Dummy Variable Regression

[Denila Jinny]

Hi,

I have some continuous variables and many categorical variables in my model. What does it mean to get an intercept that is statistically significant? Is there a problem with it or is it acceptable?

Thanks.!

[Maarten Buis]

It means that you cannot reject the hypothesis that expected outcome is 0 when all your explantory variables are 0. Whether or not that is a problem depends on the situation (usually it is not a problem).

Jinn: it has been a long tradition on this list to post under our real name (first and last). We believe that that has contributed to friendly and professional atmosphere on this list. That is the reason why the Statalist FAQ asks everybody to do the same. To change your login name you can click the "contact us" button (bottom right of the screen).

[Denila Jinny]

Maarten Buis Thanks for the reply. Is there any article that I can use to substantiate this. My supervisor tells me that the constant is significant because of omitted variable bias. How true is that?

And, I have sent a mail to the moderators requesting for a change in name. Thanks!

[Maarten Buis]

You can cite any textbook that discusses regression. Just pick the one from which you have learned regression and look where it discusses the constant.

If your supervisor really made the general statement that all regression models with significant constants suffer from omitted variable bias, then (s)he is simply wrong. Why would we expect the expected outcome when all covariates are 0 to also be 0? We can arbitrarily change the point at which the covariates are 0 (and that is often a good idea), so this can definately not be generally true.

However, there can be special situations where you have strong substantive reason to expect that the constant is 0. If your supervisor was refering to such a special case, then (s)he may well be right to be worried about a significant constant.

[Denila Jinny]

Maarten Buis Thank you so much..!

[Carlo Lazzaro]

Jinn:
assuming that you're running a linear regression, you can test for omitted variable bias via Ramsey test (type -estat ovtest- after you have obtained -regress- output).
However, it is relevant to point out that Ramsey test will not tell you which variables are omitted from your regression model (i.e., there's no white magic in it), but can give you some hints about including squared (or higher) terms and/or interactions among your predictors.

[Denila Jinny]

Carlo Lazzaro Thanks much.

[Eric de Souza]

To add to Carlo's comment, the Ramsey test will only tell you whether functions of the already included regressors should have been included. In other words, whether the functional form is correct. The name "ovtest" is a misnomer and easily leads people astray.

[Denila Jinny]

ovtest tells me i have omitted variables bias. Since my data is from a survey of millions of enterprises, I can't collect more data on that regard. Interactions or higher order terms can be considered.

Constant is said to reflect the coefficients of base categories of the categorical variables included in the data. Can that be one reason why the constant is significant?

[Carlo Lazzaro]

Denila:
I do agree with Eric.
It seems that Ramsey made a long-lasting advertising stunt in favour of -ovtest- by labelling it as an omitted variable test.
When my side-interest in econometrics started to grow and I knew almost nothing about that stuff, the idea we could rely on a clever test that works for us in pointing out what is missing among our predictors was very appealing. Later I discovered, at my own expenses, that it was only maths (no white magic, unfortunately...)
Sticking with your query, you should focus on higher terms and interaction that make sense in your own research field. By the way, skimming through the literature and see what others did in dealing with the same research topic is, as usual, a good first step.