Dynamic data visualizations as events: effects of framing and change salience on segmenting dynamic maps

TL;DR

This study explores how people segment dynamic data visualizations into events, showing that perceptual and conceptual factors influence their understanding of trends in animated maps.

Contribution

The study introduces a novel framework for analyzing how framing and change salience affect event segmentation in dynamic data visualizations.

Findings

Inter-individual segmentation agreement in dynamic maps occurs above chance and is influenced by the direction of the trend.

A three-way interaction between trend, framing, and salience significantly affects segmentation agreement in high-salience conditions.

Segmentation agreement is higher for negative trends when population decline is salient and framed as endangered.

Abstract

Event segmentation theory, which explores how individuals divide continuous experiences into discrete events, has been extensively studied in naturalistic stimuli. We investigate whether key findings generalize to animated data visualizations, specifically dynamic thematic maps. Experiment 1 showed that inter-individual segmentation agreement in dynamic maps occurs above chance levels and is influenced by the direction of the depicted trend. Experiments 2 and 3 build on these findings by systematically varying the depicted trend in maps showing population changes of fictional insect species. In addition, we examined how conceptual (framing of the species as endangered or invasive) and perceptual factors (salience of directional change) interact to shape segmentation agreement. In Experiment 2, salience was manipulated using different color scales: Saturation-based scales as the high-salience condition and hue-based scales as the low-salience condition. We found a significant three-way interaction between trend, framing, and salience: Agreement was higher when the framing matched the trend direction, but only in the high-salience condition. In Experiment 3, salience was more subtly manipulated by showing the trend either spatially clustered (high salience) or spatially distributed (low salience) across the maps. The results partly replicate the findings of Experiment 2, showing a significant interaction between trend, framing, and spatial pattern on segmentation agreement, with higher agreement for negative trends when population decline was salient and framed as endangered. These findings suggest that symbolic visualizations are subject to event segmentation processes, provided both bottom-up perceptual features and top-down conceptual expectations support the formation and updating of internal event models.

Figures