help > RE: Design Matrix and Contrast
Jul 20, 2018  05:07 PM | Andrew Zalesky
RE: Design Matrix and Contrast
Hi Anita,

you wrote "I ran a paired two-sample means test to compare the changes in significant connections between group 1 and group 2"

But this seems to contradict the fact the you did not find any significant connections between group 1 and group 2.

Note that the one-sample test will not test for differences between the groups. It will just tell you whether the connection weight is significantly different from zero.This has nothing to do with a between group difference.

It is also unclear how you corrected for multiple comparisons in Excel. Did you run a t-test separately for each connection (this requires correction for multiple comparisons), or did you run a t-test on some kind of average across connections?

In general, a paired t-test will paired t-test will offer greater power in this situation.

Andrew



Originally posted by asinha:
Andrew,

I will clarify my previous post below.

Given that I did not find statistical significance in NBS when running a paired t-test between group 1 (before) and group 2 (after), I ran one-sample t-tests on my 26 subjects using the following design matrix:

1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1

to identify the most significant connections between ROIs for group 1 and again for group 2.

Then, in Excel, I ran a paired two-sample means test to compare the changes in significant connections between group 1 and group 2 and got a p-value = 6.40*10^-5.

My question is: how is this different than running a paired t-test in NBS that accounts for repeated measures with the first design matrix in my original post? When I run that, I get a p-value of 1.0.

Regards,

Anita

Threaded View

TitleAuthorDate
asinha Jul 17, 2018
Andrew Zalesky Jul 17, 2018
asinha Jul 18, 2018
Andrew Zalesky Jul 18, 2018
asinha Jul 19, 2018
RE: Design Matrix and Contrast
Andrew Zalesky Jul 20, 2018