Hello there friend.

so this was inspired by two previous polls on the same question, one of which has unpredictable ad hoc bracketing and the other has uniformly sized bins.

they both show that most tumblr users have a follower count in the low three digits, but there are non-negligible populations with much more than that. we have a distribution where the values can span orders of magnitude, and we want to get some idea of what it looks like at various 'scales'.

A uniformly-sized bins approach has to choose between a small bin size that gets you a decent resolution for the small values where most people are, but means all the big values are collapsed into the last category without getting much of a clue of what the distribution looks like there, versus a large bin size that gets you a 'bigger picture' of the range but at very low resolution. With increasing bin sizes, you can get the benefits of both at once!

so, exponential bins. I picked an integer power just for simplicity's sake, and 2 felt like it grew too slowly (the last bin would've been "513 or more", which certainly seems too early to saturate), so i went for the next integer, 3. (I considered 4, too. seeing the data now, maybe 4 would've been a better option, but obviously i didn't have the data then and i had to make a guess)