From: Christoph Buser <buser_at_stat.math.ethz.ch>

Date: Thu 21 Jul 2005 - 01:06:41 EST

Christoph Buser <buser@stat.math.ethz.ch> Seminar fuer Statistik, LEO C13

http://stat.ethz.ch/~buser/

R-help@stat.math.ethz.ch mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Thu Jul 21 01:19:10 2005

Date: Thu 21 Jul 2005 - 01:06:41 EST

Dear Andy

Since the power of the t-test decreases when there are
discrepancies in the data to the normal distribution and there
is only a small loss of power if the data is normal distributed,
the only reason to use the t.test is his simplicity and the
easier interpretation.

Generally I'd prefer the wilcoxon test even if the data is
normal distributed.

But I agree that for a gamma distribution there is no huge loss
of power.

## example simulation:

n <- 1000

z1 <- numeric(n)

z2 <- numeric(n)

## gamma distribution

for (i in 1:n)

{

x<-rgamma(40, 2.5, 0.1)

y<-rgamma(40, 3.5, 0.1)

z1[i]<-t.test(x, y)$p.value

z2[i]<-wilcox.test(x, y)$p.value

}

## Power

1 - sum(z1>0.05)/1000 ## 0.71

1 - sum(z2>0.05)/1000 ## 0.76

##################################

## t distribution

for (i in 1:n)

{

x<-rt(40, df = 3)

y<-1 + rt(40, df = 3)

z1[i]<-t.test(x, y)$p.value

z2[i]<-wilcox.test(x, y)$p.value

}

## Power

1 - sum(z1>0.05)/1000 ## 0.76

1 - sum(z2>0.05)/1000 ## 0.91

##################################

## normal distribution

for (i in 1:n)

{

x<-rnorm(40, 0, 3)

y<-1 + rnorm(40, 1, 3)

z1[i]<-t.test(x, y)$p.value

z2[i]<-wilcox.test(x, y)$p.value

}

## Power

1 - sum(z1>0.05)/1000 ## 0.83

1 - sum(z2>0.05)/1000 ## 0.81

Regards,

Christoph Buser

Christoph Buser <buser@stat.math.ethz.ch> Seminar fuer Statistik, LEO C13

ETH (Federal Inst. Technology) 8092 Zurich SWITZERLAND phone: x-41-44-632-4673 fax: 632-1228

http://stat.ethz.ch/~buser/

Liaw, Andy writes:

> > From: Christoph Buser

* > >
** > > Hi
** > >
** > > "t.test" assumes that your data within each group has a normal
** > > distribution. This is not the case in your example.
** >
** > Eh? What happen to the CLT?
** >
** > Andy
** >
** >
** > > I would recommend you a non parametric test like "wilcox.test" if
** > > you want to compare the mean of two samples that are not normal
** > > distributed.
** > > see ?wilcox.test
** > >
** > > Be careful. Your example produces two gamma distributed samples
** > > with rate = 10, not scale = 10.
** > > rate = 1/scale.
** > > If you want to use scale, you need to specify this argument
** > > x<-rgamma(40, 2.5, scale = 10)
** > > see ?rgamma
** > >
** > > I do not see the interpretation of your result. Since you do
** > > know the distribution and the parameters of your sample, you
** > > know the true means and that they are different. It is only a
** > > question of the sample size and the power of your test, if this
** > > difference is detected.
** > > Is that something you are investigating? Maybe a power
** > > calculation or something similar.
** > >
** > > Regards,
** > >
** > > Christoph Buser
** > >
** > > --------------------------------------------------------------
** > > Christoph Buser <buser@stat.math.ethz.ch>
** > > Seminar fuer Statistik, LEO C13
** > > ETH (Federal Inst. Technology) 8092 Zurich SWITZERLAND
** > > phone: x-41-44-632-4673 fax: 632-1228
** > > http://stat.ethz.ch/~buser/
** > > --------------------------------------------------------------
** > >
** > > pantd@unlv.nevada.edu writes:
** > > > Hi R Users
** > > > I have a code which I am running for my thesis work. Just
** > > want to make sure that
** > > > its ok. Its a t test I am conducting between two gamma
** > > distributions with
** > > > different shape parameters.
** > > >
** > > > the code looks like:
** > > >
** > > > sink("a1.txt");
** > > >
** > > > for (i in 1:1000)
** > > > {
** > > > x<-rgamma(40, 2.5, 10) # n = 40, shape = 2.5, Scale = 10
** > > > y<-rgamma(40, 2.8, 10) # n = 40, shape = 2.8, Scale = 10
** > > > z<-t.test(x, y)
** > > > print(z)
** > > > }
** > > >
** > > >
** > > > I will appreciate it if someone could tell me if its alrite or not.
** > > >
** > > > thanks
** > > >
** > > > -dev
** > > >
** > > > ______________________________________________
** > > > R-help@stat.math.ethz.ch mailing list
** > > > https://stat.ethz.ch/mailman/listinfo/r-help
** > > > PLEASE do read the posting guide!
** > > http://www.R-project.org/posting-guide.html
** > >
** > >
** > > ______________________________________________
** > > R-help@stat.math.ethz.ch mailing list
** > > https://stat.ethz.ch/mailman/listinfo/r-help
** > > PLEASE do read the posting guide!
** > > http://www.R-project.org/posting-guide.html
** > >
** > >
** > >
** >
** >
** >
** > ------------------------------------------------------------------------------
** > Notice: This e-mail message, together with any attachments, contains information of Merck & Co., Inc. (One Merck Drive, Whitehouse Station, New Jersey, USA 08889), and/or its affiliates (which may be known outside the United States as Merck Frosst, Merck Sharp & Dohme or MSD and in Japan, as Banyu) that may be confidential, proprietary copyrighted and/or legally privileged. It is intended solely for the use of the individual or entity named on this message. If you are not the intended recipient, and have received this message in error, please notify us immediately by reply e-mail and then delete it from your system.
** > ------------------------------------------------------------------------------
*

R-help@stat.math.ethz.ch mailing list

https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html Received on Thu Jul 21 01:19:10 2005

*
This archive was generated by hypermail 2.1.8
: Fri 03 Mar 2006 - 03:33:51 EST
*