A Robust Optimization Approach to Reliable Statistical Inference with Variables Generated by Machine Learning

Organizations increasingly use machine learning to turn text, images, and other unstructured data into variables that inform decisions and research. But, because machine learning predictions are never perfect, the resulting data can contain errors that quietly distort statistical analyses, sometimes leading to incorrect conclusions about what truly drives important outcomes. This study introduces a robust optimization approach that helps analysts and decision makers draw more reliable insights when working with machine learning–generated data. The method is designed to strengthen the signal of real effects, reducing the influence of noisy or imperfect predictions, resulting in more trustworthy hypothesis tests and fewer missed or misleading findings. The approach also includes a simple correction step that uses a small amount of high-quality labeled data—such as a subset of manually reviewed cases—to further improve accuracy. Across simulations and a real-world example using Amazon reviews, the method consistently delivers more dependable results than common alternatives. For professionals who rely on machine learning in areas such as marketing, operations, public policy, or risk management, this framework offers a practical, transparent way to ensure that conclusions remain sound even when data sources are imperfect.

• 저널Information Systems Research
• 토픽Explainable Artificial Intelligence (XAI) · Artificial Intelligence
• DOI10.1287/isre.2023.0340
• 저자Aaron Schecter, Weifeng Li