Social robots are often implemented with facial expressions to convey emotion. However, bodily cues are powerful sources of affective information. Our study investigated whether the oscillatory cues in robots of different sizes (small non humanoid, small and large humanoid) could impact human engagement during human-robot interaction. Three groups of twenty volunteers were asked to listen to a robot telling negative emotional stories. The robots displayed bodily cues expressing negative emotions while moving at three frequencies (high, medium, and low) to display contrasting arousal levels relating to anger, stress, and sadness (Barrett, 1998). We gathered self-reported changes in affective states and perceptions of robot social traits (Russell et al., 1989, Bartneck et al., 2009). We also tracked the head, torso, and shoulder movements with 3D motion capture technology (Brossard et al., 2023). The analysis of the motion capture data showed that participants moved faster when the small humanoid robot oscillated at high and medium frequencies and moved slower when it oscillated at the low frequency. Conversely, participants inhibited all spontaneous movement when facing the large robot. Results from the questionnaires showed that the small humanoid robot was perceived as more likable (p<0.001) and more intelligent (p<0.05) than the other two robots. Finally, results of the self-reported affective states indicated a decrease in valence for participants who listened to the small humanoid robot only. Overall, we report affective resonance and engagement with small humanoid robots only. Hence, both size and bodily morphology should be considered in designing compliant social robots.