Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
st: Problems with ml max Survival analysis
From
<[email protected]>
To
<[email protected]>
Subject
st: Problems with ml max Survival analysis
Date
Fri, 8 Nov 2013 09:39:48 +0000
Working through the details of Amparo's -ml- code is beyond me right now, but I notice that the output reported for gradient length seems huge: " Gradient length = 9.21e+09". Is this a signal that something is wrong in the code?
Consider the following:
* what happens if you use numerical derivatives (i.e. without the -mlvecsum- etc gradient calculation)? Principle: start simple and get something working, and then complicate it. Following the same principle, use a smaller set of predictors, and later add others when you're sure the model is working.
* what does -ml check- show?
* What model are you actually trying to fit? Refer to a paper (full reference please). Also, what is the nature of your survival time data (continuous/discrete; what is the nature of the competing risks; etc.) Someone may already have code for the model, or a related one, and have tips about how to approach fitting it.
Stephen
------------------
Stephen P. Jenkins <[email protected]>
------------------------------
Date: Thu, 07 Nov 2013 11:43:27 +0100
From: "Amparo Nagore Garcia" <[email protected]>
Subject: st: Problems with ml max Survival analysis
Dear all,
I am trying to maximize my own likelihood function and I get the coefficients but I do not get any other output(Std. Err, z, P>Z.....)
Please, could you help me with this? Is something wrong in my program???.
Just below you could find my ml model, my program and the results that I get.
Thank you in advance.
Amparo
The objective function is the L= π h(ti|xi)^di S(ti|xi)
********************
my ml model
*******************
ml model d1 CR_stable (cr_05_stable: t0_s t_s d_s= tp* $personal $labour ,nocons), cluster(`ID')
ml max, difficult
where:
global personal "u_rate male h_skill m_skill non_manual municipio spanish_speakers no_spanish_speaker Aged_16_19 Aged_20_24 Aged_25_29 Aged_30_34 Aged_35_39 Aged_40_44 Aged_45_51 older61 responsabilities"
global labour "construction industry size_0 size_10_19 size_20_49 size_50_249 size_250 discontinuous fix_term on_call duration_1 coefte_num_1 prest_emp"
global ID "id_bis"
**************************************
The program is the following:
**************************************
capture program drop CR_stable
****Competing risk without uh
program CR_stable
version 11.2
args todo b lnf g
***Declare and define the arguments of the LL
tempvar beta1
mleval `beta1'=`b', eq(1)
local dt0_1="$ML_y1"
local dt1_1="$ML_y2"
local d1="$ML_y3"
tempvar inthaz1 last p1
*tempname p1
tempvar sumb1
sort contadorbis
by contadorbis: gen double `sumb1'=sum(`d1'*`beta1') if $ML_samp
tempvar haz1
by contadorbis: gen double `haz1'=exp(`sumb1'[_N]) if $ML_samp
by contadorbis: gen double `inthaz1'=sum(exp(`beta1'*(`dt1_1'-`dt0_1'))) if $ML_samp
by contadorbis: gen double `p1'= `haz1'*exp(-`inthaz1'[_N]) if $ML_samp
by contadorbis: gen byte `last'=(_n==_N)
mlsum `lnf'= ln(`p1') if `last'==1
*Calculate the gradient,I have to supply dlnl/dtheta and mlvecsum returns dlnL/dbetai=sum dlnl/dtheta*Xij
if(`todo'==0 | `lnf'>=.) exit
mlvecsum `lnf' `g' = `d1'-(`dt1_1'-`dt0_1'), eq(1)
end
**********
the results
**********
. ml max, difficult
initial: log likelihood = -204354.53
rescaling entire vector +++.
rescale: log likelihood = -42637.678
- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Iteration 0:
log likelihood = -42637.678
Gradient length = 9.21e+09
Step length = 0
Parameters + step -> new parameters
log likelihood = -42637.678
(initial step good)
(1) Stepping forward, step length = 0
log likelihood = -42637.678
(ignoring last step)
- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Iteration 1:
log likelihood = -42637.678
g inv(H) g' = 0
Gradient length = 9.21e+09
- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Number of obs = 658755
Wald chi2(0) = .
Log likelihood = -42637.678 Prob > chi2 = .
- ------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
- -------------+----------------------------------------------------------------
tp1 | -.81 . . . . .
tp2 | -.79875 . . . . .
tp3 | -.79125 . . . . .
tp4 | -.81125 . . . . .
tp5 | -.81375 . . . . .
tp6 | -.82125 . . . . .
tp7 | -.8125 . . . . .
tp8 | -.86125 . . . . .
tp9 | -.85625 . . . . .
tp10 | -.87375 . . . . .
tp11 | -.8875 . . . . .
tp12 | -.88 . . . . .
tp13 | -.8825 . . . . .
tp14 | -.89625 . . . . .
tp15 | -.9025 . . . . .
tp16 | -.89875 . . . . .
tp17 | -.915 . . . . .
tp18 | -.93 . . . . .
tp19 | -.92625 . . . . .
tp20 | -.93625 . . . . .
tp21 | -.96375 . . . . .
tp22 | -.92625 . . . . .
tp23 | -.925 . . . . .
tp24 | -.9475 . . . . .
tp25 | -.82375 . . . . .
tp26 | -1.12 . . . . .
tp27 | -1.00875 . . . . .
tp28 | -1.15375 . . . . .
u_rate | -.469875 . . . . .
male | .014625 . . . . .
h_skill | .01775 . . . . .
m_skill | .000375 . . . . .
non_manual | .0035 . . . . .
municipio | .00075 . . . . .
spanish_sp~s | -.01375 . . . . .
no_spanish~r | -.03125 . . . . .
Aged_16_19 | -.00875 . . . . .
Aged_20_24 | .026 . . . . .
Aged_25_29 | .04 . . . . .
Aged_30_34 | .041 . . . . .
Aged_35_39 | .04325 . . . . .
Aged_40_44 | .045 . . . . .
Aged_45_51 | .042875 . . . . .
older61 | -.0725 . . . . .
responsabi~s | .015625 . . . . .
construction | .02325 . . . . .
industry | .006 . . . . .
size_0 | -.00375 . . . . .
size_10_19 | .012125 . . . . .
size_20_49 | .0215 . . . . .
size_50_249 | .029 . . . . .
size_250 | .02475 . . . . .
discontinu~s | .223 . . . . .
fix_term | .062625 . . . . .
on_call | .115625 . . . . .
duration_1 | -3.88e-06 . . . . .
coefte_num_1 | .002375 . . . . .
prest_emp | -.08625 . . . . .
- ------------------------------------------------------------------------------
.
end of do-file
Please access the attached hyperlink for an important electronic communications disclaimer: http://lse.ac.uk/emailDisclaimer
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/