Dear STATA users,
I would like to use GLLAMM to fit a mixed response models with a multinomial
logit component and a logistic component. The multinomial component has four
categories, with dummies a2, a3, a4. For each patient, I have several
measures of the multinomial response at different times and one measure of
the binomial response.
The first patient chose 217 times the first answer, 256 times the second
answer, 39 times the third answer and 29 times the fourth answer. The patient
is ill (rep=1 when var==2).
Here are the grouped data of the first patient:
+----------------------------------------------------------------------+
| ident var rep d2 a2_d1 a3_d1 a4_d1 patt chosen wt1 |
|----------------------------------------------------------------------|
1. | 1 1 1 0 0 0 0 1 1 217 |
2. | 1 1 2 0 1 0 0 1 0 217 |
3. | 1 1 3 0 0 1 0 1 0 217 |
4. | 1 1 4 0 0 0 1 1 0 217 |
5. | 1 1 1 0 0 0 0 2 0 256 |
|----------------------------------------------------------------------|
6. | 1 1 2 0 1 0 0 2 1 256 |
7. | 1 1 3 0 0 1 0 2 0 256 |
8. | 1 1 4 0 0 0 1 2 0 256 |
9. | 1 1 1 0 0 0 0 3 0 39 |
10. | 1 1 2 0 1 0 0 3 0 39 |
|----------------------------------------------------------------------|
11. | 1 1 3 0 0 1 0 3 1 39 |
12. | 1 1 4 0 0 0 1 3 0 39 |
13. | 1 1 1 0 0 0 0 4 0 29 |
14. | 1 1 2 0 1 0 0 4 0 29 |
15. | 1 1 3 0 0 1 0 4 0 29 |
|----------------------------------------------------------------------|
16. | 1 1 4 0 0 0 1 4 1 29 |
17. | 1 2 1 1 . . . . . 1 |
- ident is the patient identifiant
- var is the response type (var=1 for the multinomial response, and var=2 for
the binomial response)
- rep is the response, ie. equals to the multinomial response (1, 2, 3, or 4)
if var==1 or equals to the binomial response (0 or 1) if var==2
- d2 is a dummy for binomial response
- d1 is a dummy for the multinomial response
- a2_d1 = a2*d1
- a3_d1 = a3*d1
- a4_d1= a4*d1
- wt1 is a weight
I have tried this GLLAMM syntax:
eq fac: a2_d1 a3_d1 a4_d1 d2
gllamm rep a2_d1 a3_d1 a4_d1 d2, expand(patt chosen o) noconst i(ident)
lv(var) fv(var) link(mlogit logit) family(binom binom) weight(wt) eqs(fac)
trace
There is an error message "could not calculate numerical derivatives flat or
discontinuous region encountered (error occurred in ML computation)"
Is it possible to use a mixed response model including a multinomial response
model ? Does the problem come from the expanded() options ? Do you know how I
can fix it ?
Thank you for your answer,
Best regards,
Maud Tournoud.
---------------
Maud Tournoud, PhD
Laboratoire Biostatistique-Sant�
Lyon, France
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
