Data collection is becoming ubiquitous: from smart phones and watches to large businesses and governments. However, collecting data is only a first step: we need to make sense of data through analysis, to extract knowledge and patterns from it, so that we can learn, reflect and improve knowledge and processes. By visually representing data we can identify broad structures and subtle nuances that help with this process. Visual analysis can benefit both from a computer’s incredible abilities to deal with number as well as a human’s experience, interpretation and intuition. In ALGOVIS we are developing ways of producing graphics of data to help people in their analysis of numbers. We use geometry and computation to present numbers in ways that allow people who study them to see structure and patterns clearly.
In many cases, data have a geographic context, such as the country, county or location to which the data relate. There are various techniques to show such data on geographically accurate maps – weather maps are an example. These are appropriate for showing geographic processes that change dynamically and for navigation and planning. Equally techniques exist for showing data entirely without geography, if the statistical properties of the data are deemed to be more important than where the data were recorded. Bar charts and bubble plots are examples. However, the middle ground of spatially informative graphics – visualizations that show data in a “roughly geographic way” so that we can consider statistics and geography concurrently – is underexplored, with only a few techniques developed in a somewhat ad-hoc manner.
In this project, we set out to combine visualization design with algorithmic rigor, to get a better understanding of what it means to show data in a spatially informative way that is “roughly” geographic. We aim to develop techniques that retain important characteristics of geography in these rough maps and using computers to explore the extents to which this is possible given different characteristics and data sets and to create these maps automatically.
The current primary result of this project is a better understanding of the spatial deformations that happen in spatially arranged small multiples. In such a “small multiples map” or “grid map”, each region of interest is represented using a simple rectangle, functioning as a container for other statistical (often nongeographic) visualization. These rectangles are the small multiples, and by arranging them in a roughly geographic way, we obtain spatially informative visual representations. By collecting and developing various metrics to capture aspects of geography (displacement, shape, topology, etc.) and measuring and optimizing these on a large variety geographic regions, we were able to establish relations between them and analyze manually crafted layouts for a sense of priority among the metrics. In particular we also looked at cases where there is more space for the rectangles than strictly necessary – that is, empty rectangles or “gaps” are used to increase the spatial fidelity of the small multiples array.
