 |
 |
Hello, listserv! I’m interested in analyzing the effect of a parametric modulator in an ROI. To do this, I am running models with one EV for the condition itself and one for the parametric modulation of the effect (the condition is problem solving blocks, the parametric modulator is mean-centered Likert-scale values representing strategy used.) However, some of the runs for some of my subjects have zero variance for the modulator (i.e., the participant indicated they used the same strategy to solve all problems within a run). This is causing a rank deficiency in the design matrix and the modulator contrast is automatically being set to zero. Since that is the case, should I (1) concatenate data across all runs and input this into the first-level, (2) exclude runs with zero variance, or (3) exclude the condition EV entirely and only input the EV modeling the parametric modulator? (1) seems like the best option as it would largely mitigate my zero variance issue - the parametric modulator would model the effect across all blocks, not just the blocks within one run, and very few subjects indicated they used the same strategy for all problems. However, I am aware that several members of this list have expressed concern about including multiple runs at the first-level due to the introduction of discontinuities in the time series at the joining points. Is there a workaround to this issue? Perhaps flagging those joining points as TRs to be scrubbed out during the analysis? I dislike options (2) because excluding the zero variance runs would remove data that are ultimately useful when considered across all runs (the entire problem set). Any input would be appreciated! Thank you! Jessica