The Statistical Methods for Health Equity Series is a monthly online series co-hosted by the Data Science for Health Equity community, the Alan Turing Institute Health Equity Interest Group, and the Department of Statistical Science at University College London.

---

We are excited to announce the next instalment of our webinar series with speaker Christoph Kern from LMU Munich.

Specific details on the topic are as below:

Topic: 

Model Design Decisions, Lazy Data Practices and Algorithmic Fairness

Abstract:

The downstream effects of machine learning and algorithmic decision making systems critically depend on the decisions made during a systems’ design, implementation, and evaluation. In this talk, we highlight how biases in training data can be mitigated or reinforced along the modeling pipeline dependent on preprocessing, analysis and evaluation decisions. We first present a set of common, yet unreflective data practices which disproportionately affect minority groups and can distort model comparisons. We next highlight how the conceptualization and measurement of protected groups introduces critical decision points in fairness evaluations. Motivated by these examples, we introduce the method of multiverse analysis for algorithmic fairness that draws on insights from the field of psychology. With this approach, we turn implicit decisions during model design and evaluation into explicit ones and demonstrate their fairness implications. By combining decisions, we create a grid of all possible “universes” of decision combinations. For each of these universes, we compute metrics of fairness and performance. We illustrate how decisions during the design and evaluation of a machine learning system can have surprising effects on its fairness and how to detect these effects using multiverse analysis

Speaker Biography:

Christoph Kern is Junior Professor of Social Data Science and Statistical Learning at the Ludwig-Maximilians-University of Munich, Project Director at the Mannheim Centre for European Social Research (MZES) and Research Assistant Professor at the Joint Program in Survey Methodology (JPSM) at the University of Maryland. His research focuses on the social impacts of algorithmic decision-making and on methodology to mitigate algorithmic unfairness and improve training data quality.

--

Please direct any questions about the webinar series to Dr Brieuc Lehmann at b.lehmann@ucl.ac.uk.