Table 1.

Summary of past MVPA studies of BOLD data.

Study	N	# Emot. categories	# Stimuli per category (unique)	Induction method	Relevant preprocessing	Feature selection	Classification algorithm
Kassam et al.24	10	9	2	Participant-generated scenario immersion	No spatial smoothing	Voxels with the most stable activation profile	Gaussian Naive Bayes
					Z-scoring
					Anatomical normalization
Kragel and LaBar25	32	7	4	Movies Music	No spatial smoothing	All grey matter voxels	Partial least squares discriminant analysis
					Mean centering
					Anatomical normalization
Saarimäki et al.29	48	5 (study 1) 6 (study 2)	10 (study 1) 6 (study 2)	Participant-generated scenario immersion Movies	No spatial smoothing	Voxels most sensitive to manipulation using ANOVA	Linear neural network with no hidden layers
					Z-scoring
					Anatomical normalization
Saarimäki et al.30	25	15	4	Narrative-guided scenario immersion	No spatial smoothing	Voxels most sensitive to manipulation using ANOVA	Linear neural network with no hidden layers
					Z-scoring
					Anatomical normalization
Wager et al.32	2159	5	Variable	Variable	Meta-analysis (peak based)	Whole brain	Bayesian spatial point process model
					Binarized data
					Anatomical normalization

These studies all claim to identify unique patterns of brain activity for specific emotion categories, yet these patterns are inconsistent across studies. Other existing MVPA studies of affect (e.g.,³⁸), and conceptual knowledge (e.g.,³⁹) are less relevant and so are not listed here.