Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: bootstrapping problem: difference between two numbers


From   "G. ter Riet" <[email protected]>
To   [email protected]
Subject   st: bootstrapping problem: difference between two numbers
Date   Wed, 23 Jul 2008 08:24:36 +0200

Dear Stata listers,
I am trying to work out how much it matters if people use different definitions of asthma in prediction models. To get rid of any effects of different choices of study populations and study designs, we perform a number of logistic regression models that use different definitions of the dependent variable (asthma) and a fixed set of predictors in a single database (= 1 population and 1 study design) that we have access to. I think I successfully bootstrapped the areas under the ROC curve of two models like this (using the auto.dta data set for illustration):

clear
sysuse auto
*GENERATE THREE BINARY DEPENDENT VARS TO MIMICK OUR THREE ASTHMA DEFINITIONS
g price_d1=(price>5000)
*g price_d2=(price>6342)
g price_d3=(price>10000)
*PROGRAM TWO LOGISTIC REGRESSIONS CONTAINING THE SAME XVARS BUT TWO DIFFERENT DEPVARS
capture program drop gs_arcade_auc
program define gs_arcade_auc, rclass
	version 10
	args y1 x1 y2 x2  
	confirm var `y1'
	confirm var `x1'
	confirm var `y2'
	confirm var `x2'
	tempname auc1 auc2
	logit `y1' `x1'
	lroc, nograph
	scalar `auc1' = r(area)
	logit `y2' `x2'
	lroc, nograph	
	scalar `auc2' = r(area)
	return scalar dif1=`auc1'-`auc2'
end

#delimit ;
bootstrap dif1=r(dif1),saving(c:\temp\gs_arcade_auc, replace) seed(2345)
reps(1000) nowarn nodots: gs_arcade_auc price_d1 weight price_d3 weight
;
estat bootstrap, all

However, when I try to bootstrap a quantity that I think clinicians are more interested in than the area under the ROC curve, namely, the difference in the number of patients above some fixed (treatment) threshold, the program won't run. I can't figure out why not. What would be a good way to program this? I used <count> since <predict> seems to save nothing in r(). Any help is greatly appreciated. Here is the code that won't run:

capture program drop gs_arcade_thresh
program define gs_arcade_thresh, rclass
	version 10
	args y1 x1 y2 x2  
	confirm var `y1'
	confirm var `x1'
	confirm var `y2'
	confirm var `x2'
	tempname treat1 treat2
	logit `y1' `x1'
*COUNT THE NUMBER OF PATIENTS WHOSE PREDICTED PROBABILITY IS GREATER THAN .75
	count if 1/(1+exp(-(_b[`_cons']+_b[`x1']*`x1')))>.75
	scalar `treat1' = r(N)
	logit `y2' `x2'
	count if 1/(1+exp(-(_b[`_cons']+_b[`x2']*`x2')))>.75
	scalar `treat2' = r(N)

	return scalar dif1=`treat1'-`treat2'

end

#delimit ;
bootstrap dif1=r(dif1),saving(c:\temp\gs_arcade_thresh, replace) seed(2345)
reps(100): gs_arcade_thresh price_d1 weight price_d3 weight
;

 
Many thanks,
Gerben ter Riet, Ac Medical Center, Amsterdam, The Netherlands
begin:vcard
n:ter Riet;Gerben
fn:Gerben ter Riet
tel;fax:+31 20 5669186
tel;work:+31 20 5664640
org:Academic Medical Center;General Practice
adr:;;Meibergdreef 15, Room J2-118;Amsterdam;;1105 AZ;The Netherlands
version:2.1
email;internet:[email protected]
title:Associate Professor
end:vcard



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index