Understanding Data Visualisations: Cognitive Processes Involved in Representing Data

The range of data visualisations used within academia and journalism is ever-increasing. With data literacy's importance recognised by the UN and the Gapminder Foundation, style tips and advice on producing effective data visualisations are also widespread (Cairo, 2016; Tufte, 1983). Whilst there is some research on perceptual and comprehension processes involved in simple graph-reading (Carpenter & Shah, 1998; Shah & Freedman, 2011), studies have frequently focused on geo-spatial maps (e.g. Ratwani, Trafton, & Boehm-Davis, 2008), and there is little empirical evidence on how individuals form coherent mental representations of complex data in graphs. It is somewhat unclear, therefore, how graph-readers make sense of what they see, and therefore how graph-creators can decide which graph best depicts their data. My research will investigate cognitive processes involved in understanding data visualisations and explore how individuals create and maintain a 'situation model' of a graph.

Given the importance of reliable and robust findings, I plan to carry out four carefully designed, large sample studies over the course of four years. In my master's and PhD projects, I will use a range of visualisations to investigate how different presentation styles influence encoding and representation. Experiments will also investigate uncertainty and ambiguity in visually presented data, and the influence of accompanying text. In addition, I will explore the visual behaviours associated with comprehension of data visualisations. Participants' eye movements, recorded during exposure to graphs in the first two PhD studies proposed here, will be analysed with machine learning algorithms (e.g. principal component analysis). This will provide an unbiased method of identifying areas of interest (like Davies et al., 2017) and emergent patterns of behaviour associated with comprehension. This data will be used to simulate experiments, generating predictions which will then be tested with highly controlled studies. For example, if absent fixations to a particular area are associated with poorer comprehension, can emphasising those areas improve comprehension, or facilitate resolution of ambiguity?

The studies outlined here will provide valuable insights into how data in graphs are mentally encoded and represented. All experiments will be linked not only by their exploration of how 'situation models' are built, but also through the comparison of visualisation formats, old and new. Together, this body of evidence will produce both theoretical implications about the representation of graphically-presented data, and practical implications relevant to journalism, academia and human-computer interaction. For example, it will highlight the effects of accompanying text and the potential for facilitating resolution of ambiguity. In addition, comprehension measures are likely to generate implications for data visualisation choice and design, highlighting the most suitable formats for presenting data, and also the possibility of making certain aspects more salient to facilitate interpretation. This research will make an important contribution to current knowledge on data visualisations, advancing understanding of graph comprehension and ultimately helping to improve design to benefit both users and creators.