Okay, they messed up something here. The number of little ASCII-art person squares depends on the size of your browser window. The squares get smaller when you make the window narrower, so it looks like it was coded to try to keep the number roughly constant.

If I make the window narrow enough, there are 10 squares in a row and 19 rows, a total of 190 squares. The number that are coloured "very conservative", "conservative", "centrist", "liberal", "very liberal", respectively, are 39, 67, 24, 31, 29.

In percentages, that's 20.5%, 35.3%, 12.6%, 16.3%, 15.3%. Roughly 56% conservatives, 32% liberals.

If I make the window really wide, I see 20 squares in a row and 13 rows, a total of 260 squares. The distribution is now 39, 100, 37, 46, 38.

In percentages, that's 15.0%, 38.4%, 14.2%, 17.7%, 14.6%. Roughly 53% conservatives, 32% liberals.

It's weird that the number of squares increases and decreases when you resize the window, and I would argue it's misleading because there's an animated transition that is obviously meaningless. But it's a lot worse that the proportions aren't consistent! All of us saw exactly 39 in the "very conservative" category, so maybe it is failing to proportionally scale that category while scaling the others?

Conclusions:

1. There's a programming bug that misrepresents the proportions.

2. The sample is significantly skewed toward conservatives.

Great work... I thought I noticed something similar on the resizing. I guess they prioritized looks over accuracy which is kind of fair although its not obvious why they would have to do it this way.

In any case, it would be good to see the actual data for this stuff.

Thanks for the correction ^^