Message/Author |
|
|
I am running a Complex Model using MLR simultaneously examining 2 independent variables and 2 dependent variables. The analysis seems to work properly for most of the tests, but for a few of tests I get the following message directly following, "THE MODEL ESTIMATION TERMINATED NORMALLY" in the output: "THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS (one of these numbers: 0.804D-17 or 0.506D-14 or 0.119D-14). PROBLEM INVOLVING PARAMETER 6." I noticed that when this message appears, I get the following results that seem suspect: Estimate = 0.000 S.E. = 0.000 Est./S.E. = -0.663 Two-tailed p-value = .507. A separate test with different variables follows a similar pattern: Estimate = 0.000 S.E. = 0.000 Est./S.E. = -0.549 Two-tailed p-value = 0.583 But another test with different variables has a significant p-value: Estimate = 0.000 S.E. = 0.000 Est./S.E. = -4.937 Two-tailed p-value = 0.000 All other parameters in each model seem to estimate correctly. Any idea as to why this is happening or any possible solutions? Let me know if you need any other information (e.g., syntax, dataset, etc.). --Matt |
|
|
It seems like the variables involved must be measured on a very small scale. You can use the DEFINE command to multiply them by a constant, for example, DEFINE: y = y*5; |
|
|
I used the DEFINE command to multiply by a constant, as you suggested. The output produced the same error message, along with the odd 0.000 standard error and estimate: THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES MAY NOT BE TRUSTWORTHY FOR SOME PARAMETERS DUE TO A NON-POSITIVE DEFINITE FIRST-ORDER DERIVATIVE PRODUCT MATRIX. THIS MAY BE DUE TO THE STARTING VALUES BUT MAY ALSO BE AN INDICATION OF MODEL NONIDENTIFICATION. THE CONDITION NUMBER IS 0.324D-11. PROBLEM INVOLVING PARAMETER 5. Other ideas? |
|
|
Please send the output and your license number to support@statmodel.com. |
|
|
I ran a choice model for men and women, LL of free and constraint (one beta is forced to be equal) : free -9716.29 ,61 DF constraint -9719.75 ,56 DF. DF difference should be 1 why is it 5? used INTEGRATION=GAUSSHERMITE MCONVERGENCE=0.01 miterations=1000. free model: MODEL: %OVERALL% soc by socc1_2 socc1_3 socc1_4 ; prod by prec1_1 quac1_1 ; hed by hedc1_1 hedc1_2 hedc1_3 ; soc with prod; prod with hed; hed with soc; choicc#1 ON prod soc hed; choicc#2 ON prod soc hed; [choicc#1 choicc#2 ]; %c#1% soc by socc1_2 socc1_3 socc1_4 ; prod by prec1_1 quac1_1 ; hed by hedc1_1 hedc1_2 hedc1_3 ; soc with prod; prod with hed; hed with soc; choicc#1 ON prod soc hed; choicc#2 ON prod soc hed; [choicc#1 choicc#2 ]; %c#2% soc by socc1_2 socc1_3 socc1_4 ; prod by prec1_1 quac1_1 ; hed by hedc1_1 hedc1_2 hedc1_3 ; soc with prod; prod with hed; hed with soc; choicc#1 ON prod soc hed ; choicc#2 ON prod soc hed; [choicc#1 choicc#2 ]; OUTPUT: standardized residual tech1 tech8; constraint model is similar but I add (1) to HED in %c#1% and %c#2% : choicc#1 ON prod soc hed (1); choicc#2 ON prod soc hed; [choicc#1 choicc#2 ]; |
|
|
Compare TECH1 for both analyses to see what the 5 degree of freedom difference is. |
|
|
thank you! I looked at tech 1, the free model is ok, but I saw in the constraint model I have constraint all 3 variables (prod, soc and hed) all have the same number..and in the stdyx results I see it fixed all three..How can I specify the command and check only one variable? thought choicc#1 ON prod soc hed (1) would fix hed only... appreciate the help! |
|
|
choicc#1 ON prod soc hed (1); |
|
|
That's exactly what I did and it constrainted all three variables (prod, soc and hed). I see it in the estimates and in the DF in tech 1...(why is that?) I really need to check only 1 variable each time but I dont know how to do so. |
|
|
should I write the command in the overall and c#1 and c#2 models? I wrote it only in the classes commands.Thanks again |
|
|
Please send the output and your license number to support@statmodel.com. If you specify choicc#1 ON prod soc hed (1); rather than choicc#1 ON prod soc hed (1); you should get what you want. |
|
|
it worked) thank you very much! |
|
Back to top |