Hm, good point. I suppose there are two parts to my assertion: 1. When you think of yourself as talking to an actual person, you are less likely to be dumb an

Hm, good point. I suppose there are two parts to my assertion: 1. When you think of yourself as talking to an actual person, you are less likely to be dumb and/or mean. 2. As HN has grown, the degree to which the average commenter considers themselves to be talking to an actual person has decreased. I was pretty sure I'd be able to find a psychology or sociology study somewhere giving some evidence on (1) or a more general case of (2) for large groups. Wildly, it looks like all of the interest is on how groups reduce your personal identity (deindividuation), not how they reduce the identity of others to you. So I guess without an appeal to prior research, the best thing would be to develop a measure of "does this comment seem like it's addressed to a real person". You could do that by manually going through and rating them, as (I believe?) you already did with quality metrics. You could also come up with some heuristics and cross-check them with your own assessment. Here are some I can think of: * Use of parent's username * Use of first names * Use of other usernames * Use of "you", "your". * Email addresses, phrases like "contact me", "email me", etc. * Questions. * Comment reply chains (ie people replying to a descendent of their own comment) * "Please", "Thanks" and "Thank you" I'm not saying that I think comments with this quality are necessarily good, but my prediction is that per (2) the proportion of such comments correlates inversely with the number of active users and per (1) that proportion correlates with a drop in comment quality.