Programming the statistical procedures from SAS

proc probit for multinomial probit

Reply
Occasional Contributor
Posts: 6

proc probit for multinomial probit

Hello,

I am using SAS 9.2 to estimate multinomial probit model. The dependent variable has the values 1, 2, 3, 4 and there are about 20 (binary and continuous) independent variables. proc probit converge the model best but it do not show me the estimates based on the variance covariance matrix. These are standard deviations and correlation coefficients of different alternatives. Please share how can I see the results of variance covariance matrix along with the parameter estimates.

Thanks

Respected Advisor
Posts: 2,655

Re: proc probit for multinomial probit

My first thought would be to try running the data through PROC GENMOD, with dist=multinomial and link=cprobit.  Adding /covb to the model statement  will present the variance-covariance matrix. Parameter estimates are standard output.  If you need to capture these in a data set, try adding:

ODS output covb=covb parameterestimates=parameterestimates;

This should produce two data sets in your work directory containing what you are looking for.  The lst file will also have these values.

Good luck.

Steve Denham

SAS Employee
Posts: 187

Re: proc probit for multinomial probit

If you are trying to fit a model to a choice response and some of the predictors are characteristics of the choices (for example, a model on travel choice (car, bus, plane) with predictors like cost, time, etc.), then this model cannot be fit by PROC PROBIT.  For discrete choice models like this, you need to use PROC MDC in SAS/ETS software:

http://support.sas.com/documentation/cdl/en/etsug/63939/HTML/default/mdc_toc.htm

Occasional Contributor
Posts: 6

Re: proc probit for multinomial probit

Thanks for your reply StatDave@sas and SteveDenham. Actually the requirement is exactly proc mdc statment but as i run proc mdc, the result misses some standard errors and p values. The data arrangement and programming is as follows:

id          mode         decision       asso1................asso2................asso3................asso4

1            1                1                 2.5                    0                       0                       0

1            2                0                  0                      0                       0                       0

1            3                0                  0                      0                       0                       0

1            4                0                  0                      0                       0                       0

2            1                0                  0                      0                       0                       0

2            2                1                  0                     1.3                     0                       0

2            3                0                  0                      0                       0                       0

2           4                 0                  0                      0                       0                       0

3           1                 0                  0                      0                       0                       0

3           2                 0                  0                      0                       0                       0

3           3                 1                  0                      0                      2.1                     0

3           4                 0                  0                      0                       0                       0

4           1                 0                  0                      0                       0                       0

4           2                 0                  0                     0                        0                       0

4           3                 0                  0                     0                        0                       0

4           4                 1                  0                     0                        0                      1.9

proc mdc data=WORK.DEMO;

model decision=c1

c2 asso2 ab2 aac2 .....................

c3 asso3 ab3 aac3 .....................

/ type=mprobit choice=(mode 1 2 3 4) unitvariance=(1 4) covest=QML;

id id;

output out=demo5;

run;

In the log window it show me:

NOTE: Convergence criterion (FCONV=2.220446E-16) satisfied.

NOTE: At least one element of the gradient is greater than 1e-3.

WARNING: The model contains a RESTRICT or BOUNDS statement.  The resulting goodness-of-fit measures may be misleading.

WARNING: The Hessian matrix is singular.




I am seeking any useful suggestions to deal with this problem.

Occasional Contributor
Posts: 6

Re: proc probit for multinomial probit

Well, I did it with proc MDC and it gives me the results. But for the results of variance covariance matrix (standard deviations and rho's) it gives me only the estimates and not the standard errors nor p values. How to get this problem? Are the results for covariates meaningful in case of missing SE for sd and rho's (while algorithm is converged)?

I anticipate your help.

Ask a Question
Discussion stats
  • 4 replies
  • 657 views
  • 4 likes
  • 3 in conversation