Wald test

[bibimiwi]

Hello,

I estimated a probit model with 7 explanatory variables of which 3 are dummy variables. Two of the explanatory variables are giving me a hard time: they have a z-statistic of -0.95 and 1.46 and p-values of 0.34 and 0.14. Therefore I concluded that they are not statistically signifcant. The Wald-test comes to the same result.
BUT if I run a reduntant variable test on those two explanatory variables, they are highly significant! (The correleation between those two variables is 0.82).
How can this be, that one time they are statistically significant and the other time not???

Thank you very much

[EViews Glenn]

What Wald test did you run?

[startz]

It is very common for two coefficients to be jointly significant when neither is individually significant. Essentially, there is something common to the two variables that matters, but because the two variables are highly correlated it isn't possible to separate out which variable is responsible for the outcomes.

[bibimiwi]

The one that eviews offers for the coefficient diagnostics (wald - Coefficient Restrictions...)

[bibimiwi]

the thing is that I estimated various model, and all of them had those 2 variables included. And in all cases but this one, I was able to distinguish between those two variables.
Does a high correlation not only start at 0.98 and not already at 0.82?

[EViews Glenn]

On the Wald test...

Yes, but which hypothesis to test did you enter?

[bibimiwi]

I am not sure if I understood your question...
I estimated a probit model.
Is this the answer to your question?

[startz]

When you went to

The one that eviews offers for the coefficient diagnostics (wald - Coefficient Restrictions...)

what command did you enter?

[bibimiwi]

the only way possible .. c(3), c(4).

I really think that I am not getting your question

[startz]

the only way possible .. c(3), c(4).

I really think that I am not getting your question

Glenn asked for the command you entered. Did you type "c(3),c(4)"? Or did you type something different involving c(3) and c(4)?

[bibimiwi]

only c(3),c(4)

[startz]

When I do that on my version nothing happens. Equal signs are required as part of the specification.

[bibimiwi]

Equal signs are required as part of the specification.

My coefficients indeed have different signs... do you think this is the problem??
Is it not possible to perform a wald test on two coefficients with different signs?

[startz]

Equal signs are required as part of the specification.

My coefficients indeed have different signs... do you think this is the problem??
Is it not possible to perform a wald test on two coefficients with different signs?

By "equal signs" I meant literally "="

[bibimiwi]

Oh I am so sorry! This was completely my fault!
Of course I mean C(3)=0 and C(4)=0