The Role of Collider Bias in Understanding Statistics on Racially-Biased Policing
Presented at the 8th Annual BayesiaLab Conference on October 27, 2020.
Abstract
Contradictory conclusions have been made about whether unarmed blacks are more likely to be shot by police than unarmed whites using the same data. The problem is that by relying only on data from ‘police encounters,’ there is the possibility that genuine bias can be hidden. We provide a causal Bayesian network model to explain this bias – which is called collider bias or Berkson’s paradox – and show how different conclusions arise from the same model and data. We also show that causal Bayesian networks provide the ideal formalism for considering alternative hypotheses and explanations of bias.
Presentation Video
About the Presenter
Steven Frazier is an experienced engineer with over 40 years of experience in manufacturing management and operations. Graduating with a BSME from Washington University St. Louis in 1979, he has managed and worked in production operations, maintenance, environmental, and safety for Procter and Gamble, ITW, Coca-Cola, Georgia Pacific, and now with OnPoint, a part of Koch Engineered Solutions. As such, he has been involved in the manufacture of a variety of products, from Cascade dishwashing detergent, flexible plastic, and stretch films, metalized films, paper, and building products such as lumber, plywood, oriented strand board, and gypsum board. In 2000, he received 3 patents related to liquid packaging and bag-in-box technology while working for Coca-Cola. He became interested in Bayesian Networks in 2014 because of their ability to predict, discover, and define cause and effect displayed in simplified graphical models. With his extensive domain knowledge, he is currently working as an advanced analytics process engineer for OnPoint, applying machine learning and Bayesian networks to help customers optimize and solve problems related to their manufacturing operations.
Steven Frazier can be described as a “geek’s geek” – throughout his career, he has been the go-to guy in learning increasingly complex applications to enhance the tool kit for production managers, for operations excellence, problem-solving analytics, and innovations to increase safety, optimize production and bottom-line profitability. He has 3 United States patents; he is a peer-reviewed co-author, and his breadth of experience spans iconic brands of Anheuser Busch, Procter and Gamble, Coca-Cola, and Georgia Pacific. With a Washington University Bachelor of Science in Mechanical Engineering, he has managed and worked in production operations, maintenance, environmental, mergers and acquisitions. As a comparative advantage, Steve has leveraged Bayesian Networks -- to increase manufacturing quality performance while maximizing profit, to examine beneficial reuse analysis, and to assess competitive product quality as an element of capturing increased market share. In his role with Georgia Pacific, he built a first-of-its-kind materials model that surpassed all other comers – both “foreign & domestic.” Today he is part of the Koch Industries OnPoint working as an advanced analytics process engineer applying machine learning to optimize customers manufacturing operations. He became a “thought leader” for Bayesialab when he gave his wife a crash course in big data Bayesian Networks to help her complete her Masters in Sustainability at Georgia Tech. Now that is real-world brand reach – Bayesian Networks applied.