Good day,
I was performing a power analysis of articles published in a journal of
management using the pwr package in R. However, it seemed to be impossible to
compute power for small, medium and large Effect Size for multiple ordinal
logistic regression output. I have tried using G*power, but it seemed to be
just useful when we have simple logistic regression output. Thus, I have tried
to simulate to calculate the power based on Greg Snow and Gung answers.
library(rms)
tmpfun <- function(n, beta1,beta2,beta3,beta4,beta5,beta6,beta7,
beta8,beta9,beta10,beta11,beta12,beta13,beta14) {
var1=log(runif(n,3,33))
var2=rbinom(n,1,0.4502)
x=1
while (x>0.96) {var3=rbinom(n,1,0.9474); x=mean(var3)}
var4=rbinom(n,1,0.2749)
var5=rbinom(n,1,0.6199)
var6=rbinom(n,1,0.4327)
var7=rnorm(n,1.11,0.29)
var8=rnorm(n,0.06,0.07)
var9=runif(n,106,536804)
var10=runif(n,0.08,3.7)
var11=runif(n,0.08,0.92)
var12=rnorm(n,96.24,113.11)
for (i in 1:n) {if (var12[i]<=0) {var12[i]=runif(1)} else
{var12[i]=log(var12[i])}}
var13=runif(n,0.05,0.88)
var14=runif(n,-1.06,0.03)
eta1 <- beta1*var1+beta2*var2+beta3*var3+beta4*var4+beta5*var5+
beta6*var6
eta2=eta1+beta7*var7+beta8*var8+beta9*var9+beta10*var10
+beta11*var11+beta12*var12+beta13*var13+beta14*var14
p1 <- exp(eta1)/(1+exp(eta1))
p1=replace(p1,is.na(p1),1)
p2 <- exp(eta2)/(1+exp(eta2))
p2=replace(p2,is.na(p2),1)
tmp <- runif(n)
y <- (tmp < p1)+(tmp<p2)
fit <- lrm(y~var1+var2+var3+var4+var5+var6)
fit$stats[5]
}
out <- replicate(10000, tmpfun(n=112,0.582,-0.176,-0.413,
-0.138,0.861,-0.942,3.481,4.542,1.059,0.322,-2.562,0.599,2.4,0.732))
mean( out < 0.05 )
Is my simulation right? How can I get power for small, medium and large ES?
Those are the questions... Thank for your time and your help.
Regards,
Adrien.
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.
