| |
 |
|
|
Science Forum Index » Space - Consult Forum » General query about t-tests...
Page 1 of 1
|
| Author |
Message |
| Parsleybrimbrim at (no spam) googlemail.com... |
 Posted: Fri May 09, 2008 4:16 am |
|
|
|
Guest
|
Hi.
Sorry for the simple, and very general question. I'm new to this and I
hope this is appropriate for this group.
I have carried out an experiment, and have data from all my test
subjects, for a variety of variables (e.g. height, weight, eye colour
etc.). I'll call this data 'ALL'.
I then split these data into several subsets (classified by e.g. age
group). These age group subsets are of differing size.
My question is: Can I compare the subsets to the ALL data using e.g. t-
test, Anova etc? I'm looking for, of course, statistically significant
differences in the data. Or is this misguided because the subsets are
all extracted from the 'same' population?
If so, what is the best means of saying something concrete about the
differences (if any) between the subsets and the ALL data?
Thanks,
Parsley |
|
|
| Back to top |
|
| Richard Ulrich... |
| Posted: Fri May 09, 2008 5:14 pm |
|
|
|
Guest
|
On Fri, 9 May 2008 07:16:52 -0700 (PDT),
"Parsleybrimbrim at (no spam) googlemail.com" <Parsleybrimbrim at (no spam) googlemail.com>
wrote:
Quote: Hi.
Sorry for the simple, and very general question. I'm new to this and I
hope this is appropriate for this group.
I have carried out an experiment, and have data from all my test
subjects, for a variety of variables (e.g. height, weight, eye colour
etc.). I'll call this data 'ALL'.
You say that you have an *experiment* , and ALL data
comprises height, weight, eye color, etc. Which of these
are the outcome of the experiment? and, Was there a
control group, or were there multiple groups?
What is the experiment?
Quote:
I then split these data into several subsets (classified by e.g. age
group). These age group subsets are of differing size.
My question is: Can I compare the subsets to the ALL data using e.g. t-
test, Anova etc? I'm looking for, of course, statistically significant
differences in the data. Or is this misguided because the subsets are
all extracted from the 'same' population?
This is "misguided" because an experiment properly
has some reason for being done. The total number of
tests suggested by and needed for that experiment would
be a tiny, tiny subset of comparing ALL versus ALL (if ALL
is big).
If you are concerned with doing "too many tests"...
well, that is a proper concern, if you want the p-values
to indicate very much. You limit the number of tests
that are directed to your hypothesis by stating beforehand
a small number of hypotheses. However, it is okay to do
as many tests as you want while you are looking for BAD
things that might mess up your tests-of-hypothesis.
Quote: If so, what is the best means of saying something concrete about the
differences (if any) between the subsets and the ALL data?
For your particular example -
Looking at Subsets is generally the wrong way to look
at AGE, since that is a continuous variable that very, very
often has linear relationships for fairly wide age ranges.
For that, you would use correlation or regression.
I think you need to search for a tutorial or FAQ on
"experimental design." Sites at UCLA and Texas might help.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Sun Sep 07, 2008 9:49 am
|
|