Evaluating data classification methods for choropleth maps in South Africa : a usability study

Abstract

Today, location is entrenched in many organisations in both the public and private sectors. Organisations, both locally and internationally, are realising the importance of location – and therefore maps – but do they know how to visually communicate this spatial knowledge? Geospatial data visualisation techniques have evolved rapidly over the past decade. Today, most geographic information system (GIS) software has a plethora of built-in spatial analysis and visualisation techniques that enable users to quickly and effortlessly visualise spatial patterns in data. Choropleth maps are among the oldest and still one of the most frequently used techniques for visualising quantitative data in a GIS. The challenge with using choropleth maps in South Africa is selecting a data classification method that effectively displays unequal and dispersed population densities. The aim of this research was to assess the suitability of different data classification methods for effectively visualising population demand using choropleth maps in South Africa. The research focused on geographic accessibility as a use case where choropleth maps are used to visualise population demand, allowing decision makers to identify over- or underserved areas for the provisioning of service centres. This was achieved with a user study. The user study included the design of an online questionnaire featuring map interpretation questions specifically related to geographic accessibility. Subsequently, the results from the user study were compared to a recommended mathematical equation that measures the error between class breaks, in a data classification method. The user study shows that respondents were more likely to provide correct answers when presented with maps using the quantiles and natural breaks (Jenks) data classification methods, suggesting that these methods are easier to interpret and analyse for understanding population distribution in South Africa. A goodness of variance fit calculation that measures the error between class breaks delivered somewhat different results. Based on these calculations, natural breaks (Jenks) and geometric interval were considered the optimal data classification methods, while logarithmic scale and quantiles were ranked lowest. Based on the results of both the user study and error calculations, a more comprehensive view of the use of data classification methods was obtained. This research emphasises the importance of including human interpretation when assessing methods or techniques used to represent spatial phenomena.