Zero-inflated dependent variable calculations

[Nur Eken]

Hello,

I am trying to run a regression on Stata where my dependent variable is skewed towards zeros. I thought that running a zero-inflated poisson regression would be the appropriate way to do it, however, I do not actually know that there is a specific variable that is responsible for the zero inflation, which does not fit with what the model is built on. Would a simple linear regression then be the appropriate choice?

Thank you.
Nur

[Clyde Schechter]

What do you mean when you say that you "do not actually know that there is a specific variable that is responsible for the zero inflation?" Do you mean that you believe that there may not be any such variables, or do you mean that you think there are such variables but you don't know which ones they might be.

If your basic belief is that there are no such variables, you can specify the -inflate(_cons)- option, and then the zero inflation will be assumed to be random with a fixed probability that does not depend on any variables at all.

If your basic belief is that there some such variables, but you are unsure which ones, then I would specify all the reasonable candidate variables in the -inflate()- option and let Stata determine the contributions of each (assuming your data set has enough observations to support models with that many variables. If not, you may have to select the ones that you think are most likely to be involved in the zero inflation, limiting to a reasonable number of such variables.)

[Nur Eken]

I meant the former, and the code worked perfectly. Thank you.

[Joao Santos Silva]

Dear Nur Eken,

Following up on Clyde useful comments, I would like to note that the fact that your dependent variable has many zeros does not justify using a zero inflated model. Moreover, you should not use a zero-inflated model if your dependent variable is not a count.

Best wishes,

Joao

[Nur Eken]

Thank you, that is very helpful. Does that mean that a simple linear regression model can also be acceptable even if there is a skew towards zero?

[Joao Santos Silva]

I would start by using a simple Poisson regression with robust standard errors.

Best wishes,

Joao

[Nur Eken]

Thank you.