scipy.stats.wilcoxon

scipy.stats.wilcoxon(x, y=None, zero_method='wilcox', correction=False, alternative='two-sided', method='auto', *, axis=0, nan_policy='propagate', keepdims=False)

Calculate the Wilcoxon signed-rank test.

The Wilcoxon signed-rank test tests the null hypothesis that two related paired samples come from the same distribution. In particular, it tests whether the distribution of the differences x - y is symmetric about zero. It is a non-parametric version of the paired T-test.

Parameters:

xarray_like

Either the first set of measurements (in which case y is the second set of measurements), or the differences between two sets of measurements (in which case y is not to be specified.) Must be one-dimensional.

yarray_like, optional

Either the second set of measurements (if x is the first set of measurements), or not specified (if x is the differences between two sets of measurements.) Must be one-dimensional.

zero_method{“wilcox”, “pratt”, “zsplit”}, optional

There are different conventions for handling pairs of observations with equal values (“zero-differences”, or “zeros”).

• “wilcox”: Discards all zero-differences (default); see [4].

• “pratt”: Includes zero-differences in the ranking process, but drops the ranks of the zeros (more conservative); see [3]. In this case, the normal approximation is adjusted as in [5].

• “zsplit”: Includes zero-differences in the ranking process and splits the zero rank between positive and negative ones.

correctionbool, optional

If True, apply continuity correction by adjusting the Wilcoxon rank statistic by 0.5 towards the mean value when computing the z-statistic if a normal approximation is used. Default is False.

alternative{“two-sided”, “greater”, “less”}, optional

Defines the alternative hypothesis. Default is ‘two-sided’. In the following, let d represent the difference between the paired samples: d = x - y if both x and y are provided, or d = x otherwise.

• ‘two-sided’: the distribution underlying d is not symmetric about zero.

• ‘less’: the distribution underlying d is stochastically less than a distribution symmetric about zero.

• ‘greater’: the distribution underlying d is stochastically greater than a distribution symmetric about zero.

method{“auto”, “exact”, “approx”}, optional

Method to calculate the p-value, see Notes. Default is “auto”.

axisint or None, default: 0

If an int, the axis of the input along which to compute the statistic. The statistic of each axis-slice (e.g. row) of the input will appear in a corresponding element of the output. If None, the input will be raveled before computing the statistic.

nan_policy{‘propagate’, ‘omit’, ‘raise’}

Defines how to handle input NaNs.

• propagate: if a NaN is present in the axis slice (e.g. row) along which the statistic is computed, the corresponding entry of the output will be NaN.

• omit: NaNs will be omitted when performing the calculation. If insufficient data remains in the axis slice along which the statistic is computed, the corresponding entry of the output will be NaN.

• raise: if a NaN is present, a ValueError will be raised.

keepdimsbool, default: False

If this is set to True, the axes which are reduced are left in the result as dimensions with size one. With this option, the result will broadcast correctly against the input array.

Returns:

An object with the following attributes.

statisticarray_like

If alternative is “two-sided”, the sum of the ranks of the differences above or below zero, whichever is smaller. Otherwise the sum of the ranks of the differences above zero.

pvaluearray_like

The p-value for the test depending on alternative and method.

zstatisticarray_like

When method = 'approx', this is the normalized z-statistic:

z = (T - mn - d) / se

where T is statistic as defined above, mn is the mean of the distribution under the null hypothesis, d is a continuity correction, and se is the standard error. When method != 'approx', this attribute is not available.

See also

kruskal, mannwhitneyu

Notes

In the following, let d represent the difference between the paired samples: d = x - y if both x and y are provided, or d = x otherwise. Assume that all elements of d are independent and identically distributed observations, and all are distinct and nonzero.

• When len(d) is sufficiently large, the null distribution of the normalized test statistic (zstatistic above) is approximately normal, and method = 'approx' can be used to compute the p-value.

• When len(d) is small, the normal approximation may not be accurate, and method='exact' is preferred (at the cost of additional execution time).

• The default, method='auto', selects between the two: when len(d) <= 50, the exact method is used; otherwise, the approximate method is used.

The presence of “ties” (i.e. not all elements of d are unique) and “zeros” (i.e. elements of d are zero) changes the null distribution of the test statistic, and method='exact' no longer calculates the exact p-value. If method='approx', the z-statistic is adjusted for more accurate comparison against the standard normal, but still, for finite sample sizes, the standard normal is only an approximation of the true null distribution of the z-statistic. There is no clear consensus among references on which method most accurately approximates the p-value for small samples in the presence of zeros and/or ties. In any case, this is the behavior of wilcoxon when method='auto': ``method='exact' is used when len(d) <= 50 and there are no zeros; otherwise, method='approx' is used.

Beginning in SciPy 1.9, np.matrix inputs (not recommended for new code) are converted to np.ndarray before the calculation is performed. In this case, the output will be a scalar or np.ndarray of appropriate shape rather than a 2D np.matrix. Similarly, while masked elements of masked arrays are ignored, the output will be a scalar or np.ndarray rather than a masked array with mask=False.

References

[1]

https://en.wikipedia.org/wiki/Wilcoxon_signed-rank_test

[2]

Conover, W.J., Practical Nonparametric Statistics, 1971.

[3]

Pratt, J.W., Remarks on Zeros and Ties in the Wilcoxon Signed Rank Procedures, Journal of the American Statistical Association, Vol. 54, 1959, pp. 655-667. DOI:10.1080/01621459.1959.10501526

[4] (1,2)

Wilcoxon, F., Individual Comparisons by Ranking Methods, Biometrics Bulletin, Vol. 1, 1945, pp. 80-83. DOI:10.2307/3001968

[5]

Cureton, E.E., The Normal Approximation to the Signed-Rank Sampling Distribution When Zero Differences are Present, Journal of the American Statistical Association, Vol. 62, 1967, pp. 1068-1069. DOI:10.1080/01621459.1967.10500917

Examples

In [4], the differences in height between cross- and self-fertilized corn plants is given as follows:

>>> d = [6, 8, 14, 16, 23, 24, 28, 29, 41, -48, 49, 56, 60, -67, 75]

Cross-fertilized plants appear to be higher. To test the null hypothesis that there is no height difference, we can apply the two-sided test:

>>> from scipy.stats import wilcoxon
>>> res = wilcoxon(d)
>>> res.statistic, res.pvalue
(24.0, 0.041259765625)

Hence, we would reject the null hypothesis at a confidence level of 5%, concluding that there is a difference in height between the groups. To confirm that the median of the differences can be assumed to be positive, we use:

>>> res = wilcoxon(d, alternative='greater')
>>> res.statistic, res.pvalue
(96.0, 0.0206298828125)

This shows that the null hypothesis that the median is negative can be rejected at a confidence level of 5% in favor of the alternative that the median is greater than zero. The p-values above are exact. Using the normal approximation gives very similar values:

>>> res = wilcoxon(d, method='approx')
>>> res.statistic, res.pvalue
(24.0, 0.04088813291185591)

Note that the statistic changed to 96 in the one-sided case (the sum of ranks of positive differences) whereas it is 24 in the two-sided case (the minimum of sum of ranks above and below zero).