On 2004-11-01, at 11.07, Joseph Coveney wrote:
Could you fit a different generalized linear model to your data? One
with
the logarithmic link instead of the logit link will give you the risk
ratio
directly. I'm not really sure what you mean by comparing the risk at
different levels of scale, but you can set each of the levels of scale
of
the covariate as the reference level for estimating the risk ratios
(and
their confidence intervals) of the other levels relative to it.
Thanks for your suggestion. It seems to be a nice solution for one set
of models however, some of the information I left out complicate it a
bit.
I have estimated a -gllamm , link(ologit) fam(binomial)- model on a
dependent variable describing events on an ordinal scale. My
independent variable of interest is a scale with a linear and a
quadratic term. I have also used the -thresh()- option to relax the
proportional odds assumption of the linear component of the scale.
Despite the above, because I'm only interested in one level of the
dependent variable (the ordinal model is only used to maximise the use
of information in data i.e. the event of interest is "borrowing
strength" from other events) and the model can be reduced to a simple
logit with one constant and two covariates (linear and quadratic).
What I want to do is to produce a table with absolute risks
(probabilities) Relative Risk Ratios (RRR), and confidence intervals
(CI) for all combinations of the scale (my independent variable) which
would give 10*10 RRRs. I also want to produce tables where I have added
1 and 2 standard deviation of the estimated random intercept to the
model, describing subjects with different event propensities.
I assume that if I predicted the probability score with CI, and
calculated RRRs from the upper CIs as one limit and RRRs from the lower
CIs as the other limit, that would not be correct? Or would it?
Any suggestions?
Thanks,
Michael
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
