statsmodels.stats.multitest.multipletests¶

statsmodels.stats.multitest.multipletests(pvals, alpha=0.05, method='hs', maxiter=1, is_sorted=False, returnsorted=False)[source]¶

Test results and p-value correction for multiple tests

Parameters:¶

pvalsarray_like, 1-d

uncorrected p-values. Must be 1-dimensional.

alphafloat

FWER, family-wise error rate, e.g. 0.1

methodstr

Method used for testing and adjustment of pvalues. Can be either the full name or initial letters. Available methods are:

• bonferroni : one-step correction

• sidak : one-step correction

• holm-sidak : step down method using Sidak adjustments

• holm : step-down method using Bonferroni adjustments

• simes-hochberg : step-up method (independent)

• hommel : closed method based on Simes tests (non-negative)

• fdr_bh : Benjamini/Hochberg (non-negative)

• fdr_by : Benjamini/Yekutieli (negative)

• fdr_tsbh : two stage fdr correction (non-negative)

• fdr_tsbky : two stage fdr correction (non-negative)

maxiterint or bool

Maximum number of iterations for two-stage fdr, fdr_tsbh and fdr_tsbky. It is ignored by all other methods. maxiter=1 (default) corresponds to the two stage method. maxiter=-1 corresponds to full iterations which is maxiter=len(pvals). maxiter=0 uses only a single stage fdr correction using a ‘bh’ or ‘bky’ prior fraction of assumed true hypotheses.

is_sortedbool

If False (default), the p_values will be sorted, but the corrected pvalues are in the original order. If True, then it assumed that the pvalues are already sorted in ascending order.

returnsortedbool

not tested, return sorted p-values instead of original sequence

Returns:¶

rejectndarray, bool

true for hypothesis that can be rejected for given alpha

pvals_correctedndarray

p-values corrected for multiple tests

alphacSidakfloat

corrected alpha for Sidak method

alphacBonffloat

corrected alpha for Bonferroni method

Notes

There may be API changes for this function in the future.

Except for ‘fdr_twostage’, the p-value correction is independent of the alpha specified as argument. In these cases the corrected p-values can also be compared with a different alpha. In the case of ‘fdr_twostage’, the corrected p-values are specific to the given alpha, see fdrcorrection_twostage.

The ‘fdr_gbs’ procedure is not verified against another package, p-values are derived from scratch and are not derived in the reference. In Monte Carlo experiments the method worked correctly and maintained the false discovery rate.

All procedures that are included, control FWER or FDR in the independent case, and most are robust in the positively correlated case.

fdr_gbs: high power, fdr control for independent case and only small violation in positively correlated case

Timing:

Most of the time with large arrays is spent in argsort. When we want to calculate the p-value for several methods, then it is more efficient to presort the pvalues, and put the results back into the original order outside of the function.

Method=’hommel’ is very slow for large arrays, since it requires the evaluation of n partitions, where n is the number of p-values.