FAIR2: A framework for addressing discrimination bias in social data science

Abstract

[EN] Building upon the FAIR principles of (meta)data (Findable, Accessible, Interoperable and Reusable) and drawing from research in the social, health, and data sciences, we propose a framework -FAIR2 (Frame, Articulate, Identify, Report) - for identifying and addressing discrimination bias in social data science. We illustrate how FAIR2 enriches data science with experiential knowledge, clarifies assumptions about discrimination with causal graphs and systematically analyzes sources of bias in the data, leading to a more ethical use of data and analytics for the public interest. FAIR2 can be applied in the classroom to prepare a new and diverse generation of data scientists. In this era of big data and advanced analytics, we argue that without an explicit framework to identify and address discrimination bias, data science will not realize its potential of advancing social justice.