1
$\begingroup$

I am studying a parameter in several organs of mice and I have two categorial factors. I would like to study the effects of those two factors on my parameter in each organ (I don't care about differences of the parameter bewteen organs). I was previously doing a two-way ANOVA for each organ but this led to multiple ANOVA tests. I think this could be a problem since it increases the risk of having false positive results. To adress the problem of multiple comparisons, could I perform a single repeated measure ANOVA test, taking organs as the within-subject factor (even though I don't want to study this factor) and the two remaining factors as between-subject factors ? In my mind, repeated measures ANOVA is mainly used to measure a parameter at different time points so I'm not quite sure about this test in this case.

Thank you !

$\endgroup$
  • $\begingroup$ Sounds legit. You can just report the effects on this factor and ignore the results. Alternatively, you could run a multivariate test. $\endgroup$ – Henrik Jun 3 '13 at 22:10
1
$\begingroup$

Yes, it does sound like the right test, if you expect different organ types to respond differently to your experimental treatments, or even if they simply have a different default values for your dependent variable (DV). (I understand you have 2 IVs and 1 DV.)

Just to clarify a couple of things:

  1. Within-subject(s) (or repeated-measure(s)) AnOVa is not just for measuring at different time points, so don't worry about that. It is commonly used differentiate between subjects, as implied by the name (and differentiating between organs in your case may be appropriate).

  2. Within-subject AnOVa is also not for studying the factor used to identify different experimental units (in your case organs), so don't worry about that either. It is not for getting output about that factor. Identifying that factor just helps improve the results.

It is meant to help account for variance that is not related to treatment levels, but that is also not random: variance due to individual differences. ...so when the amount of variance explained by your model gets compared by the amount of variance not explained, the 'individual difference variance' is kind of set aside.

In your case, it sounds like you think different organs might respond differently to the manipulations, and that seems like a reasonable assumption, so yes, it would fine (and you should benefit from better statistical power) to run the within-subject AnOVa.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.