Multivariate Visualization in Social Sciences and Survey Data
Abstract
For presentation of survey results, social science data, and other geospatial statistics requires careful attention in order to facilitate fast and accurate interpretation. Adding dimensionality can easily saturate the observer, leading to confusion instead of adding perspective. We produce over a dozen techniques to facilitate multivariate geospatial visualization, filter them with pilot groups, and then design a computer-based human experiment to evaluate their relative performance. In the experiment, the participants locate (with a mouse click) regions with extreme primary or secondary values and then later estimate numerically the values of these variables. We analyze these data with linear and logistic regression and general additive models to characterize the variance due to a learning effect, and then use general linear mixed-effects models to block out the variability due to individual participants and the independent and randomly-generated survey data used to generate the experiment plots. The effectiveness of a particular technique depends heavily on the goal of the presentation: a technique that provides relative perspective without distracting from the primary variable may not facilitate estimation that is as accurate as other techniques. Four scenarios are provided to qualify the presenter s intent. Only one technique performed poorly in all four scenarios and only one technique was average in all four; all remaining varied from very good to very bad between scenarios.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2013
- Accession Number
- ADA589793
Entities
People
- William Evans
Organizations
- Naval Postgraduate School