Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Tobit regression generating predicted values above the censored limit.

    Hi all !

    I am running the tobit model on stata 14.1 with the following command:

    tobit effscr lnta_round firmage forex freecashflow bussegcons mktsh_nsales i.year, ll(0) ul(1) vce(robust)

    The dependent variable effscr takes values between 0 and 1, however, the minimum values in the data is 0.0064. The output generated is as below:

    Tobit regression Number of obs = 1,942
    F( 23, 1919) = 20.14
    Prob > F = 0.0000
    Log pseudolikelihood = 828.36599 Pseudo R2 = -0.5368


    Robust
    effscr Coef. Std. Err. t P>t [95% Conf. Interval]

    lnta_round .0262072 .0033906 7.73 0.000 .0195575 .0328568
    firmage .0093179 .0046506 2.00 0.045 .0001973 .0184386
    forex .049268 .0063153 7.80 0.000 .0368824 .0616537
    freecashflow .034074 .0069969 4.87 0.000 .0203516 .0477964
    bussegcons .0962874 .0266129 3.62 0.000 .0440941 .1484806
    mktsh_nsales 1.262972 .2660517 4.75 0.000 .741191 1.784753

    year
    2001 -.0358751 .018999 -1.89 0.059 -.073136 .0013857
    2002 -.0491669 .0196069 -2.51 0.012 -.0876199 -.0107139
    2003 -.0578418 .0202878 -2.85 0.004 -.0976303 -.0180533
    2004 -.0581242 .0209067 -2.78 0.005 -.0991265 -.0171219
    2005 -.0116608 .0205047 -0.57 0.570 -.0518747 .0285532
    2006 -.0065574 .0200456 -0.33 0.744 -.0458709 .0327562
    2007 -.0252917 .0202699 -1.25 0.212 -.0650451 .0144617
    2008 -.0273508 .0210971 -1.30 0.195 -.0687265 .0140249
    2009 -.0048216 .0209372 -0.23 0.818 -.0458838 .0362405
    2010 .0181514 .0217199 0.84 0.403 -.0244456 .0607484
    2011 -.0150825 .0210903 -0.72 0.475 -.0564448 .0262798
    2012 -.0016375 .0210923 -0.08 0.938 -.0430038 .0397288
    2013 .0140805 .0185623 0.76 0.448 -.0223238 .0504848
    2014 -.0014518 .0206883 -0.07 0.944 -.0420256 .039122
    2015 -.0092528 .0221999 -0.42 0.677 -.0527914 .0342857
    2016 .0028913 .019862 0.15 0.884 -.036062 .0418447
    2017 .0221752 .0188914 1.17 0.241 -.0148746 .0592249

    _cons .4680239 .0452609 10.34 0.000 .3792582 .5567895

    /sigma .1433193 .0041708 .1351395 .1514992

    0 left-censored observations
    1,813 uncensored observations
    129 right-censored observations at effscr >= 1


    However, when I am generating the predicted values for Y, using the command "predict double xb, xb", I get few values (44 nos.) >1. Why is stata generating the predicted values more than the upper censored limit, I am not able to understand. Can somebody please help.

    Thank You.




  • #2
    It happens like this because you are specifying a linear prediction, but from your post it seems that you want nonlinear prediction.

    I think that to obtain the quantity that I think you want, you need to do:

    predict double xb, e(0,1)

    Comment


    • #3
      Read -tobit postestimation- to see the available options for -predict-.

      Comment


      • #4
        Thanks Joro Kolev for suggesting the reading "tobit postestimation". I tried with

        predict double xb, ystar(0,1)

        and the values matches with the predicted values obtained from other statistical package, however, I prefer Stata results due to certain flexibility available in it. It simplified my task. I hope the above estimations are correct.

        Comment


        • #5
          There are 4 possible things you can predict after fitting the tobit. You get the prediction for the subpopulation that is on the interior (i.e., in (0,1)) with e(0,1). The prediction for the censored expected value is ystar(0,1), which will also include the 0s and the 1s.

          Comment


          • #6
            Dimitriy V. Masterov Thanks.

            Comment

            Working...
            X