I am just wondering how can we make our findings statistically more robust. Structural Equation Modeling (SEM) seems to overcome some of the limitations imposed by basic statistics but it does have certain shortcomings. So I am trying to explore what else we can do statistically to make findings more credible.
Not much, II think. I suspect that dealing with a closed system in which almost everything is endogenous one has to accept that there are at least 2^N possible models, where N is the number of variables in the system. Your model may be fine but it possoble that there are better ones.
You should be careful using the word "robust" because it has a technical meaning in this context (i.e., how likely an estimation technique is to produce "correct" results despite violations of the assumptions).
As far as the credibility of SEM goes, the social science fields in which I work find it highly credible, and as you say, an improvement over more basic models. What is the source of your concern in this regard?
Hello,
Actually SEM offers quite a lot. You may build one model and execute it in one go, check if paths are sig. or not, include latent concepts along with observed items... and finally check if the model fits the data.
I think it is something to be appreciated. Of course you are very welcome to present here some specific issues and doubts.
Regards, Witold
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling