Hi Gang,
I noticed that whenever a hybrid theta equation matrix (0,1,and 2s) is specified for 1dSEM using tree or forest growth approaches that Bollen's parsimonious index (BFI) is always negatively valued and becomes increasing negatively valued with successive iterations.
It seems to me that this is a bug in 1dSEM since it calculates BFI using the chi square for the nuclear model (where 1s are specified) as the the null model chi square. If you recalculate BFI using the true null model, in other words with all parameters set to 0, BFI inverts its sign, achieves reasonable values, and asymptotes over successive iterations. Also the BFI for the first iteration beyond the nuclear model is relatively high .3-.5 rather than starting at zero. This is similar to the BFI plot shown in Stein et al. 2007 where they appear to be calculating BFI using the true null model at not the nuclear model
When a fully exploratory theta equation matrix of all 2s is specified BFI is correctly calculated.
Is this a bug or am I missing something here?
Vinny