Statalist The Stata Listserver


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Number of cutpoints in -oprobit-?


From   [email protected] (William Gould, Stata)
To   [email protected]
Subject   Re: st: Number of cutpoints in -oprobit-?
Date   Mon, 30 Apr 2007 11:09:02 -0500

Rachel <[email protected]> asks, 

> My question is why oprobit identifies /cut1 at all, rather than
> normalizing it to zero as certain other statistical packages (such as
> Limdep) do.  My understanding is that the ordered probit/logit is
> identified by the relative lengths of the intervals--the cutpoints
> themselves have no meaning. And since the length of the interval
> between infinity and the first cutpoint is always infinite, why is
> this first cutpoint identified at all?

I have two answers:

     1.  Note that the number of cutpoints is k-1, where k is the 
         number of outcomes.  Perhaps Rachel is thinking k.

     2.  Even if Rachael is thinking k-1, one can write the ordered 
         probit model with an intercept.  If one does that, then 
         only k-2 cut points are identified.

An easy way to understand -oprobit- is to use it to fit a standard 
(two outcome) probit model.  Try the following, 

        . sysuse auto, clear 

        . probit foreign mpg 
          [...]
        -------------------------------------- [...] 
             foreign |      Coef.   Std. Err.  [...]
        -------------+------------------------ [...]
                 mpg |   .0960601   .0301523   [...]
               _cons |  -2.635268   .6841458   [...]
        -------------------------------------- [...]

        . oprobit foreign mpg 
          [...]
        -------------------------------------- [...]
             foreign |      Coef.   Std. Err.  [...]
        -------------+------------------------ [...]
                 mpg |   .0960601   .0301523   [...]
        -------------+------------------------ [...]
               /cut1 |   2.635268   .6841458   [...]
        -------------------------------------- [...] 

The first thing to note is one intercept (probit), one cut point (oprobit),
which is to say, the single cut point reported by -oprobit- corresponds 
to the intercept reported by -probit-.

Next to note is that the intercept and the cut point have different signs.
That, however, is due to interpretation.  In probit, the result is
to be interpreted as 

         Pr(foreign) = Pr(.096*mpg - 2.635 + u > 0)

where u is N(0,1).  In oprobit, the result is to be interpreted as

         Pr(foreign) = Pr(.096*mpg + u > 2.635)

Thus, the two results mean the same thing.  -probit- and -oprobit- fitted the
same model, just as they should.

-- Bill
[email protected]
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index