Structural equation modelling

Structural equation modelling, often abbreviated to SEM, is a multivariate analysis technique, which pools together a range of traditional multivariate analysis methods and are then estimated simultaneously.

Whilst structural equation modelling utilises a multitude of different techniques, the most prominent are as follows:

• Causal modelling or path analysis.
• Second order factor analysis.
• Covariance structure models.
• Correlation structure models.

What sets structural equation modelling apart from singular multivariate analysis methods is that it is largely used for hypothesis testing (i.e. a confirmatory technique) rather than for exploratory purposes. In practice this means that researchers are more likely to use structural equation modelling in order to determine whether a certain model is valid rather than to try and find a model which may be of relevance.

Often, the focus of structural equation modelling is of abstract psychological variable such as attitudes and knowledge.

Examples of latent variable, which may be investigated by structural equation modelling, includes family and peer dynamics, depression and other phenomenon.

Because such abstract variables are notoriously difficult and error prone, users of structural equation modelling look to find unbiased estimates for the relations that exist between such variables. The model allows multiple measures to be related with a single abstract variable in order to check for this.

Another advantage of using structural equation modelling is that instead of merely conducting straightforward significant tests, multiple tests are used to determine the differences that exist between groups or relationships between variables. The tests deployed in structural equation modelling include chi-square and comparative fit index. The triangulation of various significance checks means that any apparent relationships or differences have been subject to more rigorous checks.

An example of when structural equation modelling may be used could be in a research project relating to psychological health to determine which of a plethora of variables (such as education, income etc.) has the largest bearing.