[R-sig-ME] logistic model with exponential decay

[David Duffy]

On Sun, 21 Dec 2008, Stijn Ruiter wrote:

> Hi,
> Thanks David, I should look into that, but I usually estimate discrete-time 
> models which are basically logistic regression models on person-years files 
> with only the records included that are "at risk". Anyway, do these packages 
> also allow for nonlinear effects the way I need them?
> Stijn
>

Well, your setup is pretty similar to "standard" bivariate survival 
problems, with your "nonlinear term" being simply time since the first 
event (the father's offence) -- there is a correlation between years at 
risk of the index person and years since his/her father offended. Your 
assumption of an exponential decay is going to induce a particular 
survivor curve.  That is, if you included number of years between the 
paternal offence and birth of the index person as a 
non-time-varying covariate, and instead changed the survival model from a 
non-proportional-hazards to say an accelerated failure time model, you may 
model the results just as well.

My $0.02, cheers, David Duffy.

-- 
| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v