Eric Teman posted on Tuesday, April 22, 2014 - 10:49 pm
I am doing a simple observed path analysis in Mplus. I understand that maximum likelihood produces biased variance estimates. For example, when computing the variances of the IV's, the computation is Sums of squares/N (rather than n-1). However, I do not understand why when computing the variance of the disturbance term it is SS/n-1 instead of SS/n. Where are we losing the 1 d.f.? Maximum likelihood is said to produce a biased estimate of the error variance as the sums of squares residuals divided by N (where as OLS divides by n-2).
Eric Teman posted on Wednesday, April 23, 2014 - 3:20 pm
Never mind. I was using the STDYX solution instead of the unstandardized solution. Sorry for the bother.