Interaction effects in the Tobit model

[Amantia Simixhiu]

Hello Everyone,

I'm using the Tobit model to estimate the effect of a series of variables on donations. In the last stage of my work I want to analyse the interaction effect between my variables of interest on donations.

For clarity, throughout my analysis I always use to the "unconditional expectation", which specifically estimates the effect of the explanatory variables on the actual dependnet variable (and not on the latent variable). While the partial effects from the "unconditional expectation" describe a more or less linear relation beten x and y, the unnconditional expectation itself does not.

So that my question, is whether for the analysis of the interaction effects, I should consider the Norton and Ai, and Greene approach similar to the porbit/logit model, or just include an interaction term and analyse the sign and signifficance of the term, which is the partial effect from the unconditional expectation, same as in the linear function?

Does any of you have any Stat experience with interaction terms in Tobit? Until now I didnt come across any paper who exactly uses the interaction terms in Tobit.

Thanks in advance and kind regards

[Joao Santos Silva]

Dear Amantia,

My advice would be to avoid the Tobit and simply use Poisson pseudo maximum likelihood (PPML). This is likely to be more appropriate and you will have no problem interpreting the interaction effects.

Best wishes,

Joao

[Amantia Simixhiu]

Hello Joao,

indeed, I will use another regression model to analyse the intreaction effects. Due to time restrictions and since partial effects of the unconditional expectation in the Tobit produce regressors qualitatvely and quantitatively similar to OLS regresors, the linear regression model is the first choice for now. Nevertheless, I will consider your proposed model once I'm done with my work and have more time to inquire on the interaction effects for the Tobit model. In general I find that there is not enough litterature concerning the Tobit, especially for two-limit dependent variables, furthermore there is a lot of unclarity (sometimes with mistakes) concerning post estimation commands, or even commands related to multivariate regression models for the two-limit Tobit model. It's is not the best of the models to work with, especially if it is a two-limit model.

Thanks for your reply and kind regards,

Amantia

[Joao Santos Silva]

Dear Amantia,

The Tobit is valid only under assumptions that I find totally implausible, so it is not a bad thing that there is not much on it ;-)

Best wishes,

Joao