Zhilin Zhang¹ (), Tatiana Aloi Emmanouil^1,2; ¹The Graduate Center, City University of New York, ²Baruch College, City University of New York

Ensemble perception has been shown to discount outlier items whose properties lie outside the distribution of the ensemble. Previous research investigated the mechanisms of outlier discounting for spatial ensembles that varied in orientation and found evidence for an iterative outlier exclusion process. In the current study, we examined whether and how outlier filtering occurs for ensembles in which objects are presented sequentially in time. In Experiment 1, participants reported their estimation of the average tilt in sets of 12 oriented lines (2 outliers) presented either sequentially (temporal ensembles) or simultaneously (spatial ensembles). To evaluate outlier filtering, we examined whether the standard deviation of their estimation to the local mean (average tilt without outliers) was smaller than that to the global mean (average tilt with outliers). The results showed outlier filtering of similar magnitude for temporal and spatial ensembles. Interestingly, for temporal ensembles, we observed reduced filtering for outliers presented later in the sequence. In Experiments 2 and 3 we further investigated this recency effect. Experiment 2 tested whether this effect is due to reduced processing time between the late outliers and the response probe. We did not find any evidence for this possibility since varying the time interval between ensemble and response had no effect on filtering. Experiment 3 tested the hypothesis that reduced filtering for late outliers was due to the lack of post-outlier context. The results supported this prediction, showing that outlier filtering was restored when appending additional non-outlier items in the sequence. Overall, this study demonstrates that outliers in temporal ensembles are discounted, although this effect is diminished for outliers appearing late in the sequence. Our findings suggest that temporal ensemble perception may involve a dynamically updating process that reinforces discounting of outliers whose deviation is confirmed by subsequently encoded items.

Acknowledgements: We thank Holland Brown, Elijah Burgos, Tanzifa Akther and Selly Kebbe for assistance in data collection. This project is supported by CUNY Professional Staff Congress grant PSC-TRADA-55-414 to Tatiana Emmanouil.