Assumptions of SEM

With any statistical technique, assumptions are made. Here are a few of the assumptions with SEM that you need to be aware of going forward:

1. Multivariate Normal Distribution of the Indicators—there is an assumption that the data has a normal distribution.
2. Dependent variables need to be continuous in SEM—while the independent variables do not have this assumption, dependent variables need to be continuous.
3. SEM assumes linear relationships between  variables.
4. Maximum likelihood estimation is the default method—maximum likelihood estimation is a technique known to provide accurate and stable results (Hair et 2009).You can use other estimations, but maximum likelihood is the default unless otherwise specified.
5. SEM assumes a complete data set—more to come on this topic later if you have missing data.
6. Multicollinearity is not present—multicollinearity makes it difficult to determine the influence of a concept if it is highly correlated with another variable in the model.

7. Adequate sample size—this is one of the challenging assumptions with SEM because this technique does require a large sample size compared to other techniques.
8. Unidimensionality of a construct—the idea that you are solely capturing a construct of interest.

Source: Thakkar, J.J. (2020). “Procedural Steps in Structural Equation Modelling”. In: Structural Equation Modelling. Studies in Systems, Decision and Control, vol 285. Springer, Singapore.