[R] classification table in logistic regression

[Frank E Harrell Jr]

Peter Flom wrote:
> Frank E Harrell Jr <f.harrell at vanderbilt.edu> wrote
> 
>> Armida,
>>
>> I regret putting CTABLE as an option on the old SAS PROC LOGIS which was 
>> a basis for PROC LOGISTIC.  Classification tables are arbitrary and 
>> misleading so I would stay away from them.
>>
>> You might build a model with and without the variable of interest and 
>> plot the two predicted probabilities against each other for more insight 
>> than what is provided by a classification table.
>>
> 
> Frank, 
> 
> I agree with 'arbitrary' but why 'misleading'?  Purely for the same sorts
> of reasons that histograms can be misleading, or for other reasons, as well?

Hi Peter,

I have an example where adding a very important variable to a model 
makes the classification accuracy worsen (it's attached if such 
attachments are allowed).   In many other cases such an accuracy 
assessment moves in the right direction but is too insensitive to real 
information.  Its statistical power is not competitive and it gets us 
too much in the habit of dichotomization.  Classification is also 
misleading in that it makes the user implicitly use a utility or loss 
function that is nonsense.

Frank

> 
> Plotting the probabilities is an excellent idea, I recommend it often, in a variety
> of situations.
> 
> Peter
> 
> Peter L. Flom, PhD
> Statistical Consultant
> www DOT peterflomconsulting DOT com
> 


-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ctable.png
Type: image/png
Size: 107923 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-help/attachments/20090604/7411ad57/attachment-0002.png>