Date of Award
May 2019
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Psychology
Committee Member
Patrick J Rosopa
Committee Member
Cindy Pury
Committee Member
Job Chen
Committee Member
Ryan Gagnon
Abstract
Though available for over 80 years, bifactor models have recently experienced a resurgence in the psychological and social sciences literature (Reise, 2012). Bifactor models provide an attractive alternative to modeling multidimensional data compared to other approaches (e.g., correlated factors model, second-order factor model). Unlike alternative models, bifactor modeling can effectively model variance that is common between all items and variance that is specific to particular subscales (i.e., specific factors) within a measure. Given its unique benefits, researchers have applied bifactor models to previously validated multidimensional inventories. Researchers have found that when using bifactor modeling on some established scales (e.g., Chen, Hayes, Carver, Laurenceau, & Zhang, 2012; Chen, West, & Sousa, 2006), that particular subscales no longer have significant loadings. This phenomenon is referred to as factor collapse (Mansolf & Reise, 2016) and occurs when a substantial amount of variance is shifted to the general factor from one or more specific factors. As bifactor models are applied to more established scales, researchers will likely continue to discover more subscales that collapse onto the general factor and that these subscales contain items that only measure the general factor. Manipulating eight independent variables in a 3 (sample size) × 3 (number of variables per specific factor) × 3 (number of specific factors) × 3 (number of collapsed factors) × 2 (presence or absence of cross-loadings) × 3 (size of specific factor loadings) × 3 (size of general factor loadings) × 2 (presence or absence of pure indicators) factorial design with 100 samples per condition, a Monte Carlo simulation was conducted in R to better understand when factor collapse occurs, if it can it be accurately detected using currently available rotation methods, and the theoretical and practical implications this has for psychological measurement. Results indicated that one rotation, the Schmid-Leiman with iterative target rotation, performed better than other rotations. Implications regarding the results of the simulation and nuances involved in the use of bifactor models are discussed.
Recommended Citation
Robertson, Stephen, "Bifactor Models and Factor Collapse: A Monte Carlo Study" (2019). All Dissertations. 2366.
https://open.clemson.edu/all_dissertations/2366