We tested how non-experts judge point probability for seven different visual representations of uncertainty, using a case from an unfamiliar domain. Participants (n = 140) rated the probability that the boundary between two earth layers passed through a given point, for seven different visualizations of the positional uncertainty of the boundary. For all types of visualizations, most observers appear to construct an internal model of the uncertainty distribution that closely resembles a normal distribution. However, the visual form of the uncertainty range (i.e., the visualization type) affects this internal model and the internal model relates to participants' numeracy. We conclude that perceived certainty is affected by its visual representation. In a follow-up experiment we found no indications that the absence (or presence) of a prominent center line in the visualization affects the internal model. We discuss if and how our results inform which visual representation is most suitable for representing uncertainty and make suggestions for future work.

Abstract-Treemaps are a well known and powerful space-filling visualisation method for displaying hierarchical data. Many alternative treemap algorithms have been proposed, often with the aim being to optimise performance across several criteria, including spatial stability to assist users in locating and monitoring items of interest. In this paper we demonstrate that spatial stability is not fully captured by the commonly used 'distance change' metric, and we introduce a new 'location drift' metric to more fully capture spatial stability. An empirical study examines the validity and usefulness of the location drift metric, showing that it explains some effects on user performance that distance change does not. Next, we introduce 'Hilbert' and 'Moore' treemap algorithms, which are designed to achieve high spatial stability. We assess their performance in comparison to other treemaps, showing that Hilbert and Moore treemaps perform well across all stability metrics.

Multiday weather forecasts often include graphical representations of uncertainty. However, visual representations of probabilistic events are often misinterpreted by the general public. Although various uncertainty visualizations are now in use, the parameters that determine their successful deployment are still unknown. At the same time, a correct understanding of possible weather forecast outcomes will enable the public to make better decisions and will increase their trust in these predictions. We investigated the effects of the visual form and width of temperature forecast visualizations with uncertainty on estimates of the probability that the temperature could exceed a given value. The results suggest that people apply an internal model of the uncertainty distribution that closely resembles a normal distribution, which confirms previous findings. Also, the visualization type appears to affect the applied internal model, in particular the probability estimates of values outside the depicted uncertainty range. Furthermore, we find that perceived uncertainty does not necessarily map linearly to visual features, as identical relative positions to the range are being judged differently depending on the width of the uncertainty range. Finally, the internal model of the uncertainty distribution is related to participants’ numeracy. We include some implications for makers or designers of uncertainty visualizations.

Abstract. Switching between windows on a computer is a frequent activity, but current switching mechanisms make it difficult to find items. We carried out a longitudinal study that recorded actual window switching behaviour. We found that window revisitation is very common, and that people spend most time working with a small set of windows and applications. We identify two design principles from these observations. First, spatial constancy in the layout of items in a switching interface can aid memorability and support revisitation. Second, gradually adjusting the size of application and window zones in a switcher can improve visibility and targeting for frequently-used items. We carried out two studies to confirm the value of these design ideas. The first showed that spatially stable layouts are significantly faster than the commonly-used recency layout. The second showed that gradual adjustments to accommodate new applications and windows do not reduce performance.

Keyboard shortcuts are generally accepted as the most efficient method for issuing commands, but previous research has suggested that many people do not use them. In this study we investigate the use of keyboard shortcuts further and explore reasons why they are underutilized by users. In Experiment 1, we establish two baseline findings: (1) people infrequently use keyboard shortcuts and (2) lack of knowledge of keyboard shortcuts cannot fully account for the low frequency of use. In Experiments 2 and 3, we furthermore establish that (3) even when put under time pressure users often fail to select those methods they themselves believe to be fastest and (4) the frequency of use of keyboard shortcuts can be increased by a tool that assists users learning keyboard shortcuts. We discuss how the theoretical notion of 'satisficing', adopted from economic and cognitive theory, can explain our results.