Hi PLS community,
I recently recieved this comment from a reviewer and am unsure how to respond:
How does task-evoked seed-based PLS factor out correlation due to task, so that what is being correlated is variability about the task induced effect. For instance, psychophysiological interaction analysis explicitly account for this by separately modelling the anticipated task effect, allowing connectivity to only be inferred from residual variance left after controlling for the task main effect. This seems an important consideration as without controlling for the task effect connectivity could be inferred spuriously. By analogy, if the output of two independent microphones were recorded simultaneously while exposed to the same sinusoidally modulated sound wave, correlation of their outputs would lead to the incorrect inference that they were "connected", however if the sinusoidal modulation were first factored out, such that only noise about that modulation remained, correlation would reveal no basis for inferring "connectivity".
Any insight into this issue would be much appreciated.
Many thanks,
Heather
The answer is that it doesn't. I understand the theoretical motivation to remove "stimulus-locked transients" but haven't seen compelling data where this is a problem for fMRI. In electrophys, it does obscure things, where the time resolution is much higher. For fMRI I would be concerned that residualization of the task-regressor would actually results in a loss of potential interactions. Of course, I also do not have evidence myself that this is the case.
Its also worth noting that in PLS, we typically don't model the HRF, so the functional connections/correlations you get that are cross-lag also speak against stimulus-locked effects, which should be at the same lag.
Finally, its worth pointing out that no matter what analysis you do, whether PLS or PPI, the inferences are stil about functional connectivity, which by construction are correlational.
Thanks, Randy!
Baycrest is an academic health sciences centre fully affiliated with the University of Toronto
Privacy Statement - Disclaimer - © 1989-2024 BAYCREST HEALTH SCIENCE. ALL RIGHTS RESERVED