Decision Analysis and Equity

• In Event: Advances in Measurement

Abstract

Before being put into practice, a risk scale is often cut into levels, for example, low, medium, and high. These levels are then used for decision making. For instance, persons scoring above the high threshold are targeted for a treatment program. Making decisions at thresholds of the risk scale is a type of classification. The usefulness of a classification is a function of the accuracy of the risk scale as well as formal decision criteria. In practical settings constraints arise that lead to a focus on different decision criteria. When setting risk scale cut points for assessments conducted in diverse populations, the decision analysis will result in more equitable classifications for different gender and ethnic groups when it considers differential validity in terms of diagnostic accuracy and calibration, parity in terms of positive predictive value, and proportionality in terms of selection ratio. Drawing on real data and previous work in decision analysis, we define these terms and discuss tradeoffs.

Authors

• Bill Dieterich, Northpointe Inc.

• William L. Oliver, Northpointe Inc.

• Tim Brennan, Northpointe Inc.

• Christina Mendoza, Northpointe Inc.