help > RE: t vs F statistic question
Oct 15, 2020  04:10 PM | Andrew Zalesky
RE: t vs F statistic question
Hi Rob,

1. The situation that you describe is possible. Under the F-test, subnetworks can comprise a combination of positive and negative associations with IQ and thus subnetworks can become larger (more connections) compared to the case where only positive or only negative associations are assessed with a t-test. What you are describing sounds reasonable. 

2. F=t^2     F stat is the square of the t stat. 

Your design matrix and contrast sound reasonable. You may want to add confounds such as age, sex, etc, if appropriate. 

Andrew

Originally posted by Rob McCutcheon:
Dear NBS users and Prof Zalesky,

I hope you can help with what I think may be a fundamental error in my understanding.

I have data acquired from 2 different scanners and want to see relationship between iq and connectivity

My design matrix is three columns, 1: Constant, 2: Scanner(1/0) 3: iq

A) If performing a t-test my contrast to identify if there are networks where connectivity strength is negatively associated with iq would be: [0 0 -1]
B) If performing a F-test my contrast to identify if there are networks where connectivity strength is either negatively or positively associated with iq would be: [0 0 1]

I have two queries
1. I think my understanding may be incorrect because if I perform analysis A (with both the positive and negative contrast) I do not get any significant results, whereas if I perform analysis B I do. And this is after adjusting the thresholds so that both F and t-test approaches give the same size networks.
2. How does one convert between F and t statistic?

Many thanks!

Rob

Threaded View

TitleAuthorDate
Rob McCutcheon Oct 15, 2020
Rob McCutcheon Oct 15, 2020
Andrew Zalesky Oct 16, 2020
RE: t vs F statistic question
Andrew Zalesky Oct 15, 2020