Hello everyone,

I am trying to fit a multilevel SEM model in Stata 15.1, and it just won't converge. I have tried all the different integration methods suggested by the manual, as well as reducing the number of integration points, but it doesn't work. I have built the model step by step, so to use the previous as starting values. I was very slow to start, but in the last step I have added the multilevel variable, and now I keep getting the message "initial values not feasible r(1400); "everytime straight after "Fitting full model". It won't even start converging.

I have two independent variables: one categorical (PT, with 4 categories set as 1,2,3,4) and one dummy (V, 0-1).
All my three dependent variables (HH, EH, AvEffort) and my two mediators (CSE and ICM) are continuous.

The code is:
. gsem (HH -> H_1, family(gaussian) link(identity)) (HH -> H_4, family(gaussian) link(identity)) (HH -> H_6, family(gaussian) link(identity)) (HH -> H_10, family(gaussian) link(identity)) (EH -> H_5, family(gaussian) link(identity)) (EH -> H_7, family(gaussian) link(identity)) (EH -> H_9, family(gaussian) link(identity)) (CSEm -> HH, ) (CSEm -> EH, ) (CSEm -> CSE1, family(gaussian) link(identity)) (CSEm -> CSE6R, family(gaussian) link(identity)) (CSEm -> CSE15R, family(gaussian) link(identity)) (CSEm -> ICMq, family(gaussian) link(identity)) (SCF2 -> HH, ) (SCF2 -> EH, ) (SCF2 -> CSEm, ) (SCF2 -> avgEffort, family(gaussian) link(identity)) (SCF2 -> ICMq, family(gaussian) link(identity)) (1b.PT -> SCF2, family(gaussian) link(identity)) (2.PT -> SCF2, family(gaussian) link(identity)) (3.PT -> SCF2, family(gaussian) link(identity)) (4.PT -> SCF2, family(gaussian) link(identity)) (ICMq -> avgEffort, family(gaussian) link(identity)) (V -> SCF2, family(gaussian) link(identity)) (V -> avgEffort, family(gaussian) link(identity)) (M1[t] -> avgEffort, family(gaussian) link(identity)), covstruct(_lexogenous, diagonal) latent(HH EH CSEm M1 ) nocapslatent

I know it's a complicated model, but would anyone have any ideas?

THANK YOU in advance!