Closed venkatachalapathy closed 5 years ago
@srikanthmudigonda and @WheezePuppet , are there standard tests for this? And in what packages?
Hmm, ginis are ratios, does that mean two population proportion test? Asking for a friend.
On Mon, Jun 3, 2019, 15:04 venkatachalapathy notifications@github.com wrote:
@srikanthmudigonda https://github.com/srikanthmudigonda and @WheezePuppet https://github.com/WheezePuppet , are there standard tests for this? And in what packages?
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/WheezePuppet/specstar/issues/37?email_source=notifications&email_token=ALNPB246HMZN6TMWMT3LCHTPYVTK3A5CNFSM4HSDTKZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODW2MA7Q#issuecomment-498385022, or mute the thread https://github.com/notifications/unsubscribe-auth/ALNPB25ON45FVLZUNXGSKDTPYVTK3ANCNFSM4HSDTKZA .
https://cran.r-project.org/web/packages/dineq/index.html
https://cran.r-project.org/web/packages/dineq/index.html
both seem to do it (using the proportion test like what you (jeff) suggested) via confidence intervals but since we just removed RCall
dependencies, we have either create tests from scratch or code it ourselves. But the ineq
package we used don't seem to have this feature.
and also this one
https://www.rdocumentation.org/packages/DescTools/versions/0.99.19/topics/Gini
Surely we're not claiming that standard statistical tests like the ones you learn about in STAT 101 are not available here?? https://github.com/JuliaStats (See, for instance, https://github.com/JuliaStats/HypothesisTests.jl) If we actually need to call R from Julia in order to run a freaking t-test, I'll jump off the highest bridge in my field of view with no parachute.
You don't have to jump of the bridge, Stephen. t-tests will do for now. I was kind of expecting some special test just for Gini comparisons. t-tests usually assume normally distributed data. I do not know enough about Gini to know if they are asymptotically normally distributed. But OK.... we can make do with what we have ;)
Hmm, I took a gander at that link... Not so sure about the t test straight up. Looks like the distribution for ginis has a funky srandard error. Worst case, can you export what you need to a file and just run R on the file?
On Tue, Jun 4, 2019, 17:43 venkatachalapathy notifications@github.com wrote:
You don't have to jump of the bridge, Stephen. t-tests will do for now. I was kind of expecting some special test just for Gini comparisons. t-tests usually assume normally distributed data. I do not know enough about Gini to know if they are asymptotically normally distributed. But OK.... we can make do with what we have ;)
— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/WheezePuppet/specstar/issues/37?email_source=notifications&email_token=ALNPB242VGN5QC7PPYP5OCTPY3OYJA5CNFSM4HSDTKZKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODW56X3A#issuecomment-498854892, or mute the thread https://github.com/notifications/unsubscribe-auth/ALNPB27AGEDY5EOPTBHNFDDPY3OYJANCNFSM4HSDTKZA .
Ah right, normally distributed data, which might well not be true for Ginis. Good point. So, I see that https://github.com/JuliaStats/HypothesisTests.jl also supports Wilcoxon and Mann-Whitney tests -- would these be more appropriate?
(Btw, my "jump off a bridge" comment was not related to t-tests being necessarily the correct thing, but rather the idea that we would have to call out to R for something like this. I would be astonished if whatever the correct test is for this were not implemented natively in Julia.)
R
library DescTools
This should do the job for now.
I'm closing this as we have enough of this figured out to complete phase 1
and pushed pushed the two-sample version to phase 2+
.
Once we have Gini coefficients from different conditions, we may have to compare them statistically. The issue is to create such tests or compare histograms directly with standard methods.