We have released a new technical report: “Auditing for Bias in Ad Delivery Using Inferred Demographic Attributes”, available at https://arxiv.org/abs/2410.23394.
From the abstract:
[Imana23c, figure 3]: Detecting racial skew with BISG-based inference is less sensitive (shown by the lower test statistic Z) than either knowing true-race, or using our improved version that reflects potential inference error. More samples and larger underlying skew make the range of confusion smaller, but do not eliminate it.
Auditing social-media algorithms has become a focus of public-interest research and policymaking to ensure their fairness across demographic groups such as race, age, and gender in consequential domains such as the presentation of employment opportunities. However, such demographic attributes are often unavailable to auditors and platforms. When demographics data is unavailable, auditors commonly infer them from other available information. In this work, we study the effects of inference error on auditing for bias in one prominent application: black-box audit of ad delivery using paired ads. We show that inference error, if not accounted for, causes auditing to falsely miss skew that exists. We then propose a way to mitigate the inference error when evaluating skew in ad delivery algorithms. Our method works by adjusting for expected error due to demographic inference, and it makes skew detection more sensitive when attributes must be inferred. Because inference is increasingly used for auditing, our results provide an important addition to the auditing toolbox to promote correct audits of ad delivery algorithms for bias. While the impact of attribute inference on accuracy has been studied in other domains, our work is the first to consider it for black-box evaluation of ad delivery bias, when only aggregate data is available to the auditor.
This technical report is joint work of Basilial Imana and Aleksandra Korolova (both of Princeton) and John Heidemann (USC/ISI). This work was supported by the NSF via CNS-1956435, CNS-2344925, and CNS-2319409 (the InternetMap project).
Experiments showing educational ads for for-profit schools are disproportionately shown to Blacks at statistically significant levels. (from [Imana24a], figure 4).
Digital ads on social-media platforms play an important role in shaping access to economic opportunities. Our work proposes and implements a new third-party auditing method that can evaluate racial bias in the delivery of ads for education opportunities. Third-party auditing is important because it allows external parties to demonstrate presence or absence of bias in social-media algorithms. Education is a domain with legal protections against discrimination and concerns of racial-targeting, but bias induced by ad delivery algorithms has not been previously explored in this domain. Prior audits demonstrated discrimination in platforms’ delivery of ads to users for housing and employment ads. These audit findings supported legal action that prompted Meta to change their ad-delivery algorithms to reduce bias, but only in the domains of housing, employment, and credit. In this work, we propose a new methodology that allows us to measure racial discrimination in a platform’s ad delivery algorithms for education ads. We apply our method to Meta using ads for real schools and observe the results of delivery. We find evidence of racial discrimination in Meta’s algorithmic delivery of ads for education opportunities, posing legal and ethical concerns. Our results extend evidence of algorithmic discrimination to the education domain, showing that current bias mitigation mechanisms are narrow in scope, and suggesting a broader role for third-party auditing of social media in areas where ensuring non-discrimination is important.
This paper is a joint work of Basileal Imana and Aleksandra Korolova from Princeton University, and John Heidemann from USC/ISI. We thank the NSF for supporting this work (CNS-1956435, CNS- 1916153, CNS-2333448, CNS-1943584, CNS-2344925, CNS-2319409, and CNS-1925737).
I would like to congratulate Dr. Basileal Imana for defending his PhD at the University of Southern California in August 2023 and completing his doctoral dissertation “Methods for Auditing Social Media Algorithms in the Public Interest”.
Basi at his PhD hooding in May 2023 with his thesis advisors.
From the abstract:
Social-media platforms are entering a new era of increasing scrutiny by public interest groups and regulators. One reason for the increased scrutiny is platform-induced bias in how they deliver ads for life opportunities. Certain ad domains are legally protected against discrimination, and even when not, some domains have societal interest in equitable ad delivery. Platforms use relevance-estimator algorithms to optimize the delivery of ads. Such algorithms are proprietary and therefore opaque to outside evaluation, and early evidence suggests these algorithms may be biased or discriminatory. In response to such risks, the U.S. and the E.U. have proposed policies to allow researchers to audit platforms while protecting users’ privacy and platforms’ proprietary information. Currently, no technical solution exists for implementing such audits with rigorous privacy protections and without putting significant constraints on researchers. In this work, our thesis is that relevance-estimator algorithms bias the delivery of opportunity ads, but new auditing methods can detect that bias while preserving privacy.
We support our thesis statement through three studies. In the first study, we propose a black-box method for measuring gender bias in the delivery of job ads with a novel control for differences in job qualification, as well as other confounding factors that influence ad delivery. Controlling for qualification is necessary since qualification is a legally acceptable factor to target ads with, and we must separate it from bias introduced by platforms’ algorithms. We apply our method to Meta and LinkedIn, and demonstrate that Meta’s relevance estimators result in discriminatory delivery of job ads by gender. In our second study, we design a black-box methodology that is the first to propose a means to draw out potential racial bias in the delivery of education ads. Our method employs a pair of ads that are seemingly identical education opportunities but one is of inferior quality tied with a historical societal disparity that ad delivery algorithms may propagate. We apply our method to Meta and demonstrate their relevance estimators racially bias the delivery of education ads. In addition, we observe that the lack of access to demographic attributes is a growing challenge for auditing bias in ad delivery. Motivated by this challenge, we make progress towards enabling use of inferred race in black-box audits by analyzing how inference error can lead to incorrect measurement of skew in ad delivery. Going beyond the domain-specific and black-box methods we used in our first two studies, our final study proposes a novel platform-supported framework to allow researchers to audit relevance estimators that is generalizable to studying various categories of ads, demographic attributes and target platforms. The framework allows auditors to get privileged query-access to platforms’ relevance estimators to audit for bias in the algorithms while preserving the privacy interests of users and platforms. Overall, our first two studies show relevance-estimator algorithms bias the delivery of job and education ads, and thus motivate making these algorithms the target of platform-supported auditing in our third study. Our work demonstrates a platform-supported means to audit these algorithms is the key to increasing public oversight over ad platforms while rigorously protecting privacy.
Basi’s PhD work was co-advised by Aleksandra Korolova and John Heidemann, and supported by grants from the Rose Foundation and the NSF (CNS-1755992, CNS-1916153, CNS-1943584, CNS-1956435, and CNS-1925737.) Please see his individual publications for what data is available from his research.
Overview of our proposed platform-supported framework for auditing relevance estimators while protecting the privacy of audit participants and the business interests of platforms.
Concerns of potential harmful outcomes have prompted proposal of legislation in both the U.S. and the E.U. to mandate a new form of auditing where vetted external researchers get privileged access to social media platforms. Unfortunately, to date there have been no concrete technical proposals to provide such auditing, because auditing at scale risks disclosure of users’ private data and platforms’ proprietary algorithms. We propose a new method for platform-supported auditing that can meet the goals of the proposed legislation. The first contribution of our work is to enumerate the challenges and the limitations of existing auditing methods to implement these policies at scale. Second, we suggest that limited, privileged access to relevance estimators is the key to enabling generalizable platform-supported auditing of social media platforms by external researchers. Third, we show platform-supported auditing need not risk user privacy nor disclosure of platforms’ business interests by proposing an auditing framework that protects against these risks. For a particular fairness metric, we show that ensuring privacy imposes only a small constant factor increase (6.34x as an upper bound, and 4x for typical parameters) in the number of samples required for accurate auditing. Our technical contributions, combined with ongoing legal and policy efforts, can enable public oversight into how social media platforms affect individuals and society by moving past the privacy-vs-transparency hurdle.
A 2-minute video overview of the work can be found here.
This paper is a joint work of Basileal Imana from USC, Aleksandra Korolova from Princeton University, and John Heidemann from USC/ISI.
We published a new paper “Auditing for Discrimination in Algorithms Delivering Job Ads” by Basileal Imana (University of Southern California), Aleksandra Korolova (University of Southern California) and John Heidemann (University of Southern California/ISI) at TheWebConf 2021 (WWW ’21).
From the abstract:
Skew in the delivery of real-world ads on Facebook (FB) but not LinkedIn (LI).
Comparison of ad delivery using “Reach” (R) and “Conversion” (C) campaign objectives on Facebook. There is skew for both cases but less skew for “Reach”.
Ad platforms such as Facebook, Google and LinkedIn promise value for advertisers through their targeted advertising. However, multiple studies have shown that ad delivery on such platforms can be skewed by gender or race due to hidden algorithmic optimization by the platforms, even when not requested by the advertisers. Building on prior work measuring skew in ad delivery, we develop a new methodology for black-box auditing of algorithms for discrimination in the delivery of job advertisements. Our first contribution is to identify the distinction between skew in ad delivery due to protected categories such as gender or race, from skew due to differences in qualification among people in the targeted audience. This distinction is important in U.S. law, where ads may be targeted based on qualifications, but not on protected categories. Second, we develop an auditing methodology that distinguishes between skew explainable by differences in qualifications from other factors, such as the ad platform’s optimization for engagement or training its algorithms on biased data. Our method controls for job qualification by comparing ad delivery of two concurrent ads for similar jobs, but for a pair of companies with different de facto gender distributions of employees. We describe the careful statistical tests that establish evidence of non-qualification skew in the results. Third, we apply our proposed methodology to two prominent targeted advertising platforms for job ads: Facebook and LinkedIn. We confirm skew by gender in ad delivery on Facebook, and show that it cannot be justified by differences in qualifications. We fail to find skew in ad delivery on LinkedIn. Finally, we suggest improvements to ad platform practices that could make external auditing of their algorithms in the public interest more feasible and accurate.
This paper was awarded runner-up for best student paper at The Web Conference 2021.
The data from this paper is upon request, please see our dataset page.
![[Imana23c, figure 3]: Detecting racial skew with BISG-based inference is less sensitive (shown by the lower test statistic Z) than either knowing true-race, or using our improved version that reflects potential inference error. More samples and larger underlying skew make the range of confusion smaller, but do not eliminate it.](https://ant.isi.edu/blog/wp-content/uploads/2024/11/Imana23c_icon.png){kind=icon}
![Experiments showing educational ads for for-profit schools are disproportionately shown to Blacks at statistically significant levels. (from [Imana24a], figure 4).](https://ant.isi.edu/blog/wp-content/uploads/2024/06/Imana24a_icon.png){kind=icon}
