From categorization to quantification: Stage-dependent neural and behavioral mechanisms of facial emotion recognition

Facial emotion recognition (FER) is a core process in social communication, involving the identification of emotion categories and the estimation of intensity. Although both are essential for adaptive social functioning, whether they rely on shared or partially differentiated neural processes remains unresolved. To address this issue, the present study combined event-related potential (ERP) measures with trial-by-trial brain-behavior modeling using (generalized) linear mixed-effects models, i.e., (G)LMMs, in 44 healthy adults performing sequential emotion categorization and intensity rating on morphed faces varying in category (happiness, anger, fear, and sadness) and intensity (25%, 50%, 75%, and 100%). Behaviorally, categorization reflected categorical distinctions, whereas intensity estimation revealed emotion-specific patterns that did not align with categorization. At the neural level, ERP results revealed a temporally ordered pattern. Early (P1, N170) and mid-latency (EPN) components were modulated by category, while sensitivity to intensity emerged only at the late stage (LPP), indicating a temporal difference between structural encoding and evaluative integration. Brain-behavior modeling further revealed stage-specific predictive associations. In categorization, early and mid-latency components were negatively associated with performance, while the late components (LPP) indicated positive predictive associations; in intensity estimation, these associations varied with emotional category and intensity. Together, these findings suggested that within a fixed sequential judgment paradigm, intensity-related evaluation showed behavioral and brain-behavior association profiles that differed from those observed for categorization, without implying fully independent processes. By leveraging trial-by-trial brain-behavior modeling, this study uncovers stage-dependent predictive associations that traditional condition-average analyses cannot capture.