wiheto
commented
4 years ago Since each stimulation run is short and will only result with ca 10 time-points per stimulation run. Deriving functional connectivity estimates on just 10 time-points is ill-advised. So a deviation in the preregistration is going to be needed.
There are two possibilities to change this analysis. The options are:
-
Use a leave-n-out strategy. The n that would be left out is the entire stimulation period. This will create differences in connectivity per stimulation period. Positive: this keeps a measure per stimulation period. Negative: This changes the topographical properties (connectivity values = variation from mean, instead of r-values). What PC means when calculating PC on these types of connectivity values is not apparent in their meaning.
-
Use the displacement values from analysis 1 and relate them to PC at stimsite. A similar statistical model becomes PC_stimsite ~ displacement(PC_eson) + displacement(z_oson) where all runs are included. Displacement (es-on - es-off) would need to be used here since the variation is no longer calculated over multiple runs. This will become a linear model instead of logic and all runs will be included. Positive: much easier to understand. Negative: Removing the subject layer in the model is going to be necessary (but could be attempted) and repeated measures from same run is lost.
-
Alternatively, do as planned which would be a more noisey estimates as ten values for connectivity is too small.
I think 2 is the best bet here. But will think about it.
[PROC] Functional connectivity will be calculated for time-points that correspond to “stim-on” and “stim-off” periods. Functional connectivity will be calculated with Pearson correlations. For COMP-1, all stim-on and stim-off periods will be concatenated, and the functional connectivity will be calculated for each. For COMP-2, functional connectivity will be calculated for each stimulation period during a run.