Author: Sonja Kelly, Mehrdad Mirpourian
Publisher: Women’s World Banking
Publication Year: 2021
Summary: The following report discussed where gender-based bias originates and how to mitigate such biases in the emerging digital credit space. The report discussed the data collection process on emerging credit platforms and what data would be collected. It also introduced 3 potential origins of bias in algorithms, including sampling bias, labeling bias, and outcome proxy bias. They also came up with three stages to measure data and model fairness. The most impressive part was the fairness classification they used based on the classification by Verma and Rubin (2018). Statistical definition of fairness is based on the likelihood of a prediction and its outcome belonging to type I or type II error. The authors classified 18 types of fairness, including 12 statistical measures and three Similarity-Based Measures.
