Christophe Bontemps (Statistical Institute for Asia and the Pacific, United Nations)

How to lie with graphics

vendredi 5 mars 2021, 11h00 - 12h00


Graphical methods have proved for a long time to be powerful tools both for data analysis and for data-based communication. However, data visualization can create distorted views of the underlying data. Voluntarily, or involuntarily, many students, but also researchers, data analysts, data journalists, etc. make mistakes in their practices. The result is that the visual tests performed for data analysis, when the analyst wants to study or question a dataset, can be misleading to a wrong conclusion. The structure of the data, patterns and (ir) regularities, groups, trends, outliers observed on the screen may be misunderstood, leading to serious errors in the understanding of the underlying phenomena. Even worse, when used as a tool for communication and, as such, as a visual language, some graphics may present spurious relationships, false proportions and biased views. In this session, we show through real-life examples and case studies, how to manipulate some of the principles and rules of data visualization in order to deliberately play and then « lie » with graphics. Our goal here is not Machiavellian but pedagogical: by showing the power of some bad practices, we preach for better ones.