correlation between the presence of ride-hailing companies and road traffic, which would distort the results of an econometric analysis. [2] [HD(21] Matthew Tarduno suggests instead using the fact that ride-hailing companies decide to cease operations. In 2015, the city of Austin proposed tightening regulations on these companies. These new regulations were opposed by Lyft and Uber, but on May 7, 2016, following a local referendum, 56% of the population rejected the two companies’ counterproposal. The companies then decided to cease operations in Austin. According to Matthew Tarduno, exploiting this unexpected withdrawal of ride-hailing companies would make it possible to obtain a variation in ride-hailing activity that is not correlated with the level of road traffic.

Thanks to this « natural experiment, » it became possible to implement two empirical strategies. The first consisted of comparing road traffic just before and just after the withdrawal of Lyft and Uber. For example, comparing road traffic in the first week of May 2016 (when Lyft and Uber were still operating) with that of the second week of May (when they had withdrawn from the market). The second method consists of comparing the evolution of road traffic on the same day between 2015 and 2016 before and after the companies’ withdrawal. For example, it is possible to compare the change between April 1, 2015, and 2016 (when the companies were present on both dates) and between June 1, 2015, and 2016 (when they were only present in 2015). The change in April allows us to measure the « normal » change in road traffic between 2015 and 2016, and the change in June allows us to measure the normal change plus the effect of the withdrawal of ride-hailing companies. The difference between these changes therefore allows us to measure the impact of the withdrawal of Lyft and Uber on road traffic.

Both methods are based on the assumption that nothing other than the withdrawal of Lyft and Uber would explain a systematic difference in travel time between the two periods. However, the methods differ in terms of the effect measured. The first method measures a « local » effect, immediately after the decision. The second method measures a more global effect, which would take into account, for example, the fact that drivers may—after noticing a possible change in road traffic—choose new routes and thus alter the level of traffic on each section.

Empirical results

Figures 1 and 2 illustrate the second method. Figure 1 shows the average time required to travel 1 mile before and after the withdrawal of Uber and Lyft. It shows that the average times and, above all, the change over time between 2015 and 2016 are relatively similar before May. After May 7, however, the average times begin to differ. This suggests that the departure of Uber and Lyft had an impact on road traffic: the time required to travel 1 mile decreased. Furthermore, this graph also suggests that using the change in travel times prior to the withdrawal of these companies makes sense and therefore that the experiment is valid[6].

Graph 1: Average traffic before and after the withdrawal of Uber and Lyft

Source: Matthew Tarduno, The congestion costs of Uber and Lyft, Journal of Urban Economics, Volume 122, 2021, p.7

Figure 2 shows the average effect of the withdrawal on an hourly basis. It reveals that while the withdrawal of Uber and Lyft appears to be associated with a decrease in road traffic (the time needed to travel 1 mile decreases), this effect is generally small (on average, around 0.1 minutes) and rarely statistically significant at the 5% threshold. Furthermore, the effects of these companies’ withdrawal seem to be concentrated around midday (11 a.m., 12 p.m., and 1 p.m.) rather than during rush hour when there are the most motorists on the road (and therefore when more drivers would be affected by a change in road congestion).

Figure 2: Hour-by-hour effect

Source: Matthew Tarduno, The congestion costs of Uber and Lyft, Journal of Urban Economics, Volume 122, 2021, p.6

Conclusion: the effect of congestion on well-being

The above estimates suggest that Uber and Lyft have caused slight slowdowns on Austin’s roads, but can we conclude that the services offered by these companies are « socially harmful »?

Thanks to the above estimates, it is possible to provide a (very) partial initial answer to this question. The societal cost of the slowdowns generated by Uber and Lyft can be calculated as the additional time required to travel 1 mile due to these companies (the value estimated above), multiplied by the number of miles traveled by Austin residents, multiplied by the value of time. Using other sources and studies to obtain this value of time and the number of miles traveled, Matthew Tarduno calculates that the total cost of traffic delays caused by these companies would be between $33 million and $52 million (per year).

However, estimates by Cohen et al. (2016) suggest that the gain to consumers using Lyft and Uber in Austin could be between $47 million and $73 million per year. As such, and taking into account the broad approximations in these calculations, the benefits to consumers and the costs of these services appear to be relatively close. This means that there is a transfer of welfare from private motorists to Uber and Lyft customers. It also suggests that imposing restrictions on these companies would not significantly increase the welfare of the general population. Again, it would be more of a transfer of welfare from users of these services to other motorists.

References:

Matthew Tarduno, The congestion costs of Uber and Lyft, Journal of Urban Economics, Volume 122, 2021, ISSN 0094-1190, https://doi.org/10.1016/j.jue.2020.103318.

Peter Cohen, Robert Hahn, Jonathan Hall, Steven Levitt and Robert Metcalfe, USING BIG DATA TO ESTIMATE CONSUMER SURPLUS:THE CASE OF UBER, NBER working paper, 2016, Working Paper 22627 http://www.nber.org/papers/w22627