Read our COVID-19 research and news.


A massive study of taxi drivers in Beijing suggests that turning down some paying customers is profitable.

EPA European Pressphoto Agency B.V./Alamy Stock Photo

Having trouble hailing that taxi? This could be why

We've all been there: You're waiting for your Uber or Lyft driver to pick you up. You've got just enough time to make your meeting. But then the ride gets canceled and now you're definitely going to be late. Did you just get turned down by a driver that is searching for better fares? The results of a massive study of taxi drivers in Beijing support that suspicion: Avoiding certain passengers based on their destination is profitable. As companies like Uber and Lyft become the de facto public transportation system in many places, this profit-motivated bias will leave some people stranded on the curb.

The results come from GPS records shared by more than 12,000 taxi drivers in Beijing, who agreed to give researchers route data for 2 months in 2012. The two researchers who dove into this data set, Sihai Zhang and Zhiyang Wang, computer scientists at the University of Science and Technology of China in Hefei, both have had plenty of bad taxi experiences. "Sometimes I've been dumped by my taxi drivers after I tell them my destination," says Zhang, or "the drivers find some excuse, such as 'I am now off duty' or 'My taxi is out of gas.'" And they wondered: From the driver's point of view, how could it make sense to turn down a paying customer?

At first glance, there wasn't that much information in those 2 billion rows of data: They only reveal the start and stop locations of taxi trips. Whether a particular driver turned down one passenger in favor of another was never recorded. But the researchers realized that they could infer what they dub "passenger avoidance" by stepping back and looking at the drivers as a population.

Here's the idea: If every driver in Beijing were just riding around randomly and picking up the first passengers they saw, then all passengers would have an equal chance of hailing a cab, and all drivers would make the same amount of money, on average. But of course, that's not what happens. One of the biases that changes the picture is innocent: Taxis congregate at taxi stands, so the start location of trips aren't randomly distributed but concentrated at places like train stations and stadiums where more people need to be picked up. Yet when they looked at the destinations, another bias emerged: Some drivers were far more likely than chance to take someone to a busy pickup location rather than to some remote location in the city. The reason for that skew, Zhang and Wang inferred, is that those drivers must have been turning down passengers. Otherwise all drivers would have a similar spread of daily trips. But can turning down passengers really be profitable? After all, an empty taxi is making no money.

To test that, the team estimated how much money each driver would have gotten from each of the millions of trips using Beijing's standard taxi fare rates. And sure enough, some drivers were making money far more efficiently than others. The top quartile of drivers earned about $80 per day on average—far higher than that of the average Beijing resident—whereas the bottom quartile made a tenth of that. Then, the scientists compared the trips of the high earners with those of the lower earners.

The results reveal why taxi drivers in Beijing might be choosy: Those who stuck to trips between major pickup areas netted far more money. It turns out that trips to remote places, no matter how long the drive, pay less over the course of the day because the drivers waste time getting back to dense areas. Beijing taxi drivers who made the most money avoided about one in every 12 passengers, the team reports this week in PLOS ONE. How profitable is it? Perhaps as much as $0.75 per rejected passenger, says Zhang, though it's "very hard to tell" because it is a statistical inference.

So next time your ride gets canceled, you're justified in being miffed: You are likely being inconvenienced for a profit.

Although the study relies on correlations rather than catching the drivers red-handed, "I'm convinced," says Vsevolod Salnikov, a computer scientist at the University of Namur in Belgium who studies human mobility. "The problem is significant and exists not only for classical taxi transport, but for [Uber and Lyft] as well." But these phone app–based taxi services do have the power to protect passengers, he says. "They can easily punish or encourage drivers based on their behavior." Because the company can actually see when a driver avoids a passenger—the app handles all the communication between them—it can charge the driver a fee for every ride turned down. So next time you're waiting to get picked up, an algorithm may have your back.