A Systematical Review of the Use of Structural Equation Modeling in Social Work Journals

Background and Purpose: The use of structural equation modeling (SEM) in social work research has increased over the past few decades. An earlier systematic review by Guo, Perron, and Gillespie (2009) covering top-ranked social work journals between January 2001 and January 2007 found that only 32 studies used SEM methods. SEM is a powerful multivariate technique integrating measurement approaches and structural approaches and has been applied to various areas such as psychology, education, health sciences, and management over the last 30 years. It has advantages over other techniques since the term covers a family of related procedures including regression, path analysis, general linear modeling, exploratory factor analysis, and confirmatory factor analysis (CFA). Apart from being useful in testing theories, SEM has the flexibility to examine relationships between latent and observed variables across groups and occasions. This study builds on the earlier one by Guo and colleagues and its purpose was to assess the use of SEM in social work journals between February 2007 to date.  

Methods: The study targeted the top 39 journals related to the field of social work provided by the 2010 Journal Citation Reports Social Science Edition. A critical inclusion criteria for the articles in the study meant that one of the authors had to have either a graduate of PhD degree in social work or be affiliated by a school or department of social work. We used SPSS 17.0 for all coded items to reflect how SEM is applied in social work research; what journals use SEM; as well as establish if the articles have improved in quality since Guo and colleagues recommendations for best practices in SEM reporting.

Results: We found that the use of SEM has indeed increased over the past few years, with 60 studies using SEM in 16 journals between February 2007 to March 2012. Descriptive statistics also showed that 45% of the articles used SEM with both measurement models and structural models while 36 % of the articles used CFA. Most articles used MPlus (22%), AMOS (14%), and LISREL (14%) for analyses. SEM. Maximum likelihood was the most frequently used (19%) method of estimation while about16% of the articles did not report their method of estimation. Only 29% of the articles modified their initial models for any reason and all articles reported model fits using multiple fit indices.     

Conclusion and Implications: Both the quantity and quality of publications using SEM continues to increase in social work journals. We expect that this will continue as more social work doctoral programs offer the courses or as more students take SEM electives from other departments. However, studies did not report important information related to SEM application as suggested by Guo and colleagues such as colinearity, univariate and multivariate normality, the process of modification, competing models, and missing data analysis process. Further implications on SEM education in social work and application will be discussed.