Message/Author 

Nick posted on Wednesday, April 27, 2016  10:33 am



This is a 2 mediator serial multilevel model with diary data in long format. I keep getting warnings of "nonpositive definite matrix" & "Model Est Reached Saddle Point". Does this syntax look correct? We control for lagged variables due to autocorrelation in diary data. %WITHIN% Y X M1 M2 Y_lag M1_lag M2_lag; M1 ON X (aw); M2 ON M1 (bw); M2 ON X; Y ON M2 (cw); Y ON M1; Y ON X; Y M2 M1 ON Y_lag M2_lag M1_lag; X WITH Y_lag M2_lag M1_lag; %BETWEEN% Y X M1 M2 Y_lag M2_lag M1_lag; M1_lag ON X(ab); M2 ON M1 (bb); M2 ON X; Y ON M2 (cb); Y ON M1; Y ON X; Y M2 M1 ON Y_lag M2_lag M1_lag; X WITH Y_lag M2_lag M1_lag; [Y X M1 M2 Y_lag M1_lag M2_lag]; 

Nick posted on Wednesday, April 27, 2016  10:40 am



I understand double posting is frowned upon, but it may be helpful to add that this same model ran fine without any warnings before the inclusion of the lagged control variables. For these, each variables is lagged one day. Thanks in advance. 


Please send the output that has a problem and your license number to support@statmodel.com. 


I am trying to run a 111 multilevel mediation using weekly survey data in which observations are nested within individuals. The model includes a level 1 control variable. When I run the below model, there are no error messages. However, when looking at the output, the estimates (including S.E.s, pvalues, etc.) are exactly the same for in the x and con variables within in the first regression statement: m ON x con. This seems odd. It is my understanding that control variables should be included on all ON regression statements so I am unsure how to correct this. Thus, A) Is there a way to control for con in the m ON x statement without duplicate estimates in the output happening? B) Or if I take the control out of this equation and only include it in the following statements: y ON m con; y ON x con;  Would this still control for con in the model? It is theoretically important for con to be controlled for throughout the model. CLUSTER IS id; ANALYSIS: TYPE IS TWOLEVEL RANDOM; MODEL: %WITHIN% m ON x con(aw); y ON m con(bw); y ON x con; %BETWEEN% x m y con; m ON x con(ab); y ON m con(bb); y ON x con; MODEL CONSTRAINT: NEW(indb indw); indw=aw*bw; indb=ab*bb; OUTPUT: TECH1 TECH8 CINTERVAL; 


Please send the output and your license number to support@statmodel.com. 

Back to top 