Some Remarks on Extrapolation

Michael Lavine
Duke University

Nov 30 1989

Many standard techniques for model selection, model fitting and diagnostics can be inappropriate in extrapolation problems. We illustrate some pitfalls using a data set analyzed by Dalal, Fowlkes and Hoadley (1989) (DFH) -- the data on space shuttle O-ring damage available at the time of the decision to launch the shuttle Challenger. The aim of the analysis is to estimate the probability of failure for a launch at 31 degrees, the temperature at which the Challenger was launched, and to use that estimate to decide whether the launch should have been postponed. We conclude that the launch should have been postponed, and that the decision is robust to modeling choices.


Decision Analysis, Diagnostics, Extrapolation


PDF icon 1990-05.pdf