Analyses

Structural equation modelling

Fit a path model that can include latent factors.

Edit on GitHub

When to use it

Use structural equation modelling when the claim is a set of paths, not only a measurement model. Factors are defined with =~. Regressions between factors or observed variables use ~. A measurement-only model is confirmatory factor analysis.

Assumptions

Observed variables in the spec must be numeric columns. Rows with a missing value on those columns are dropped. Each factor needs enough indicators to be identified, and the first loading on a factor is fixed at 1. A model that is not identified, or that names a column that is not in the data, fails with an error rather than a blank table.

Running it in Tensr

Multivariate → SEM. In chat: “SEM with F2 predicted by F1.”
Write the model, up to 8000 characters.
Optionally list the columns. If you leave them out, Tensr reads observed names out of the spec.

Options

Prop

Type

Reading the output

Focus measured by item1 to item3, then exam score predicted from Focus and study hours, on 96 people.

ToOpFromEstimateStd. Errp-value
item1~Focus1——
item2~Focus1.386840.476729p = .004
item3~Focus1.403770.496276p = .005
score~Focus-0.3093371.74098p = .859
score~hours2.239960.655468p < .001

Structural equation model with semopy fit indices. Structural equation model with semopy fit indices. Metrics: CFI = 1.036; RMSEA = 0; SRMR = 0.077.

Reporting (APA 7)

Structural equation model with semopy fit indices. Report the estimate in the table. This procedure is not summarised by one p-value.

Coming from SPSS

The product menu is Multivariate → SEM. SPSS fits path models in Amos. =~ defines a factor and ~ defines a regression, as in lavaan.