AW: st: Cox Competing Risks Model (Cox CRM)

[Stefan Göke <stefan.goeke@whu.edu>]

Steve,

many thanks. I solved it now. I was not aware of stratifying commands like "stcox stratum#(c.sex c.thick), strata(stratum)" where you really get one coefficient per risk type in the output. Adding the small constant to break the ties and then clustering by id as you suggested works fine in my eyes. Thanks!

May be for the sake of others: I can only recommend "stcompadj" by Enzo Coviello available in the SSC archives. It contains a very nice explanation how to implement the data augmentation method.

Best

Stefan

-----Ursprüngliche Nachricht-----
Von: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von Steven Samuels
Gesendet: Donnerstag, 28. Juli 2011 05:31
An: statalist@hsphsun2.harvard.edu
Betreff: Re: st: Cox Competing Risks Model (Cox CRM)


If you use the stratified version of the Lum/McNeil model, as I suggested, then the log-likelihood of the combined model will be the total of the separate log-likelihoods. Notice the coefficients of the third are those in the first two.  


Steve
sjsamuels@gmail.com


*************************************
sysuse auto, clear

eststo clear
stset mpg
eststo M1: qui  stcox turn length if foreign==0, nohr eststo M2: qui  stcox turn length if foreign==1, nohr eststo M3: qui stcox c.turn#ibn.foreign c.length#ibn.foreign, nohr strata(foreign) estadd scalar Log_Likelihood =e(ll): M1 M2 M3 estout M1 M2 M3,  cells(b se) stats(Log_Likelihood)

di    -131.9868  +  -42.78481   

esttab M1 M2 M3,  title(Separate & Combined Models) label nonumbers mtitles("Domestic" "Foreign" "Combined") stats(Log_Likelihood)
********************************

On Jul 27, 2011, at 8:46 AM, Stefan Göke wrote:

Dear statalisters,

I am aware that a competing risks can be implemented e.g. via a separate Cox models or via augmenting the data before applying a Cox model. 

Reading “He et al. (2010) A Competing Risks Analysis of Corporate Survival in Financial Management” and “Dickerson et al. (2003) Is attack the best form of defence? A competing risks analysis of acquisition activity in the UK in Cambridge Journal of Economics”, I am curious to know whether somebody on the list knows how to implement those models in Stata. In fact, the authors present results for three competing risk types where each covariate has a separate coefficient per risk type. However, LL, Chi2 are only presented once, implying that it is only one model the authors estimate. The only way I know to do this is via three separate models that would have one LL, Chi2 etc. each.

Here is an excerpt of table 4 p. 1713 from He et al (2010) to illustrate my question. Their three risk types are bankruptcy, M&A, privatized. –LL is only reported once.

Cox CRM Model Without Heterogeneity	
	Bankruptcy	M&A	Privatized	
Profitability	−3.0515∗∗∗	1.1989	4.9439∗∗∗	
Growth	−0.9699	0.3095∗∗	0.3105∗∗∗	
Size	0.5173∗∗∗	0.194	0.5300∗∗∗	
Turnover (market)	−0.0051	−0.0011∗∗	−0.0019∗∗	
Turnover (stock)	−0.2964	0.0361∗∗∗	0.0042	
Stock performance	−24.74∗∗∗	12.14∗∗	−5.928	
FCF	0.3675	10.9	15.02∗∗∗	
Tax	31.58∗	26.49	−5.8077	
q-ratio	0.1767∗∗∗	0.1231	−0.4449∗∗∗	
Leverage	−0.1151	−1.3721	−1.3365∗	
Investment	−0.3223	−1.4312	0.0621	
				
- ln L 458.44				


Many thanks for your consideration

Stefan

--
Stefan Göke
Ph.D. Candidate
Department of Management
University Paderborn



*
*   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/


*
*   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/