Authors

Ruofeng Wen

Type

Text

Type

Dissertation

Advisor

Zhu, Wei | Wu, Song | Wang, Xuefeng | Yang, Yuanyuan.

Date

2015-12-01

Keywords

Statistics | Conditional Dependence, Error-in-Variables, Graphical Model, Mixed Network, Probit Factor Model

Department

Department of Applied Mathematics and Statistics.

Language

en_US

Source

This work is sponsored by the Stony Brook University Graduate School in compliance with the requirements for completion of degree.

Identifier

http://hdl.handle.net/11401/76382

Publisher

The Graduate School, Stony Brook University: Stony Brook, NY.

Format

application/pdf

Abstract

Learning and visualizing complex causal or dependence structure among various variables is of great interest in applied science. Probabilistic Graphical Models, including Bayesian Networks and Markov Random Field, are well-developed tools for such problems, particularly when variables are either all categorical or continuous. In the first and major part of this text, we proposed a novel graphical structure learning approach, the MIxed Sparse FActor Network (MISFAN), to accommodate categorical and continuous variables seamlessly in one sparse Probit latent factor model. Such a network bridges the gap among latent variable models, traditional multivariate analysis and graphical models, can visualize the underlying interaction and clustering in a more extensive and succinct way, and simultaneously presents certain causal hypothesis as in a Bayesian Network, along with local conditional dependence structure as in a Markov Random Field. Another independent application of latent variable models is from the Error-in-Variable (EIV) perspective. EIV considers the intrinsic and mostly inevitable measurement error affecting the latent true predictors of a regression model. Although proved to be inconsistent and biased on data with measurement error, Ordinary Least Square still dominates for its simple computation and interpretation, while the EIV models seem to be daunting and confusing for its diverse formulations. We intend to give a clear, systematic and unified description with novel geometric insight for the common EIV models in the second part of the text. Additionally, model caveats, parameter specification and alternative estimation approaches are discussed for practical interests. | 74 pages

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.