An official website of the United States government
Information access systems, such as search and recommender systems, often use ranked lists to present results believed to be relevant to the user’s information need. Evaluating these lists for their fairness along with other traditional metrics provide a more complete understanding of an information access system’s behavior beyond accuracy or utility constructs. To measure the (un)fairness of rankings, particularly with respect to protected group(s) of producers or providers, several metrics have been proposed in the last several years. However, an empirical and comparative analyses of these metrics showing the applicability to specific scenario or real data, conceptual similarities, and differences is still lacking. We aim to bridge the gap between theoretical and practical application of these metrics. In this paper we describe several fair ranking metrics from the existing literature in a common notation, enabling direct comparison of their approaches and assumptions, and empirically compare them on the same experimental setup and data sets in the context of three information access tasks. We also provide a sensitivity analysis to assess the impact of the design choices and parameter settings that go in to these metrics and point to additional work needed to improve fairness measurement.
more »
« less
Measuring Group Advantage: A Comparative Study of Fair Ranking Metrics
Ranking evaluation metrics play an important role in information retrieval, providing optimization objectives during development and means of assessment of deployed performance. Recently, fairness of rankings has been recognized as crucial, especially as automated systems are increasingly used for high impact decisions. While numerous fairness metrics have been proposed, a comparative analysis to understand their interrelationships is lacking. Even for fundamental statistical parity metrics which measure group advantage, it remains unclear whether metrics measure the same phenomena, or when one metric may produce different results than another. To address these open questions, we formulate a conceptual framework for analytical comparison of metrics.We prove that under reasonable assumptions, popular metrics in the literature exhibit the same behavior and that optimizing for one optimizes for all. However, our analysis also shows that the metrics vary in the degree of unfairness measured, in particular when one group has a strong majority. Based on this analysis, we design a practical statistical test to identify whether observed data is likely to exhibit predictable group bias. We provide a set of recommendations for practitioners to guide the choice of an appropriate fairness metric.
more »
« less
- Award ID(s):
- 2007932
- PAR ID:
- 10251285
- Date Published:
- Journal Name:
- Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society (AIES ’21)
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
A variety of fairness constraints have been proposed in the literature to mitigate group-level statistical bias. Their impacts have been largely evaluated for different groups of populations corresponding to a set of sensitive attributes, such as race or gender. Nonetheless, the community has not observed sufficient explorations for how imposing fairness constraints fare at an instance level. Building on the concept of influence function, a measure that characterizes the impact of a training example on the target model and its predictive performance, this work studies the influence of training examples when fairness constraints are imposed. We find out that under certain assumptions, the influence function with respect to fairness constraints can be decomposed into a kernelized combination of training examples. One promising application of the proposed fairness influence function is to identify suspicious training examples that may cause model discrimination by ranking their influence scores. We demonstrate with extensive experiments that training on a subset of weighty data examples leads to lower fairness violations with a trade-off of accuracy.more » « less
-
Statistical parity metrics have been widely studied and endorsed in the AI community as a means of achieving fairness, but they suffer from at least two weaknesses. They disregard the actual welfare consequences of decisions and may therefore fail to achieve the kind of fairness that is desired for disadvantaged groups. In addition, they are often incompatible with each other, and there is no convincing justification for selecting one rather than another. This paper explores whether a broader conception of social justice, based on optimizing a social welfare function (SWF), can be useful for assessing various definitions of parity. We focus on the well-known alpha fairness SWF, which has been defended by axiomatic and bargaining arguments over a period of 70 years. We analyze the optimal solution and show that it can justify demographic parity or equalized odds under certain conditions, but frequently requires a departure from these types of parity. In addition, we find that predictive rate parity is of limited usefulness. These results suggest that optimization theory can shed light on the intensely discussed question of how to achieve group fairness in AI.more » « less
-
In this paper, we interrogate whether data quality issues track demographic group membership (based on sex, race and age) and whether automated data cleaning — of the kind commonly used in production ML systems — impacts the fairness of predictions made by these systems. To the best of our knowledge, the impact of data cleaning on fairness in downstream tasks has not been investigated in the literature. We first analyse the tuples flagged by common error detection strategies in five research datasets. We find that, while specific data quality issues, such as higher rates of missing values, are associated with membership in historically disadvantaged groups, poor data quality does not generally track demographic group membership. As a follow-up, we conduct a large-scale empirical study on the impact of automated data cleaning on fairness, involving more than 26,000 model evaluations. We observe that, while automated data cleaning is unlikely to worsen accuracy, it is more likely to worsen fairness than to improve it, especially when the cleaning techniques are not carefully chosen. Furthermore, we find that the positive or negative impact of a particular cleaning technique often depends on the choice of fairness metric and group definition (single-attribute or intersectional). We make our code and experimental results publicly available. The analysis we conducted in this paper is difficult, primarily because it requires that we think holistically about disparities in data quality, disparities in the effectiveness of data cleaning methods, and impacts of such disparities on ML model performance for different demographic groups. Such holistic analysis can and should be supported by data engineering tools, and requires substantial data engineering research. Towards this goal, we discuss open research questions, envision the development of fairness-aware data cleaning methods, and their integration into complex pipelines for ML-based decision making.more » « less
-
Group fairness is measured via parity of quantitative metrics across different protected demographic groups. In this paper, we investigate the problem of reliably assessing group fairness metrics when labeled examples are few but unlabeled examples are plentiful. We propose a general Bayesian framework that can augment labeled data with unlabeled data to produce more accurate and lower-variance estimates compared to methods based on labeled data alone. Our approach estimates calibrated scores (for unlabeled examples) of each group using a hierarchical latent variable model conditioned on labeled examples. This in turn allows for inference of posterior distributions for an array of group fairness metrics with a notion of uncertainty. We demonstrate that our approach leads to significant and consistent reductions in estimation error across multiple well-known fairness datasets, sensitive attributes, and predictive models. The results clearly show the benefits of using both unlabeled data and Bayesian inference in assessing whether a prediction model is fair or not.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1145/3461702.3462588
