 |
 |
Hello SPMers, I have a seemingly strange issue regarding the SIGN of the plotted contrast estimates for effects of interest vs 2nd level contrasts within an ROI. I’m very new to this, so I hope someone out there can help. I’ve combed through the archives but haven’t found an answer to my question. I have a 2x3 design, within-subjects experiment as follows: Condition (condition1, condition2, control) Set size (2, 4 items) C1s1, c1s2, c2s1, c2s2, c3s1, c3s2. I’ve modeled a separate regressor for each condition at the 1st level, and then a flexible factorial model at the 2nd level with condition and subjects as factors. One of my a priori ROIs is in the dorsomedial prefrontal cortex (dmPFC). When I run a contrast of task vs control at the 2nd level [1 1 1 1 -2 -2], I get a large cluster in the dmPFC. Plotting the effect of interest in this region’s peak shows positive contrast estimates in condition 2, and negative contrast estimates in condition 1 and the control condition. HOWEVER, as this is an ROI, I wanted to compare condition 1 with condition 2, and so I modeled this specific contrast at the 2nd level only [1 1 -1 -1 0 0], and used a svc over an 8mm sphere in this region (note: the ROI coordinates were defined based on a recent meta-analysis, not on the task vs control contrast, although they happen to be very close to each other). Now, there is a significant cluster, and the peak plotted contrast estimate is positive, which I would think suggests greater activity in condition 1 compared to condition 2. To sum, plotting the effect of interest in an ROI shows positive estimates in condition 2, but not condition 1, whereas modeling a main effect (condition 1 – condition 2) using a svc shows a positive contrast estimate value. I know this isn’t a coding issue as this does not appear in any other main effect contrast, and I’ve double checked 1st level modeling for each subject. Does this have to do with how baselines are estimated for the different contrasts? Which plot is “true”? Many thanks, Kris