Addressing Hidden Biases: Enhancing Equity through Causal Modeling in Big Data Evaluations

Selection bias, a pervasive yet often overlooked issue in big data science evaluations, arises when systematic differences between participants impact program engagement. In this presentation, we will reveal how causal modeling can address selection bias in social program data, ensuring equitable treatment by focusing on contextual predictors rather than demographic factors as selection criteria. We will explore the innovative use of machine learning algorithms to identify historical natural experiments within matched groups, effectively controlling selection bias and uncovering strategies that maximize outcomes. Additionally, we will describe a method for detecting potential social biases within each group and assessing treatment disparities based on demographic factors. This presentation aims to illuminate the challenge of social biases in program administration and provide a deeper understanding of how causal modeling using machine learning can expose and address them.