Re: st: Speed of ROC analysis

[Gary Longton <glongton@fhcrc.org>]

Shaw, Jim (NIH/NCI) wrote:

> Has anyone developed faster versions of roctab or roccomp (using DeLong et
> al.'s estimated standard errors)?  The versions available in Stata 8 (as
> well as in earlier versions of Stata) run very slowly when used with large
> samples (e.g., 500-1000 observations).  For example, in simulations I have
> been performing, it generally takes about 24 hours to complete 1000
> repetitions of roccomp in samples consisting of 1000 observations each.  I
> suspect the problem is related to the fact that all possible pairwise
> comparisons are being made between scores for true-positive and
> true-negative subjects.  
For the single test variable case (-roctab-), I have written an 
alternative (-emroc-) which should calculate faster AUC se's (per 
DeLong, Delong, and Clark-Pearson) though the SE calculation does not 
include the correction for ties that -roctab- does, and does not include 
other SE alternatives.  I believe the DeLong SE calculation in -roctab- 
is bogged down because it loops through observations.

I've never posted/submitted -emroc- to the SSC because it was in many 
respects redundant of -roctab- (the first version of -emroc- was written 
at about the time when Stata and Mario Cleves made -roctab- and the 
corresponding suite of other ROC programs available).

If interested in the single test case, the emroc.ado and emroc.hlp files 
can be obtained from:

      http://www.fhcrc.org/labs/pepe/book/

the .toc and .pkg files necessary for automatic installation are 
unfortunately not yet there, but you can download the files in any case.

- Gary





*
*   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/