The Society for Social Work and Research

2014 Annual Conference

January 15-19, 2014 I Grand Hyatt San Antonio I San Antonio, TX

Structural Equation Modeling's Role in Social Work Research

Friday, January 17, 2014: 10:00 AM
HBG Convention Center, Room 003B River Level (San Antonio, TX)
* noted as presenting author
Natasha Bowen, PhD, Associate Professor, University of North Carolina at Chapel Hill, Chapel Hill, NC

Background and Significance

Rigorous quantitative methods, combined appropriately with theory and design, are necessary for the building of a science of social work. Unless social work researchers master, employ, and participate in ongoing advances in quantitative methods, their work will continue be underrepresented in influential publications outside of social work. Structural equation modeling (SEM) is one important example of a sophisticated analysis method that is recognized by researchers in multiple disciplines as superior to conventional analysis procedures, such as, regression and exploratory factor analysis. SEM offers numerous advantages to the study of social work topics and interventions.


The SEM framework provides superior modeling options for social work research involving scales (multiple items used to measure a complex construct), mediation effects, moderated effects with categorical variables, and longitudinal or repeated measures. Confirmatory factor analysis (CFA) creates latent variables that represent only shared variance among multiple indicators of a construct of interest. Sophisticated tests of partial or full invariance of measurement across groups and time can also be conducted to build evidence of the validity and reliability of scores from measures. The error-free latent variables obtained with CFA can be used to test hypotheses involving multiple simultaneous equations, that is, models with one or more mediated effects, more than one dependent variable, and reciprocal effects. Repeated measures data can be used to test mediation hypotheses with appropriate controls for scores of mediators and dependent variables at prior time points. Repeated measures and longitudinal data can also be modeled as linear and non-linear trajectories defined by latent intercept and slope terms. Recent developments in SEM allow researchers to simultaneously accommodate clustered, ordinal, non-normal, and weighted data, and data with missing values. Person-centered mixture modeling offers a new analytic tool for data reduction that helps practitioners understand unique characteristics of different clusters of clients. As these procedures become commonplace in other fields, the ability of social work researchers to have their studies published and respected outside of social work will become increasingly dependent on their willingness to master the best quantitative methods. 



The best social work research and practice has strong theoretical foundations. Consistent with the role of theory in social work, the SEM literature emphasizes the importance of testing only models that are theoretically justifiable. Using SEM not only reminds social work researchers of the critical role of theory in science, but provides systematic procedures for building theory.



Effective social work practice involves understanding complex, multilevel developmental and intervention processes. The appropriate modeling of such processes requires the use of advanced and emerging statistical procedures. For social work research to remain relevant to social work practice, and to become more visible and respected among the social sciences, we must have renewed commitment to the study and application of the best statistical modeling techniques. SEM is a flexible and powerful analysis approach that is appropriate for a majority of the research questions and datasets available to social work researchers.