help
489 Subscribers 
help > repeated measure contrast weight
Showing 1-1 of 1 posts
Sep 18, 2017 12:09 PM | Xiaodi Lin - Baylor College of Medicine
repeated measure contrast weight
Hi Alfonso,
I am using CONN version 16.a doing a repeated measure analysis with two groups: patients and controls.
When I was trying to understand how to setup the contrast weight for a paired t-test, I found a reference from the forum, which was the entitled "import two session data" from Dec 1, 2013 between you and Mengyu Tian. I followed your recommendation by selecting 19 individual patients and the patients group for both time scans (Patients_Time 1 and Patients_Time2). Then I set up the contrast weights as [zeros(1,19) 1 -1].
The results from the negative contrast parametric and nonparametric analyses were similar, except that the nonparametric results included many (more than 20) small clusters. I've never seen so many significant small clusters in Conn before, so I am wondering if I did the contrasts correctly. Please see the attachment 1 and 2.
The results from the positive contrast parametric and nonparametric analysis are quit different. The parametric result showed none significant, but the nonparametric analysis showed significant. Please see the attachment 3 and 4.
Meanwhile one of my colleagues also looked at the same analysis by selecting only the patients' group Time1 and Time2 scans in the Subject effects, then she used contrast weights of [1 -1] in the Between-subjects contrast. The results showed in attachment5 (they had a cluster similar to one of the clusters in the negative analyses).
My questions are:
1. The version of Conn used in the 2013 message is likely different from the one I am using now, which is 16a. I just wanted to confirm that the contrast weights should still be [zeros(1,19) 1 -1]. If so, is my colleague's contrast weights of [1 -1] incorrect?
2. Why there is no significant result in parametric analysis but there is significant result in non-parametric analysis?
Thank you very much for your help,
Xiaodi
I am using CONN version 16.a doing a repeated measure analysis with two groups: patients and controls.
When I was trying to understand how to setup the contrast weight for a paired t-test, I found a reference from the forum, which was the entitled "import two session data" from Dec 1, 2013 between you and Mengyu Tian. I followed your recommendation by selecting 19 individual patients and the patients group for both time scans (Patients_Time 1 and Patients_Time2). Then I set up the contrast weights as [zeros(1,19) 1 -1].
The results from the negative contrast parametric and nonparametric analyses were similar, except that the nonparametric results included many (more than 20) small clusters. I've never seen so many significant small clusters in Conn before, so I am wondering if I did the contrasts correctly. Please see the attachment 1 and 2.
The results from the positive contrast parametric and nonparametric analysis are quit different. The parametric result showed none significant, but the nonparametric analysis showed significant. Please see the attachment 3 and 4.
Meanwhile one of my colleagues also looked at the same analysis by selecting only the patients' group Time1 and Time2 scans in the Subject effects, then she used contrast weights of [1 -1] in the Between-subjects contrast. The results showed in attachment5 (they had a cluster similar to one of the clusters in the negative analyses).
My questions are:
1. The version of Conn used in the 2013 message is likely different from the one I am using now, which is 16a. I just wanted to confirm that the contrast weights should still be [zeros(1,19) 1 -1]. If so, is my colleague's contrast weights of [1 -1] incorrect?
2. Why there is no significant result in parametric analysis but there is significant result in non-parametric analysis?
Thank you very much for your help,
Xiaodi