As discussed here, on June 21, 2022, the United States Department of Justice (DOJ) filed a lawsuit and on June 27, 2022, obtained a settlement agreement with Meta Platforms Inc. (Meta previously known as Facebook) to resolve allegations that Meta’s housing advertising system discriminated against Facebook users in violation of the Fair Housing Act (FHA). On January 9, 2023, the DOJ advised the court that, pursuant to the settlement, the parties reached an agreement on the Variance Reduction System (VRS) compliance metrics. These metrics are intended to reduce variances in the delivery of housing ads for sex and estimated race/ethnicity so that the actual audience that sees an ad will more closely reflect the eligible target audience.
As background, the DOJ’s complaint specifically alleged that Meta enabled and encouraged advertisers to target their housing ads by relying on characteristics such as race, color, religion, sex, disability, familial status, and national origin (Protected Characteristics) to decide which Facebook users would be eligible to receive certain housing ads. Further, the DOJ alleged Meta’s ad delivery system used machine-learning algorithms that relied in part on Protected Characteristics to help determine which subset of an advertiser’s targeted audience would actually receive a housing ad. The DOJ alleged both disparate treatment and disparate impact discrimination.
The parties’ agreement provides that:
- Meta will use the VRS so that the actual ad audience more closely reflects the eligible audience.
- The VRS works by showing an ad to a large group of people and then measuring their aggregate age, gender, and estimated race and ethnicity. That measurement is compared to measurements of the population of people who would have been eligible to see the ad. If there is a variance, the system corrects for this as the ad is shown to more people.
- The agreement provides for certain allowed variances in sex and estimated race/ethnicity. Those allowed variances decrease over time. For example, for housing advertisements with more than 1,000 ad impressions, the VRS variance for sex by April 30, 2023 is less than 10% for 82.6% of ads. By December 31, 2023, it is less than 10% for 91.7% of ads.
- Measurements for estimated race/ethnicity will be based on information estimating race/ethnicity using a privacy-enhanced version of the Bayesian Improved Surname Geocoding (BISG) methodology. The use will be limited to aggregate measurements and subject to privacy protections to protect individual identification.
- Meta will be subject to court oversight and regular review of its compliance with the settlement agreement through June 27, 2026.
As part of the original settlement:
- Meta agreed to stop using the ad targeting tool known as “Special Ad Audience” for housing, employment, and credit advertisements that used a machine-learning algorithm that considered Protected Characteristics in finding Facebook users who shared similarities with an advertiser’s source audience.
- Meta agreed to pay a $115,054 civil penalty, which is the maximum penalty currently available under the FHA.
The allowable variances agreed to by the parties will likely be used as benchmarks by the court and DOJ in monitoring Meta’s compliance efforts. More importantly, this lawsuit and settlement are extremely critical barometers for the very unsettled issue of how regulators will evaluate targeted advertising for housing and other products with anti-discrimination statutes, like consumer credit. Here is what we see as the implications for the industry of the settlement becoming final:
- For regulated entities that were reluctant to advertise on Facebook specifically, the fact that the company and the DOJ have reached an agreement should signal that the level of regulatory risk in advertising on that platform has been reduced. This will likely come as a relief to institutions that wanted to use the Facebook platform for advertising, but may have worried about doing so with the litigation not yet fully resolved.
- The agreement reached to reduce “variances” in persons to whom advertisements are displayed appears to involve the use of an automated system to measure and then correct such variances, and reduce them to levels below the thresholds set forth in the letter sent to the court. It seems likely that other targeted advertising providers (social media platforms, web search platforms) may experience pressure from either regulators or customers to adopt similar measures.
- Likewise, the allowable variances agreed to by the parties seem likely to be used as benchmarks by regulators in the future, particularly the Consumer Financial Protection Bureau (CFPB). The CFPB noted in its March 16, 2022 press release announcing modifications to its UDAAP Examination Manual (which massively expanded the CFPB’s interpretation of UDAAP to include discrimination concepts) that it would review targeted advertising and marketing for evidence of discrimination for the first time. While it was unclear at the time how the CFPB would assess whether advertising or marketing would violate its new UDAAP interpretation, it seems likely to us that the CFPB will now be inclined to use, or try to use, these VRS compliance metrics as a roadmap for evaluating targeted advertising and marketing in examinations, supervision, and investigations going forward.
There is, however, still one area of uncertainty. The DOJ’s complaint alleged the direct use of Protected Characteristics in determining users who saw housing advertisements, but the violations were pled as both disparate treatment and disparate impact violations. The remedy, as communicated to the court on January 9, seems to adopt a very disparate impact-centric measurement for compliance. Does this mean that disparate impact principles in general should be applied to advertising (if that is possible), or is this remedy simply a convenient way of reversing the disparate treatment the DOJ alleged in its complaint? We don’t have an answer to that question.
Our team will, of course, continue to monitor the issue of targeted advertising as it evolves.