Robust Prediction in a Monotonic World
Michael Dougherty
University of Maryland

While data analytic tools have clearly inspired theoretical innovation within psychology, the reverse has rarely occurred: there are few instances of psychological theory inspiring novel analysis tools. In this talk, I introduce a computational framework that serves the dual purpose of both describing human judgment and decision-making and modeling statistical relationships. Instead of developing the computational framework from the perspective of existing analytic or statistical tools, our framework was inspired by theoretical advances in understanding human judgment and decision-making. The algorithm, dubbed the General Monotone Model (GeMM) blends ideas from the areas of cognitive science, knowledge discovery and data mining, and statistics. Using both simulated and real data, I illustrate GeMM’s ability to effectively recover latent data structures and predict new observations while exhibiting extraordinary robustness to non-linearity.