Missing Wald chi2 statistic in probit model with robust standard errors.

[Vincent La]

Hi all,

I am encountering an error with probit models I have never seen before. Unfortunately, the data is restricted, so I cannot post the data or even a sub-sample of the data here for people to see. In addition, I cannot recreate the error using a fake data set. Thus, I'm hoping that someone else has seen this "error" before and can offer a quick explanation/solution.

I am running a probit model with robust standard errors with 4 independent variables:

probit y x1 x2 x3 x4, vce(robust)

The probit model converges and gives me a table of coefficients. However, the wald chi-squared statistic (and also the associated p-value) is missing. In addition, when I run the exact same probit model without the robust-adjusted standard errors, the model does in fact converge and give me a LR statistic.

Why is it the case that the wald chi-squared statistic shows up as missing with robust standard errors? Is this a sign that something troubling is going on with my data/model?

In the end, for my analysis, I'm not too concerned about the wald chi-squared statistic; I'm more interested in the coefficients. However, I'm more concerned that the missing wald statistic is a sign that something more salient is lurking in the background that could cause problems.

Thank you in advance for your help!

Vincent

[Carlo Lazzaro]

Vincent:
this old thread might give you some guidance: http://hsphsun3.harvard.edu/cgi-bin/...ticle-203.html

[Vincent La]

Hi Carlo. I saw that thread, but I didn't find it particularly helpful. I think also I don't know enough about probit models and how heteroskedastic-robust standard errors can sometimes go wrong. Indeed, when I run probit models without vce(robust) I do get a statistic for the model test. But after using vce(robust) I do not get a wald statistic.

Does this mean my heteroskedastic-adjusted variance-covariance matrix is not full rank or is some way deficient to calculate a wald statistic?

[Carlo Lazzaro]

Vincent.
as a temptative answer, I would point you out to [ U ] 20 Estimation and postestimation commands, Example 19 in Stata 13.1 .pdf manual.

[Clyde Schechter]

Your situation is a bit unusual: this problem of missing F or chi2 statistic typically arises with vce(cluster robust), and has various causes. But let me just throw out a hunch here. Are any of your predictor variables dichotomous variables that takes on the value 1 for only a single observation and 0 for all the others (in your estimation sample)? (Or the other way around?) Singleton dummies will make the sandwich vce estimator singular.

If it isn't that, I don't have any other ideas to offer and, this post having been active several days now without resolution, you might want to raise the question with Stata technical support.

[Vincent La]

Hi Clyde,

Thank you for your response. While there are dummy predictor variables, none take on the value 1 for only a signal observations and 0 for all others (or the other way around). I'll think about the issue more and perhaps contact Stata technical support if the need arises.

Thank you all for your advice!

Vincent

[Panika Jain]

Hi Vincent,

I am also facing the similar problem with probit model. How to tackle it?. Is there any particular way to deal with it.

[Sayoree Gooptu]

Your situation is a bit unusual: this problem of missing F or chi2 statistic typically arises with vce(cluster robust), and has various causes. But let me just throw out a hunch here. Are any of your predictor variables dichotomous variables that takes on the value 1 for only a single observation and 0 for all the others (in your estimation sample)? (Or the other way around?) Singleton dummies will make the sandwich vce estimator singular.

If it isn't that, I don't have any other ideas to offer and, this post having been active several days now without resolution, you might want to raise the question with Stata technical support.

I have used state dummies in the regression (i.state) but my main variable of interest is a district level variable. That's why I used vce(cluster district). But still, my wald statistic is not being reported. Is this a similar kind of problem to the ones mentioned in the aforementioned posts? How do I tackle this problem?

Any suggestion would really be very helpful.

[Clyde Schechter]

Before you "tackle this problem," first determine whether it is really a problem. If the Wald statistic you are not seeing is the one for the regression as a whole, this usually makes no difference. That statistic is just a test of the hypothesis that all of the model coefficients are zero. That hypothesis is typically of no interest anyway, and its absence is not a problem--so just ignore it and move on.

If one of your research goals is to specifically test the hypothesis that all of the model coefficients are zero, then you do have a problem. To get help solving it, though, you would have to post back with the complete regression command and its output, as well as example data (example data rich enough to run the regression command on.) In the event that you do this, be sure to use the -dataex- command to show the example data. If you are running version 17, 16 or a fully updated version 15.1 or 14.2, -dataex- is already part of your official Stata installation. If not, run -ssc install dataex- to get it. Either way, run -help dataex- to read the simple instructions for using it. -dataex- will save you time; it is easier and quicker than typing out tables. It includes complete information about aspects of the data that are often critical to answering your question but cannot be seen from tabular displays or screenshots. It also makes it possible for those who want to help you to create a faithful representation of your example to try out their code, which in turn makes it more likely that their answer will actually work in your data.

When asking for help with code, always show example data. When showing example data, always use -dataex-.

[Sayoree Gooptu]

Okay. Thanks a lot.

I just had another question. How do I post regression results on statalist?

[Clyde Schechter]

Copy them from your Results window or log file to your computer's clipboard. Then come over here to the Forum editor. Click on the # button. (If you don't see a # button, click on the A button and it, along with several other buttons, will appear.) You will now see code delimiters in the edit window. Paste from the clipboard between the code delimiters. That's it. When you hit "Post Reply" the Statalist software will convert the code to a fixed width typeface, properly aligned.

[Sayoree Gooptu]

Thank you.
Since my question is related to a heckprobit model results which I had asked in a different thread, I have posted this and some other queries related to heckprobit in that thread. I have added example data using dataex. I hope that is not a problem. Sorry for any inconvenience.