MerQuR+Statistics Glossary

Statistics Glossary → Measurement model and structural model

Measurement model and structural modelconcept

In structural equation modelling, the part that defines how latent variables are measured by observed indicators (the measurement model) and the part that defines the directed relationships among latent variables (the structural model).

The measurement model states which items measure each latent construct (e.g. job satisfaction) and how strongly they load on it. In lavaan-style syntax it is written as 'F =~ x1 + x2 + x3', and on its own it is a confirmatory factor analysis. The structural model contains the regression relationships among the latent variables ('Y ~ X1 + X2'). Path analysis is the special case with no latent variables, consisting only of a structural model. Because structural relationships are estimated between latent variables freed of measurement error, they are less attenuated than regressions on observed total scores.

The two-step approach of Anderson and Gerbing (1988) recommends first testing the fit and validity of the measurement model, and only then adding the structural paths. Structural results built on a poorly fitting measurement model cannot be interpreted. In covariance-based SEM the indicators are usually reflective (the construct causes the items), while PLS-SEM also allows formative indicators (the items form the construct). Changing the model on the basis of modification indices is data-driven and should be confirmed in a new sample.

An everyday example

A researcher thinks that workload increases burnout and that burnout increases the intention to quit. In the measurement model each of the three constructs is defined by four questionnaire items, and the analysis checks whether the items measure their constructs well. The structural model then estimates the paths 'burnout ~ workload' and 'intention to quit ~ burnout'.

In MerQur

In confirmatory factor analysis, 'Number of factors' defines the measurement model, 'Structural paths (SEM)' the directed relationships between latent factors and 'Additional model syntax' residual covariances and cross-loadings. 'Model syntax' (=~, ~, ~~) in structural equation modelling and 'Model syntax' in PLS-SEM (reflective =~, formative <~, structural ~) build the measurement and structural parts together. In path analysis, by contrast, 'Template' and 'Model text' (~ path, ~~ covariance) define directed relationships among observed variables only.

Analyses where this concept appears

Confirmatory Factor Analysis (CFA)Structural Equation Modeling (SEM)Path AnalysisPLS-SEM (Partial Least Squares Path Modeling)

Related concepts

Model fit indices (χ², CFI, TLI, RMSEA, SRMR)Model identification (std.lv, unit loading)Measurement invariance (configural, metric, scalar, strict)