[R] Tobit model on unbalanced panel

[Arne Henningsen]

On 23 November 2010 23:32, Achim Zeileis <Achim.Zeileis at uibk.ac.at> wrote:
> On Tue, 23 Nov 2010, Liang Peng wrote:
>
>> Appreciate any suggestions regarding how to fit an unbalanced panel data
>> to
>> a Tobit model using R functions. I am trying to analyze how real estate
>> capital expenditures (CapEx) are affected by market conditions using a
>> panel
>> Tobit model. The CapEx is either positive or 0, so it is censored. The
>> data
>> are unbalanced panel, including the CapEx of about 5000 properties over
>> about 40 quarters, with the starting and ending quarters differ across
>> properties.
>>
>> In case you are not familiar with the term "Tobit", the model is
>> essentially
>> the following. The "true" value of the CapEx of property i in quarter t,
>> Y*[i,t], is determined by a property fixed effect (dummy) a[i],
>> explanatory
>> variables x[I,t], and an error: Y*[i,t]=a[i]+b*X[i,t]+u[i,t].  Further,
>> the
>> observed CapEx, Y[i,t], equals Y*[i,t] if Y[i,t]>0 and equals 0 otherwise.
>> Now I observe Y[i,t] and X[i,t], and the purpose of this analysis is to
>> estimate coefficients b. Thanks for any suggestions!
>
> I think that the "censReg" package might be worth exploring, see
>
>  http://CRAN.R-project.org/package=censReg
>
> I haven't looked at the package in detail yet, but it says that it is
> concerned with "estimation of censored regression (Tobit) models with
> cross-section and panel data".

Yes, the censReg function/package can estimate random effects tobit
models with balanced and unbalanced panel data. However, the
estimation with *unbalanced* panel data has not been thoroughly tested
yet. If you experience any problems, please report them via the
"tracker" of the "sampleSelection" project on R-Forge [1]. Thanks!

[1] https://r-forge.r-project.org/tracker/?group_id=256

Best regards,
Arne

-- 
Arne Henningsen
http://www.arne-henningsen.name