Hacker News new | ask | show | jobs
by bialpio 1323 days ago
Some metrics aren't linear so w/o knowing more about Gini coefficient, my first thought is "I have no idea if the difference is significant or not". Can someone ELI5 this so that I can build an intuition for what "1 unit of Gini" means?
1 comments
ch4s3 1323 days ago
It's a curve reflecting income (not wealth) share of a population against a line of perfect equality, which is a 45 degree angle. A low disparity hugs the line and a high disparity hugs the X and Y axis. Gini = A/(A + B) where A is area over the curve and B is the area under the curve. So an increase of 0.1 in the gini number reflects a larger A.

It's not a very good way to measure what it is trying to measure[1].

[1] https://en.wikipedia.org/wiki/Gini_coefficient#Limitations

link
bialpio 1323 days ago
I think the main problem is the lack of intuition of what "1 Gini means", except the "lower is better". Is difference between coefficient of 10 & 11 the same as difference between 30 and 31? The poster to which I responded said that "32 vs 41 is not far apart" - is it? Is difference between 10 and 19 the same as difference between 32 and 41 (delta is the same)? How about between 0 and 9?
link
ch4s3 1323 days ago
That was me, and they aren't that far apart. Is 41% of the area under a lot more than 32%? Not really, see this example[1] Norway here has a coefficient of 27.5 and the US was at 41.2 but the graphs are barely different except that you can see that the US has a sharp bend on the far right hand side. The gini formula is really sensitive to that in a way that doesn't tell us much.

[1] https://wol.iza.org/uploads/articles/495/images/IZAWOL.462.g...

link