Nice post, Shreyas Raghavan; thanks.

One other technique is binning the data. You can do it on a grid, of course, or also a hexagon tiling:

There can be some big problems with binning too, especially if you don’t have enough data or there are natural clusters or thresholds in it that get wiped away: then it’s easy to tell various stories, just based on the arbitrary bin sizes/positions. But if it’s more continuous data, it can work out.

Of course, we can always summarize the data or show a model instead. It’s a big choice going in to a project: do I show all the data, or some part of it?