A Likelihood-Based Approach to Detecting Aberrant Individuals in Confirmatory Factor Analytic Models
Public Deposited- Last Modified
- March 19, 2019
- Creator
-
Sideris, John
- Affiliation: College of Arts and Sciences, Department of Psychology and Neuroscience
- Abstract
- The study presented here was intended to develop and provide a relatively simple method for detecting aberrant observations in confirmatory factor analysis (CFA). This method exploited a by-product of Full Information Maximum Likelihood (FIML) estimation of these models, the log-likelihood produced for each individual observation. This score, after adjusting for missing data, indexed the degree to which a model fits for a specific individual. A simulation study was run to test this index, labelled adj_lli. Data were simulated under varying levels of covariance structure, proportion of aberrant data, and proportion of missing data. Each cell had 200 samples with n = 200. Additionally, adj_lli was compared to three existing methods: Reise and Widaman's (1999) INDCHI, Yung's (1997) method for detecting outliers in mixture models, and Bollen's A, a general multivariate method (1987). Results indicated that adj_lli was effective in detecting outliers and offered some advantages over three other methods.
- Date of publication
- August 2006
- Keyword
- DOI
- Resource type
- Rights statement
- In Copyright
- Advisor
- Panter, Abigail
- Degree
- Doctor of Philosophy
- Degree granting institution
- University of North Carolina at Chapel Hill Graduate School
- Graduation year
- 2006
- Language
Relations
- Parents:
This work has no parents.
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
umi-unc-1512.pdf | 2019-04-12 | Public | Download |