AFNI Message Board

History of AFNI updates  

February 26, 2003 03:57PM
Hi Sara:

1) Both Anovas and t-tests can be used for both between and within subjects analysis. A within subjects t-test is called a paired t-test, while a within subjects ANOVA is called a repeated measures anova. Your design sounds like a repeated measures anova. You have 3 conditions and all subjects were presented all 3. To test this with a repeated measures anova use 3dANOVA2 -type 3 for a mixed effects model. Your fixed effect is your experimental manipuulation (3 conditions; -alevels 3) and your random effect is your subjects (all 10 subjects -blevels 10). You could use the coefficient from each condition (baseline corrected to get percent change from baseline as you mentioned you wanted this; and blurred).

2) The differences between these two approaches is that when using the GLT you are doing the test within a subject, using the error term for the regression model. When doing a t-test across subjects you are using the within-subject variability for each condition as the error term. In theory you should get the same regions when doing both, however, it is possible to find a region that shows the GLT effect in all subjects, but that doesnt show up in the t-test. This can happen if individual subjects are showing different directionality in the GLT effets. For example, a voxel in subject 1 may show condition 1>condition2 while in subject 2 that same voxel shows condition2>condition1. This would pass threshold for the individual GLT. The anova on this voxel across these two subjects would have the differences between condition1 and 2 averaging to zero; no mean difference, so it wouldnt show a significant effect.

3) You are right, you need to create an additionaly thresholding step for the conjunction map like to consider voxels to be active if only all subjects show them (3dmerge option -1clip 9.5) or maybe just the majority (-1clip 5.5). You would then describe the results as two thresholding steps a) the initial GLT threshold value, and b) the conjunction map thereshold.

4) I am not sure how to get area under the curve if your maxlag is 0. My familiarity with 3deconvolve is mainly when using it as a deconvolution program instead as a straight regression program as you are using it. I am not sure if the -iresp option works with maxlag =0. I would post that question on the message board to get answers from someone who is more familiar with this.

Hope this helps,

Subject Author Posted


Sara February 21, 2003 03:09PM

Re: 3dttest

Christine Smith February 21, 2003 05:29PM

Re: 3dttest

Sara February 25, 2003 11:00AM

Re: 3dttest

Christine Smith February 26, 2003 03:57PM

Re: 3dttest

Sara March 04, 2003 12:54PM

Sorry, only registered users may post in this forum.

Click here to login