Glitch in Kruskal-Wallis test?

... Moreover, you should not analyze proportions in this way, which treats
.5 = 2/4 or .5 = 2000/4000 identically. As David said, you need to work
with a statistician.

Bert Gunter

"The trouble with having an open mind is that people keep coming along and
sticking things into it."
-- Opus (aka Berkeley Breathed in his "Bloom County" comic strip )


On Sat, Dec 22, 2018 at 7:32 AM David L Carlson <dcarlson at tamu.edu> wrote:

You may need to spend some more time with the statistician who needs to
see your data. It is not clear if you have a two sample test or a paired
sample test. Kruskall-Wallis expects data for each observation, not grouped
data. Without the observations, the test cannot compute the sample size and
the degrees of freedom. You have run kruskal.test separately on each
sample. The kruskal.test is designed for comparing two or more samples.

----------------------------------------
David L Carlson
Department of Anthropology
Texas A&M University
College Station, TX 77843-4352

-----Original Message-----
From: R-help <r-help-bounces at r-project.org> On Behalf Of Jenny Liu
Sent: Saturday, December 22, 2018 7:10 AM
To: Michael Dewey <lists at dewey.myzen.co.uk>
Cc: r-help at r-project.org
Subject: Re: [R] Glitch in Kruskal-Wallis test?

Hi Michael,

Thank you for your reply! I'm testing the difference in proportions. Temp
is temperature, and Prop is the proportion of insect pupae that survived
at
that temperature. I was told by a statistician that the K-W was
appropriate
for testing proportions, but perhaps you know of an alternative? I have
already tested for heteroscedasticity using the Breusch-Pagan test.

Thanks again,
Jenny



On Dec 22, 2018 7:38 AM, "Michael Dewey" <lists at dewey.myzen.co.uk> wrote:

Dear Jenny

What exactly do you think you are testing here? You are telling K-W you
have seven groups each with a single value which is not the usual
situation for K-W.

Michael


On 22/12/2018 04:58, Jenny Liu wrote:

Hi everyone,
I have been running a K-W test with the attached data, PupMort1. My

code:

kruskal.test(Prop~Temp,data=PupMort1)
However, I found that I get the exact same result when I change the

x-values, as

in the attached data PupMort2.
Test run with PupMort1Kruskal-Wallis rank sum testdata:  Prop by Temp

Kruskal-Wallis chi-squared = 6, df = 6, p-value = 0.4232
Test run with PupMort2Kruskal-Wallis rank sum testdata:  Prop by Temp

Kruskal-Wallis chi-squared = 6, df = 6, p-value = 0.4232
Does anybody know why this is happening?
Thank you!
Jenny