Back to Annual Meeting Page		American Public Health Association 133rd Annual Meeting & Exposition December 10-14, 2005 Philadelphia, PA

Abstract #101229

George J. Knafl, PhD and Margaret Grey, DrPH. School of Nursing, Yale University, PO Box 9740, New Haven, CT 06536-0740, 203-785-6280, knaflg@ohsu.edu

We have developed methods for objectively evaluating exploratory and confirmatory factor analysis models using k-fold likelihood cross-validation (LCV) as well as SAS macros to support these methods. We demonstrate these methods through analyses of baseline data from a clinical trial, the ABCs (Adolescents Benefit from Control) of Diabetes Study, studying adolescents with type 1 diabetes. These examples cover a variety of aspects of factor analysis modeling including choosing the number of factors, comparing alternative factor extraction methods, evaluating item-scale allocations suggested by rotations, adjusting an item-scale allocation, and comparing alternate specifications for inter-scale covariance. For example, the FACES instrument has 30 items generating two summated scales measuring subjects' perception of their families' adaptability and cohesion. For FACES item responses from the ABC study, the associated 2-factor model has the best LCV score among models with 0-10 factors, indicating that the recommended number of factors is an effective choice for these data. This holds for each of 12 different factor extraction methods, suggesting that the factor extraction method may have little effect on the choice of the number of factors. Using maximum absolute rotated loadings to uniquely allocate FACES items to scales, the same item-scale allocation is generated for each of 10 different rotation schemes, suggesting that the rotation scheme may have little effect on scale development. These rotation-suggested summated scales distinctly outperform the recommended FACES scales. Both sets of scales can be improved by treating them as dependent and also by estimating their loadings.

Learning Objectives: At the conclusion of the session, the participant (learner) in this session will be able to

• Articulate how a k-fold likelihood cross-validation can be used to evaluate factor analysis models.
• Describe the results of applying such methods to data for adolescents with type 1 diabetes.

Keywords: Biostatistics, Outcomes Research

Related Web page: nursing.yale.edu/faculty/knafl/factoranalysis.html

Presenting author's disclosure statement:

I wish to disclose that I have NO financial interests or other relationship with the manufactures of commercial products, suppliers of commercial services or commercial supporters.

Recorded presentation