I have model that, theoretically, is a "2-2-2". It's done with aggregated data (n=228) for the X and M and with Y measured ate team level only. It works fine.
As per a reviewer's request, I have to run a MLSEM (Nwithin = 1480; Nbetweeb = 228). I am using the following code:
USEVARIABLES ARE grupo Perf PI1 PI2 PI4 TWE1 TWE2 TWE3 TWE4 TWE5 TWE6 TWE7 TWE8 TWE9; BETWEEN IS Perf; CLUSTER IS grupo;
ANALYSIS: TYPE IS TWOLEVEL RANDOM;
MODEL: %WITHIN% IP BY PI1 PI2 PI4; TWE BY TWE1 TWE2 TWE3 TWE4 TWE5 TWE6 TWE7 TWE8 TWE9; %BETWEEN% TIP BY PI1 PI2 PI4; TTWE BY TWE1 TWE2 TWE3 TWE4 TWE5 TWE6 TWE7 TWE8 TWE9; Perf ON TTWE(b); Perf ON TIP; TTWE ON TIP(a);
MODEL CONSTRAINT: NEW(ab); ab = a*b;
OUTPUT: TECH1 TECh8 CINTERVAL;
I get the following error message:
THE H1 MODEL ESTIMATION DID NOT CONVERGE. CHI-SQUARE TEST AND SAMPLE STATISTICS COULD NOT BE COMPUTED. INCREASE THE NUMBER OF H1ITERATIONS.
I have tried to increase the number of iterations but the error remains.
Also, the estimates are VERY different from my original ("2-2-2") model. One path (TWE ON PERF), which is positive and significant originally, becomes negative and nonsignificant now...
Can you help me understand why? Thank you in advance.
Your setup looks ok but I think you should first sort out why you get that message - perhaps you have very low coverage (a lot of missing data). If you like, send your output to Support along with your license number.
You don't get a model test of fit with this error message. Estimates are trustworthy only if you don't have a lot of missing data. If you have data missing by design, there are solutions. We need to see the output to say more - you can send to Support along with your license number.