Tobit and Fixed effects

[Olive Bat]

Can we perform tobit regression for panel data .. i want tobit for fixed effects.. is there any particular command.. i read xttobit .. but i assumed its for random effects only..

[Carlo Lazzaro]

Olive:
you're right: -xttobit- allows -re- specification only.
Usually, when some options is not available for a given Stata command, is for good theoretical reasons.

[Olive Bat]

Thanks for the reply Carlo .. what should i do if i need to run Tobit for FE any options or solutions... mine is an expenditure data and i am DID for my data so Tobit with FE

[Carlo Lazzaro]

Olive:
I do not know whether -tobit- with -i.panelid- can do the trick.

[Olive Bat]

Okay .. thanks.. if anyone could help...Jeff Wooldridge ...any suggestions..

[Jeff Wooldridge]

I need more information. The unit is the individual? How big are N and T? Balanced panel? Do you have control variables?

[Jeff Wooldridge]

I need more information. The unit is the individual? How big are N and T? Balanced panel? Do you have control variables?

[Scott Merryman]

You might find Bo Honore's pantob useful: https://www.princeton.edu/~honore/stata/index.html

Honoré, Bo E. (1992): "Trimmed Lad and Least Squares Estimation of Truncated and Censored Regression Models with Fixed Effects," Econometrica, Vol. 60, No. 3, Pages 533-565.

[Olive Bat]

Jeff Wooldridge Sir.. Its a balanced panel..the data is at household level... the two time period are 2011 and 2016; N-7100.. it contains both treatment and control.. (I use a DID.. because my expenditure data is truncated .. i need to use a tobit (fixed effects)

[Olive Bat]

Jeff Wooldridge Sir.. Its a balanced panel..the data is at household level... the two time period are 2011 and 2016; N-7100.. it contains both treatment and control.. (I use a DID.. because my expenditure data is truncated .. i need to use a tobit (fixed effects) and i do have control variables too..(related to household head and household level characteristics )...Could you suggest me something Jeff Wooldridge...

[Olive Bat]

Dear Jeff Wooldridge .. Its a balanced panel..the data is at household level... the two time period are 2011 and 2016; N-7100.. it contains both treatment and control.. (I use a DID.. because my expenditure data is truncated .. i need to use a tobit (fixed effects) and i do have control variables too..(related to household head and household level characteristics )...Could you suggest me something can i use a Poisson fixed effect ...

[Jeff Wooldridge]

I agree with Scott that you can try Honore's method. The drawback is it does not give the effect on the mean or median of the variable you observe; it gives the effect on the median of the underlying latent variable, which is not usually of interest.

I don't think your expenditure data are either censored or truncated. These both imply unobservability. I'm guess you mean there is a corner at zero for some expenditure categories. That's different even though the statistical solutions are similar.

In a DID setting it is perfectly fine to start with a linear model estimated by fixed effects. But this is almost the same as just putting in a dummy for 2016, a dummy for treatment, and the interaction. In fact, without other control variables, it would be identical. In other words, doing "fixed effects" in the linear case in your setting is the same as the simple DID. Therefore, you can do the same thing in Tobit and then use the margins command to get the average treatment effect.

With covariates, include them along with the time averages, as a correlated random effects alternative. I discuss this in my MIT Press book.

Here is the code:

xtset id year
gen w = treat*d2016
egen x1bar = mean(x1), by(id)
...
egen xkbar = mean(xk), by(id)
tobit y i.d2016 i.treat i.w x1 ... xk x1bar ... xkbar, ll(0) vce(cluster id)
margins, dydy(w) predict(ystar(0,.))

Having said this, I would use the Poisson FE estimator, too, as it is completely robust. It does imply a different conditional mean than the Tobit, so your coefficients will have different interpretations.

xtpoisson y i.d2016 i.w x1 ... xk, fe vce(robust)

JW

[Olive Bat]

Dear Jeff Wooldridge thank you so much.. let me work it out.

[Kaushik Borah]

Prof. Jeff Wooldridge how to implement CRE Tobit, if we have an unbalanced panel? Moreover, if we have T=3 (Suppose for 2011, 2016 and 2019), how to specify the command? My scenario is not a DID.

[Jeff Wooldridge]

See my 2019 Journal of Econometrics paper. It only works on the complete cases. Why is the panel unbalanced? Are you missing entire years? Are you missing data on some variables?