Vitalik Buterin
@vitalik.eth
Medians are vulnerable to some interesting statistical paradoxes that means are not. For example, consider this scenario: * The median family owns $100k in real estate * The median family owns $100k in DOGE * The median family net worth is $100k * Nobody has any debt This is possible! Challenge: figure out how.
17 replies
62 recasts
206 reactions
Daniel Lombraña
@teleyinex.eth
A really good bok about statistics https://www.statisticsdonewrong.com/ For these scenarios we should be using the mode.
1 reply
0 recast
0 reaction
shazow
@shazow.eth
Wouldn't mode also fail to capture an internally consistent cross section for the same reasons?
1 reply
0 recast
0 reaction
Daniel Lombraña
@teleyinex.eth
Yes and no. If you get the mode you will get them by grouped most common values. Then you can place those values in quartiles and have a better idea about the population. My intention with the mode was to get the most common value in the population.
1 reply
0 recast
0 reaction
shazow
@shazow.eth
I think the criticism comes from "the average person does not exist", vitalik points out that we can get similar edge cases with medians, same issue would apply with mode: If we take the mode across multiple metrics of a family, the combined family may not be internally consistent.
1 reply
0 recast
1 reaction
shazow
@shazow.eth
Once we start looking beyond singular cross-sections, that's a whole different ballgame. :) The issue is that a lot of policy decisions are done by data reduced to singular scalars, not complex multi-modal datasets that are carefully considered.
1 reply
0 recast
0 reaction
