I have fitted a structural equation model with 28 parameters (500 samples, 12 observed variables). there is a discrepancy between the output of MPlus to that of my own code for estimating parameter CIs.
to estimate CI for a parameter, that value of the parameter that increased the model chi-square value by 3.84 units (compared to value at original parameter estimate) was taken as the 95% confidence interval. model chi-square value was estimated as (N-1)*val, where N was sample size and val was value of the fit function (ML).
would be grateful if you could throw light on possible reasons for discrepancy in output between MPlus and my code.
Anonymous posted on Friday, October 14, 2016 - 1:22 pm
I am curious if there is a way to constrain the lower bound for confidence limits to be >= zero. I am aware of how to constrain specific parameters to zero (using MODEL CONSTRAINT), but wish to constrain the lower CI bound so that it does not take on a negative value. This has interpretive utility, as the parameter is a variance proportion, which cannot be less than zero.