Message-Id: <9902021423.AA10340@svs.dk>
Date: Tue, 02 Feb 1999 15:22:29 +0000
To: yudi@hermes.ucd.ie
From: Bendix Carstensen <bxc@svs.dk>
Subject: Re: [R] glm null deviance
In-Reply-To: <0F6J00ICI2FY7V@hermes.ucd.ie>
At 12:52 02-02-99 , you wrote:
>Question to the devel:
> shouldn't the cloglog =3D log(-log(p))?
> rather than log(-log(1-p)) as implemented?
No cloglog is an abbreviation for Complementary log - log.
The model emerges from gruped survival:
Suppose for the sake of the argument that the failure rate is
lambda and that objects are observed over an interval of length
1. Then the probability of failure is:
=20
p =3D 1 - exp(-lambda*1)
If you want to implement a proportional hazards model, the relvant
parameter from a linear poit of view is eta=3Dlog(lambda), hnce the model:
=20
p =3D 1 - exp(-exp(eta))
or
log(-log(1-p)=3Deta
Hence the Complementary log-log.
If you want the other model just use:
glm(1-y~x,family=3Dbinomial(link=3Dcloglog))
and keep track og your signs!
Bendix Carstensen
\\\|///
\\ - - //
( @ @ )
+--------------------------o00o-(_)-o00o-----------------------+
| Bendix Carstensen |
| Statistiker tlf: 35 30 01 00 |
| Dansk Zoonosecenter dir: 35 30 01 39 |
| Statens Veterin=E6re serumlaboratorium fax: 35 30 01 20 |
| B=FClowsvej 27 e-mail: bxc@svs.dk |
| 1790 K=F8benhavn V Ooooo |
+-------------------------ooooO----( )-----------------------+
( ) ) /
\ ( (_/
\_)
-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !) To: r-help-request@stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._