CASE WESTERN RESERVE UNIVERSITY



STATISTICS COLLOQUIUM


Title: Simulation - Extrapolation Estimation for Calculated Variables

Speaker: John Holcomb
Department of Mathematics and Statistics Youngstown State University

Time: Friday, October 18,1996
3:30 pm - refreshments
4:00 pm - talk

Location:
Room 327, Yost Hall

Abstract

Calculated variables arise when a set of directly measured variables are combined algebraically to obtain new quantities of interest. This technique arises quite often in the medical sciences. When the set of direct measurements contain error, estimating the slope between calculated variables can become very difficult. In this situation, measurement error components of the directly measured variables will appear in both the calculated independent and dependent variables. This causes an artificial correlation resulting from the undesired measurement error. In addition, the measurement error of the predictor variable will no longer be independent of the error free quantity. This paper adapts the simulation-extrapolation techniques developed by Cook and Stefanski (1994) to two cases of calculated variables. The results of this adaptation are compared to a traditional method of moments estimator in a large scale simulation.