.-
help for ^aucsize^ (version 1.1)
.-
Power for one-sample AUC
------------------------
^aucsize^, ^mdb(^#^) sdb(^#^) md(^#^) sd(^#^) nd(^#^) ndb(^#^)^
[^auc^null^(^#^)^ ^a^lpha^(^#^)^ ^ns^im^(^#^)^ ^de^long ^mp^ ^lo^git
^d^ots ^nodis^play ^res^file^(^filename^)^ ^replace^ ]
Description
-----------
^aucsize^ determines power to detect AUC improvement relative to a specified
null. The AUC test distribution is based on the non-parametric AUC
estimate and s.e. for test value samples drawn from specified normal
distributions for ^nd^ diseased ( ~N(^md^,^sd^) ) and ^ndb^ non-diseased
( ~N(^mdb^,^sdb^) ) subjects. Rejection of the null AUC for each sample
is based on the lower one-sided 1-^alpha^ level confidence bound for the
sample AUC estimate.
returned results:
r(power_W) - power based on Wald interval rejection
r(power_L) - power based on logit(auc)-interval rejection (if specified)
Options
-------
^aucnull(^#^)^ specifies the null AUC. The default is 0.5 if not specified.
^nsim(^#^)^ specifies the number of simulated study samples to be drawn for
determination of power. The default is 500 samples if not otherwise
specified.
^alpha(^#^)^ specifies the type I error rate. the default is
1 - ($S_level/100.), i.e. .05 if the level has not been set. See @level@
^delong^ specifies that the AUC standard error for constructing the lower
confidence bound be calculated by the method of DeLong, Delong, and
Clarke-Pearson. The default method is that of Hanley and McNeil.
The AUC and standard error estimates are obtained from @roctab@ in either
case.
^mp^ specifies that the AUC standard error be calculated using MP's version of the
DeLong, Delong, Clarke-Pearson estimator. This is the method used in @emroc@.
The default method is that of Hanley and McNeil if neither ^mp^ nor ^delong^
are specified.
^logit^ includes power calculations using the logit(auc) based lower confidence
bound in addition to the default Wald interval based calculations.
^nodisplay^ suppresses output display (in case where only returned results are
desired).
^dots^ optionally displays a dot on the screen for each simulated sample drawn.
^resfile(^filename^)^ creates a Stata file (^.dta^ file) with the simulation
results.
^replace^ requests that if an existing file is specified for saving simulation
results it should be overwritten.
Remarks
-------
Persons interested in reproducibility of results should set the random-number
seed by typing set seed # before running ^aucsize^; see help @generate@.
Setting the seed prior to otherwise identical runs using the Hanley-McNeil vs
DeLong estimators will not result in the same sequence of samples (beyond the
first sample) - I suspect this has something to do with -roctab- using the
random number generator for one or the other options?
Specifying the DeLong standard error estimator may result in very long
computation times if large sample sizes are specified, as the -@roctab@-
subroutine for the DeLong standard error computation includes a loop across
observations.
References
----------
Author
------
Gary Longton, Fred Hutchinson Cancer Research Ctr.
glongton@@fhcrc.org
Also see
--------
On line: help for @roctab@, @emroc@ if installed,
@aucbs@ if installed, @rocsize@ if installed, @scrsize@ if installed,
@bsample@ or @bstat@, @postfile@.