Sampling characteristics in the mean emotion processing of multiple faces

Previous studies have shown that people can quickly and accurately extract the mean emotion from multiple faces, but debate remains about how many faces are involved in this process. This study investigated the sampling characteristics in ensemble coding of facial expressions. Experiment 1 compared the traditional subset paradigm with an adapted version that included filler faces to simulate unsampled items. Results revealed that while fillers reduced averaging performance, they did not impair sampling efficiency. Approximately four faces were sampled from the set of 16 faces. Experiments 2a and 2b examined the effect of spatial arrangement and found little evidence that spatial grouping reliably affects sampling efficiency after controlling for face eccentricity. Experiments 3a and 3b investigated whether mean emotion perception depends on facial features or holistic processing. Results showed that presenting isolated facial features reduced averaging performance and sampling efficiency relative to intact faces, whereas face inversion reduced averaging precision without affecting sampling efficiency. In conclusion, the mean emotion processing of multiple facial expressions relies on capacity-limited sampling and benefits from holistic facial information, while sampling efficiency is relatively insensitive to spatial arrangements.