Continuous Time-Series Prediction in Autistic and Non-Autistic Adults

In this study, we seek to examine the ability of autistic adults recruited through SPARK and non-autistic adults recruited through Prolific to predict the next value of an evolving time series. Participants complete two sequential blocks of a time-series prediction task, with an easier block presented first and a harder block presented second. In each block, participants make 24 successive predictions about the next value in a time series. In each trial, they view the previously observed portion of the series and use a continuous mouse-controlled slider to make their prediction of the next value. The amount of available history increases by one time point on each successive trial within a block. Performance is assessed using absolute prediction error. We hypothesize that autistic participants will show larger prediction errors than non-autistic participants, particularly in the harder block.