I am desperate need of help, and no source I’ve found has simple answer answers to these questions:
In a CFA, what are the important output measures you need to look at? What about a MIMIC model? And what about an SEM?
I know a bit about structural vibration and response to loadings (including seismic), but I’m not familiar with CFA or MIMIC (and although I can deduce what SEM means, I’ve never seen it in acronym form). If you type them out I may be able to point you to some different resources than the ones you’re looking at.
I think the OP is talking about this, which is not an engineering topic. I wish I could help, but I’m not familiar with the subject.
Thanks; just to be sure though, Structual Equation Modeling is a psychological modeling technique; it has nothing to do with engineering or earthquakes. CFA stands for Confirmatory Factor Analysis, MIMIC stands for Multiple Indicators and Multiple Causes.
My apologies. I’ve never heard of it, sorry.
Structural equation modeling has been around for well over 80 years (if you count the very early work done by Spearman and Thurstone). One of the hallmarks of it is that it incorporates so-called latent variables (variables like “intelligence”, “positive attitude,” “confidence”, “ease of use” etc.) with their associated “manifest” measures (things we can observe).
The typical latent variable model used in psychology today is called the “reflective model.” That means you, the researcher, posit that some manifest measures are caused by an underlying latent variable. For example, suppose you’re studying intelligence (the latent variable). You assume that “intelligence” causes observable scores on tests of mental ability (maybe a verbal and a quantitative reasoning test). The theory is that changes in the latent variable (intelligence) cause changes in the observed measures (the scores on the test). Exactly how you model that is a bit more complex.
SEM is done mostly with computers today because you need to do some heavy-duty computations on high-dimension matrices. One of the best known programs is called LISREL (written by a guy named Joreskog). It’s been around for nearly 30 years. Another big one is Amos.
SEM is closely related to factor analysis, except that SEM is called a confirmatory technique, in that the researcher (you) propose a model that implies a certain type of covariance matrix structure. The model can be tested statistically using a chi-square type test. The null hypothesis is the model you proposed is the true model and the alternative is that it isn’t. Researchers would “like” to not reject the null hypothesis because that implies the researcher’s theory is plausible (if the assumptions are right, that is, which is a whole other story that has spawned volumes of research).
If you want a good intro, I would recommend any book by Ken Bollen or Peter Bentler.
Obviously, there are still a lot of details to doing this, but I hope this gives you some insight.