Thanks... I made those changes and still got missings with the 'no
constant' specification (log included below). I have read the Gould and
Scribney book but haven't found anything about why the constant/no constant
specification would change whether the model works.
Any further ideas would be much appreciated,
Thanks again,
Deborah.
program define ml_move2;
1. args lnf theta1 theta2;
2. quietly replace `lnf' = ln( (1 - normprob(`theta1' +
`theta2'))*probmax + (1 - normprob
> (`theta2')*(1-probmax)) ) if $ML_y1==1;
3. quietly replace `lnf' = ln( normprob(`theta1' + `theta2')*probmax +
normprob(`theta2')*
> (1-probmax) ) if $ML_y1==0;
4. end;
. ml model lf ml_move2 (leavejob = alt_ret pv_ira, nocons )
> (pension esitaxprice jobtenure esi spesi
> age35_45 age45_55 hispanic black race_oth
> school_1-school_3 firm_x firm24 firm25_99
> occ_2-occ_7 occ_x ind_1-ind_4 ind_6 ind_x urban union
> hlth_x hlth_gd hlth_fp logann logtxinc
> secondearn kid kid2 married female ira_x jobten_x
> year84 year90 year96)
> [pweight=irawgt] if age1 >= 25 & age1 <= 55 ;
. ml maximize;
initial: log likelihood = -30971.235
alternative: log likelihood = -21880.591
rescale: log likelihood = -20492.416
rescale eq: log likelihood = -20304.363
Iteration 0: log likelihood = -20304.363
Iteration 1: log likelihood = -17187.65
Iteration 2: log likelihood = -17036.272
Iteration 3: log likelihood = -17034.944
Iteration 4: log likelihood = -17034.944 (not concave)
Number of obs =
53011
Wald chi2(0) =
.
Log likelihood = -17034.944 Prob > chi2 =
.
----------------------------------------------------------------------------
--
| Robust
leavejob | Coef. Std. Err. z P>|z| [95% Conf.
Interval]
-------------+--------------------------------------------------------------
--
eq1 |
alt_ret | 815.4468 . . . .
.
pv_ira | .000559 . . . .
.
-------------+--------------------------------------------------------------
--
eq2 |
pension | .7307576 .0258399 28.28 0.000 .6801123
.781403
esitaxprice | -.0503448 .1366193 -0.37 0.712 -.3181137
.2174242
jobtenure | .0550601 .0023604 23.33 0.000 .0504338
.0596864
[ADDITIONAL VARIABLES SUPPRESSED]
_cons | -.6282337 .0767639 -8.18 0.000 -.7786882
-.4777793
----------------------------------------------------------------------------
--
-----Original Message-----
From: Renzo Comolli [mailto:[email protected]]
Sent: Friday, May 02, 2003 10:22 AM
To: [email protected]
Subject: Re: st: Maximum Likelihood
Dear Deborah,
I have done very little maximum likelyhood, so I am not sure I can help you
fully. There is something I can tell you for sure, though.
You cannot write
args Li x1 x2
You must write
args lnf theta1 theta2
And you must keep on doing so throughout the script.
I found the book
Maximum Likelihood Estimation with Stata by William Gould, William Sribney
to be extremely useful (it is written for version 6.0 but ML has changed
very little, if at all, since then)
Suggestion to Stata: I found the description of ML in the manuals base
reference (for version 7.0) too short. Just one little example, I could not
find reference to the issue mentioned here.
Best,
Renzo Comolli
-----Original Message-----
[only bits of the original message]
I am trying to use the maximum likelihood procedures in Stata and am getting
some bizarre results that suggest that I do not fully understand how the lnf
feature works. I'd be very grateful for some illumination.
thanks,
deborah.
------------------------------------
lf version, no constant option:
program define ml_move;
1. args Li x1 x2;
2. quietly replace `Li' = ln( (1 - normprob(`x1' + `x2'))*probmax + (1
- normprob(`x2')*(1
> -probmax)) ) if $ML_y1==1;
3. quietly replace `Li' = ln( normprob(`x1' + `x2')*probmax +
normprob(`x2')*(1-probmax) )
> if $ML_y1==0;
4. end;
. ml model lf ml_move (leavejob = alt_ret pv_ira, nocons )
> (pension esitaxprice
> jobtenure
> esi spesi
> age35_45 age45_55
> hispanic black race_oth
> school_1-school_3 firm_x firm24
> firm25_99 occ_2-occ_7 occ_x ind_1-ind_4 ind_6 ind_x urban
union
> hlth_x hlth_gd hlth_fp
> logann logtxinc
> secondearn
> kid kid2
> married female ira_x jobten_x
> year84 year90 year96)
> [pweight=irawgt] if age1 >= 25 & age1 <= 55 ;
. ml maximize;
initial: log likelihood = -30971.235
alternative: log likelihood = -21880.591
rescale: log likelihood = -20492.416
rescale eq: log likelihood = -20304.363
Iteration 0: log likelihood = -20304.363
Iteration 1: log likelihood = -17187.65
Iteration 2: log likelihood = -17036.272
Iteration 3: log likelihood = -17034.944
Iteration 4: log likelihood = -17034.944 (not concave)
Number of obs =
53011
Wald chi2(0) =
.
Log likelihood = -17034.944 Prob > chi2 =
.
----------------------------------------------------------------------------
--
| Robust
leavejob | Coef. Std. Err. z P>|z| [95% Conf.
Interval]
-------------+--------------------------------------------------------------
--
eq1 |
alt_ret | 815.4468 . . . .
.
pv_ira | .000559 . . . .
.
-------------+--------------------------------------------------------------
--
eq2 |
pension | .7307576 .0258399 28.28 0.000 .6801123
.781403
esitaxprice | -.0503448 .1366193 -0.37 0.712 -.3181137
.2174242
[OTHER VARIABLES SUPPRESSED]
_cons | -.6282337 .0767639 -8.18 0.000 -.7786882
-.4777793
----------------------------------------------------------------------------
*
* 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/
*
* 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/
