Development build. This page describes main, not a released package. The latest published Lodestar.Stats is 0.4.0 — read its documentation.

Home › Stats › Hypothesis tests

Compares two paired samples by the ranks of their differences.

public static TestResult Paired(ReadOnlySpan<double> x, ReadOnlySpan<double> y, ZeroMethod zeroMethod = ZeroMethod.Wilcox, Alternative alternative = Alternative.TwoSided, Continuity continuity = Continuity.None, ExactMethod method = ExactMethod.Auto, NanPolicy nanPolicy = NanPolicy.Propagate)

Parameters — x is the first measurement of each pair. y is the second measurement of each pair, in the same order as x. zeroMethod says what to do with pairs whose difference is zero. alternative says which tail the p-value covers. continuity says whether the normal approximation gets the half-unit correction. method chooses the exact null distribution, the exhaustive permutation test, its normal approximation, or a choice between them. nanPolicy says what to do with a NaN; scipy's nan_policy, defaulting to NanPolicy.Propagate.

Returns — TestResult: the signed-rank statistic and the p-value. Two-sided, the statistic is the smaller of the two rank sums; under Alternative.Less or Alternative.Greater it is the sum of the positive ranks, which can be the larger one — scipy's wilcoxon reports the same.

Exceptions — ArgumentException when the two samples differ in length, are empty, or nanPolicy is NanPolicy.Raise and either sample holds a NaN. ArgumentOutOfRangeException when method is ExactMethod.Exact and the ranked sample exceeds 500 values.

Example — seven pairs, two of them unchanged: what the three zero methods disagree about.

using Lodestar.Stats;

double[] before = [12.0, 9.0, 15.0, 11.0, 8.0, 14.0, 10.0, 13.0];
double[] after = [9.0, 6.0, 16.0, 11.0, 4.0, 15.0, 10.0, 17.0];

TestResult wilcox = Wilcoxon.Paired(before, after);
TestResult pratt = Wilcoxon.Paired(before, after, ZeroMethod.Pratt);
TestResult zsplit = Wilcoxon.Paired(before, after, ZeroMethod.ZSplit);

double wilcoxW = wilcox.Statistic;   // => 8.5
double prattW = pratt.Statistic;     // => 14.5
double zsplitW = zsplit.Statistic;   // => 16

Remarks — this delegates to OneSample on the pairwise differences x[i] - y[i]; Paired(x, y, ...) and OneSample(differences, ...) agree exactly wherever the differences agree, the same relationship TTest.Paired has to TTest.OneSample.

Two of the seven pairs above are unchanged, and the three zero methods read that differently: Wilcox drops both pairs before ranking the rest; Pratt ranks them alongside everything else but excludes their ranks from the sums that follow; ZSplit ranks them the same way and keeps them, splitting their rank sum evenly between the positive and negative totals. All three land on different statistics here — the ZeroMethod page has the full three-way comparison, including the p-values.

method here refuses past a lower bound than MannWhitney.Test's: ExactMethod.Exact is only honoured up to 500 ranked values, because the exact null distribution's total, 2^n, overflows a double to infinity past n = 1023 and every p-value it divides would silently become exactly zero rather than throwing. ExactMethod.Auto never reaches that bound on its own — it is unconditionally asymptotic above 50 values, exact only below that and free of both ties and zeros, and falls to an exhaustive permutation test at 13 values or fewer when ties or zeros rule out the plain exact table.

Under NanPolicy.Omit the two inputs are filtered together: an index is kept only when neither side holds a NaN, so a pair survives or neither value does.

Applies to — net10.0, netstandard2.0.

See also — Wilcoxon.OneSample, TTest.Paired for the parametric counterpart, ZeroMethod, MannWhitney.Test, the Python equivalence table.