scikit_posthocs.posthoc_steel

scikit_posthocs.posthoc_steel(a: list | ndarray | DataFrame, val_col: str | None = None, group_col: str | None = None, control: str | None = None, alternative: Literal['two-sided', 'less', 'greater'] = 'two-sided', p_adjust: str | None = None, sort: bool = False, to_matrix: bool = True) → Series | DataFrame

Steel’s many-to-one rank test [1], a nonparametric alternative to Dunnett’s test for comparisons of several treatment groups against one control group.

Parameters:

  • a (Union[list, np.ndarray, DataFrame]) – An array, any object exposing the array interface or a pandas DataFrame.

  • val_col (str, optional) – Name of a DataFrame column that contains dependent variable values (test or response variable). Values should have a non-nominal scale. Must be specified if a is a pandas DataFrame object.

  • group_col (str, optional) – Name of a DataFrame column that contains independent variable values (grouping or predictor variable). Values should have a nominal scale (categorical). Must be specified if a is a pandas DataFrame object.

  • control (str, optional) – Name of the control group within the group_col column. Must be specified if a is a pandas DataFrame.

  • alternative (['two-sided', 'less', or 'greater'], optional) – Whether to get the p-value for the one-sided hypothesis (‘less’ or ‘greater’) or for the two-sided hypothesis (‘two-sided’). Defaults to ‘two-sided’.

  • p_adjust (str, optional) – Method for adjusting p values across the treatment-vs-control comparisons. See statsmodels.sandbox.stats.multicomp for details.

  • sort (bool, optional) – Specifies whether to sort DataFrame by group_col or not.

  • to_matrix (bool, optional) – Specifies whether to return a DataFrame or a Series. If True, a DataFrame is returned with some NaN values since it’s not a pairwise comparison. Default is True.

Returns:

result – P values.

Return type:

pandas.Series or pandas.DataFrame

Notes

This implements the normal-approximation variant of Steel’s test: each treatment group is compared to the control with a Mann-Whitney U test (scipy.stats.mannwhitneyu), and the resulting p values across treatments may be combined with p_adjust. PMCMRplus’s steelTest instead looks up exact critical rank-sum values from a fixed table (balanced designs only, n <= 20, k <= 9, alpha = 0.05); that table is not reproduced here.

References

Examples

>>> import scikit_posthocs as sp
>>> import pandas as pd
>>> x = pd.DataFrame({"a": [1,2,3,5,1], "b": [12,31,54,62,12], "c": [10,12,6,74,11]})
>>> x = x.melt(var_name='groups', value_name='values')
>>> sp.posthoc_steel(x, val_col='values', group_col='groups', control='a')