Newey-West standard errors in probit regression

[th87]

Hello All,

I require the use of Newey-West standard errors when running a probit regression. From what I have read on the forum, it seems the nwest function seems the most appropriate. However, having used it, and then viewing the predicted probabilities that the dependent variable is equal to one, I am getting negative values.

The help article on the nwest function, states "nwest produces Newey-West standard errors for coefficients estimated by OLS regression".

Is there a way to produce Newey-West standard errors for coefficients estimated by maximum likelihood, such as the probit model?

Many thanks in advance.

Terrence

[Jeff Wooldridge]

The nwest command works properly to compute the standard errors. However, when you predict after nwest you do not get the predicted probabilities. Instead you get the predicted linear index, xi*b. These, of course, can take negative and positive values. In old versions of Stata, this was also the default after probit, but now the default is to report the probabilities. It appears nwest is behind regarding that convention.

You can get the fitted probabilities using probit (if those are of interest) and use nwest to get the proper standard errors. I assume this is a time series application. JW

[th87]

Hi Jeff,

A very informative answer. I cannot thank you enough for your help.

Kind regards,

Terrence