June 2, 2016

Uber, trust, and price discrimination

On May 17^th, NPR interviewed Keith Chen, head of economic research at Uber. Among other things, he discussed some interesting behaviors of users and drivers toward surge pricing. This information has emerged from the large volumes of data that Uber collects. These include:

1. Diminishing impact. When Uber first introduced surge pricing, going from 1x (regular fare) to 1.2x (20% more than regular fare) resulted in a 27% drop in requests. But after some time, there was only a 7% drop in requests for the same surge level, suggesting that people are getting used to surge pricing and accepting it.
2. Round numbers have strong effects. For example, going from 1.9 to 2.0 reduces demand more than 6 times as much as going from 1.8 to 1.9. And, people would rather pay a surge multiplier of 2.1 rather than 2. Chen suggests that this is because with 2.1, people feel there must be a smart algorithm behind it, whereas with 2.0 they feel that prices have just been arbitrarily doubled and they are being gouged.
3. Drivers respond too. The Uber data shows drivers tend to extend their shifts if surge pricing is happening compare to same situation.
4. Desperation. How much battery is left on a traveler’s cell phone can help to predict whether people going to accept surge pricing or not.

Regarding the last point, Chen assures us that Uber does not use cell phone charge information to manipulate the prices. However, there is no clear explanation about Uber’s surge pricing algorithm, so it is impossible to determine whether the surge price is fair or not. This raises the question of whether there is price discrimination based on other learned behavior. Besides battery information, it might be possible for Uber to price discriminate based on a passenger’s history of surge acceptance or any other type of information such as gender, age, or how long they have been using Uber.

I did an experiment to look for any variation in pricing for exact same pick up locations for different people. I used an IPhone belong to myself (a female in her 20s who uses Uber frequently), and an Android phone belong to a male in his 30s who has never used Uber before and just registered for the experiment. I tried different battery levels for each one of them (The battery level is visible in the screen shots). I also tried different locations such as near Boston, near Los Angeles, and Tacoma, WA. As expected, I could not find any differences between offered prices. Here are some of the screenshots.

Figure 1. Near Boston. IPhone Charge: 8%, Android Charge: 68%

Figure 2. Redondo Beach, near Los Angeles, IPhone Charge: 14% Android Charge: 4%

Figure 3. Near Tacoma WA, IPhone Charge: 26% Android Charge: 28%

Figure 4. Tacoma, WA IPhone Charge: 17% Android Charge 13%

Previously, Northeastern University researchers tried to reverse engineer how Uber’s surge pricing works in San Francisco and Manhattan. One of their findings was that people who stand on each side of the street or with some distance apart might get different surge pricing which implies that price discrimination is mostly based on the location rather than characteristics of the users and users may be able to avoid it by moving.

It is scary how much information Uber can collect about its clients and the ways it can use them, but it is not the only app service with such capabilities. A lot of popular apps can collect user’s information and use them for their own benefit or sell them. For example, Google Maps probably has much more information about users’ trips, and Amazon has detailed data on shopping behavior which they could use for price discrimination as well. The Northeastern University study explains that since the surge pricing algorithm is not transparent, it would be vulnerable to manipulation. Both users and drivers can find strategies to avoid or create surge pricing. Perhaps because of Uber’s controversial history of abusing customers’ data, as recounted by Naked Security , and the black-box algorithm used for surge pricing, people seem to have a harder time trusting that they would not use this data collection for their benefit.

While our experiment is by no means dispositive, it did not turn up any evidence of Uber price discriminating based on phone OS, battery charge, or traveler characteristics, based on the two accounts I tested.