Discussion: View Thread

  • 1.  Tobit distribution

    Posted 03-12-2012 09:53
    Hi all,

    I was wondering if anyone has had any experience working with predicted values from a Tobit distribution? Specifically, I'm trying to make sure I understand how to interpret predicted values, both uncensored and censored. I'm working with a dataset that is essentially left-censored at 0 and follows a continuous distribution to the right of that.

    In SAS, I'm using PROC LIFEREG and following the example they give here:
    http://support.sas.com/documentation/cdl/en/statug/63962/HTML/default/viewer.htm#statug_lifereg_sect035.htm

    Basically, what I'm trying to figure out is how to determine whether an observation crosses the zero threshold or hurdle. Is it whether or not the uncensored predicted value is greater than 0? Or is there some other method?

    I've also considered some kind of zero-inflated approach, like zero-inflated gamma (eg, http://listserv.uga.edu/cgi-bin/wa?A2=ind0805A&L=sas-l&D=0&P=33115), but it seems like Tobit might be the way to go.

    Any suggestions or guidance would be appreciated.

    Thanks,
    Gabe Farkas



    -------------------------------------------
    Gabriel Farkas
    -------------------------------------------


  • 2.  RE:Tobit distribution

    Posted 03-12-2012 21:05

    Economists often use Tobit regression ( Tobit is named after James Tobin, who won the Nobel in Economics).

    The derivation in Tobin's original paper is in two steps, a regression part for values above the threshold and a function for the indicator of above/below the threshold.
    Tobin, J. 1958. Estimation of relationships for limited dependent variables. Econometrica 26: 24-36.


    STATA and SAS provide a way to prepare predictions
    http://support.sas.com/documentation/cdl/en/statug/63033/HTML/default/viewer.htm#statug_lifereg_sect035.htm

    http://www.ats.ucla.edu/stat/stata/dae/tobit.htm

    -have fun!

    -------------------------------------------
    Chris Barker, Ph.D.
    President - San Francisco Bay Area Chapter of the American Statistical Association
    www,barkerstats.com

    ---
    "In composition you have all the time you want to decide what to say in 15 seconds, in improvisation you have 15 seconds."
    -Steve Lacy
    -------------------------------------------