If you specify either the PRL=WALD or PRL=BOTH option in the FIT statement in PROC MODEL, and you impose bounds or restrictions that are a function of more than one parameter, then the Wald profile likelihood confidence limits about the constrained parameters may be incorrect.

For example:

BOUND A &gt 1;

produces correct Wald profile likelihood confidence limits when the PRL=WALD or PRL=BOTH option is used, since the BOUND is a function of only a single parameter. However, the reported Wald confidence limits for

BOUND A &gt B;

or

RESTRICT A + B &gt C;

may be incorrect, since these constraints involve multiple parameters.

Note that only the Wald profile likelihood confidence limits associated with the constrained parameters may be incorrect. The confidence limits associated with any free-estimated parameters in the model will be correct.

To circumvent the problem, specify PRL=LR on the FIT statement in order to obtain likelihood ratio-based confidence limits.