Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: FW: ML for logit/ologit

 From Fernando Rios Avila To statalist@hsphsun2.harvard.edu Subject Re: st: FW: ML for logit/ologit Date Wed, 16 Nov 2011 07:57:13 -0500

```Hi Thomas
I cought the bug in your program.
THe key is that because u are using a non linear combination of X u
need to declare it in such a way that you can freely manipulate it on
stata.
Check the following code

program drop logittest
program define logittest

args lnf b1 rho
tempvar lng
qui {
gen double `lng' = ln(invlogit(`b1'*((\$ML_y2^(1-`rho')-1)/(1-`rho'))))
if \$ML_y1==1
replace `lng' =  ln(invlogit(-`b1'*((\$ML_y2^(1-`rho')-1)/(1-`rho'))))
if \$ML_y1==0
replace `lnf' = `lng'
}
end

ml model lf logittest  (b1:y x=) (rho:), maximize

for the rho to be bounded between 0 and 1 you can do the following

program define logittest

args lnf b1 R
tempvar lng  rho
qui {
gen double `rho'=exp(`R')/(1+exp(`R'))
gen double `lng' = ln(invlogit(`b1'*((\$ML_y2^(1-`rho')-1)/(1-`rho'))))
if \$ML_y1==1
replace `lng' =  ln(invlogit(-`b1'*((\$ML_y2^(1-`rho')-1)/(1-`rho'))))
if \$ML_y1==0
replace `lnf' = `lng'
}
end

ml model lf logittest (y x=) (), maximize init(b)
On Wed, Nov 16, 2011 at 7:23 AM, Nick Cox <n.j.cox@durham.ac.uk> wrote:
> Two completely separate points are bundled together here.
>
> 1. If your program declares that you are feeding it arguments, then those arguments must be supplied (unless you trap non-supply).
>
> 2. A parameter that must be bounded should be re-parameterised so that the bounds are respected. Some usual tricks are collected at
>
> http://www.stata.com/support/faqs/stat/interval-constraints.html
>
> logit(rho/3) is unbounded for 0 < rho < 3.
>
> I recommend very detailed study of the documentation on -ml- and the associated book from Stata Press when writing a first non-trivial -ml- program.
>
> Nick
> n.j.cox@durham.ac.uk
>
> Thomas Murray (Department of Economics)
>
> This makes sense -  rho needs to be defined in the program.  I need rho to converge to a value that maximises the log likelihood.  What I want to do is set a range of possible values for rho (between zero and three specifically) and find a point estimate using ML.  Is this possible to define?
>
> Nick Cox
>
> You need to supply the right number of arguments too. Here it looks as if rho is undefined because you are not feeding the program with enough arguments.
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/
```