Rating
Title	Approach to estimating relative goodness of fit with application to the diagnostic interview scheduled in measuring depression for selected race/ethnic groups in the United States
Author	Sadler, Zara Ellis
Date	2004
Affiliation/College	College of Graduate Studies
Department	Biometry and Epidemiology
Extent	ix, 143 leaves : ill.; 29 cm
Source	W 4A S2 2004
Subject.MeSH	Academic Dissertations Academic Dissertations--South Carolina Diagnostic and Statistical Manual of Mental Disorders Depressive Disorder, Major Factor Analysis, Statistical Depression--ethnology Mexican Americans African Americans European Continental Ancestry Group Suicide Nutrition Surveys
Collection	MUSC Theses and Dissertations
Contributing Institution	Waring Historical Library (MUSC)
Rights	Copyright is held by the author.
Language	English
Type	Text
Format	application/pdf
Media Type	Manuscripts
Resource Identifier	mtd2004_sadler_ze
Abstract	The specific aims of this dissertation research are to 1) evaluate and compare the factor structures and item functioning of lifetime major depressive symptoms from NHANES III Diagnostic Interview Survey (DIS) for Non-Hispanic (NH) Whites, NH Blacks, and Mexican Americans (MA) (manuscript 1); 2) develop an approach for comparing independent fit statistics obtained from confirmatory factor analyses fitted within race-ethnic groups (manuscript 2) when the data are binomially distributed; and 3) apply the approach from Specific Aim 2 to compare ethnic specific factor structures identified in Specific Aim 1 for the three race/ethnic groups (manuscript 3). Regardless of race/ethnic group, the exploratory factor analyses performed retained two factors, namely a neurovegetative factor (F1) and a suicidality factor (F2). Of 16 items measuring lifetime depressive symptoms, only "worthlessness" exhibited differential item functioning (DIF) by race/ethnic groups. Misclassification of this item did not explain differences in prevalence. Racial/ ethnic differences in prevalence in NHANES III are not explained by differences in factor structure or DIF. Although the two factors retained did not significantly differ for the three race/ethnic groups, these two factors do differ from the eight groups of symptoms upon which the DSM-III criteria are based. Hence there may be increased potential for under-diagnoses given the scoring algorithm may be incorrect. We explored the large and small sample properties of the bootstrapped variance estimates for the Comparative Fit Index (CFI) using both Mplus, designed for categorical data, and SAS, designed for continuous data. While we found expected bias using SAS, we found minimal bias using Mplus regardless to whether we used large or small samples. We also mis-specified the factor structure and found similar results. Our results suggest that similar results. Our results suggest that bootstrapping techniques, without adjustment for bias, can be used to obtain confidence intervals on the goodness of fit (GOF) measures for the CFI. While we suggest that investigators calculate bias when computing confidence intervals for GOF measures, we expect Mplus results to be robust in most situations. As previously, the overlapping confidence intervals suggest no difference in the exploratory two-factor models for NH Whites, NH Blacks and MAs was detected.

Approach to estimating relative goodness of fit with application to the diagnostic interview scheduled in measuring depression for selected race/ethnic groups in the United States

Description

Tags

Comments