I have been researching this topic for some time now and was after some opinions or comments.

Are there any situations where a path analysis (a form of SEM) might be more appropriate than a full SEM? Specifically, if measures used are standardised and often very long making a measurement model extremely complex and subsequently the increase in the parameters to be estimated places a greater demand on sample size, I wondered whether it would be justifiable to use path analysis as a compromise? The focus is on the relationships between the 'latent variables' Which I understand are treated as observed in a path analysis.

Measurement models account for the fact that measurement will never be perfect but many tests performed in SPSS, correlational and tests of difference, I assume are on this basis I.e. that there is no error in measurement?

I have mixed feelings on this and the literature is equally mixed. Many articles say they are using SEM but do not discuss the specification of measurement models and makes me wonder whether they are using path analysis instead. A recent course I have been on suggests that this is acceptable but again I appreciate there are limitations and wanted to reach out to others that have used these techniques.

I hope this makes sense. I suppose I want to make sure that I have considered all avenues and feel confident in my decisions.


1 Answer 1


1) There are surely many questions for which a path analysis is sufficient.

2) Increasing complexity or increasing sample size required (etc.) are not by themselves valid reasons to avoid inclusion of latent variables.

3) Latent variables are not "treated as observed" in a path analysis - it's more accurate to say that a path analysis is an SEM without any latent variables included (somewhat semantic, but I think it's an important distinction)

4) Many models do in fact assume no measurement error, linear regression being the most abused culprit.

5) Many people probably do say they are using SEM when they are in fact using path analysis. This isn't that egregious, because path analysis can certainly be viewed as a type of SEM "sub-model" (for lack of a better word).

Your choice to include latent variables in the model should really be driven by the theory you are testing and data you have available. If you choose to use latent variables, you should certainly do some analyses of your respective measurement models before moving onto the full SEM where these are related to other variables. Without more detail on your specific problem, it's hard to give more detailed advice.

  • $\begingroup$ 2) This is debatable. 3) How is this more accurate, and the OP's phrasing inaccurate? 5) How is it egregious at all? Path analysis is SEM of variables for which data points are supplied, not estimated. Measurement models such as common factor models are also fit with "full SEM", which subsumes factor analysis. $\endgroup$ Apr 9, 2014 at 14:59

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.