Tobit regression using panel data

[Max Drese]

Hello everyone,

I am trying to fit a tobit model on my panel data. I am trying to find the relation between the occurrence of a natural disaster and the level of remittances. The problem is that often the remit variable has the value of 0. So according to the literature, a tobit model would be appropriate. I used the following command to estimate the model.

xttobit remit nd_t, ll(0)

Where nd_t is a binary variable for a natural disaster at time t. The problem now is that STATA keeps on iterating when fitting the full model. For instance

Iteration 1365: log likelihood = -13661.049 (not concave)

It keeps giving the same value for the log likelihood (-13661.049). Does anyone have any idea what could be wrong?

With kind regards,
Max

[Phil Bromiley]

You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, sample data using dataex.

It is odd that you're getting this with such a simple model. It might be a problem just having one dummy on the rhs. Do you have additional control variables that you could include? You might want to run it with xtreg with random effects to see if you really have explanatory power.

[Joao Santos Silva]

Dear Max Drese,

That will happen if nd_t = 1 only when remit = 0; could that be the case?

Anyway, Tobit is not appropriate in this context; I would suggest you use Poisson regression with robust (clustered) standard errors.

Best wishes,

Joao