The Collections
  
   

Meta-Analytic Structural Equation Modeling (MASEM):

This collection brings together and organizes methodological work relevant to meta-analytic structural equation modeling (MASEM). Loosely speaking, MASEM entails using data from multiple samples of subjects to analyze features of a structural equation model (SEM) for subject-level variables. Popular SEMs include regression and path models for only observed/manifest variables, various factor models (e.g., exploratory, confirmatory, hierarchical, higher-order) and close relatives (e.g., principal components), and path models involving latent variables. Research syntheses that involve MASEM share many characteristics with conventional reviews that include meta-analysis, such as distinct but interdependent stages of responsible reviewing as well as statistical tasks and issues (e.g., extracting data, combining and comparing estimates, assessing or modeling heterogeneity). Substantive applications of MASEM, however, present unique methodological opportunities and challenges, many of which stem from the potential richness and complexity of SEMs and relevant data compared to their simpler counterparts in routine meta-analysis applications.

Curator:

Dr. Adam Hafdahl earned his Ph.D. in Quantitative Psychology from UNC-Chapel Hill and a M.A. in Statistics from Washington University. He joined the quantitative faculty in Psychological Sciences at the University of Missouri-Columbia and later pursued an independent consulting career. He has authored and co-authored several methodological articles in peer-reviewed journals mainly on the topic of meta-analysis. He has served as a statistical consultant on NIH- and NSF-funded projects, pursues his own research interest and enjoys collaborating on various professional activities (e.g., refereeing journal articles and Campbell Collaboration systematic reviews).

Eligibility Criteria

The above methodological concepts, procedures, and issues, as well as others related to MASEM, have been addressed in numerous journal articles, dissertations, book chapters, conference papers, and other types of scholarly contributions. This collection on MASEM methodology is meant to be a fairly comprehensive compilation of that literature. Our only explicit criteria for including a work was that a non-trivial portion of it addresses matters relevant to MASEM methodology. We have not deliberately excluded work based on publication language or dissemination date, but we recognize that those factors may have influenced our ability to find eligible work and assess its relevance. Also, we have not deliberately excluded work based on an assessment of its perceived value or utility (e.g., quality, importance), though such factors were considered in organizing and presenting records.

Special Core Topics

Within the methodological literature on MASEM, several narrower topics have received notable attention. Here we identify six of these used to help organize this collection’s records.

  • Correlation Matrices: Many approaches to MASEM begin by comparing or combining correlation matrices across studies; several authors have developed or evaluated meta-analytic procedures for those preliminary tasks.
  • Two-Stage SEM: One approach to MASEM, often called two-stage SEM (TSSEM), entails using SEM to first combine correlation or covariance matrices and then analyze the focal SEM using the first stage’s results.
  • Functions of Effect Sizes: Another approach to MASEM uses meta-analytic techniques for functions of effect sizes, such as when the focal SEM’s parameters can be expressed as a function of a correlation matrix.
  • Missing Data: Substantive applications of MASEM often involve missing data, such as when correlation matrices are combined across studies but at least one study reports only a subset of its matrix’s correlations.
  • Exploratory Factor Analysis: Distinct techniques have been proposed or employed when the focal SEM involves exploratory factor analysis, based on an unrestricted factor model; some of these techniques entail comparing factor-analytic results among studies, while others entail combining correlation or covariance matrices across studies before factoring.
  • Multiple Regression: Special procedures may be appropriate when the focal SEM involves multiple regression, with an observed outcome variable regressed on two or more observed predictors (e.g., X → Y ← Z for predictors X and Z); a typical challenge in such MASEM applications is that different studies’ partial coefficients (e.g., slopes) are based on different sets of predictors or covariates.
 

 

Dr. Adam Hafdahl