We create a simulated city environment with drivers, passengers and various parameters which we will introduce throughout the section. Our goal is to simulate a diverse range of cities and real-world traffic conditions, thus we derive our model parameters from real-world data sets.Taxis in our city drive along a grid, moving one block at each time step, with constant speed. In our basic setting, both request and drop-off locations are most likely to be in the city center, in line with previous studies on real-world data39,40,41,42. The matching algorithm between drivers and requests is similar to the algorithm that Uber and most taxi companies use: passengers are matched with the closest available car. Additional parameters capture various real-world scenarios such as city layout, changes in supply and demand, driver strategies, and different settings of the matching algorithm. The pricing scheme is similar to that of UberX in Boston43, and fuel costs are accounted for44. We run all simulations for what equals a 40-hour workweek (see Section 3 for details), and calculate income as the earning over the time a driver spends online. Initially, the closest available driver is assigned to the longest waiting passenger (see details of the nearest algorithm in Section 3), and every driver remains at the drop-off location of their latest passenger.
According to interviews with Uber drivers15, the main concerns of drivers are their hourly wage, the overall rates with which they operate, and the number of hours they need to work until they reach a given daily/weekly income target. In our simulation, inequality will manifest as a varying hourly wage, Rolstoelvervoer Zevenkamp | Zorgtaxi Rotterdam 010 – 818.28.23 as drivers spend the same amount of time online. In real life, the same often manifests as a highly varying number of hours needed to achieve a given amount of income. Since Uber drivers have limited possibilities to communicate with each other (in contrast to traditional taxi companies, see p. 9415, it prohibits them from comparing their wages and strategies. Therefore, while drivers might not recognize it, their lack of information and the information asymmetry towards to company are the greatest problems in such a system. Thus, our goal is to create a system of full information, helping us investigate the distribution of incomes across workers for various settings.We use two variables to calibrate the supply and the demand in our system: the supply d (density) is defined as the number of taxis per square kilometer, while R captures demand-to-supply ratio, which is the fraction of the demanded travel distance over the supplied travel distance (see Section 3 for details). To pick parameter spaces for d and R that cover actual real-world scenarios, we calculate the number of cars and passengers using empirical data from the NYC Taxi and Limousine Commission45.
The dots in Fig. 1A depict each hour of the year 2013 in NYC in the d−R parameter space. On top, in white, we place an example day, 15 January 2013, with its hourly parameter values. We see that even within one day and one city, both parameters strongly vary. While such detailed data is not available from other cities, we analyzed aggregate statistics as well as extreme events such as public transportation failures, strikes, or bad weather conditions (see Section 3). We conclude that the taxi density range d∈[3,30] [1/km2] and the demand-to-supply ratio range R∈[0.1,1] will capture a wide range of city sizes, seasonal changes, varying traffic at different times of the day, and even extraordinary events that cause sudden changes in the demand or supply.