Advances the Study of Measurement Invariance: Bayesian Estimation of Confirmatory Factor Analysis Modeling

Schedule:
Saturday, January 17, 2015: 8:00 AM
Preservation Hall Studio 10, Second Floor (New Orleans Marriott)
* noted as presenting author
G. Lawrence Farmer, PhD, Associate Professor, Fordham University, New York, NY
BACKGROUND:  Multigroup Confirmatory Factor Analysis represents an important analytical tool for social work researchers seeking to develop and examine evidence of the cross-group validity of our measurement tools.  One issue that has motived advances in the study of measurement invariance is the need to reduce the labor intensiveness  of the process, especially when more than two groups are involved (Asparouhov & Muthen, 2013; Muthén & Asparouhov, 2013).  Bayesian CFA modeling (Dutch Mplus Users Group, 2012; Muthén & Asparouhov, 2013) is an examples of an advance in the use of a general latent variable modeling framework  (Muthén, 2002) to the study of measurement invariance that has been developed to address both the labor intensiveness  and  expansion of the  number of groups that have presented problems for standard multiple group confirmatory factor analysis.

PURPOSE:  Using Mplus 7.11 this study will introduce participants to the estimation of CFA models using a Bayesian framework.  Data from the wave I of the ADD Health Study will be used to evaluate the measurement invariance of multidimensional measure of well-being across African American, Asian American, Hispanic American and European American Adolescent males and female study participants.  The measurement of multidimensional well-being was ground in a dual factor model of mental health.  The dual-factor model stresses the need to conceptualize well-being as a balance among  positive affect, negative affect and life-satisfaction  (Lyons, Huebner, Hills, & Shinkareva, 2012).  The study of a variety of positive youth outcomes has been found to be strongly associated with youth’s well-being (Lynch, Lerner, & Leventhal, 2013), therefore the variable has become an important outcome variable evaluation the outcomes of youth development programs.                              

STUDY OBJECTIVES:  To evaluate the measurement invariance of the multidimensional measure of well-being across eight groups of adolescences. To demonstrate how Bayesian multiple group confirmatory factor analysis provides a more efficient approach to evaluation measurement invariance than traditional multiple group CFA in situation with many groups.  

FINDINGS:   Across the eight groups partial metric and scalar invariance was found.   Both metric and scalar invariance was found primary across the gender groups. Those items focused on the affective dimensions of well-being demonstrated consistent invariance between males and females, but between ethnic groups. 

IMPLICATIONS:  Bayesian multiple group confirmatory factor analysis provides an efficient alternative to traditional CFA in those situations when measurement invariance among many groups is the focus of the research.  The seeking to measure well-being across gender groups need to be aware of the potential for measurement invariance to be present.