Regression modeling: Latent structure, theories and algorithms

The topics of this thesis stem from two EPA/NISS (Environmental Protection Agency/National Institute of Statistical Sciences) projects, which require the use of available data to make risk assessment, estimate uncertainty and suggest future studies. Based on the heterogeneous and batch correlated nature of the data, the thesis invents some new regression modeling methods, provides theoretical background for these newly developed and some other existed ad hoc modeling techniques, and develops associated algorithms. The modeling techniques include scaled link in the class of generalized linear model, newly developed aspects of conditional and marginal modeling techniques, and latent modeling of nonzero control (baseline) regression model. We have Monte-Carlo-Newton-Raphson Algorithm, Gibbs Sampler, EM algorithm and algorithm to evaluate weighted sum $\chi\sp2$ quantile. The associated theories are provided. In scaled link model, some sensitivity studies are made.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Statistics

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2011

Author dc:creator

• Xie, Minge

Contributors dc:contributor

• Simpson, Douglas G.

• Biology, Biostatistics
• Statistics
• Environmental Sciences

Statement dc:rights

• Copyright 1996 Xie, Minge

Language dc:language

eng

Identifier

AAI9625216
(UMI)AAI9625216

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/20609