Portal de Periódicos da CAPES

Measuring the impact of ride‐hailing firms on urban congestion: The case of Uber in Europe

2021; Elsevier BV; Volume: 100; Issue: 5 Linguagem: Inglês

10.1111/pirs.12607

1435-5957

Xavier Fageda,

Urban Transport and Accessibility

This paper examines the impact of Uber, the world's largest ride-hailing firm, on congestion. Drawing on data from European cities for the period 2008 through 2016, I find a negative impact of Uber on congestion. The estimated impact in the baseline regression is −3.5 percentage points, but it is higher in cities that do not impose strong regulatory restrictions to ride-hailing services. In addition, the negative impact of Uber on congestion is only statistically significant in denser cities. The Uber effect is gradual given that its impact increases over time. Finally, I find suggestive evidence that the potential endogeneity bias underestimates the negative effect of Uber on congestion. Este documento explora el impacto de Uber, la mayor empresa de servicios de reserva de taxis del mundo, en la congestión de tráfico. A partir de datos de ciudades europeas para el período 2008 a 2016, se encontró un impacto negativo de Uber en la congestión de tráfico. El impacto estimado en la regresión de referencia es de –3,5 puntos porcentuales, pero es mayor en las ciudades que no imponen fuertes restricciones normativas a los servicios de reserva de taxis. Además, el impacto negativo de Uber en la congestión de tráfico sólo es estadísticamente significativo en las ciudades más densas. El efecto de Uber es gradual, ya que su impacto aumenta con el tiempo. Por último, se encontró evidencia que sugiere que el sesgo potencial de endogeneidad subestima el efecto negativo de Uber sobre la congestión de tráfico. 本稿では、世界最大の配車サービス会社であるUberの渋滞への影響を検討する。2008~2016年の欧州の都市のデータから、Uberが渋滞に悪影響を与えていることを発見した。ベースライン回帰で推定した影響は、−3.5%ポイントであるが、配車サービスに強い規制を課していない都市ではもっと高い値がみられる。さらに、渋滞に対するUberの負の影響は、人口密度の高い都市においてのみ統計的に有意である。Uberの影響は、時間の経過とともに大きくなるものであり、徐々に拡大するものである。また、潜在的な内生性バイアスが、Uberの渋滞に対する負の効果を過小評価することを示唆するエビデンスが認められる。 Uber, Lyft, and other ride-hailing firms have reshaped mobility in many cities across the globe; and, at the same time, their success has generated a considerable number of economic, social and legal controversies, including debates about working conditions, safety, quality standards and unfair competition. Hence, many cities have banned or imposed restrictions on the activity of ride-hailing firms. One of the main controversies concerns their impact on congestion, the focus of this present study. Urban congestion results in traffic jams that affect both drivers and pedestrians, who have to put up with increasing levels of gridlock, noise and pollution. INRIX and Centre for Economics and Business Research carried out a study in 2014 to estimate the economic impact of the delays caused by traffic jams in the UK, France, Germany, and the US (INRIX and Cebr, 2014). Congestion costs represented $200 billion in the four countries (around 0.8% of their joint GDP). Furthermore, the relationship between congestion and pollution is well-documented, with prolonged car circulation at reduced speeds having a notable effect on the emission of polluting substances (Barth & Boriboonsomsin, 2008; Beaudoin et al., 2015; Parry et al., 2007). Congestion would also seem to have a negative impact on road safety outcomes, especially for the most congested cities (Albalate & Fageda, 2021). In this paper, I examine the impact of the largest ride-hailing firm in the world today—Uber—on urban congestion. The analysis is undertaken for cities in Europe (EU) between 2008 and 2016. The relationship between ride-hailing firms and congestion is a priori unclear. On the one hand, users of ride-hailing firms may benefit from shorter waiting times, greater flexibility, and lower prices than users of traditional taxi services. Hence, congestion could increase due to demand induction (i.e., car trips, which without ride-hailing services, would not have been made). On the other hand, ride-hailing firms match passengers with drivers via websites and mobile apps so this reduces the number of unoccupied cars driving around looking for customers. Thus, the higher efficiency of ride-hailing firms to match demand with supply may reduce congestion. Finally, it is unclear whether ride-hailing firms compete with private cars, with public transit or with both. In this regard, the limited evidence about the impact of ride-hailing firms on congestion in United States is not conclusive. Using data for a large sample of US cities, Li et al. (2016) find evidence that Uber entry has decreased congestion costs. Ward et al. (2019) find that ride-hailing services appears to cause a decline in state per capita vehicle registrations by 3%, but results regarding travel distances, gasoline consumption, and several air pollutants are not conclusive. Finally, case studies for San Francisco and New York provide evidence that in these large cities ride-hailing services may have worsen congestion (Anderson, 2014; Qian et al., 2020). Here, I seek to add to this literature by providing a direct test of the impact of Uber on congestion using data of European cities between 2008 and 2016. In this regard, I examine the different effects of Uber on congestion based on the regulatory environment in which it operates. Taking into account that Uber is constrained to operate in the pre-booking segment in the considered period, I can distinguish two regulatory environments in my sample of cities; (i) a highly restrictive environment that imposes quantitative restrictions and strong qualitative requirements; and (ii) a low restrictive environment that as much only imposes some specific qualitative requirements. Thus, I exploit the variability in the regulatory environment in which Uber operates to capture the intensity in the use of its services. I should expect a more intense use of Uber services in those cities where the regulatory environment imposes less restrictions. I also examine whether the different effects of Uber on congestion are based on the relatively levels of public transport provision, income and population density. In particular, the impact of Uber could be higher in denser cities given that density is a strong predictor of the probability of Uber entry and it can be expected that denser cities have a better endowment of public transportation. One potential threat to my identification strategy is that Uber presence may be correlated with the previous levels of congestion in the city. To deal with this concern, I include city specific time trends so that I control for the fact that each city may have its own linear time trend. Furthermore, I apply a propensity score matching procedure so that I pair treated and control observations that have similar levels of congestion and similar values of the covariates in the initial year of the considered period. Finally, I apply an event study analysis in which I capture the effect of Uber over time. Another potential threat to my identification strategy is that Uber entry may be correlated with unobserved time varying factors that also have an influence on congestion. To deal with this potential endogeneity bias, I use a subsample based on cities that are eventually treated. The potential endogeneity bias for this subsample should be more modest given that such bias would be related with the time of entry and not with the decision of entry or not. Furthermore, I apply an instrumental variables procedure using as main instrument the regulatory environment for Uber alike services that it is in force in the sample cities. I find a negative and significant impact of Uber on congestion, an outcome that is consistent across all regressions in the empirical analysis. The estimated impact in the baseline regression is 3.5 percentage points. In addition, I find that less restrictions on Uber lead to a stronger negative impact on congestion. In highly restrictive environments, the impact of Uber is weak. In contrast, such impact is significant when it operates in a regulatory environment that does not impose quantitative restrictions nor strong qualitative requirements. Furthermore, I find that the negative impact of Uber on congestion is only statistically significant in denser cities where the presence of Uber is most common. Finally, I find suggestive evidence that the potential endogeneity bias underestimate the negative effect of Uber on congestion. The rest of this paper is organized as follows. In the next section, I review the main mechanisms that may explain the relationship between Uber and congestion. Then, I provide details about the regulatory environment in which Uber operates in Europe. After that, I explain the empirical equation that I estimate and discuss my identification strategy. In the following section, I outline the sample used and provide relevant information about the data employed in the empirical analysis. Then, I present the results of the econometric estimates. The last section is devoted to the discussion and concluding remarks. Uber could affect congestion through different mechanisms. Several studies have found evidence that the arrival of ride-hailing services has led to a substitution effect away from traditional taxi services. For example, Wallsten (2015) and Brodeur and Nield (2018) find fewer taxi trips as Uber grows in New York City, Contreras and Paz (2018) find that ride-hailing firms have reduced taxi ridership in Las Vegas and Berger et al. (2018) found that taxi drivers experienced a relative decline in earnings due to Uber's entry into a new US market. Outside the US, Nie (2017) finds that Uber entry has reduced taxi ridership in Shenzhen, while Chang (2017) shows that it has reduced the number of operating miles of taxi drivers in Taiwan. Thus, we may expect that the arrival of ride-hailing services will have a strong impact on the for-hire car with drivers' sector (including traditional taxi and ride-hailing services). In this regard, Uber's presence could have a negative effect on congestion. Ride-hailing firms may promote an increase in the efficiency of the for-hire car with drivers' sector with the consequent decrease in congestion. Uber services imply the use of a technology that efficiently matches demand and supply. Indeed, the use of apps to match demand with supply should reduce the cruising externality associated with empty cars looking for customers. In this regard, Cramer and Krueger (2016) examine the efficiency of ride-hailing services by comparing the vehicle utilization rate of UberX drivers and taxi drivers in five major urban areas in the US. Similarly, Kong et al. (2020) compare the vehicle utilization rate of Didi (a Uber alike service) and taxi services in Chengdu (China). Both studies find that the vehicle utilization rate of ride-hailing services is higher than that of taxi services. The entry of Uber and other ride-hailing firms may incentivize taxi firms to use similar apps for their own services. In Europe, examples of taxi-hailing apps include MyTaxi (now Free Now) and Taxify (now Bolt). Another potential mechanism that might have an influence on congestion by means of promoting efficiency is the dynamic price-system (surge pricing) used by ride-hailing services (Chen & Sheldon, 2015; Cohen et al., 2016). In contrast to traditional taxi services where prices are usually the same for both peak and off-peak hours, ride-hailing firms charge higher prices in periods of excess demand and lower prices in periods of excess supply. Higher peak-hour prices could deviate demand to off-peak periods. Therefore, we can expect a negative effect of Uber's presence on congestion should it significantly promote greater efficiency in the entire for hire car with drivers sector. Indeed, ride-hailing firms may have increased the utilization of vehicles due to the use of online apps and surge prices that match more efficiently demand and supply. Furthermore, there may be a substitution effect away from private cars as it is found in the study by Ward et al. (2019). Hence, we can expect a negative effect of Uber's presence on congestion if this substitution effect is relevant. However, ride-hailing firms may also lead to an increase in the size of the for-hire car with drivers sector. Uber users may benefit from shorter waiting times, greater flexibility, and lower prices (particularly in periods of less demand) than users of taxi services. Hence, the size of the for-hire car with drivers sector could increase due to demand induction. Thus, we can expect a positive effect of Uber's presence on congestion if it promotes an increase of the size of the for-hire car with drivers' sector. In addition, there may be a substitution effect away from public transportation. Thus, we could expect an increase in congestion due to Uber's presence if this substitution effect is relevant. Evidence on this line for United States is inconclusive. Hall et al. (2018) find evidence that Uber may complement public transit, especially in larger cities. The positive effect of Uber on public transit might lie in the fact that it helps overcome the "last-mile problem" attributable to the fixed routes and fixed schedules of public transit services. Nelson and Sadowsky (2019) find that the positive effect of Uber on public transit largely disappeared with the entry of Lyft. They attribute this result to the fact that price competition between the two firms appears to have made making the whole trip with one of them more attractive. Taken together, Uber's presence will have a negative effect on congestion if the increased efficiency it promotes and the replacement effect of private cars is more relevant than demand induction and the public transport substitution effect. This is the main hypothesis that we test in the empirical analysis. Furthermore, the effect of Uber on congestion should be greater in cities where the intensity in the use of its services is higher. We can expect that the intensity in the use of Uber is higher in those cities that impose fewer regulatory restrictions to ride-hailing firms. Furthermore, our data clearly shows (see details below) that Uber's presence is more common in denser cities. Finally, Uber could be more popular in less affluent cities given its ability to charge lower prices than traditional taxi services. In addition to identifying Uber's net effect on congestion, the empirical analysis also tests the hypothesis that Uber's effect on congestion should be stronger in cities that impose fewer regulatory restrictions and in denser and less affluent cities. A substitution effect from public transport to Uber alike services should be less relevant in cities with more public transport options. Thus, we may expect better outcomes of Uber's presence in terms of congestion in those cities with a higher endowment of public transport infrastructures. This is another hypothesis that is tested in the empirical analysis. In short, I examine Uber's net effect on congestion and the potential heterogeneous impacts according to city's attributes and the regulatory environment in which it operates. Table 1 provides the list of cities used in the empirical analysis. The sample includes 130 cities from 19 different countries in the European Union and United Kingdom adding up 1162 observations. The time span covered is from 2008 to 2016. Figure 1 provides a map with the cities included in the analysis differentiating between cities with Uber entry and cities with no Uber entry in the considered period. Map of the sample cities Note: Yellow circles mark treated cities (ie; cities with Uber entry) and red circles mark control cities (ie; cities with no Uber entry) I exploit the variability in the regulatory context for Uber alike services in Europe to capture the intensity in the use of Uber services. I should expect a more intense use of Uber services in those cities where the regulatory environment imposes less restrictions. In this regard, the number of licences available for ride-hailing firms may be capped, while technical requirements on drivers, cars and the service to weaken competition are frequently imposed (Rienstra et al., 2015). The study of Frazzani et al. (2016) prepared for the European Commission provides detailed information about the regulation of the hired cars with driver sector in all EU member states up to 2016. The regulation of ride-hailing firms like Uber present two common features across EU: (i) the driver obligation to perform the service based on a prior reservation (for a pre-arranged fixed fare); and (ii) the driver obligation to return to the place of business after each ride, except if there is a prior reservation. The main purpose of these two obligations is to avoid ride-hailing firms picking up passengers on the street. In the period considered, the only exception is in Northern Ireland where since 31 May 2016, ride-hailing firms are able to ply for hire in certain areas of Belfast. Thus, competition is in theory restricted to the pre-booked segment although the waiting times may be so low in some countries that competition between taxis and Uber alike services may be de facto very intense. Furthermore, almost all EU Member States regulate the for-hire car with drivers' sector at the national level, except Belgium and the United Kingdom. In other countries, the national legislation is supplemented by regional legislation and local regulation. However, the regulation is mainly based on national legislation for the period considered even in semi-federal states like Spain. While cities within a country may impose or relax the regulatory restrictions set at the national level, I did not identify a different regulatory environment between cities from the same country (with the exception of Belgium). Table 2 provides the details of the regulatory environment for ride-hailing firms on all countries for which I have at least one city in my sample. While I can distinguish three regulatory environments, the activity of Uber should be particularly constrained in a regulatory environment that imposes quantitative restrictions and strong qualitative requirements. Professional qualification test; 3 years of driving experience Financial guarantee (EUR 18,400 per vehicle) Professional card, morality conditions, the professional qualification and the creditworthiness of the applicant. Suitability of the vehicle In Brussel and Walloon region, minimum 3 hours and EUR 90 Professional card, training + exam Financial standing, technical requirements for vehicles (minimum length and height, or hybrid vehicles) Some countries impose quantity restrictions so that the licences for Uber alike services may be capped in function of different parameters like the population of the city or the number of taxi licences. The imposition of quantitative restrictions is always accompanied with strong specific qualitative requirements like minimum price and time of the trip in Belgium, financial plan and credit line in Denmark, minimum pre-booking time in Greece, obligation to have a garage in Italy, obligation to keep the contract with the passenger in the vehicle in Germany and Spain or to have a car rental contract in Portugal. Second, some countries do not impose quantitative restrictions, but they impose specific qualitative requirements like financial standing of drivers or technical requirements of vehicles. Finally, some countries do not impose quantitative restrictions and ride-hailing firms are subject to the same qualitative requirements than taxi services. Table 2 also shows the total number of cities included in the sample for each country and the number of cities with Uber services in at least some year of the considered period. In very restrictive environments, Uber is only offering services in the largest cities of the country. This is the case for example of Germany and Spain. The only exception would be Italy with a large number of cities with Uber services in 2014. However, in many cases these are services that were offered by Uberpop (that do not use professional drivers) that were banned in 2015. In less restrictive environments, Uber's expansion goes beyond the most populated cities. This is the case for example in Poland, the UK or especially France. Table 2 also reports information on the regulation for taxi services. In all countries, taxi drivers must pass official exams and other qualitative requirements may include financial capability and medical fitness. The need to pass official exams implies that, in some countries, the regulation may be stricter for taxi services than for ride-hailing firms. Bearing this in mind, some countries impose stringent regulations like quantitative restrictions or fixed fares to taxi services. In this regard, a cap in the number of licenses is imposed for taxi services in all countries where ride-hailing services are also subject to stringent regulations. Regarding the countries where ride-railing firms only need to meet some specific qualitative requirements, we may find cases where taxi services are not subject to strict regulations (Poland, Wallonia) and others where taxi services are subject to quantitative restrictions (France, Prague) or fixed fares (Austria, Budapest). Finally, a liberal environment for ride-hailing firms usually implies a liberal environment for taxi services (Ireland, the Netherlands, Slovakia, Sweden, London). However, quantitative restrictions are imposed in Finland, Romania and United Kingdom (except London). Figure 2 shows the number of cities with Uber presence in at least one year of the considered period according to the different regulatory environment in which it operates. A strict regulatory context for Uber alike services is almost always associated with a strict regulation for taxi services. In contrast, the regulation of taxi services is more flexible in around a third of the cities where the regulation for Uber is also more flexible. Note that the sample does not include cities highly liberal in the regulation of taxi services but with a strong regulation in terms of ride-hailing services. Regulatory environment for cities with Uber services Note: This figure shows information for cities with Uber services in at least one year of the considered period. High regulation_Uber means that Uber services are subject to quantitative restrictions and strong qualitative requirements. Low regulation_Uber means that Uber services are subject to some specific qualitative requirements and Liberal_Uber means that Uber is not subject to quantitative restrictions nor specific qualitative requirements. High regulation_taxis means that taxi services are subject to quantitative restrictions and Low regulation_taxis means that taxi services are not subject to quantitative restrictions. The figure provides the number of cities with Uber presence according to the different regulatory environment in which it operates The dependent variable (Congestion) measures the additional time as a percentage that a vehicle needs for any trip in the city compared to a situation characterized by free traffic flow. Data have been obtained from TomTom (https://www.tomtom.com/en_gb/trafficindex). To identify the impact of Uber, I use a dummy variable that takes a value of one if Uber offers any of its services (UberX, UberBlack, UberPop, etc.) in the city, and 0 otherwise. The dates of entry and exit of Uber have been obtained from local newspapers and from Uber's own press releases. As control variables, I take into account different attributes of cities as potential drivers of congestion. The appropriate spatial unit of analysis is the city given that Uber entry in the core city of the urban area does not necessarily imply its entry in all municipalities within the urban area. Thus, control variables used refer to the most disaggregated spatial unit possible based on data availability. First, I include the population density that is measured by the number of inhabitants per square kilometre at the city level. The relationship between urban density and congestion is unclear as denser cities are characterized by a lower number of vehicle/kilometres travelled but traffic is concentrated in fewer points (Ewing et al., 2014, 2018; Sarzynski et al., 2006; Su, 2010). Data for population have been obtained from Eurostat, while data for the size of the city in terms of square kilometres have been collected from city councils' websites. Second, I also consider the income by incorporating the regional GDP per capita (Income variable). The income variable is the GDP per capita at the NUTS 3 level. The relationship between income and congestion is again unclear. Although a positive relationship seems to make sense as the number of car trips in richer cities is typically higher, it is also true that richer cities have better infrastructures (including roads and different types of public transportation) that could mitigate congestion. Data for this variable have been obtained from Eurostat. The quality of public transportation networks is also taken into consideration. Since comparable data for urban buses are not available, I incorporate a comprehensive measure of the urban rail systems in terms of total kilometres of rail lines per square kilometre (Rail_length variable), which includes metro, light trains, trams, and local trains. This variable captures an important source of variability in my sample as we may find cities with a dense network of metro lines and cities with no rail services available. A recent review by Beaudoin et al. (2015) suggests that better public transportation options can help in reducing congestion, with the magnitude of such effects being specific of each particular location. Data for this variable have been obtained from Urban rail (http://www.urbanrail.net/), World Metro database (http://mic-ro.com/metro/table.html) and operators' websites. I also consider a variable that measures the number of registered cars per inhabitant. Data for this variable have been obtained from Eurostat and it is at the NUTS 2 level. We may expect more congestion in cities with more cars per inhabitant. This variable has the limitation that the geographic unit with available data focuses on the surrounding region and not the city. The amount of precipitation (Rain variable) at the city level is also included as a regressor. Data is for the precipitation sum with unit 0.01 Precipitation (mm) and it have been obtained from Tank et al. (2002). We may expect more congestion in more rainy cities. All regressions include city fixed effects and year fixed effects. Most regressions also include the interaction between city and year fixed effects. In this regard, the estimation of a city fixed effects model implies the identification of changes from one period to another because it is based on the within-transformation of the variables as deviations from their average. Therefore, it is the most appropriate method to evaluate the effect of Uber on congestion. Note also that city fixed effects control for omitted, time-invariant factors correlated with the variables of interest. Furthermore, I add year dummies to control for yearly effects that are common to all cities. Finally, I include city specific time trends so that I control for the fact that each city may have its own linear time trend. Some potential threats may distort the identification of the impact of Uber on congestion. The dummy variable for Uber exploits the variability between cities with and without Uber services, but it does not capture the intensity of Uber use that can differ substantially from one city to another with Uber services. Furthermore, Uber entry may be correlated with the levels of congestion in the city. Finally, Uber entry may be correlated with unobserved time varying factors that also have an influence on congestion. To capture the intensity of Uber use, I examine the different effects of Uber on congestion based on the regulatory environment in which Uber is operating. We could expect that the intensity of use of Uber is higher in those cities that impose less restrictions to ride-hailing firms. Another potential threat to my identification strategy is that the treatment (Uber presence) may be correlated with the explanatory variables and with the previous levels of congestion in the city. To deal with this concern, I include city specific time trends so that I control for the fact that each city may have its own linear time trend. Furthermore, I apply the logic of differences in differences with matching that is a common methodology employed within the treatment evaluation framework (see Angrist & Pischke, 2009; Gertler et al., 2016 for details). Indeed, to identify the causal effect of the treatment status on outcomes we need to compare comparable cities. Thus, I apply a propensity score matching procedure and re-estimate Equation (1) with the observations that have common support. Matching procedures eliminate possible bias by pairing observations in the treated cities (with Uber services in at least one year of the considered period) with control cities (with no Uber services in any year of the considered period) having similar characteristics. In a second step, I match the observations in the treated and control groups with respect to the propensity score, using the first nearest neighbour algorithm. This algorithm matches treated observations with the control observations that have the closest propensity score. Then, I drop all the observations without common support and re-estimate Equation (1) using the reduced matched sample. Furthermore, I also reframe the Equation (1) in which the Uber dummy variable is now defined as so that equals 1 if Uber enters in the city k years from year t. This definition implies that k = 0 represents the first year following the entry of Uber, k = −1 is the year prior to treatment and k = 1 is the year after treatment. With this event study analysis, I can examine the timing of the Uber effects after treatment, and I can test for parallel trends between treat