Add Bartlett's and Levene's tests for homogeneity of variances #216
Conversation
Shimuuar
left a comment
Shimuuar
left a comment
There was a problem hiding this comment.
Thank you for PR! I'll be busy whole April and will need to read about both tests. So thing will probably move slowly.
There're few comments on things that caught my eye,
| -- Degrees of freedom and chi-squared distribution | ||
| df = k - 1 | ||
| chiDist = chiSquared df | ||
| pValue = mkPValue $ 1 - cumulative chiDist tStatistic |
There was a problem hiding this comment.
complCumulative should calculates 1-CDF without loss of precision
AFAIR χ² doesn't implement it but it may at some point
Statistics/Test/Levene.hs
Outdated
| zbari = map Sample.mean deviations | ||
| zbar = sum (zipWith (*) zbari (map fromIntegral ni)) / fromIntegral n | ||
|
|
||
| numerator = sum [fromIntegral (ni !! i) * (zbari !! i - zbar) ** 2 | i <- [0..k-1]] |
There was a problem hiding this comment.
!! is generally smell since it's O(n). zbari and ni should probably be vector
|
Hi @Shimuuar! Thanks for the helpful feedback!
Thanks again, and no worries at all about things moving slowly — I really appreciate your time and review whenever you get a chance. |
Shimuuar
left a comment
Shimuuar
left a comment
There was a problem hiding this comment.
In general PR is fine.
complCumulative – I noticed that chiSquared currently doesn’t implement complCumulative. So for now, I’ve added a TODO comment in the code noting that this could be replaced with complCumulative when it becomes available.
It's not implemented so default implementation will be used instead. And if class method gets implemented it will be used automatically, no need to change anything in test implementation.
Please also add simple unit test in Tests.Parametric that checks both test compute approximately same test statistics and p-value as python/R/whatever.
Statistics/Test/Levene.hs
Outdated
| levenesTest :: Double -- ^ Significance level (alpha) | ||
| -> Center -- ^ Center calculation method | ||
| -> [V.Vector Double] -- ^ Input samples | ||
| -> Either String TestResult |
There was a problem hiding this comment.
Function should return Statistics.Test.Types.Test as Bartlett do,
There was a problem hiding this comment.
I have now reworked Levene's Test to return Test instead of TestResult.
Hi @Shimuuar! I've pushed the requested changes:
Please let me know if any further adjustments are needed! |
|
Well. It needs quite a bit of improvements but it would be easier to just merge it as is and fix it later |
2 participants
Summary
Changes
Statistics.Test.BartlettStatistics.Test.Leveneexposed-modulessection in thestatistics.cabalfile.changelog.mdto reflect the addition.Notes for Reviewers