Tobit Model

[tripti sharma]

Hello,

I am trying to run a tobit model with the amount of equity released (non negative) as the dependent variable and a set of demographic and economic variables as explanatory variables. Where, demographic variables (age and marital status) are dummies and the economic variables (household income, unsecured debts, property wealth, outstanding mortgage and liquid assets) are 'self reported' values. This is a panel data. On running the code,

[xi: tobit Equity i.wave Age55_64 Age65_74 Age75+ MarStat1 MarStat2 MarStat3 HHInc UnSecDebt PropWlth OSMortg TotLiqAsset, vce(robust) ll]

I am getting massive standard errors for dummy variables. The standard errors for economic variables seem fine. For example:

[Variable Coef. Std. Err.
Age55_64 7742.762 9887.739
HH Income .2881224 .1006127]

My query is that why are the standard errors for dummy variables so huge and how should one fix this? Is it because there are so many zeros in the dependent variable? Only a handful (close to 1% out of 92461 observations) answered this question in the survey.

I would request you to respond to my query.

Thanks

[Carlo Lazzaro]

Tripti:
welcome to the list.
Please note that your output is unreadable.
Please see the FAQ on how using CODE delimiters to post what you typed and what Stata gave you back. Thanks.

[Joao Santos Silva]

Dear Tripti,

It is likely that the problem is exactly what you guessed. Suppose that your dummies are only equal to 1 when the dependent variable is zero; in that case you cannot even estimate its coefficient. This may be what is happening and Stata is just achieving spurious convergence. If the case is not that extreme but the dummy has few 1s when the dependent variable is positive, it will be difficult to estimate the parameter with precision and that may be what is happening.

I would also add that using Tobit here may not be the best option in this context because the Tobit relies on very strong assumptions. Have you considered using Poisson regression?

Best wishes,

Joao