A Comparison of Visualizations for Identifying Correlation over Space and Time

Abstract

Small Multiples (juxtapose location) Single Map (juxtapose time) Cartogram (visual encodings use map features) Proportional Symbol Map (visual encodings use symbols on top of map features) GlyphSM DorlingSM Barchart1M (a) (b) (c) Fig. 1. The three visualizations compared in our study. (a) Dorling cartograms as small multiples, (b) proportional symbols (circles) on maps as small multiples, and (c) proportional symbols (bar charts) on a single map. In this example, each map shows the values of two artificially-created variables over four years. In each case, both variables have an overall positive correlation (Pearson correlation coefficient ≥ 0.75) and no monotonic evolution.

A Comparison of Geographical Propagation Visualizations

“…There seems to be a speed/accuracy trade-off at play, possibly related to the complexity of Glyph Map, which likely requires more effort to interpret. Our results confirm that Glyph Map, the only technique that juxtaposes time [55] in this study, performs worst for tasks involving the comparison of consecutive time-steps across regions (H2, H4). Indeed, the information is dispersed over the entire representation, making the tasks cognitively demanding.…”

“…Our focus, however, is not on techniques that visualize the movements of entities, but rather on techniques that represent the spatio-temporal dynamics themselves. We organize them according to Peña-Araya et al's categorization [55], in which geo-temporal visualizations either juxtapose time or juxtapose location.…”

“…Regardless of the diversity of contexts in which animation and small multiples have been compared, there is no study in the literature about the effectiveness of both techniques at supporting the analysis of geotemporal movement: available studies do not consider geography and use fewer time steps ( [5,6,31,12,57]), or focus on tasks and data not related to spatio-temporal dynamics ( [11,39]). Furthermore, studies that compare other geo-temporal visualizations (e.g., [37,43,55]) focus on representing trends over space and time, and are therefore less related to our work. Detecting change over time is a common task on these studies, but their results cannot be directly applied to propagation as analysis goals and movement patterns differ.…”

“…First, we could have used cartograms instead of choropleth maps. While some cartogram techniques can be useful when encoding more than one variable per geographical feature [52,55], they also have weaknesses that make them ill-suited to our con-text (at least as a default choice): the abstract representation approximates topology, can be disorienting, and makes animated maps very unstable when mapping data to area (which would have been the main reason to investigate cartograms in the first place), no matter the type of cartogram [51].…”

Feature Driven Combination of Animated Vector Field Visualizations

“…Results: We base our analysis on 95% bias‐corrected and accelerated (BCa) bootstrap confidence intervals [Dra16], using 10000 replications. Confidence intervals have been used in many recent HCI and visualization studies [BBB∗19, PAPB19], and have been preferred over p‐values as the latter may lead to dichotomous thinking [BD17]. We compare techniques by calculating the confidence intervals of their pairwise difference [Cum14] adjusted for multiple comparisons using Bonferroni correction.…”