Skip to content

Logistic regression problem

6 messages · milicic.marko, Milicic B. Marko, Frank E Harrell Jr +2 more

milicic.marko
# 
I have a huge data set with thousands of variable and one binary
variable. I know that most of the variables are correlated and are not
good predictors... but...

It is very hard to start modeling with such a huge dataset. What would
be your suggestion. How to make a first cut... how to eliminate most
of the variables but not to ignore potential interactions... for
example, maybe variable A is not good predictor and variable B is not
good predictor either, but maybe A and B together are good
predictor...

Any suggestion is welcomed
2 days later
Milicic B. Marko
# 
The only solution I can see is fitting all possib le 2 factor models enabling
interactions and then assessing if interaction term is significant...


any more ideas?
Milicic B. Marko wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Frank E Harrell Jr
    

    

      
      #
    

  


  

      
Milicic B. Marko wrote:

            

      
      
      
    

        
Please don't suggest such a thing unless you do simulations to back up 
its predictive performance, type I error properties, and the impact of 
collinearities.  You'll find this approach works as well as the U.S. 
economy.

Frank Harrell

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Jason Jones Medical Informatics
    

    

      
      #
    

  


  

      
So...I wouldn't suggest the trying all possible logistic models approach either and I'm not sure exactly what your goals are in modeling.

However, I've been fiddling around with the variable importance (varimp) functions that come with the randomForest and party packages.  The idea is to get an idea of which independent variables are likely to be useful and then to focus on those variables (identified as being of high importance) with more attention than you could spend on the whole set.

A general advantage of the recursive partitioning approach is that it deals fairly nicely with interactions and collinearity.

Theoretically, the recursive partitioning approaches should be able to deal with missing values (often a problem with large datasets), but I have been unable to apply this with the variable importance functions.

Let me know if you require more details.  You can check out http://www.biomedcentral.com/1471-2105/9/307 for a couple examples of variable importance.


Jason Jones, PhD
Medical Informatics
j.jones at imail.org
801.707.6898


-----Original Message-----
From: r-help-bounces at r-project.org [mailto:r-help-bounces at r-project.org] On Behalf Of Frank E Harrell Jr
Sent: Tuesday, September 30, 2008 2:54 PM
To: Milicic B. Marko
Cc: r-help at r-project.org
Subject: Re: [R] Logistic regression problem

        
Milicic B. Marko wrote:

            

      
      
      
    

        
Please don't suggest such a thing unless you do simulations to back up
its predictive performance, type I error properties, and the impact of
collinearities.  You'll find this approach works as well as the U.S.
economy.

Frank Harrell

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        
--
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University

______________________________________________
R-help at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

  
  


  


      

    

    

      
      

        


  

        

      Bernardo Rangel Tura
    

    

      
      #
    

  


  

      
Em S?b, 2008-09-27 ?s 10:51 -0700, milicic.marko escreveu:

            

      
      
      
    

        
milicic.marko

I think do you start with a rpart("binary variable"~.)
This show you a set of variables to start a model and the start set to
curoff  for continous variables

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Frank E Harrell Jr
    

    

      
      #
    

  


  

      
Bernardo Rangel Tura wrote:

            

      
      
      
    

        
I cannot imagine a worse way to formulate a regression model.  Reasons 
include

1. Results of recursive partitioning are not trustworthy unless the 
sample size exceeds 50,000 or the signal to noise ratio is extremely high.

2. The type I error of tests from the final regression model will be 
extraordinarily inflated.

3. False interactions will appear in the model.

4. The cutoffs so chosen will not replicate and in effect assume that 
covariate effects are discontinuous and piecewise flat.  The use of 
cutoffs results in a huge loss of information and power and makes the 
analysis arbitrary and impossible to interpret (e.g., a high covariate 
value:low covariate value odds ratio or mean difference is a complex 
function of all the covariate values in the sample).

5. The model will not validate in new data.

Frank