I am using R to run a latent variable SEM model and my indicators are all on a 7-point scale. From what I gather, in the instance of categorical variables with 5+ categories it is fine to use ML estimation with a large enough sample. My variables are non-normal and I'm wondering if using MLR estimation or ML with bootstrapping would be better? An advantage of MLR would be that it provides robust fit indices as well as standard errors, while ML with bootstrapping will only correct standard errors. I've also read something about the CIs obtained with the bootstrapping method being more accurate, but I'm really not sure and at least from what I can find not much compares the methods.