At dataviz.shef.ac.uk we generally focus on practical matters such as how to create plots using Python and best practice for choosing accessible colours for visualisation. But as researchers and practitioners we also need to be conscious of the wider social impact which can arise from visualising data and consider how to do so responsibly. This post was inspired by two recent Open Access publications; firstly, Data Visualisation in Society (2020, Amsterdam University Press) edited by Sheffield’s Professor Helen Kennedy and Professor Martin Engebretsen from the University of Agder. This book covers a wide range of topics looking at the social aspects of visualisations, from the politics and semiotics of visualisations through to the need for new literacies to understand them. Secondly, Data Feminism (2020, MIT Press) by Catherine D'Ignazio and Lauren F. Klein which explores different aspects of data science from an intersectional, feminist perspective. It is framed around 7 principles which include examining power relationships, rethinking the process of classifying data, context, valuing different types of knowledge and making the labour involved in data science visible.
this does not mean they are more true, in the sense that they offer a more objective representation of the world.
Both books draw attention to the power of data visualisations, especially as a form of communication which is trusted more than written accounts, but as Kennedy and Engebretsen observe “this does not mean they are more true, in the sense that they offer a more objective representation of the world” (p. 24). Data is rarely ‘raw’, it is the consequence of conscious decisions by those who collected the data and the social structure where it was collected. For example, D’Ignazio and Klein highlight a case where historic data from a range of social services was used to train a model to predict child abuse. As poorer families had previously had more interactions with public services there was more data about them in the databases used, so the model over-estimated the levels of abuse in this community compared to wealthier communities (p. 39). This presents a challenge and places a responsibility on those visualising data to consider what is being communicated, intentionally or not, through a visualisation. Traditionally the response to this has been to avoid evoking emotion and attempt to be ‘neutral’, something D’Ignazio and Klein are critical of as likely to lead to representing the perspective of the dominant groups in society (pp. 75-6). Instead they suggest, citing the works of Donna Haraway, Sandra Harding and Linda Alcoff, that instead of aiming for a false neutrality researchers should situate their visualisations and the perspective the maker is coming from to help frame visualisations (p. 83). By doing so it is possible to help viewers understand the context of a visualisation and better assess biases inherent in many datasets.
None of this is to suggest that data visualisations are not important in communicating research but it is important to recognise that they are “abstractions and reductions of the world, the result of human choices, social conventions, and technological processes and affordances, relating to generating, filtering, analysing, selecting, visualizing and presenting data.” (Kennedy & Engebretsen, p. 22). Visualisations have great power, as Kennedy & Engebretsen observe, they “often serve as a main entry point to data for non-experts” (p. 20), so what steps can be taken to ensure that they are used responsibly and also understood by the intended audience?
Firstly, as makers of visualisations, there is an onus on us to consider both the context and the audience (D’Ignazio and Klein, p. 91). This includes ensuring there is sufficient contextual information to make the viewer aware of any gaps in the dataset, being explicit about the biases we all bring to our work, and considering whether it is legible to the intended audience. Audience understanding is also highlighted by Kennedy and Engebretsen who call for data visualisation literacy to avoid another digital divide between those who can critically evaluate visualisations and those who can’t. Both books draw attention to the power of visualisations and thus the need to be careful and thoughtful when creating them.
Kennedy, H., & Engebretsen, M. (2020). Introduction: The relationships between graphs, charts, maps and meanings, feelings, engagements. In Kennedy H. & Engebretsen M. (Eds.), Data Visualization in Society (pp. 19-32). Amsterdam: Amsterdam University Press. doi: 10.2307/j.ctvzgb8c7.7
D’Ignazio, C. & Klein, L. (2020). Data Feminism. MIT Press.