Dissertation/Thesis Abstract

Analysis of longitudinal latent-variables with two sources of measurement error
by Nussbaum, Amy Elizabeth, Ph.D., Southern Methodist University, 2015, 185; 3740030
Abstract (Summary)

One theory of personality development is Latent State-Trait theory, which states that personality trait measurements can be influenced by both traits, or enduring characteristics of an individual, and states, which can change on both internal and external circumstances. Personality trait measurements can have long term clinical and practical implications; hence, the stability of such measurements is extremely important. Partitioning the overall variance into a state-error component and a measurement error component can lead to more accurate predictions and better information about latent trait levels. We will explore techniques for analyzing longitudinal latent variable models of continuous data. These methods include parameter estimation using method of moments estimators, as well as matrix norming methods and maximum likelihood estimation, as well model selection via split-sample and ten fold cross-validation. After developing these methods, we can apply moment-reconstruction prediction techniques to find predictions of the latent variables. Then, we will develop methods to correct for discretized data in order make inferences about latent trait levels. These methods include discretizing mechanisms, marginal imputation, correlation reconstruction, and an implementation of the EM algorithm for parameter estimation. Then, we will apply our statistical methods to a motivating data set from Chmielewski and Watson, 2009, consisting of five trait level measurements on 440 individuals at two different time points.

Indexing (document details)
Advisor: Potgieter, Cornelis J.
Commitee: Cao, Jing, Chmielewski, Michael, Stokes, Lynne
School: Southern Methodist University
Department: Statistical Science
School Location: United States -- Texas
Source: DAI-B 77/05(E), Dissertation Abstracts International
Source Type: DISSERTATION
Subjects: Statistics, Quantitative psychology
Keywords: Grouped data, Latent variables, Linear models, Personality, State error
Publication Number: 3740030
ISBN: 9781339319926
Copyright © 2018 ProQuest LLC. All rights reserved. Terms and Conditions Privacy Policy Cookie Policy
ProQuest