statsmodels.sandbox.stats.multicomp.MultiComparison.allpairtest

MultiComparison.allpairtest(testfunc, alpha=0.05, method='bonf', pvalidx=1)[source]

run a pairwise test on all pairs with multiple test correction

The statistical test given in testfunc is calculated for all pairs and the p-values are adjusted by methods in multipletests. The p-value correction is generic and based only on the p-values, and does not take any special structure of the hypotheses into account.

Parameters:

testfunc : function

A test function for two (independent) samples. It is assumed that the return value on position pvalidx is the p-value.

alpha : float

familywise error rate

method : string

This specifies the method for the p-value correction. Any method of multipletests is possible.

pvalidx : int (default: 1)

position of the p-value in the return of testfunc

Returns:

sumtab : SimpleTable instance

summary table for printing

errors: TODO: check if this is still wrong, I think it’s fixed.

results from multipletests are in different order

pval_corrected can be larger than 1 ???