Hard-to-Study Risk/Benefit Forecasting Issues

• Activity Number: 151
• Type: Invited
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Risk Analysis
• Author(s): Michael E. Tarter*+
• Companies: University of California
• Keywords: experimental subject replacement ; loglogistic ; lognormal ; order statistic ; stopping rule ; threshold parameter

Abstract

High on any list of practical difficulties is the need to choose truncated models of asymmetric underlying densities. Even in the univariate case these models must be formulated in terms of four or more parameters; hence the motivation for a new model-free procedure for detecting density edge removal.

To detect stump presence, two sets of sample trigonometric moments are compared. The first set contains members of moderate order while the second set contains members of high order. The rationale for this new approach stems from the relationship between the rate of convergence of a curve's Fourier expansion and rate at which the curve's border derivatives approach a common value. For example, for representational purposes when a Normal density with intact tails is expanded a handful of trigonometric moments will suffice. Conversely, a Venus de Milo-like and/or lopsided bell shaped curve will tend to have a sequence of Fourier coefficients that converges slowly.