Clustering
Use covariance of goal scores to cluster OHI regions, similar to how covariance of stock prices produced this visualization of stock market structure:
- NAs. Some regions have NAs for goals, such as TR/LE/AO for unpopulated regions (eg Jarvis Island).
- Interpretation. For a given cluster we want to be able to quantitatively describe the drivers for similarity within a cluster and the differences between clusters.
- Static 2D. For manuscript sake, we need a compelling 2D static image.
- Term Contour plots Per predictor term driving the clustering differences, the clusters could be plotted with a term in contour, similar to a GAM bivariate term plot. For exmaple, see Figures 4-6 in Schick et al (2011).
-
Dynamic Graph. Would be cool to see a simple graph to start with one node per cluster with each node showing the average flower plot.
- Perhaps some subtle visual techniques could describe the mean value (hard edge to flower plot) and standard error (darker shading around mean towards center and out to edge).
- A cluster could be clicked and the flower plots for all constituent regions would pop out.
