“As you can see in Figure 1…” may well be the most frequently made claim in science. Unlike claims involving data, statistics, models and algorithms, those relating to visualisations are rarely evaluated or verified. So how do data scientists understand visualisations’ effectiveness and expressiveness when they are developing these graphs, maps and networks?
Explaining the science
This project is using three user research methods to understand how people visualise data in terms of workflows comprised of ‘decision-making’, ‘making queries’ and ‘coding and making’ the visualisation. These methods operate at different scales and depths, and will produce a variety of views into the visualisation work of data scientists which will direct the design and development of visualisation solutions.
Designing effective visualisations goes far beyond selecting a graph, its scales and a ‘pretty’ style. Effective visualisations must negotiate sensitivities and interactions between visual elements (e.g. encodings, coordinate systems, guides, annotations), data (e.g. characteristics, transformations, partitions), and the discriminator function, which in this case is the perceptual and cognitive systems of humans. Despite their criticality, these methodological and design considerations are rarely surfaced, limiting the value extracted from visualisations. What does 'Figure 1' actually visualise?
The project's end goal is to develop tools which enhance people’s capacity to visualise data, by letting them see what can and can’t be seen in the visualisation. The outcomes will be made available in R and Python which are used widely and intensively in data science. The goal is to further develop the substantial graphics functionality of these programming languages, into a greater capacity to visualise data and understand the properties of those visualisations.
Visualisation, formerly known as plotting data or creating maps, is an everyday activity in data science, having a pivotal role in exploratory data analyses, making sense of data and communicating information. Yet visualisation may be something we have received far less training in. As such, it is unlikely that one has scoured the visualisation literature to understand the impact of our design choices. By closing the epistemological and methodological distance between crucial visualisation concepts and the project team's work, the WAYS projects aims to help any data scientist.