Visualizing Georeferenced Data: Representing Reliability of Health Statistics

TLDR

Human visual pattern recognition underlies effective data visualization, yet it can mislead when georeferenced health data reliability is ignored. The study applies an integrated cognitive‑semiotic framework to develop and evaluate three visualization methods that convey reliability of georeferenced health statistics. One method pairs separate maps for data and reliability, while two other methods combine them on a single map, allowing comparison of adjacent versus coincident representations. The coincident, visually separable approach—color for data and texture overlay for reliability—enables users to detect unreliable units without obscuring mortality clusters, whereas a visually integral design hampers cluster detection and independent assessment of reliability.

Abstract

The power of human vision to synthesize information and recognize pattern is fundamental to the success of visualization as a scientific method. This same power can mislead investigators who use visualization to explore georeferenced data—if data reliability is not addressed directly in the visualization process. Here, we apply an integrated cognitive-semiotic approach to devise and test three methods for depicting reliability of georeferenced health data. The first method makes use of adjacent maps, one for data and one for reliability. This form of paired representation is compared to two methods in which data and reliability are spatially coincident (on a single map). A novel method for coincident visually separable depiction of data and data reliability on mortality maps (using a color fill to represent data and a texture overlay to represent reliability) is found to be effective in allowing map users to recognize unreliable data without interfering with their ability to notice clusters and characterize patterns in mortality rates. A coincident visually integral depiction (using color characteristics to represent both data and reliability) is found to inhibit perception of clusters that contain some enumeration units with unreliable data, and to make it difficult for users to consider data and reliability independently.