Designing More Effective Data Visualizations

Complex datasets often need to be communicated visually to non-experts to inform their understanding or decision making about the topic at hand. This is a key area of Human Computer Systems research on data visualisation. The design choices used by those creating data visualisations can bias how the datasets are subsequently interpreted - and what decisions a viewer might subsequently make.

When producing a graph, journalists have many degrees of freedom in their design, such as colour, line weight, labelling and more. Historically, data visualisation experts such as Edward Tufte have made recommendations on use of these features, such as using harmonious colour palettes and removing redundant features (Tufte 2001, Kirk 2016). However, empirical testing of how these graphical design principles affect graph comprehension is limited. To what extent does variation in graph design impact perceived magnitude of an issue?

Recent research the group at UoM have published in venues such as IEEE Transactions on Visualization and Computer Graphics, Behaviour and Information Technology, and at The ACM Conference on Human Factors in Computing Systems (CHI) shows that axis truncation (Bradley et al. 2023), opacity of plotted data points (Strain et al., 2023), and colour choices in choropleth maps (Bradley et al., 2023) can all systematically influence the inferences that viewers make when presented with a variety of different data visualisations. The proposed PhD project will examine how a variety of additional design elements (including data source) come to have an influence on the interpretation of datasets associated with a variety of real-world topics (e.g., climate change, CO2 emissions etc) adopting both quantitative and qualitative approaches.

Methodology Study 1: Interviews and Social Media Research

To identify design features to empirically investigate, Study 1 will involve a review of the most viewed and shared data visualisations in the news, on platforms including X, Facebook and Reddit*. A preliminary database systematically classifying these data visualisations according to their type, use of colour, and other design features will be created.

During this time, semi-structured interviews with members of the general public will be carried out. (Gill et al. 2008). After obtaining ethical approval, interviews will involve a discussion on the participants' news reading habits, understanding and consumption of data visualisation, and their opinions on data visualisation design. Features of data visualisations that may have been missed will be reviewed, and additional sources and platforms investigated.

By the end of this study, trends will have been identified in the design of data visualisation in the news. These findings will form the basis for Study 2.

*Note: Some platforms such as X have recently limited researcher access to its API. If this social media review becomes unfeasible as a result, study 2 can instead be informed by the qualitative study's findings. Study 2: Empirical HCI Research - Testing Visualisation Design Features

The second part of this study investigates how design features identified in Study 1 affect comprehension of data visualisations. Qualtrics will be used to obtain a large (n > 200) sample for a between-subjects online study (Serdar et al. 2021). For each feature of interest (e.g. use of saturated vs unsaturated colours) two identical visualisations will be created using the R package ggplot2, differing only in that feature.

In an experiment run on Pavlovia, each participant will be given one of either data visualisation for each feature of interest. Participants will rate their perceived magnitude of the issue illustrated in the graph, using a 0-100 scale (Voutilainen et al. 2015). Demographic information will be collected but participants will remain anonymous in accordance with GDPR requirements. For each data visualisation pair, statistical tests (mixed models) will be carried out using R