The Gini overweights changes in inequality that happen in the middle of the income distribution, which is exactly not what we should be looking at, given how stable the middle is.
I'm not sure why it even matters that its supposedly "better"; its not like you can't look at multiple measures, or that distribution is something simple that can be adequately measured by a single unidimensional measure.
Palma intentionally obscures all inequality within a certain range to focus on the extremes, Gini is sensitive across the whole spectrum. Why not use both together?
GoodBadStats praises the intuitive nature of the Palma, but has concerns about the rest of the argument in favor of it.
Especially the part where the author argues Palma can safely ignore the middle 50% because it never changes, but then says Gini is over-sensitive to changes (which apparently don't exist?) in the middle 50%.
Probably good to have several measures kicking around, or maybe any one of them is just as good as any other. If there's a relationship in some study between one of these indices and some other factor (famers per capita, land area of streams, whatever), it's unlikely to suddenly disappear just because you use another index to approximate the same thing.
3 comments
[ 2.4 ms ] story [ 15.9 ms ] threadExcept when it isn't.
Palma intentionally obscures all inequality within a certain range to focus on the extremes, Gini is sensitive across the whole spectrum. Why not use both together?
Especially the part where the author argues Palma can safely ignore the middle 50% because it never changes, but then says Gini is over-sensitive to changes (which apparently don't exist?) in the middle 50%.
Some comments here: http://goodstatsbadstats.com/?p=1836
The Hoover Index / Robin Hood Index is another interesting measure: http://en.wikipedia.org/wiki/Hoover_index
Probably good to have several measures kicking around, or maybe any one of them is just as good as any other. If there's a relationship in some study between one of these indices and some other factor (famers per capita, land area of streams, whatever), it's unlikely to suddenly disappear just because you use another index to approximate the same thing.