Thomas Liu v. Uber Technologies, Inc.

NOT FOR PUBLICATION FILED UNITED STATES COURT OF APPEALS JUN 24 2024

FOR THE NINTH CIRCUIT MOLLY C. DWYER, CLERK U.S. COURT OF APPEALS

THOMAS LIU, individually and on behalf Nos. 22-16507 of all others similarly situated, 22-16712 Plaintiff-Appellant, D.C. No. 3:20-cv-07499-VC v. MEMORANDUM* UBER TECHNOLOGIES, INC., Defendant-Appellee.

Appeal from the United States District Court for the Northern District of California Vince Chhabria, District Judge, Presiding Argued and Submitted December 7, 2023 San Francisco, California

Before: COLLINS, FORREST, and SUNG, Circuit Judges.

Plaintiff Thomas Liu appeals the district court’s dismissal of this putative class action for failure to state a claim on which relief may be granted. See FED. R. CIV. P. 12(b)(6). We have jurisdiction under 28 U.S.C. § 1291. We affirm.

I

Because this case was dismissed at the pleadings stage, we take the following well-pleaded allegations of the operative complaint as true. See Shields v. Credit One Bank, N.A., 32 F.4th 1218, 1220 (9th Cir. 2022). Uber, a

* This disposition is not appropriate for publication and is not precedent except as provided by Ninth Circuit Rule 36-3.

transportation company that connects drivers with riders via a mobile app, uses a “star rating system” whereby passengers are asked to rate their drivers on a scale of one to five after each ride. Uber terminates, or “deactivates,” drivers who fall below a “minimum average star rating,” which “has frequently been set very high.” In 2015, Liu was terminated as an Uber driver in the San Diego area when his average star rating fell below 4.6.

Liu, who is “Asian and from Hawaii and speaks with a slight accent,” filed this putative class action in 2020, alleging that Uber’s use of the star rating system in making driver termination decisions discriminates against non-white drivers. In particular, Liu alleges that Uber’s reliance on the star rating system allows passengers’ racial discrimination against non-white drivers to influence Uber’s termination decisions. Liu asserted race discrimination claims under both Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e-2, and California’s Fair Employment and Housing Act (“FEHA”), CAL. GOV’T CODE § 12940, and he invoked theories of both disparate impact and disparate treatment. The district court dismissed all claims with prejudice under Rule 12(b)(6), and Liu timely appealed.

II

Under Federal Rule of Civil Procedure 8, Liu’s complaint “must contain sufficient factual matter, accepted as true, to ‘state a claim to relief that is plausible *3 on its face.’” Ashcroft v. Iqbal, 556 U.S. 662, 678 (2009) (quoting Bell Atl. Corp. v. Twombly, 550 U.S. 544, 570 (2007) (emphasis added)); see also Mattioda v. Nelson, 98 F.4th 1164, 1174–75 (9th Cir. 2024) (holding that “the Iqbal/Twombly standard” applies to a disability-based “hostile-work-environment claim” under the Rehabilitation Act). Because there are alternative ways to establish a claim of racial discrimination, no particular method of establishing a discrimination claim— such as the prima-facie-case framework set forth in McDonnell Douglas Corp. v. Green, 411 U.S. 792 (1973)—is mandatory at the pleading stage. Swierkiewicz v. Sorema N.A., 534 U.S. 506, 511 (2002) (noting, for example, that “if a plaintiff is able to produce direct evidence of discrimination, he may prevail without proving all the elements of a prima facie case”). Instead, the standard to survive a motion to dismiss is simply whether, in light of the requirements of the substantive law invoked, the plaintiff has pleaded sufficient “factual content that allows the court to draw the reasonable inference that the defendant is liable for the misconduct alleged.” Iqbal, 556 U.S. at 678.

Accordingly, reviewing de novo, see Campanelli v. Bockrath, 100 F.3d 1476, 1479 (9th Cir. 1996), we proceed to consider whether Liu pleaded sufficient facts to support his claims of disparate impact and disparate treatment.1

1 Given that our review is de novo, we need not further address Liu’s contention that the district court improperly applied a heightened pleading standard in evaluating his claims.

A

To state a claim for discrimination under Title VII and the FEHA based on a disparate impact theory, a plaintiff must plausibly allege: (1) a “significant disparate impact on a protected class or group”; (2) “specific employment practices or selection criteria at issue”; and (3) “a causal relationship between the challenged practices or criteria and the disparate impact.” Bolden-Hardge v. Office of Cal. State Controller, 63 F.4th 1215, 1227 (9th Cir. 2023) (citation omitted). Assuming arguendo that Liu has adequately pleaded a specific employment practice—viz., “Uber’s use of its star rating system to terminate drivers”—we conclude that he has failed to plead sufficient facts to raise a plausible inference that this practice is causally related to a “significant disparate impact” on non-white drivers. In arguing for a contrary conclusion, Liu relies on three categories of allegations, but we conclude that, even taking them together, they fall short of Iqbal’s standards.

First, Liu alleges that he experienced “hostile” discriminatory treatment from Uber passengers, including that riders “cancell[ed] ride requests after he had already accepted the ride and the rider was able to view his picture.” However, the complaint itself alleges that riders rate drivers “after each ride,” and Liu pleaded no facts that would plausibly explain how riders who did not use his services could contribute to his Uber rating. Liu also alleged that he “noticed passengers appearing hostile to him,” including “riders asking where he was from in an *5 unfriendly way.” But the bare allegation that Liu sometimes thought passengers used an “unfriendly” tone does not support a plausible inference that any passenger discrimination in rating him was sufficiently pervasive to drive down his overall Uber rating.

Second, Liu’s complaint cites what the district court characterized as a “broad body of social science literature cataloguing the pervasive effects of racial bias in situations where customers rate or value the services they are receiving.” The complaint notes that Uber itself had relied on the racial-discrimination concerns presented in such literature in previously defending its since-abandoned decision to disallow tipping on its app. This literature raises an important concern about rating systems, and it may support an inference of a discriminatory causal relationship if Uber’s rating system is producing a significant racial disparity in terminations. But even assuming that, in an appropriate case, reliance on publicly available reports and studies providing relevant evidence of real-world conditions may provide a basis for plausibly inferring a statistical disparity with respect to a particular defendant, that is not the case here. The cited materials in Liu’s complaint lack sufficient data concerning relevant actual conditions to provide a non-speculative basis for plausibly inferring that any such significant disparity is actually occurring with respect to Uber.

Third, the operative complaint describes the results of a survey of Uber *6 drivers conducted by Liu’s counsel concerning whether the drivers were terminated due to low “star ratings” on the Uber app.2 The complaint states that, “[i]n November 2021, Plaintiff’s counsel sent a survey by electronic mail to approximately 20,000 Uber drivers (who are clients of Plaintiff’s counsel).” This survey “asked the drivers whether they had been deactivated by Uber based upon their star ratings, and it asked them to identify their race.” The complaint alleges that approximately 20% of the drivers who received the survey responded, with the following results: Liu’s complaint summarizes the chart as follows:

2 As we have held, “statistics are not strictly necessary” to plead a viable disparate impact claim. Bolden-Hardge, 63 F.4th at 1227. Where, as here, a complaint does include allegations concerning statistics, we must assess those allegations under Iqbal’s standards, just as we do any other allegations offered in support of an allegedly plausible inference of liability.

As shown above, 17.4% of white respondents indicated that they had been deactivated by Uber based on star ratings. In contrast, 24.6% of Asian respondents, 24.1% of Black respondents, and 24.9% of respondents who identified their race as “Other” than the choices provided indicated that they had been deactivated by Uber based on star ratings. Only 16.9% of Latinx respondents indicated that they had been deactivated by Uber based on star ratings.

The complaint asserts that Dr. Mark Killingsworth, a professor in the Rutgers University Department of Economics, “examined the survey responses and found the results to be highly statistically significant that race is associated with Uber drivers in the survey reporting that they had been deactivated based on their star ratings.”

The complaint further alleges that “Plaintiff’s counsel sent a follow-up email to the survey respondents who had answered ‘no’” to the question whether they had been deactivated based on star ratings, in order “to clarify whether or not they had been deactivated for any reason.” The complaint describes the results of that further survey as follows:

Of the respondents who answered “no” to the survey (and responded to the follow-up request for clarification), 51.7% indicated that they had not been deactivated and 48.3% indicated that they had been deactivated for reasons other than star ratings. Of the drivers who answered “no” to the survey, 56.5% responded to the follow-up request for clarification.

For several reasons, we agree with the district court that the allegations concerning counsel’s survey are insufficient to raise a plausible inference that there *8 is a significant racial disparity in star-ratings-based terminations among Uber drivers.

The crucial element of a “disparate impact” claim requires a showing “that an employer uses ‘a particular employment practice that causes a disparate impact on the basis of race, color, religion, sex, or national origin.’” Ricci v. DeStefano, 557 U.S. 557, 577 (2009) (quoting 42 U.S.C. § 2000e-2(k)(1)(A)(i)). Here, the particular employment practice that is alleged to produce a racial disparity is “Uber’s use of its star rating system to terminate drivers.” But the survey described in the operative complaint does not actually show that non-white drivers are terminated due to low star ratings at different rates than white drivers. As the district court explained, the survey failed to compare, for each racial group, the number of drivers of that race who were terminated due to low star ratings against the total number of drivers of that race in the entire survey pool (assuming arguendo that, at the pleading stage, the entire survey pool is a reasonable proxy for the entire driver population). Because the survey says nothing about the composition of the overall population of Uber drivers from which these star-based-terminated drivers were drawn, it says nothing about whether Uber terminates white drivers due to the challenged practice at different rates than non-white drivers. Consequently, the survey fails to provide any plausible basis for finding a “disproportionately adverse effect on minorities.” Id. Indeed, as the district court *9 recognized, because the survey used the incorrect denominator, the survey could show a disparity even if the challenged practice did not actually have a disparate impact:

An example illustrates the point. Imagine 100 white drivers and 100 Black drivers. Assume that, on average, there exists no difference in star ratings between the white and Black drivers—in other words, no disparate impact. Imagine that Uber deactivates 20 white drivers: 5 due to their star rating and 15 for other reasons. And imagine that Uber deactivates 10 Black drivers: 5 due to their star rating and 5 for other reasons. In this scenario, 50% of Black drivers (5 out of 10) will answer “yes” to the question posed by the survey (“If you have been deactivated by Uber, was it because your star ratings were too low?”), while 25% of white drivers (5 out of 20) will answer “yes” to the same question. That result exists even though Uber deactivated a far lower percentage of Black drivers (10%) than white drivers (20%) and even though there exists no difference in the average star rating between Black and white drivers. Indeed, Uber deactivated the same percentage of white and Black drivers due to their star ratings in this hypothetical. By only asking drivers who have been deactivated whether Uber deactivated them due to their star rating, the survey misses the point.

On top of this fundamental defect, the survey has obvious deficiencies that preclude drawing a plausible inference of disparate impact liability. The follow-up survey showed that more than half of the respondents who had answered “no” to the key survey question (“If you have been deactivated by Uber, was it because your star ratings were too low?”) had actually not been terminated at all, thereby indicating that most respondents had been confused by the question. As a result of this confusion, the survey ended up comparing a set of respondents who said they *10 had been terminated to a set of respondents that included both persons terminated for other reasons as well as a large number of persons who were not terminated at all. The resulting apples-to-oranges comparison means that the survey question was so poorly framed that it did not even accomplish its declared goal of comparing star-ratings-based terminations to terminations based on other grounds. A further design flaw—which the complaint itself candidly acknowledged—is that the survey’s use of the term “Latinx” apparently caused numerous respondents who identify as Latino or Hispanic to “check[] ‘Other’ in response to the survey,” and that made it impossible to draw any “meaningful” conclusions about the survey’s “Latinx” and “Other” numbers.

Because these allegations, taken together, do not support a plausible inference that there is a significant racially disparate impact in driver termination rates that is causally linked to Uber’s use of customer ratings in making termination decisions, we affirm the district court’s dismissal of Liu’s disparate impact claims. And because Liu was afforded three opportunities to amend the complaint and did not seek a further opportunity to amend either in the district court or in this court, we affirm the dismissal of these claims with prejudice. See Unified Data Servs., LLC v. FTC, 39 F.4th 1200, 1208 (9th Cir. 2022).

B

“A disparate-treatment plaintiff must establish that the defendant had a *11 discriminatory intent or motive for taking a job-related action.” Wood v. City of San Diego, 678 F.3d 1075, 1081 (9th Cir. 2012) (citation omitted); see also Godwin v. Hunt Wesson, Inc., 150 F.3d 1217, 1220 (9th Cir. 1998). Liu failed to allege facts that would support a plausible inference that Uber intended to discriminate against non-white drivers in using the star rating system to make termination decisions. Liu’s complaint asserted that an inference of intentional discrimination arises from Uber’s decision to persist in using the star rating system even though Uber “is aware that passengers are prone to discriminate in their evaluation of drivers,” an awareness shown by Uber’s prior reluctance to allow tipping. But we have squarely held that “[i]t is insufficient for a plaintiff alleging discrimination under the disparate treatment theory to show that the employer was merely aware of the adverse consequences the policy would have on a protected group.” Wood, 678 F.3d at 1081 (citation omitted). Liu contends that an inference of intentional discrimination is further supported by the nature of the alleged classwide disparate impact that is attributable to Uber’s practice, see Atonio v. Wards Cove Packing Co., 810 F.2d 1477, 1480 (9th Cir. 1987), but for the reasons we have explained, no such disparity has been adequately pleaded here.

AFFIRMED.