Evaluation of visualisations of geographically weighted regression, with perceptual stability

Spatial analysis (Statistics);Geographic information systems

Burke, Tommy ; Demšar, Urška ; Demšar, Urška

University:University of St Andrews

Department:Earth & Environmental Sciences and Geography & Sustainable Development (Schools of)

关键词: Spatial analysis (Statistics);    Geographic information systems;   

Others  :  https://research-repository.st-andrews.ac.uk/bitstream/handle/10023/15680/TommyBurkePhDThesis.pdf?sequence=2&isAllowed=y

来源: DR-NTU

【 摘 要 】

Given the large volume of data that is regularly accumulated, the need to properly manage,efficiently display and correctly interpret, becomes more important. Complex analysis of datais best performed using statistical models and in particular those with a geographical elementare best analysed using Spatial Statistical Methods, including local regression. Spatial StatisticalMethods are employed in a wide range of disciplines to analyse and interpret data where it isnecessary to detect significant spatial patterns or relationships. The topic of the researchpresented in this thesis is an exploration of the most effective methods of visualising results.A human being is capable of processing a vast amount of data as long as it is effectivelydisplayed. However, the perceptual load will at some point exceed the cognitive processingability and therefore the ability to comprehend data. Although increases in data scale didincrease the cognitive load and reduce processing, prior knowledge of geographical informationsystems did not result in an overall processing advantage.The empirical work in the thesis is divided into two parts. The first part aims to gain insight intovisualisations which would be effective for interpretation and analysis of GeographicallyWeighted Regression (GWR), a popular Spatial Statistical Method. Three different visualisationtechniques; two dimensional, three dimensional and interactive, are evaluated through anexperiment comprising two data set sizes. Interactive visualisations perform best overall,despite the apparent lack of researcher familiarity.The increase in data volume can present additional complexity for researchers. Although theevaluation of the first experiment augments understanding of effective visualisation display,the scale at which data can be adequately presented within these visualisations is unclear.Therefore, the second empirical investigation seeks to provide insight into data scalability, andhuman cognitive limitations associated with data comprehension.The general discussion concludes that there is a need to better inform researchers of thepotential of interactive visualisations. People do need to be properly trained to use thesesystems, but the limits of human perceptual processing also need to be considered in order topermit more efficient and insightful analysis.