Negative SCR amplitudes in PsPM

[cherepashka-94]

Hello,

I am analysing SCR data using PsPM (GLM SCR model). However, in my output, most of the reconstructed response amplitudes (β estimates) are negative, across both negative and neutral image conditions. I am trying to understand what is the reason for that and whether something in my preprocessing pipeline needs adjustment.

I collected EDA using BIOPAC and preprocessed in AcqKnowledge before importing to PsPM.
My preprocessing steps were: 1. Resample the signal (to ~62.5 Hz) 2. Median smoothing (63 samples) 3. Low-pass FIR filter at 1 Hz. I also later added a 0.05 Hz high-pass filter (as recommended for GLM in the manual) but it looks like it doesn't solve the issue with the negative values .

Originally, I used the phasic EDA channel generated in AcqKnowledge. I now understand this may not be appropriate for PsPM, so I switched to using the tonic EDA signal as input instead. However, when I re-ran several files using the tonic signal (with the steps above), I still see mostly negative reconstructed amplitudes.

Also, in my study, participants first learned the image pairs (multiple exposures), and only after that I recorded EDA during emotional assessment. So the stimuli were not novel at the time of SCR recording. I am wondering whether habituation could explain negative amplitudes.

My questions:

-Could predominantly negative β estimates simply reflect very weak SCR due to habituation to the stimuli?

-For PsPM GLM SCR, should I use the tonic signal as input (and let PsPM estimate the phasic component internally)?

-If tonic input is correct, are the preprocessing steps above appropriate, or should I modify them?

Thank you very much for any clarification!

Anna

[dominikbach]

This relates to #865

In addition:

• weak SCR should not cause negative but near-zero beta estimates
• for PsPM, it is always advisable to use the original raw SCR signal, not split into "tonic" and "phasic" components (which is essentially just a filter)
• in any case, the "tonic" input is not the right type of data, and the "phasic" (i.e. high-pass filtered) signal will be more appropriate