Running CFA binary variables return low factorloading

[Do Huy Phuc]

Good morning, I'm running a SEM using estimator DWLS in R. I'm wondering when I tested for validity, their factor loading are very low (0.1 - 0.3), although all of my goodness of fits (CFI, TLI, RMSEA,...) are very good Can you help to explain and how can I fix this?

[Clyde Schechter]

Fix what? Nothing you say suggests there is any problem.

CFI, TLI, and RMSEA are measures of fit. They quantify how well the model fit the data (where fit is, in this instance, measured by the match between the modeled and observed covariance matrices.) If your fit measures are good and the loadings of some of your indicator variables are small, that just means that, in fact, there isn't a particularly strong association between those indicators and the latent factor. Remember also that, by default, in CFA, everything is scaled to the 1st indicator in the factor. So a loading of 0.15 says very little, if anything, about the absolute association between an indicator and the latent factor: it says that the strength of that association is 15% as large as that of the first indicator.

By the way, this is a Stata forum, not an R forum. And while statistical questions, which yours is, are always welcome here, if you are looking for help with using R, this would not be the best place to find it.