question about sampling frequency. help!

[rlollumc]

I’m planning a parameter-level validation study using NeuroKit2 across different experimental tasks. The goal is simple summary metrics rather than deep modeling. Specifically, for each period (e.g., baseline relax vs. stress task) and for each device (reference vs. test), I want to compute:

• Average number of skin conductance responses (SCRs).
• Average skin conductance level (SCL).

I’ll then compare devices (e.g., are SCR counts during stress higher/lower on the test device relative to the reference?) and check whether both devices show the same directional changes across conditions (e.g., SCL and SCRs increase from baseline to stress on the reference device - does the test device show that too?).

My reference device samples at 32 Hz, and the test device samples at 4 Hz. I’m fine downsampling the reference to 8 Hz (or 4 Hz if necessary), but I’d prefer not to upsample the test device. Is NeuroKit2’s eda_process() pipeline appropriate for data sampled at 4 Hz? I’ve seen mixed answers and would like clarification.

Thanks!

[DerAndereJohannes]

Hi! Thanks for your comment. I would suggest to always test out any pipeline on your own data to see if it is suitable. For example for a 4Hz example:

import neurokit2 as nk
import matplotlib.pyplot as plt

TIME_SECONDS=60
SAMPLING_RATE=4
SCR_NUMBER=5

signal = nk.eda_simulate(TIME_SECONDS, scr_number=SCR_NUMBER, sampling_rate=SAMPLING_RATE, random_state=1234)

signals, info = nk.eda_process(signal, sampling_rate=SAMPLING_RATE)

nk.eda_plot(signals, info)

plt.show()

creates the following:

Note that 4Hz is indeed quite low, so it will not capture the timings that accurately and you are losing quite a bit of detail in your signal. In fact, 4Hz may be too low overall. What device are you using?

For any other issue, if you could specify which comments especially concern you, I would be happy to help further.