Announcement

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

  • Postestimation with tobit: predicting censored variable

    I am using tobit to get a simple censored prediction of income using data from the American Community Survey. The data contain some negative values corresponding to negative income such as business losses. I want a censored prediction of income so I use the postextimation command predict, ystar() in the following way:

    tobit faminc i.age4 i.education i.adults3 black other i.year, ll(0)
    predict xb
    predict ystar, ystar(0,.)

    I have two questions
    1) when I use the above specification including weights as provided by ACS ([aweight = perwt]) the distribution of censored predicted values is orders of magnitude above the actual distribution, with the same variance.
    2) when I use the above specification as written, predict, ystar(0,.) is supposed to give me y* = 0 if xb + u < 0, however the predicted values are E(y*|y>0), and so seems to be censored above zero, so I end up with no zero predicted values, although my sample contains about 2% zeros.


    I have found the following related discussion in the archive but I'm still stuck:

    http://www.stata.com/statalist/archi.../msg00524.html

    I am using Stata version 12. thank you very much for your help!

    Yulya


  • #2
    This is the Sandbox forum, where people can try out posting features like Latex. Questions about Stata and statistics should go in the General Forum.
    Steve Samuels
    Statistical Consulting
    [email protected]

    Stata 14.2

    Comment

    Working...
    X