While reading a bit of Yihui’s blog I found something that I’ve been hoping to hear from another statistician for a long time …

Typically I ignore any questions on Structural Equation Modeling (SEM) or factor analysis, since I’m not convinced of their usefulness at all. I know little about time series and do not like econometrics. I have little interest in quantitative research in social sciences.

SEM and factor analysis lead to very few helpful inferences. On top of this, I find the estimation to be very fragile, the computation to be prohibitive, power analysis to be damn near impossible, and the path diagrams to be completely uninterpretable.

And that is my assessment as a trained statistician that’s even taken a class on SEMs and factor analyses.

I *do* have interest in applying quantitative research to social sciences, and it is something I’m
doing increasingly at work. But I’ve yet to encounter a social science problem that seems to
*require* a SEM.

And anyway, a SEM is just a multi-level model written incorrectly

Possibly of note, there were ZERO sessions on SEM at JSM this year.