Dissertation
Finding optimal factor analytic representations of the structure of personality
University of Iowa
Doctor of Philosophy (PhD), University of Iowa
Summer 2022
DOI: 10.25820/etd.006720
Abstract
Factor analysis has served as an essential tool for understanding the nature of personality and other constructs. Over the years, factor analysis techniques have evolved from simple exploratory analyses to more sophisticated approaches that include: (a) confirmatory factor analyses (CFAs) that allow for direct testing of hypothesized models; (b) estimation procedures that adjust for non-normal distributions, correct for ordinal-level data, and enhance model convergence; (c) alterations of models to take method effects (e.g., negative phrasing of items) into account; and (d) extensions of simple structure models to allow for small cross-loadings for non-targeted items. The purpose of this dissertation was to investigate the effects of estimation method, negative item wording, and factor loading constraints on model fit indices and magnitude of factor loadings for correlated factor, hierarchical, and bifactor models of personality using a sample of 447,500 respondents who completed the International Personality Item Pool – Neuroticism, Extraversion, and Openness (IPIP-NEO-120) questionnaire. The IPIP-NEO-120 is available free of charge, computer-administered, brief yet comprehensive in coverage, and amenable to evaluation of multiple types of factor models representing domain and nested facet scores congruence with the Big Five Factor model.
Overall, results across the three types of factor models highlighted the importance of controlling for scale coarseness, accounting for method variance, allowing small off-target loadings when using maximum likelihood (ML) and robust weighted least squares estimation (WLSMV), and including informative priors when using Bayesian estimation. When properly configured, Bayesian with informative priors and WLSMV estimation provided noticeably better model fits than did Bayesian with non-informative priors and ML estimation. The best fits were yielded by bifactor, correlated factor, and hierarchical models, respectively, but excellent fits were achieved for each type of model when relationships were appropriately represented. Performing these analyses also significantly expanded evidence of reliability and validity for IPIP-NEO-120 domain and facet scores.
Taken collectively, findings from this dissertation support models of the structure of personality within several theoretical frameworks and contribute to ongoing research in refining factor analytic techniques. Methods demonstrated here are widely applicable to self-report measures in general and may serve as templates for future investigations into other multidimensional psychological constructs. Recommendations for future research into factor analytic techniques and further refinements of personality and other self-report measures are discussed. Code in Mplus for analyzing all models is included in an appendix at the end of the dissertation.
Details
- Title: Subtitle
- Finding optimal factor analytic representations of the structure of personality
- Creators
- Hyeri Hong
- Contributors
- Walter Vispoel (Advisor)Stephen Dunbar (Committee Member)Catherine Welch (Committee Member)Terry Ackerman (Committee Member)Megan Foley-Nicpon (Committee Member)
- Resource Type
- Dissertation
- Degree Awarded
- Doctor of Philosophy (PhD), University of Iowa
- Degree in
- Psychological and Quantitative Foundations
- Date degree season
- Summer 2022
- DOI
- 10.25820/etd.006720
- Publisher
- University of Iowa
- Number of pages
- xv, 411 pages
- Copyright
- Copyright 2022 Hyeri Hong
- Language
- English
- Description illustrations
- Tables, charts
- Description bibliographic
- Includes bibliographical references (pages 182-209).
- Public Abstract (ETD)
The purpose of this dissertation was to investigate the effects of estimation method, negative item wording, and factor loading constraints on model fit indices and magnitude of factor loadings for correlated factor, hierarchical, and bifactor models of personality using a sample of 447,500 respondents who completed the International Personality Item Pool – Neuroticism, Extraversion, and Openness (IPIP-NEO-120) questionnaire. The IPIP-NEO-120 is available free of charge, computer-administered, brief yet comprehensive in coverage, and amenable to evaluation of multiple types of factor models representing domain and nested facet scores congruence with the Big Five Factor model.
Results across the three types of investigated factor models highlighted the importance of controlling for scale coarseness, accounting for method variance, allowing small off-target loadings when using maximum likelihood (ML) and robust weighted least squares estimation (WLSMV), and including informative priors when using Bayesian estimation. The best fits were yielded by bifactor, correlated factor, and hierarchical models, respectively, but excellent fits were achieved within each type of model when properly configured.
Findings from this dissertation support several theoretical representations of the structure of personality and contribute to ongoing research in refining factor analytic techniques. Methods demonstrated here are widely applicable to self-report measures in general and may serve as templates for future investigations into other multidimensional psychological constructs. Recommendations for future research into factor analytic techniques and further refinements of personality and other self-report measures are discussed. Code in Mplus for analyzing all models is included in an appendix at the end of the dissertation.
- Academic Unit
- Psychological and Quantitative Foundations
- Record Identifier
- 9984284951602771
Metrics
1 File views/ downloads
69 Record Views