Design and Evaluation of Visualizations for Large Value Ranges in Multiple Time Series

This paper investigates the complex issue of visualizing large value<br/>ranges in multiple time series. We propose the design spaces for this<br/>composed visualization. In two crowdsourced user studies, we test seven<br/>designs: Three state-of-the-art designs, three extensions to existing<br/>designs, and one novel design. We assess five tasks: Maximum and<br/>minimum identification, value discrimination, difference estimation, and<br/>slope assessment. Our results show novel findings: For the minimum<br/>task, where values in low orders of magnitude have to be identified, our<br/>novel height-stack line chart yields the best results. For slope assessment<br/>and all tasks where the maximum value can be used as a proxy for the<br/>correct answer (maximum, discrimination, and estimation), the linear line<br/>graph shows comparable results to all other designs. Moreover, the use<br/>of visual mapping to color supports the perception of mantissa and<br/>magnitude variations. Unexpectedly, our results indicate that increasing<br/>the number of time series does not generally reduce the accuracy of<br/>estimation, discrimination, and identification. Our findings are domain<br/>independent. They provide useful insights for designers seeking to<br/>visualize large value ranges in multiple time series, e.g., for financial,<br/>medical, or meteorological data.