Ben Meuleman & David Rudrauf (2021), IEEE Transactions on Affective Computing 12(1):189–202 — DOI-10.1109/TAFFC.2018.2864730
Key takeaways
- Why passive induction fails (§1.1). Pictures and films make motivation and behaviour “ambiguous or inappropriate”. Hypotheses “that involve all components of emotion simultaneously, such as componential synchronization” cannot be tested, because “Testing these hypotheses is prohibited in emotion induction paradigms that do not engage adequately all mental and bodily subsystems simultaneously” (ms p. 2).
- Design (§2). Seven commercial games for the HTC Vive, chosen for one or two dominant emotions. 53 participants, each playing 1–6 games, with a questionnaire after each: CoreGRID-derived physiology, motivation and feeling items, added appraisal and regulation items, the Geneva Emotion Wheel, and a body manikin.
- Clusters (§3.2). Item clustering gave a fear cluster (danger, unpleasantness, escape and stop tendencies, arousal, regulation by looking away or reminding oneself it is virtual) and a joy cluster (pleasantness, tendencies to stay absorbed, explore, repeat). A gaming cluster lacked physiology and feeling items and was not counted as an emotion.
- Appraisal predicts best (§3.3). The appraisal-only model fits fear (marginal R² 61.3%) and joy (25.7%) better than any other single component by AIC. Danger and pleasantness are the strongest predictors.
- Multi-componential means across observations. An emotion counted as “fully multi-componential” if its intensity had “At least one significant association with each emotion component” across the observations (ms p. 7, note 2). Only fear, joy and pleasure qualified.
- Self-report throughout (§4). “our study relied entirely on self-report, even for objectively measurable emotion components such as physiology” (ms p. 13).
Standing in the record
Filed on 2026-10-23 for the record’s question of how to measure which ordering is the potential in a negotiation (QUESTION-268), as set out in NOTE-1097. It gives no per-episode measure of component synchronization. It argues that active induction is a precondition for one, and its componentiality criterion is a pattern of associations across a sample. Its mild-fear cluster, fear appraisal together with a strong tendency to explore, is a case of competing tendencies in one episode, but the data cannot say which prevailed. Mohammadi and Vuilleumier (LIT-1443) cite it as the one prior study with the same aim.
It is affective-computing methodology for emotion induction, not an instruction for machine-learning practice.