That said, it still surprises me that these map makers (all of them! not just this one) don't take the simpler approach of drawing all dots in a random order - e.g. get rid of this "first all blue dots, then green, etc" behavior, and pick the next dot's color randomly, while maintaining the distribution. Then you'd have an unbiased "top layer" regardless of the size of your dots (up to a point, obviously).
(hard to describe what I'm thinking of... basically `[red, red, ... * 25, green, green, ... * 10, orange, orange].shuffle.draw` rather than just `[].draw`)
Otherwise, just look around at densely populated areas. Like this asian-heavy area: http://cl.ly/image/1z0Q1T1V2V0N (edit 2: turned off smoothing, that's actual pixels) note that immediately-surrounding areas have blue dots, but not that one. Meanwhile, you can see less than one dot sized dots for orange and blue poking out, but no orange or blue complete dots on top of the red. Look at other dense areas, you'll see the same thing - some colors are exclusively on top of others, and it's always the same ones. (edit: for context, that block is in the middle here, in San Francisco: http://cl.ly/image/3m3N3f423S0M)
There isn't enough room for them to have fit one person per pixel, so they're stacked (somewhat). This favors the ones stacked on top.
This one, for instance, appears to draw (bottom up) "White", "Black", "Hispanic", "Other", "Asian" - this over-represents the races higher up on the list, especially in densely populated areas. That solid block of red? It probably has other colors underneath, but you can't see them, and they don't tell you this.
This is one of the nicer maps like this that I've seen, but excluding that info is blatantly misleading.