Tobit estimation technique with panel data

[miret]

Hello,

Does the Tobit estimation technique work well with Panel data? I read somewhere that EViews "does not take into account the Panel structure of the data" and am not too sure what that means. Does this refer to a problem in EViews or to a problem in the Tobit estimation technique in general?

And if Tobit does work well with Panel data: Is it possible to include cross-sectional or time fixed effects? Since EViews does not provide this option, should I then simply construct dummy variables myself and include them in the equations to be estimated?

Thank you so much in advance for your answer.

Miret

[EViews Glenn]

What it means is that EViews doesn't take any account of the panel structure of your data (apart from lags) so that there are no in-built provisions for individual effects and the like.

You can add dummy variables yourself, but you should note that the asymptotics of the problem are going to require T to go to infinity for consistency for cross-effects models, or N to go to infinity for consistency for time-effects models.

Alternately, there are some panel tobit models in the literature, but EViews doesn't currently support them as built-in estimators.

[miret]

Ok, thank you. I understand the issue with the fixed effects.

Regardless, how should I interpret the estimated coefficients using Tobit in EViews? Say, for example, the dependent variable is number of working hours of women and an explanatory variable is years of schooling.

Thank you in advance.

[EViews Glenn]

There is nothing different in the interpretation from a usual tobit...But you are making the assumption that the identity of an observation (in time or in cross-section space) is irrelevant (the usual iid is extended to all observations in the workfile).

[miret]

Sorry for asking again on this issue - I am new to Tobit models...

Can I compute Tobit marginal effects in EViews? I saw there is an older post (2008) on how to compare OLS and Tobit coefficients with a specific example taken from Greene, but perhaps there is now a specific command in EViews to compute these?

[EViews Glenn]

No. There is nothing specific.