RE: st: FW: ML for logit/ologit

["Thomas Murray (Department of Economics)" <t.l.murray@bham.ac.uk>]

I have run the set trace and it is clear there is a problem with the including an argument within the squared term (I've pasted the relevant part below).  I am sure this is the bug but I do not know why Stata doesn't like it.

I am confident all the quotations are correct in the program.

Many Thanks,

Tom

- mata: Mopt_search()
        ------------------------------------------------------------------------------------------------ begin logittest ---
        - version 12.0
        - args lnf b1 b2 x mu
        - tempvar lng
        - qui {
        - gen double `lng' = ln(invlogit(`b1'*((`x'^(1-`rho')-1)/(1-`rho')))) if $ML_y1==1
        = gen double __000007 = ln(invlogit(__000006*((^(1-)-1)/(1-)))) if educ1==1
unknown function ^()
          replace `lng' = ln(invlogit(-`b1'*((`x'^(1-`rho')-1)/(1-`rho')))) if $ML_y1==0
          replace `lnf' = `lng'



-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Nick Cox
Sent: 16 November 2011 10:17
To: 'statalist@hsphsun2.harvard.edu'
Subject: RE: st: FW: ML for logit/ologit

I don't think the example calls for such a sweeping generalization. People write programs to maximise non-linear functions in Stata [sic] all the time. 

There is a bug in Thomas' code. I can't spot it and there is a lingering doubt about use of quotation marks from previous posts. Using -set trace- might help him find where it is. 

Nick
n.j.cox@durham.ac.uk 

Yuval Arbel

Several years ago I came across the same problem with estimation of non-linear functions in STATA. The solution I found is to use MATLAB for these objectives of non-linear functions. I believe that compared to STATA, MATLAB will give you a much better flexibility to define non-linear functions.

2011/11/16 Thomas Murray (Department of Economics) <t.l.murray@bham.ac.uk>:

> Does anyone have any advice about the problem described below?

Thomas Murray (Department of Economics)

> I need to use the ML function to create a program for an ordered logit  log likelihood function.  Before writing the code for the ologit model I've been trying to practice with a computationally simpler logit regression.  This is what I've written so far:
>
> program define logit2_lf
> version 10.1
> args lnf xb
> tempvar lnlj
> qui {
> gen double `lnlj' = ln(invlogit( `xb')) if $ML_y1 == 1 replace `lnlj' 
> = ln(invlogit(-`xb')) if $ML_y1 == 0 replace `lnf' = `lnlj'
> }
> end
>
> However, the model I wish to estimate is not simply y = a + bx.
>
> I want to estimate the following: y = a + b((x^(1-ρ)-1)/(1-ρ))
>
> where ρ is a parameter to be determined.
>
> I have been trying to incorporate this into the logit code above (by replacing the xb function) along the lines of:
>
> program drop logittest
> program define logittest
> version 12.0
> args lnf b1 x rho
> tempvar lng
> qui {
> gen double `lng' = ln(invlogit(`b1'*((`x'^(1-`rho')-1)/(1-`rho')))) if $ML_y1==1 replace `lng' =  ln(invlogit(-`b1'*((`x'^(1-`rho')-1)/(1-`rho')))) if $ML_y1==0 replace `lnf' = `lng'
> }
> end
>
> ml model lf logittest ( y = x)
> ml max
>
> However I get the error 'unknown function ^()' - it doesn't like my attempted manipulation of x using powers.
>
> My question - is it possible to impose this particular form of x into a logit (or ordered logit) function?  Am I just making a computational mistake in my programming language?

*
*   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/