When one model variable is an exact linear combination of other model variables, PROC MDC produces incorrect degrees of freedom and standard errors. Additionally, PROC MDC provides no indication that the computed parameter estimates reflect one biased solution among infinitely many solutions.

To avoid the problem, check for linear dependencies among the model variables using another procedure such as PROC REG (which provides informative output about any detected collinearity) prior to invoking PROC MDC. For example, suppose your MDC syntax is:

proc mdc data=work.a;
   id pid;
   model y=x1 x2 x3 x4 x5/type=clogit choice=(mode);
   run;

Then submit the following code and check the resulting output to see if one variable is a linear combination of other variables:

proc reg data=work.a; model y=x1 x2 x3 x4 x5/ noint; run;

Finally, eliminate such a linear dependency (by removing the offending variable from the model) prior to invoking PROC MDC.