Re: [R] BMA, logistic regression, odds ratio, model reduction etc

2011-04-21 Thread khosoda
Thank you for your comment. I forgot to mention that varclus and pvclust showed similar results for my data. BTW, I did not realize rms is a replacement for the Design package. I appreciate your suggestion. -- KH (11/04/21 8:00), Frank Harrell wrote: I think it's OK. You can also use the Hmi

Re: [R] BMA, logistic regression, odds ratio, model reduction etc

2011-04-20 Thread Frank Harrell
I think it's OK. You can also use the Hmisc package's varclus function. Frank 細田弘吉 wrote: > > Dear Prof. Harrel, > > Thank you very much for your quick advice. > I will try rms package. > > Regarding model reduction, is my model 2 method (clustering and recoding > that are blinded to the out

Re: [R] BMA, logistic regression, odds ratio, model reduction etc

2011-04-20 Thread khosoda
Dear Prof. Harrel, Thank you very much for your quick advice. I will try rms package. Regarding model reduction, is my model 2 method (clustering and recoding that are blinded to the outcome) permissible? Sincerely, -- KH (11/04/20 22:01), Frank Harrell wrote: Deleting variables is a bad i

Re: [R] BMA, logistic regression, odds ratio, model reduction etc

2011-04-20 Thread Frank Harrell
Deleting variables is a bad idea unless you make that a formal part of the BMA so that the attempt to delete variables is penalized for. Instead of BMA I recommend simple penalized maximum likelihood estimation (see the lrm function in the rms package) or pre-modeling data reduction that is blinde

[R] BMA, logistic regression, odds ratio, model reduction etc

2011-04-20 Thread khosoda
Hi everybody, I apologize for long mail in advance. I have data of 104 patients, which consists of 15 explanatory variables and one binary outcome (poor/good). The outcome consists of 25 poor results and 79 good results. I tried to analyze the data with logistic regression. However, the 15 variabl