Jason Cohen wrote:
>
> Hi,
>
> I'm trying to finish my dissertation and a person on my committee raised
> the question of me having to do a Bonferroni correction. Here is the
> situation. I have 5 subjects that I have tested at 5 different times
> for 2 different conditions (cued and uncued). When I do a repeated
> measures ANOVA, I get the the difference between the two conditions over
> those 5 seperate times is statistically significant. Now, I also have a
> baseline condition that I want to compare the cued and uncued conditions
> to at each of the 5 times individually. For example, I want to compare
> Baseline value to cued at time 1, then Baseline to cued at time 2, etc.
> To do these comparisons, I use a paired t-test and a significance level
> of 0.05. The person on my committee suggests that the p-value should be
> adjusted to be 0.01 to correct for atleast 5 comparisons of cued, and
> the same when uncued is compared to baseline.
>
> I hope what I said makes sense. I'm not a statistician but I could use
> help, because from my reading on Bonferroni, I don't think I have to
> make those corrections. If you have suggestions, please e-mail me at
> [EMAIL PROTECTED]
It depends on what you want to find out. If you expect the cued condition to
outperform the uncued then the main effect of condition seems sufficient to
provide evidence for this. What can the t tests add to this (given the small
n) unless you expect the different time points to show different effects? The
bonferonni correction (and paired t tests) seems relevant only if there is
some theoretical reason to follow the infividual time points. On the other
hand a specific contrast might well be a more sensible and more powerful way
to test any predictions.
Thom
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
