Fig. 1.

Step 1: comparison of a linear model for evoked SCR, using different response functions as explained in Section 2. Lower Log Bayes Factors (LBF) indicate higher model evidence for the target model. Upper panel: predictive validity; i.e., ability of estimated SN amplitudes to predict a known sympathetic state, for three contrasts from two experiments, expressed in LBF as negative log likelihood difference between the model in question and a reference model. Several peak scoring methods (PS) are added for illustrative purposes as null models (left of the dashed line). Lower panel: Model Log evidence of the within-subject model, expressed as difference in AIC between the target model and our benchmark model, summed over participants. Abbreviations: PS: SPR (1–4 s) amp – peak scoring amplitude according to the SPR recommendations, using a 1–4 s post-stimulus onset window; PS: SPR (1–4 s) mag – peak scoring magnitude according to the SPR recommendations, using a 1–4 s post-stimulus onset window; PS: SPR (1–3 s) amp – peak scoring amplitude according to the SPR recommendations, using a 1–3 s post-stimulus onset window; PS: SPR (1–4 s) mag – peak scoring magnitude according to the SPR recommendations, using a 1–4 s post-stimulus onset window; PS: peak/baseline – peak scoring magnitude, substracting a 1 s pre-stimulus baseline from the maximum value within a 1–4 s post-stimulus window; SCRF – skin conductance response function (benchmark method); SCRF/time deriv. – skin conductance response function with time derivative; SCRF/time and disp deriv. – skin conductance response function with time and dispersion derivative; FIR 15 s – uninformed finite impulse response function with 15 timebins of 1 s duration; FIR 30 s – uninformed finite impulse response function with 30 timebins of 1 s duration; Cosine nth order – cosine basis set of nth order; SRF – subject-specific response function.