Hi Doug et al,
*I think this email got lost in the cracks before the holidays, so I am reposting*
We have been working on a linear regression-type analysis (see attached analysis.info for analysis and slopepar for a sample paradigm file) that involves task-related activation in a working memory test. For each participant, there are four conditions: 1, 3, 5, and 7 items that need to be remembered. The question that we would like to ask is: for each voxel, does the slope of activation (i.e., the slope of the best-fit line as you move from 1 to 3 to 5 to 7 items, as calculated separately for each subject) differ from zero? The second-level analysis would determine if there is a significant difference in activation-related slope between groups.
Based on previous discussion, the way we have been doing this is to modify the paradigm file to create the following contrast:
Condition 2: activation during working memory, but weighted by condition so the 1 item condition is weighted as 1, the 3 item condition as 3, etc. Condition 0: fixation The contrast is Condition 2 minus Condition 0.
However, this method of weighting does not appear to answer the question about slope, but rather focuses on a weighted average.
Consider the following scenario, involving 5 subjects (A through E):
The SLOPE column is calculated using the slope function of the spreadsheet (i.e., slope of best fit line, using values of 1...3...5...7 for the x-axis, and, say, for subject A, the values of 2...4...6...8 for the y-axis).
The WEIGHTED column is calculated in the way that I think FS-FAST is doing it, which is as follows: ((1 * 1 item activ. value) + (3 * 3 item activ. value) + (5 * 5 item activ. value) + (7 * 7 item activ. value))/4
The SLOPE and WEIGHTED values (regardless of their different scale) are not really comparable. For example, for Subject A, the SLOPE value is 1 and the WEIGHTED value is 58, while for Subject E, the SLOPE value is 1.1 and the WEIGHTED value is 28.
So when looking at the paradigm file, take the lines
76.57 2 3.4 5 213.42 2 2.29 7
Does this mean that the activation at the first timepoint is accounted for 5-fold and the second timepoint 7-fold? If this is the case then this analysis may not best represent what we are looking for.
Is there a way to modify the paradigm file so that the analysis can focus on slope of the best fit line as opposed to the weighted average?
Thanks, Josh
Forwarded from: Adam Nitenson, B.S. Brain Genomics Lab Massachusetts General Hospital nitenson@nmr.mgh.harvard.edu Phone: 617-643-3215