Hi.
I wonder if someone could spare a moment to validate my theory.
Given a logistic regression model, if the assumption that the risks interact
multiplicatively were to be violated for a given explanatory parameter xi
(i.e. the parameter influences the end result by a method different to the
other parameters, thus making the influence linear), how would you detect
this? How obvious would the flawed assumption be in the end model?
I believe that this would be quite obvious as a result of poor model fit and
a wide confidence interval for the RR for parameter xi. The wide confidence
interval would be a result of some the observed data with few parameters
other than xi requiring the RR to be large, as opposed to the observed data
with numerous parameters including xi requiring the RR to be small.
Does this sound plausible?
With thanks,
Henry Clout ([log in to unmask])
--
No virus found in this outgoing message.
Checked by AVG Anti-Virus.
Version: 7.0.300 / Virus Database: 266.4.0 - Release Date: 22/02/2005
|