Difference between "truncreg" and "simarwilson" command

[Shubham Sharma]

Hello all,
I am trying to assess the performance of the municipalities from India. For the same, I have computed super-efficiency DEA. I further wish to compute second stage DEA and identify the factors that influence the performance of the municipalities. A popular method, found in this regard is "simarwilson" introduced by Simar and Wilson (2007). However, when I tried to do so, STATA gave a warning that some scores are not bounded in the (0,1] interval and hence, dropped a few observations and presented results for the rest but when I used the "truncreg" command I got results for all the observations. Moreover after reading Stata Journal st0585, I came to know simarwilson can be used only when the dependent variable lies in the (0,1] interval. Since both the "truncreg" and "simarwilson" command compute truncated regression, I wished to know the difference between the two. Moreover, is there any issue if I have used "truncreg" instead of "simarwilson" command to conduct the truncated regression for my analysis.


Thank You

Kind Regards

Shubham

[Harald Tauchmann]

Dear Shubham:
In terms of the estimated coefficients, there is no difference between truncreg and simarwilson, but the estimated standard errors are different. simarwilson implements the parametric bootstrap suggested in Simar & Wilson (2007). This procedure for standard error estimation addresses the issue that DEA efficiency scores - that enter the regression model as left-hand-side variables – are usually computed from the same sample of data, which is inconsistent with the assumption that the observations are independently sampled from the population. This is discussed in detail in the Stata Journal article st0585.
Best wishes,
Harald