Alleging Classwide Racial Discrimination Can Be Uber-Challenging

If you’re an Uber rider, you’re probably familiar with Uber’s requests that you rate your driver. It’s a five-star rating system. Five is the best. One is the worst.

As it turns out, those ratings are vitally important to Uber drivers. Uber requires its drivers to maintain a minimum star rating.

This post is about how that rating system prompted a class action on racial discrimination — and about what a recent decision in that case teaches about how to prove classwide discrimination.

In 2015, Uber drivers in San Diego had to maintain a rating of at least 4.6 stars. A San Diego Uber driver named Thomas Liu fell below 4.6, so Uber deactivated his contract.

Mr. Liu is of Asian descent. His putative class action alleged that Uber’s rating system discriminated against him because of his race. He argued specifically that (1) prospective riders cancelled their rides after they saw his profile picture, and (2) riders who accepted his rides would sometimes ask, in an unfriendly way, where he was from.

Mr. Liu grounded his allegations in Title VII of the Civil Rights Act of 1964. Title VII offers multiple legal protections for employees, including protection against racial discrimination. Mr. Liu alleges that Uber’s star-rating system has a disparate impact on drivers of Asian descent.

When Uber moved to dismiss Mr. Liu’s lawsuit, the federal district court in San Francisco twice gave Mr. Liu leave to amend. Uber moved to dismiss each of those amended complaints as well.

In its order on Uber’s second motion to dismiss, the court urged Mr. Liu to “make a more sophisticated effort at the front end to develop a plausible factual basis in support of the assertion that terminations at Uber occur on a racially disparate basis.” The court noted that Uber’s business model made it difficult to develop that factual basis because an Uber driver cannot easily observe race-based disparities in how Uber treats it drivers.

In response to the court’s order, Mr. Liu’s lawyers conducted a survey of 20,000 Uber drivers who are clients of those lawyers. The survey asked the Uber drivers the following question: “If you have been deactivated by Uber, was it because your star ratings were too low?” The survey participants were then asked to check a box indicating their race.

According to Mr. Liu’s second amended complaint, the survey results showed a statistically significant disparity between white drivers and those who identify as Black, Asian, or “other.”

In the court’s view, were those survey results sufficient to state a plausible claim?

No, for two reasons.

First, the survey used the term “Latinx” as a racial category. The court concluded that this wording might have skewed the survey results because some drivers may not have known what the term “Latinx” means. The complaint itself showed that that many respondents marked “other” because of confusion over the term “Latinx.”

Second, and more significantly, the court held that the survey’s methodology was fundamentally flawed. Specifically, the survey asked only for drivers who had already been deactivated from Uber. As the court explained, by asking for drivers who had already been deactivated, the survey used the wrong denominator. The correct denominator would be a sample of Uber’s total driver population.

The court explained the nature of this error through an example. Imagine, the court said, that the survey had 200 respondents—100 white drivers and 100 Black drivers. Assume that the average star rating was the same for all drivers. Then assume that Uber deactivated 20 white drivers and 10 Black drivers, with 5 white drivers and 5 Black drivers deactivated because of star ratings, and the other 15 white drivers and 5 Black drivers deactivated for other reasons.

The court explained that, in this hypothetical, the star-rating system did not have a disparate impact on Black drivers. The same percentage of white drivers and Black drivers — five percent of each — were deactivated because of star ratings. But by using the number of deactivated drivers as the denominator, one might assert that deactivation was twice as likely for Black drivers (five out of ten) as for white drivers (five out of twenty).

The good news for Mr. Liu is that although the court dismissed his complaint for a third time, it did so without prejudice. And on June 20, Mr. Liu filed his fourth iteration of the complaint.

Mr. Liu’s third amended complaint is substantively identical to the second amended complaint — with one exception. The new complaint has a footnote that explains that the survey in fact included drivers who had been deactivated and those who had not. The footnote says that Mr. Liu’s counsel emailed the survey respondents to confirm this point, and that the responses he received showed that over half had not been deactivated.

Time will tell whether this footnote cures the problem. Even if the footnote keeps Mr. Liu’s case alive, the court’s decision still provides valuable lessons to class-action practitioners.

First, the decision teaches that even if a disparate-impact theory might seem plausible in the abstract, Rule 12 still requires a plausible factual basis to show how the theory will be proven. Reliance on studies and academic materials — which Mr. Liu’s complaints included—might not be enough.

Second, a court is likely to be attentive to this burden even if the plaintiff lacks access to information that might be important to establishing the factual basis. The Liu decision provides a strong incentive for a plaintiff to perform as much pre-complaint legwork as possible to assess disparate impact.

Finally, Liu shows that in data-intensive class actions, motions to dismiss might look, in tennis terms, more like a rally than like an ace.


Keith Matier, a rising second-year law student at University of Pennsylvania and summer associate at Robinson Bradshaw, contributed to this post.