UC Berkeley

Machine learning can be used to help guide and regulate adjudicator decisions, increasing the reliability and overall quality of decision making. The first chapter provides an analytic and normative overview of what I refer to as ``statistical precedent." It explains how statistical models of previous decisions can help assess and improve the reliability of an adjudication system. The subsequent chapters elaborate on and empirically illustrate two of the techniques introduced in the first chapter. Chapter two, using an original dataset of Ninth Circuit Court of Appeals decisions, presents a method for estimating the amount of inter-judge disagreement. Chapter three, using an original dataset of California parole hearings, demonstrates the potential of synthetically crowdsourced decision making.