Speakers: Emily and Miguel

Title: Can You See The Change? Change Point Detection Using Visual Inference

Abstract: Detecting change points is crucial in analyzing time series data and single-subject designs. This study investigates factors influencing change point detection through visual perception by employing a visual inference experiment. We focused on four characteristics of a transitional mean shift in the data: the shift magnitude, shift direction, change point location, and overall variance. A total of 210 participants were tasked with identifying the plot with the mean shift among null plots in a lineup panel. We used a factorial experiment with a balanced incomplete block design to assign factor combinations for participants to evaluate. Participants were found to identify higher shift magnitudes more accurately than lower shift magnitudes. Change point data with higher variances had lower identification rates. Direction and change point location effects impacted identifying change point scenarios with lower signal-to-noise ratios. Participants indicated varying reasons for selection across correct and incorrect data plot identifications. Additionally, confidence in selection was positively associated with identification accuracy for change plots with higher signal-to-noise ratios. These insights highlight the complexities of change point detection through visual inference and emphasize the multifaceted nature of human perception in identifying subtle changes within data.