Anonymous posted on Tuesday, September 07, 2004 - 10:12 pm
(1) I fit a 1-factor model to a number of categorical items, say F BY Q1-Q20. (2) I then regress the factor on a covariate, F ON COVAR. (3) I now regress on COVAR after centering it (COVAR - mean). In all three cases the loadings remain quite similar but the thresholds with the uncentered COVAR are shifted considerably by a constant. Is this what one would expect and if so how should we understand (or what should we read to understand) the effect of COVAR scaling.
bmuthen posted on Wednesday, September 29, 2004 - 4:48 pm
This outcome is to be expected since the scales of the thresholds for the indicators depend on the means of the variables predicting the indicators - in this case the factor. In case (1) and (3) the factor has mean zero, while in case (2) the factor mean is a function of the covariate.
I ran a CFA model with censored indicators (3 below and 2 above). However, in the standardized (STDYX) section of the output, there are only coefficients, but not S.E. or p values. Why is that? And How can I get S.E. and p values for the standardized coefficients (I need to report them in my manuscript)?