It does not in general make sense to eliminate a main effect
and keep an interaction in the model. This is vilolating the
"principle of marginality".
A very good exposition of this and other probelems related to
linear models can be found in Bill Venables paper
"Exegeses on linear models" to be found on:
http://www.stats.ox.ac.uk/pub/MASS3/Exegeses.pdf
Best
Bendix Carstensen
----------------------
Bendix Carstensen
Senior Statistician
Steno Diabetes Center
Niels Steensens Vej 2
DK-2820 Gentofte
Denmark
tel: +45 44 43 87 38
mob: +45 30 75 87 38
fax: +45 44 43 07 06
[log in to unmask]
www.biostat.ku.dk/~bxc
----------------------
> -----Original Message-----
> From: A UK-based worldwide e-mail broadcast system mailing
> list [mailto:[log in to unmask]] On Behalf Of Raphael Fraser
> Sent: Tuesday, October 18, 2005 1:27 AM
> To: [log in to unmask]
> Subject: QUERY: Survival Analysis
>
>
> Dear All,
>
> I am modelling time to onset of a particular disease with
> three covariates x1, x2, x3 and the interaction x1*x3 using
> Survival Analysis. All the covariates except x1 was
> significant during univariate analysis but was kept in the
> model due to its importance based on other research. In the
> final model: HR =
> exp(b1x1+b2x2+b3x3+b4x1*x3) only x1 was not significant where
> HR is hazard ratio. Any suggestions on where to go from here?
> Since x1*x3 is significant and x1 was not, How does one
> handle such a situation?
>
> Raphael
>
|