survival analysis-competing events model

[Maguin, Eugene]

An analysis problem is coming up that involves program discharge to one of three states: more restrictive environment, no change (same as current program), less restrictive environment. If I ignore time to discharge, I expect to be able to use either an ordinal logistic/probit or a multinomial logistic (spss plum or nomreg). But, suppose I want to bring in time to discharge and convert the model to a competing events model. Spss has no competing events capability but Allison, I think, showed that discrete time survival could be done with using logistic regression. I was thinking that the same method could be applied when using nomreg. Am I thinking incorrectly? Can anyone offer any references for this sort of analysis?

Thanks, Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Valerie Villella]

Thank you for your email. Please note that I will be out of the office from Monday May 7 and returning Wednesday May 9, 2012. I will respond to my emails upon my return on Thursday May 10, 2012.

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Poes, Matthew Joseph]

Would this use a 2 stage least squares model then?  I just went to a talk looking at something that was functionally similar where they talked about binary, multinomial, and probit creating problems in 2 SLS models, and sure enough in their own analysis, some of the DV's analyzed showed reversed coefficients from the expected, so they switched to a 2 stage residual included model, which was found by others to provide unbiased estimates.  In their analysis it also appeared to be unbiased, as the direction of effects were congruent with theory.  My only comment might be to carefully check the directionality of coefficients for the time variables to ensure that all are consistent with what would be expected.

How much of this model likely falls into error variance?  My understanding is that the biggest challenge here is when there is a large amount of unmeasured characteristics in the residual, which can wreak havoc on the estimation of coefficients and thus any decent interpretation.  My own experience has shown that sometimes in the model building phase, coefficients change direction unexpectedly in these kinds of situations, and it takes very thoughtful consideration of the factors which may be unmeasured, and how they might cause this.  My experience has been pretty limited though.

Matthew J Poes
Research Data Specialist
Center for Prevention Research and Development
University of Illinois
510 Devonshire Dr.
Champaign, IL 61820
Phone: [phone]
email: [hidden email]


-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Maguin, Eugene
Sent: Monday, May 07, 2012 9:36 AM
To: [hidden email]
Subject: survival analysis-competing events model

An analysis problem is coming up that involves program discharge to one of three states: more restrictive environment, no change (same as current program), less restrictive environment. If I ignore time to discharge, I expect to be able to use either an ordinal logistic/probit or a multinomial logistic (spss plum or nomreg). But, suppose I want to bring in time to discharge and convert the model to a competing events model. Spss has no competing events capability but Allison, I think, showed that discrete time survival could be done with using logistic regression. I was thinking that the same method could be applied when using nomreg. Am I thinking incorrectly? Can anyone offer any references for this sort of analysis?

Thanks, Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Bruce Weaver]

Hi Gene.  I just took a look in my copy of Singer & Willett (2003) to see if they offered any suggestions.  The following is from chapter 15 (p. 592).

--- start of excerpt ---

Armed with an event-time variable and multiple censoring indicators, model fitting is easy.  All you do is fit the same model to the same data set several times, once for each censoring indicator.  Under the assumption of noninformativeness, the likelihood functions for each event are separable, so you can estimate parameters using separate, but parallel, analyses.  Although you may be tempted to include a different set of predictors in the multiple models, we caution against this approach.  Use of identical predictors increases the tenability of the noninformativeness assumption and facilitates comparison of estimates, as we describe below.  

--- end of excerpt ---

They illustrate with a data set showing tenure of Supreme Court Justices, who left office by either dying or retiring (with some still in office, of course).  The data set appears to be available via the UCLA textbook examples website:

   http://www.ats.ucla.edu/stat/spss/examples/alda/chapter15/aldaspssch15.htm

HTH.

Maguin, Eugene wrote

An analysis problem is coming up that involves program discharge to one of three states: more restrictive environment, no change (same as current program), less restrictive environment. If I ignore time to discharge, I expect to be able to use either an ordinal logistic/probit or a multinomial logistic (spss plum or nomreg). But, suppose I want to bring in time to discharge and convert the model to a competing events model. Spss has no competing events capability but Allison, I think, showed that discrete time survival could be done with using logistic regression. I was thinking that the same method could be applied when using nomreg. Am I thinking incorrectly? Can anyone offer any references for this sort of analysis?

Thanks, Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD