Abstract:
|
The correct specification of dimensionality is fundamental to the selection of appropriate measurement models and valid interpretations of parameter estimates. Thus it is critical to detect model misspecification in terms of dimensionality based on model fit indices. This simulation study investigates the sensitivity of model fit indices to bifactor model misspecification and evaluates structural coefficients bias if misspecification is not detected. Design factors include data structure (9 items with 3 group factors; 36 items with 3 group factors; 36 items with 12 group factors), group factor loading magnitude (0.3, 0.4, 0.5, 0.6), general factor loading magnitude (0.3, 0.4, 0.5, 0.6, 0.7), and type of misspecification (ignoring the general factor; ignoring the group factors). Overall the chi-square statistic outperforms RMSEA and SRMR. When the general factor is ignored, all fit indices are insensitive to misspecification and structural coefficients are negatively and severely biased. Implications for applied and methodological researchers are discussed.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.