Search
   >> Home >> Resources & support >> FAQs >> Obtaining robust standard errors for tobit
Note: This FAQ is for Stata 9, and older versions of Stata. It is not relevant in Stata 10 and newer versions because the vce(robust) option may be specified with tobit.
In Stata 9, use the vce option with the tobit command to obtain estimates of the standard errors using nonparametric bootstrap or jackknife.

How can I get robust standard errors for tobit?

Title   Obtaining robust standard errors for tobit
Author James Hardin, StataCorp
Date February 1997; updated May 2005

The tobit command does not have the robust option; however, intreg does.

Since intreg is a generalization of cnreg (which is itself a generalization of tobit), you can use intreg to obtain the results that you want. Below is an illustrative example.

 . sysuse auto, clear

 . tobit mpg price weight, ll(20)
 
 Tobit regression                                  Number of obs   =         74
                                                   LR chi2(2)      =      61.85
                                                   Prob > chi2     =     0.0000
 Log likelihood = -119.03328                       Pseudo R2       =     0.2062
 
 ------------------------------------------------------------------------------
          mpg |      Coef.   Std. Err.      t    P>|t|     [95% Conf. Interval]
 -------------+----------------------------------------------------------------
        price |   .0002594    .000289     0.90   0.372    -.0003166    .0008355
       weight |  -.0084336   .0012104    -6.97   0.000    -.0108465   -.0060207
        _cons |    43.4248   2.838216    15.30   0.000     37.76692    49.08268
 -------------+----------------------------------------------------------------
       /sigma |   4.578498   .5572115                      3.467717    5.689279
 ------------------------------------------------------------------------------
   Obs. summary:         38  left-censored observations at mpg<=20
                         36     uncensored observations
                          0 right-censored observations
 
 . gen mpg2 = mpg
 
 . replace mpg2 = . if mpg<=20
 (38 real changes made, 38 to missing)
 
 . gen mpg3 = mpg
 
 . replace mpg3 = 20 if mpg<=20
 (35 real changes made)
 
 . intreg mpg2 mpg3 price weight
 
 Fitting constant-only model:
 
 Iteration 0:   log likelihood = -166.49203  
 Iteration 1:   log likelihood = -150.82635  
 Iteration 2:   log likelihood = -149.95827  
 Iteration 3:   log likelihood = -149.95786  
 Iteration 4:   log likelihood = -149.95786  
 
 Fitting full model:
 
 Iteration 0:   log likelihood = -141.72401  
 Iteration 1:   log likelihood =  -119.8515  
 Iteration 2:   log likelihood = -119.03468  
 Iteration 3:   log likelihood = -119.03328  
 Iteration 4:   log likelihood = -119.03328  
 
 Interval regression                               Number of obs   =         74
                                                   LR chi2(2)      =      61.85
 Log likelihood = -119.03328                       Prob > chi2     =     0.0000
 
 ------------------------------------------------------------------------------
              |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
 -------------+----------------------------------------------------------------
        price |   .0002594    .000289     0.90   0.369    -.0003069    .0008258
       weight |  -.0084336   .0012104    -6.97   0.000     -.010806   -.0060612
        _cons |    43.4248   2.838255    15.30   0.000     37.86193    48.98768
 -------------+----------------------------------------------------------------
     /lnsigma |   1.521371   .1217052    12.50   0.000     1.282833    1.759909
 -------------+----------------------------------------------------------------
        sigma |   4.578498   .5572271                      3.606844    5.811907
 ------------------------------------------------------------------------------
 
   Observation summary:        38  left-censored observations
                               36     uncensored observations
                                0 right-censored observations
                                0       interval observations

The above example shows how the intreg command is used to obtain the same results as the tobit command. To obtain robust standard errors, you now only have to add the robust option to the intreg command above.

The Stata Blog: Not Elsewhere Classified Find us on Facebook Follow us on Twitter LinkedIn Google+ Watch us on YouTube