The correlation coefficient is actually about 0.23 according to the paper. It just looks really correlated when they bin it by deciles but that's a visual artefact.

If you plotted individual people in a scatter plot this is what 0.23 correlation looks like: http://www3.cmp.uea.ac.uk/~jrk/distribution/correlations/cor... i.e. you can barely tell any positive relationship visually.