Confidence Intervals for Testing Disparate Impact in Fair Learning

This work addresses the need for more reliable statistical testing in fair machine learning, particularly for researchers and practitioners evaluating group disparities, though it is incremental as it builds on existing indexes.

The authors tackled the problem of quantifying disparate impact in machine learning by deriving the asymptotic distribution of statistical indexes, promoting the use of confidence intervals for testing group disparate impact. They illustrated the importance of using confidence intervals over single values in examples.