An improved algorithm for model-based analysis of evoked skin conductance responses

• Published in: Biological psychology Vol. 94; no. 3; pp. 490 - 497
• Main Authors: Bach, Dominik R., Friston, Karl J., Dolan, Raymond J.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.12.2013; Elsevier; Elsevier Science B.V
• Subjects: Algorithms; Behavioral psychophysiology; Biological and medical sciences; Electrodermal activity (EDA); Emotions - physiology; Fundamental and applied biological sciences. Psychology; Galvanic skin response (GSR); Galvanic Skin Response - physiology; General linear convolution model (GLM); Generative model; Humans; Model inversion; Psychology. Psychoanalysis. Psychiatry; Psychology. Psychophysiology; Psychophysiology; Skin conductance responses (SCR); Software; General linear convolution model (GLM); Model inversion; Galvanic skin response (GSR); Electrodermal activity (EDA); Skin conductance responses (SCR); Generative model; Electrodermography; Activity; Inversion; Algorithm; Linear model; Electrical conductance; Skin
• ISSN: 0301-0511; 1873-6246; 1873-6246
• DOI: 10.1016/j.biopsycho.2013.09.010

•We improve predictive validity of a general linear convolution method to analyse evoked SCR.•A constrained individual response function provides highest predictive validity.•This IRF is realised by a canonical SCRF together with its time derivative.•A high pass filter of 0.05Hz cut-off frequency is optimal for analysis.•Non-linear models better reconstruct the observed time-series but have lower predictive validity. Model-based analysis of psychophysiological signals is more robust to noise – compared to standard approaches – and may furnish better predictors of psychological state, given a physiological signal. We have previously established the improved predictive validity of model-based analysis of evoked skin conductance responses to brief stimuli, relative to standard approaches. Here, we consider some technical aspects of the underlying generative model and demonstrate further improvements. Most importantly, harvesting between-subject variability in response shape can improve predictive validity, but only under constraints on plausible response forms. A further improvement is achieved by conditioning the physiological signal with high pass filtering. A general conclusion is that precise modelling of physiological time series does not markedly increase predictive validity; instead, it appears that a more constrained model and optimised data features provide better results, probably through a suppression of physiological fluctuation that is not caused by the experiment.