Modeling multivariate populations with translation systems
Abstract
The main objective of this research is the development of regression-based procedures to fit continuous univariate and multivariate cumulative distribution functions to sample data sets. Because of its applicability in a wide variety of situations and because it is easily extended to multivariate distributions, the Johnson translation system of probability distributions was selected as the context for the methodological development. Extensive Monte Carlo studies were conducted to characterize the performance of the least squares methods relative to existing methods such as moment matching and $L\sb\infty$ norm estimation. The results show that the least squares procedures are computationally efficient and yield high-quality fits in a wide range of practical situations. A secondary objective of this research is the implementation of the compared fitting procedures in efficient and portable software. Details are given for the interactive software package FITTR which has been designed to facilitate the modeling of univariate and multivariate data sets.
Degree
Ph.D.
Advisors
Wilson, Purdue University.
Subject Area
Operations research|Industrial engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.