The Friendship Paradox

en.wikipedia.org

27 points·chirau··4 comments

Would this in general be true of any attribute that is significantly skewed? I.e., if you remove someone at random from a pool of n individuals, it's more likely to be someone at the lower range, and so the remaining (n-1) persons will be higher on average?

I'm also reminded of the paper making the rounds a couple of years ago showing that uniform allocation of wealth, combined with simple exchange processes, in the presence of randomness, can lead to very skewed wealth distributions independent of anything else.

Any chance you remember what the paper was called? I'd love to read it.

Ugh. Unfortunately I can't find it anywhere (there's a couple of things like this lately, that I remember being discussed a lot on HN and being covered in the news, and now have disappeared in the ocean of the internet).

However, these discuss or demonstrate very similar things (especially the first two):

https://www.scientificamerican.com/article/is-inequality-ine...

https://medium.com/swlh/extreme-wealth-inequality-is-inevita...

https://www.osti.gov/biblio/1474734

It's weird to me I can't find it because it was picked up in some large news and economics outlets. Basically, there was an agent-based simulation with really simple conditions that showed that even when everyone starts out with the same income, and you have some really simple exchanges of income that are essentially random, you end up with income distributions that look really similar to actual income distributions in the US.

I think the Scientific American piece discusses some of the same phenomena, and a similar simulation is discussed in the Medium piece. The paper that was going around was very simple in its assumptions, though, and very clearly written.

Average is higher than median. But if you compare two friends, by symmetry you can’t tell who has more friends.

Reply on news.ycombinator.com