Multivoxel Pattern Analysis Does Not Provide Evidence to Support the Existence of Basic Emotions

Abstract

Saarimaki et al. (2015) published a paper claiming to find the neural "fingerprints" for anger, fear, disgust, happiness, sadness, and surprise using multivariate pattern analysis. There are 2 ways in which Saarimaki et al.'s interpretation mischaracterizes their actual findings. The first is statistical: a pattern that successfully distinguishes the members of one category from the members of another (with an accuracy greater than that which might be expected by chance) is not a "fingerprint" (i.e., an essence); it is an abstract, statistical summary of a variable population of instances. The second way in which Saarimaki et al.'s interpretation mischaracterizes their results is conceptual: their findings do not actually meet the specific criteria for basic emotion theory. Instead, their findings are more consistent with a theory of constructed emotion. In our view, Saarimaki et al. is elegant in method and important in that it demonstrates empirical support for a theory of emotion that relies on population thinking; it is also an example of how essentialism-the belief that all instances of a category possesses necessary features that define what is, and what is not, a category member-contributes to a fundamental misunderstanding of the neural basis of emotion.

Decoding the Nature of Emotion in the Brain

“…Proponents of constructionist models argue that this variability indicates that multivariate classifiers do not learn the essence of emotion categories [54,56] but instead differentiate among populations of emotional instances sampled within a study. Thus, disparities in patterning across studies could be driven by differences in induction procedures, analytical methods, and inclusion of different emotions and varying numbers of emotions in each study.…”

“…These meta-analytic findings were interpreted by the authors as supporting constructionist models of emotion [52] because the patterns of activity that predicted each emotion category spanned multiple intrinsic brain systems (see also [54]). However, direct model comparisons to rule out alternative interpretations based on categorical theories were not conducted.…”

Nature of Emotion Categories: Comment on Cowen and Keltner

“…Furthermore, instances of different emotion categories have considerable similarity in their features. This is true when measuring affective properties in emotional experiences (e.g., [6]), when measuring facial actions that serve as emotional expressions [7], when measuring autonomic nervous system (ANS) features of emotion [8], and when measuring the neural patterns associated with emotion categories [9]. …”

Categories and Their Role in the Science of Emotion

“…For example, different proteins can catalyze the same reaction of enzymes (Edelman & Gally, 2001; Tononi, Sporns, & Edelman, 1999), different antibodies can bind to the same antigen (Edelman, 1974), different genotypes can produce the same phenotype (Edelman & Gally, 2001; Tononi, Sporns, & Edelman, 1999), different neurons can give rise to the same network (Marder & Taylor, 2011; Tononi, Edelman, & Sporns, 1998; Tononi, Sporns, & Edelman, 1999), and different patterns of network interaction can give rise to the same behavior (Price & Friston, 2002). Emotion categories have degenerate instances (varying in their associated facial configurations, autonomic configurations, and appraisals) and any emotional instance (whether you call it an emotion or an emotional episode) can be caused by a different pattern of different domain general system interactions implemented as different brain states (see Clark-Polner et al, in press). …”