Liat wrote:
>
> Hi!
>
> Can anyone help?
>
> I have a data on scent marking rate in a rodent. There are 6 pairs
> from 6 different territories.
> How can I test whether variation is higher between pairs than within?
> Is ANOVA appropriate in this case and how should it be used?
>
> My data look like the following:
> Terr Sex Rate
> 1 F 0.04
> 1 M 0.63
> 2 F 2.36
> 2 M 2.32
> 3 F 0.98
> 3 M 2.06
> 4 F 0.64
> 4 M 0.79
> 5 F 0.28
> 5 M 0.32
> 6 F 0.11
> 6 M 0.31
How is "rate" defined? This may affect the appropriate method of
analysis, and may require some additional information to be used.
Some brief preliminary thoughts:
(-1) You surely do not want to test "whether variation is higher
between pairs than within". You do not have 2 reps of a six-level
one-way design, you have one rep of a two-by-six two-way design, and
male-female difference cannot be used to infer observed between-pairs
variation under the null hypothesis. Sexual dimorphism is very common
and if it exists here it will cause an exaggerated estimate of
population variance, while it will not contribute at all to
between-pairs variance (ignoring the possibility of gay hamsters.)
What you want to do is to find out whether either or both of
between-pairs variation and between-sexes variation is of a size that
you can (a) be reasonably sure is not due to chance and (b) give a damn
about.
(0) You probably do not have enough data. In particular, there are a
couple peculiarities (see below) that suggest that ANOVA may not be
apppropriate. The usual "innocent-until-guilty" logic of hypothesis
testing does not apply here - you need enough data to show that your
approach IS valid. Despite all-too-common practice...
(1) The lack of values between 1.0 and 2.0 suggests that you might have
a bimodal distribution, in which case a wholly categorical approach:
> Terr Sex High-marking?
> 1 F 0
> 1 M 0
> 2 F 1
> 2 M 1
> 3 F 0
> 3 M 1
> 4 F 0
> 4 M 0
> 5 F 0
> 5 M 0
> 6 F 0
> 6 M 0
may describe the main effect. (But see (0).)
(2) The data extend very close to a "natural boundary" at 0 suggesting
transformation might be of some value (though perhaps not,
in light of (0) and (1).)
-Robert Dawson
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
