Re: Toxicological experiments: Pair t-test or unpaired t-test or Mann-whetney test?

On 4 May 2007 09:14:24 -0700, "minimausviet@xxxxxxxxxxxxxx"
<minimausviet@xxxxxxxxxxxxxx> wrote:

Dear All,

This is perhaps a basic question about statistic but I have struggled
to find the answer so far.
In my experiments, I used two groups of animals from the same species,
same source to do the experiments. One was control group and the other
was treated group (with heavy metal). Two groups were run parallel and
I took samples at different time points. Each sampling time, 3 animals
from each group were taken and sacrifice to measure heavy metal
concentrations in tissues. I wanted to compare the means of heavy
metal in the two groups.

1) Is the fact of an eventual difference here
a foregone conclusion? Should this be a MEASUREMENT
problem rather than a TEST problem?

2) Are you using the log(concentration) rather than the
raw units? Biological concentrations often need a
log transformation to keep the variances similar, and
to model the biological systems.


For me, these samples are independent and have no particular
relationship, but I am still not very sure about my decision. I also
try with paired t test but I got the note from the programe that
"With these data, the pairing (or matching) appears not to be
effective. The unpaired test may be more appropriate.". Then I tried
with unpaired t test but it also did not work because SDs are not
equal. After all I used non-parametric Mann-whetney test and the
results seem to be ok.

Could anyone out there please give me your advices and comments about
this. Thank you very much for your time and help!

If the increases fit a linear trend, you might compare the
two regression lines, if you simply want a test. If there is
some other *shape* of the response curve that is expected,
you could test for that shape in particular.

However, you seem to be aiming for "when" in your later post -

[snip some]
when results come out significant different and there are interactions
between variables, I need to follow up a significant effects in one-
factor ANOVA and post test. Because it is important to know what level
of second variable the variable in question has a significant effect.

That is probably not a *wise* statement to try for.
Is the effect generally increasing (or decreasing)?

"What is a minimum detectable dose?" - is an arbitrary
choice, statistically, if you are looking at two curves
that diverge, and you have data at intervals. It is probably
safe to *assume* that a large enough sample would
detect an arbitrarily small amount, until the doses are
rather smaller than background noise. Example: an X-ray
at the airport (that new, naked-viewing technology) does
expose you to radiation -- but it can be said to be less
than you get on a short flight, so it is 'small'.

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.