This is such a detailed article but it's giving me weird vibes.
For instance there are all these drops to near-zero in the histograms at .28, .46, .56 for no clear reason, and the article doesn't even consider that noteworthy.
The "Men Like ratio (y) vs ratio (x)" has an inexplicable wall around .33 which I could only explain with some sort of product limitation maybe? But I really wish it was explained what artifacts the product introduces.
Since there's a spike followed by a drop, it seems like some of the data points are "misattributed" to the neighboring bucket.
Since it happens at the same place in each graph (eg a spike at 0.28-0.29, followed by a drop at 0.29-0.30) I wonder if it's some kind of number-theoretic effect from the fact it's actually a ratio of integers. For example, with less than 20 views there's no way to get to the 0.29-0.30 bucket, but 4 ways to get into the 0.28-0.29 bucket. Hmm.
Definitely points to some rounding error, aliasing in the data. It would be fixed by making the buckets larger. No reason for the buckets to be that small.
For instance there are all these drops to near-zero in the histograms at .28, .46, .56 for no clear reason, and the article doesn't even consider that noteworthy.
The "Men Like ratio (y) vs ratio (x)" has an inexplicable wall around .33 which I could only explain with some sort of product limitation maybe? But I really wish it was explained what artifacts the product introduces.