Re: st: RRR with CI from logit model

[Michael Ingre <Michael.Ingre@ipm.ki.se>]

On 2004-11-02, at 16.05, Leonelo Bautista wrote:

> On the other hand, I fail to see why you want to calculate relative 
> risks
> from a logistic model.
I have estimated an ordinal (relaxed proportional odds) model with 
-gllamm- and I want to use the estimated parameters to predict 
probabilities and estimate RRRs comparing different levels of my 
independent variable. Despite the ordinal approach, after estimating 
the coefficients, my problem can be reduced to a simple logit with two 
covariates and a constant to predict the outcome. As far as I can tell, 
it is not possible to estimate my model with a log link. But I will 
appreciate any suggestions.

Predicting probabilities (with confidence intervals) are straight 
forward. Calculating the point estimate of RRR is also no problem. 
However, I also want to calculate confidence intervals.

If I use -nlcom- , I get the point estimate and confidence intervals 
(CI) but with the paradoxical result that small (very small) 
differences in risks are significant with narrow CIs and large 
differences in risks are NOT significant with HUGE CIs. The paradox can 
also be illustrated with auto.dta. Try it to see for your self (code 
was posted earlier: 
http://www.stata.com/statalist/archive/2004-11/msg00049.html).

On 2004-11-02, at 17.26, Ron�n Conroy wrote:


> Relative risks make so much more sense than odds ratios
I totally agree. That's why I'm trying to estimate RRRs with my logit 
model. RRRs make a nice addition to the predicted probabilities. I also 
have the objective to show that individuals differ in their subject 
specific RRRs because of differences in overall propensity (the random 
intercept).

I appreciate any suggestions.

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/