Omitted variable bias

[Feruza Avezova]

Hi folks!

ovtest and linktest showed that my regression suffers from omitted variable bias. I've tried to include some other variables, and although the coefficients of significant variables do not change, the tests show that problem is not solved. I know this may sound stupid, but do you know any measure to tackle the omitted variable bias if I do not know which variable causes this bias?

Thank you beforehand!

[Joao Santos Silva]

Dear Feruza,

I am afraid that, like many other researchers, you are misinterpreting these tests: they are not tests for omitted variable but tests for functional form misspecification. (Paraphrasing you, how could these tests check for the omission of a variable that we do not know what it is?)

So, if you are rejecting the null, that suggests that either you are not accounting for some nonlinearities or you need to change the way your variables enter the model. For example, you may need to include interactions of to transform your regressors.

Best regards,

Joao

[Feruza Avezova]

These is extremely helpful! Thanks a lot!

[Carlo Lazzaro]

Feruza:
As an aside to Joao's helpful advice, you may want to skim through the literature of your research field and see which predictors were included in the regression model by other researchers who faced the same research topic.

[Feruza Avezova]

Dear Carlo,

Thanks a lot! That's exactly what I did but the tests still show the persistence of omitted variable. Then I've tried including different model specifications, included squared values of some of independent variables and it worked. Thank you)