Testing heteroskedasticity in probit

[Erwin]

Hi,

How do I test heteroskedasticity and distributional assumption of a probit model?

[thesis]

If I try to run the normality (heteroskedasticity) test in a probit model. However i get the error message: insufficient observations. But I have more than 15000 observations included in my regression. Is there another manner to test for normality?

[startz]

What test did you use?

[thesis]

Histogram - Normality Test

[startz]

I don't understand why it didn't run, unless you have a perfect fit. But this test doesn't have anything to do with heteroskedasticity and one doesn't expect the raw residuals from a Probit to be normal. After all, the LHS variable is discrete.

[thesis]

Sorry, I look like a bit stupid now, haha. To make myself more clear:

I need to test for heteroskedasticity if I want to use a binary> Probit model right?

However in the menu of probit, the White test is not displayed. So how do I test for heteroskedasticity?

(I thought the normality test would maybe display some more diagnostics, that's why I tried that as well)

[startz]

You're asking perfectly reasonable questions! EViews doesn't have a built-in test, but if you look under Heteroskedasticity in the index of the Help system they show you how to do it.

[thesis]

Hey Startz, thanks a lot for showing me this feature in Eviews. I was looking up all the information in the userguides only.

I tried to take the steps, described in this help menu.

However: 'Next, the dependent variable in the test regression may be obtained as the standardized residual. Select Proc/Make Residual Series... and select Standardized Residual. We will save the series as BRMR_Y. '

This new serie only generates NA's in my case.

Later on,in the step: 'i]Then run the artificial regression by clicking on Quick/Estimate Equation..., selecting Least Squares, and entering:

brmr_y brmr_x (psi*(-xb)*fac)[/i]' I get an error message with 'unsufficient number of observations' and I think this has to do with the NA's of the BRMR_Y serie

So why do I get these NA values? How can I solve this problem?!

[startz]

The NA's are probably the key, but I'm afraid I haven't the foggiest idea why you're getting them.

[thesis]

Hey Startz,

Does it make sense that the ordinary and generalized residual types do not give NA scores? These types work properly. ( Unfortenately I don't need those)

Do you have some more advice to find out how I got these NA scores?!

Thanks!

[startz]

No idea at all

[thesis]

MMMM in that case I'm stuck!

[thesis]

Does anybody have any new insights?!

[toa]

Hi there.
Could someone kindly help out this newby to eviews and econometrics please?
I want to estimate a probit model and test for heteroskedasticity of the form: σi2 = σ2z. My variables are y (a dummy variable as the dep vble), and educ and minority (another dummy variable). z = educ. I don't trust the results of the heteroskedasticity regressions I've run as the log-likehood stats are twice the amount of the normal probit model.

Appreciate any help.
Cheers

[toa]

Sorry there the 2's in the form of heteroskedasticity in my previous post are squares (e.g. sigma squares).

Thanks