I am trying to write a d0 maximum likelihood code for
estimating a split-population duration model with
gamma unobserved individual heterogeneity and
log-logistic hazard. The estimation strategy finds
first initial values from streg and set the initial
probability of never fail to 2%. Then the ml code is
used. I already checked that my log-likelihood is
correctly written and the ml check is ok. But still
Stata finds that the log-likelihood function is not
concave and does not converges. Having this in mid,
and in the spirit to find my potential mistake, I
tried to create a d0 code for the standard duration
model with gamma unobserved individual heterogeneity
from lloghet_glf. Basically I did no changes to the
code but take away the part that refers to d1 and d2
methods. Running this alternative program I learned
that Stata does the same: it keeps saying that the
log-likelihood function is not concave and it does not
converges. Hence, it seems that the problem is not in
the d0 code. Why is this is happening? Is it that a d1
(or maybe a d2) method is required whenever a
log-logistic model is to be estimated?
Could you give me some hint on what further action
should I take?
I sending to you a do file with the simplified version
of the lloghet_glf code so that you can see what is my
problem.
Thank you,
ALFONSO MIRANDA
University of Warwick
-----------------Do-File------------------------------
use http://www.stata-press.com/data/r7/cancer, clear
set more off
stset studytime, fail(died)
streg drug age, d(logl)
matrix B0 = e(b)
matrix cr = -1
matrix colnames cr = ln_the:_cons
matrix B1=(B0,cr)
capture program drop loglgaspl_ll
program define loglgaspl_ll
version 7
args todo b lnf
tempvar beta lngamma lntheta
mleval `beta' = `b', eq(1)
mleval `lngamma' = `b', eq(2) scalar
mleval `lntheta' = `b', eq(3) scalar
local t = _t
local d = _d
quietly {
scalar `lngamma'=cond(`lngamma'<-20,-20,`lngamma')
scalar `lntheta'=cond(`lntheta'<-20,-20,`lntheta')
tempname th ga
tempvar a b c
scalar `ga' = exp(-`lngamma') /* gamma really 1/gamma
*/
scalar `th' = exp(`lntheta')
gen double `a' = 1/`th'+`d' if $ML_samp
gen double `b' = 1+(`t'*exp(-`beta'))^`ga' if $ML_samp
gen double `c' = ln(1+`th'*ln(`b')) if $ML_samp
mlsum `lnf' = -`a'*`c' + `d'*(`ga'*(ln(`t') - /*
*/ `beta') - `lngamma' - ln(`b'))
if `todo'==0 | `lnf'==. { exit }
}
end
ml model d0 loglgaspl_ll (_t: _t = drug age) (ln_gam:)
(ln_the:)
ml init B1
ml maximize, search(off)
exit
-----------------------------------------------------
__________________________________________________
Do You Yahoo!?
Yahoo! Finance - Get real-time stock quotes
http://finance.yahoo.com
*
* 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/
