gibbon lab posted on Thursday, January 05, 2012 - 8:55 am
I am trying to carry out chi-square difference tests in a two group analysis, where categorical endogenous variables are involved. For one of the path, after the equality constraint between the two groups is imposed, Haywood case happens. That is, one of the endogenous variables has a negative residual variance. In this case, can I fix the variance of that variable to 1 or some other number? Thanks a lot.
If the test says that the two coefficients are equal, you could fix the negative residual variance to zero if it is small and not significant. This finding would, however, make be have reservations about my model.