Wald test

[startz]

You might want to post screen shots of the two tests that disagree. Wald tests of linear restrictions will agree with redundant variable tests in a linear regression, but need not in a nonlinear operation such as probit. However, it is unusual for the results to be very different.

[EViews Glenn]

The idea that both startz and I are getting at is that when you compute the Wald test for

Code: Select all

c(3)=c(4)=0

you are testing the joint hypothesis that C(3) and C(4) are equal to zero. The individual t-statistics are giving you the individual probabilities for the two equalities. It is not uncommon for joint testing to reject a hypothesis while the individual hypotheses do not reject as described by startz above.