[color=#000000]Dear Mary & Xiaodi,[/color]
[color=#000000]Sorry for the late reply (and sorry the designs are a bit complicated in this case)[/color]
[color=#000000]For the Time2-Time1 effect in controls I would select [/color]Time1_Control, Time2_Control, and the 5 "within" effects for these subjects, and enter a contrast [-1 1 0 0 0 0 0].
Similarly for the same effect within patients I would select Time1_TBI, Time2_TBI, and the 19 "within" effects for these subjects, and enter a contrast [-1 1 zeros(1,19)].
Last for the time-by-group interaction I would select Time1_Control, Time2_Control, Time1_TBI, Time2_TBI, and the 24 "within" effects, and enter a contrast [-1 1 1 -1 zeros(1,24)].
For all of these analyses I would recommend the "two-sided" tests (unless you have a priori hypotheses about the directionality of these effects).
Let me know if you run into any issues and/or if you would like me to clarify any of the above
Best
Alfonso
[i]Originally posted by Mary Newsome:[/i][quote]Dear Alfonso,
We have a repeated measures design with subjects who were scanned at different TRs. As a result, we attempted to take your advice posted on Dec 1, 2013 to Mengyu Tian in the Conn forum, which worked fine. https://www.nitrc.org/forum/message.php?msg_id=9119
However, rather than looking at only Time1 vs Time 2 differences across all subjects (as in the Tian circumstance), we're interested in looking at any potential difference between controls and between patients. So we changed the covariates to:
Time1_Control
Time2_Control
Time1_TBI
Time2_TBI
within1
within2
.....
within24
We have 24 pairs, that includes 5 controls and 19 patients.
When doing the Time1_Control vs. the Time2_Control comparison, we
1) selected Time1_Control, Time2_Control, and only the 5 control subjects. Then in the "between-subjects contrast" field, we entered [1 -1 zeros(1,7)]. When we clicked "enter", this came up:
[-1 1 0 0 0 0 0; 0 -1 1 0 0 0 0; 0 0 -1 1 0 0 0; 0 0 0 -1 1 0 0; 0 0 0 0 -1 1 0; 0 0 0 0 0 -1 1]
This didn't seem to make sense. So then we tried something else and
2) defined the contrast manually with:
[1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0]
The results from this seemed to have a regular pattern, although we don't know if it's really reflecting what we want. We attached a jpg of the results from the Results Explorer. We also noted that this contrast produced an F, rather than a T, test.
We also tried:
3) [1 -1 zeros(1,24)], and the resulting patterns looked irregular/not right. In case it's relevant, in the title bar of the Results Explorer gui, it said, "Time1_Control(1).Time2_Control(-1).within1(0).within2(0).within3(0)...within18/dmn.LLP/spm.mat" (We think it might have stopped at "within18" there just wasn't room to list all 24.)
Thank you for your answer,
Xiaodi and Mary[/quote]
[color=#000000]Sorry for the late reply (and sorry the designs are a bit complicated in this case)[/color]
[color=#000000]For the Time2-Time1 effect in controls I would select [/color]Time1_Control, Time2_Control, and the 5 "within" effects for these subjects, and enter a contrast [-1 1 0 0 0 0 0].
Similarly for the same effect within patients I would select Time1_TBI, Time2_TBI, and the 19 "within" effects for these subjects, and enter a contrast [-1 1 zeros(1,19)].
Last for the time-by-group interaction I would select Time1_Control, Time2_Control, Time1_TBI, Time2_TBI, and the 24 "within" effects, and enter a contrast [-1 1 1 -1 zeros(1,24)].
For all of these analyses I would recommend the "two-sided" tests (unless you have a priori hypotheses about the directionality of these effects).
Let me know if you run into any issues and/or if you would like me to clarify any of the above
Best
Alfonso
[i]Originally posted by Mary Newsome:[/i][quote]Dear Alfonso,
We have a repeated measures design with subjects who were scanned at different TRs. As a result, we attempted to take your advice posted on Dec 1, 2013 to Mengyu Tian in the Conn forum, which worked fine. https://www.nitrc.org/forum/message.php?msg_id=9119
However, rather than looking at only Time1 vs Time 2 differences across all subjects (as in the Tian circumstance), we're interested in looking at any potential difference between controls and between patients. So we changed the covariates to:
Time1_Control
Time2_Control
Time1_TBI
Time2_TBI
within1
within2
.....
within24
We have 24 pairs, that includes 5 controls and 19 patients.
When doing the Time1_Control vs. the Time2_Control comparison, we
1) selected Time1_Control, Time2_Control, and only the 5 control subjects. Then in the "between-subjects contrast" field, we entered [1 -1 zeros(1,7)]. When we clicked "enter", this came up:
[-1 1 0 0 0 0 0; 0 -1 1 0 0 0 0; 0 0 -1 1 0 0 0; 0 0 0 -1 1 0 0; 0 0 0 0 -1 1 0; 0 0 0 0 0 -1 1]
This didn't seem to make sense. So then we tried something else and
2) defined the contrast manually with:
[1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0; 1 -1 0 0 0 0 0 0 0]
The results from this seemed to have a regular pattern, although we don't know if it's really reflecting what we want. We attached a jpg of the results from the Results Explorer. We also noted that this contrast produced an F, rather than a T, test.
We also tried:
3) [1 -1 zeros(1,24)], and the resulting patterns looked irregular/not right. In case it's relevant, in the title bar of the Results Explorer gui, it said, "Time1_Control(1).Time2_Control(-1).within1(0).within2(0).within3(0)...within18/dmn.LLP/spm.mat" (We think it might have stopped at "within18" there just wasn't room to list all 24.)
Thank you for your answer,
Xiaodi and Mary[/quote]