Hi all,

I am attempting to run a cross-lagged panel model. I have two continuous variables (depression & body image dissatisfaction) at two time points. I also want to add a cluster variable for school. I am able to successfully run the basic model using the following code:

gsem (DS_1 -> ds_t1, ) (DS_1 -> DS_2, ) (DS_1 -> ED_2, ) (DS_2 -> ds_t3, ) ///
(ED_2 -> diss_t3, ) (ED_1 -> diss_t1, ) (ED_1 -> DS_2, ) (ED_1 -> ED_2, ), ///
covstruct(_lexogenous, diagonal) latent(DS_1 DS_2 ED_2 ED_1 ) reliab(ds_t3 .8 ds_t1 .8 diss_t3 .8 diss_t1 .8 ) nocapslatent


However, when I attempt to add the cluster variable for school using the following code:

gsem (DS_2 -> ds_t3, ) (DS_1 -> ds_t1, ) (DS_1 -> DS_2, ) (DS_1 -> ED_2, ) ///
(ED_2 -> diss_t3, ) (ED_1 -> diss_t1, ) (ED_1 -> DS_2, ) (ED_1 -> ED_2, ) (M1[school_id] -> ds_t3, ),
> covstruct(_lexogenous, diagonal) latent(DS_2 DS_1 ED_2 ED_1 M1 ) ///
reliab(ds_t3 .8 ds_t1 .8 diss_t3 .8 diss_t1 .8 ) nocapslatent from (b)


I get the following error message

Iteration 0: log likelihood = -5990.2821 (not concave)
cannot compute an improvement -- discontinuous region encountered
r(430);


Is anybody any to advise on how I may address this error?

Thank you!