org.hipparchus.stat.inference.MannWhitneyUTest Maven / Gradle / Ivy
/*
* Licensed to the Apache Software Foundation (ASF) under one or more
* contributor license agreements. See the NOTICE file distributed with
* this work for additional information regarding copyright ownership.
* The ASF licenses this file to You under the Apache License, Version 2.0
* (the "License"); you may not use this file except in compliance with
* the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.hipparchus.stat.inference;
import org.hipparchus.distribution.continuous.NormalDistribution;
import org.hipparchus.exception.LocalizedCoreFormats;
import org.hipparchus.exception.MathIllegalArgumentException;
import org.hipparchus.exception.MathIllegalStateException;
import org.hipparchus.exception.NullArgumentException;
import org.hipparchus.stat.ranking.NaNStrategy;
import org.hipparchus.stat.ranking.NaturalRanking;
import org.hipparchus.stat.ranking.TiesStrategy;
import org.hipparchus.util.FastMath;
/**
* An implementation of the Mann-Whitney U test (also called Wilcoxon rank-sum test).
*/
public class MannWhitneyUTest {
/** Ranking algorithm. */
private NaturalRanking naturalRanking;
/**
* Create a test instance using where NaN's are left in place and ties get
* the average of applicable ranks. Use this unless you are very sure of
* what you are doing.
*/
public MannWhitneyUTest() {
naturalRanking = new NaturalRanking(NaNStrategy.FIXED,
TiesStrategy.AVERAGE);
}
/**
* Create a test instance using the given strategies for NaN's and ties.
* Only use this if you are sure of what you are doing.
*
* @param nanStrategy specifies the strategy that should be used for Double.NaN's
* @param tiesStrategy specifies the strategy that should be used for ties
*/
public MannWhitneyUTest(final NaNStrategy nanStrategy,
final TiesStrategy tiesStrategy) {
naturalRanking = new NaturalRanking(nanStrategy, tiesStrategy);
}
/**
* Ensures that the provided arrays fulfills the assumptions.
*
* @param x first sample
* @param y second sample
* @throws NullArgumentException if {@code x} or {@code y} are {@code null}.
* @throws MathIllegalArgumentException if {@code x} or {@code y} are zero-length.
*/
private void ensureDataConformance(final double[] x, final double[] y)
throws MathIllegalArgumentException, NullArgumentException {
if (x == null ||
y == null) {
throw new NullArgumentException();
}
if (x.length == 0 ||
y.length == 0) {
throw new MathIllegalArgumentException(LocalizedCoreFormats.NO_DATA);
}
}
/**
* Concatenate the samples into one array.
*
* @param x first sample
* @param y second sample
* @return concatenated array
*/
private double[] concatenateSamples(final double[] x, final double[] y) {
final double[] z = new double[x.length + y.length];
System.arraycopy(x, 0, z, 0, x.length);
System.arraycopy(y, 0, z, x.length, y.length);
return z;
}
/**
* Computes the Mann-Whitney
* U statistic comparing mean for two independent samples possibly of
* different length.
*
* This statistic can be used to perform a Mann-Whitney U test evaluating
* the null hypothesis that the two independent samples has equal mean.
*
* Let Xi denote the i'th individual of the first sample and
* Yj the j'th individual in the second sample. Note that the
* samples would often have different length.
*
* Preconditions:
*
* - All observations in the two samples are independent.
* - The observations are at least ordinal (continuous are also ordinal).
*
*
* @param x the first sample
* @param y the second sample
* @return Mann-Whitney U statistic (maximum of Ux and Uy)
* @throws NullArgumentException if {@code x} or {@code y} are {@code null}.
* @throws MathIllegalArgumentException if {@code x} or {@code y} are zero-length.
*/
public double mannWhitneyU(final double[] x, final double[] y)
throws MathIllegalArgumentException, NullArgumentException {
ensureDataConformance(x, y);
final double[] z = concatenateSamples(x, y);
final double[] ranks = naturalRanking.rank(z);
double sumRankX = 0;
/*
* The ranks for x is in the first x.length entries in ranks because x
* is in the first x.length entries in z
*/
for (int i = 0; i < x.length; ++i) {
sumRankX += ranks[i];
}
/*
* U1 = R1 - (n1 * (n1 + 1)) / 2 where R1 is sum of ranks for sample 1,
* e.g. x, n1 is the number of observations in sample 1.
*/
final double U1 = sumRankX - ((long) x.length * (x.length + 1)) / 2;
/*
* It can be shown that U1 + U2 = n1 * n2
*/
final double U2 = (long) x.length * y.length - U1;
return FastMath.max(U1, U2);
}
/**
* @param Umin smallest Mann-Whitney U value
* @param n1 number of subjects in first sample
* @param n2 number of subjects in second sample
* @return two-sided asymptotic p-value
* @throws MathIllegalStateException if the p-value can not be computed
* due to a convergence error
* @throws MathIllegalStateException if the maximum number of
* iterations is exceeded
*/
private double calculateAsymptoticPValue(final double Umin,
final int n1,
final int n2)
throws MathIllegalStateException {
/* long multiplication to avoid overflow (double not used due to efficiency
* and to avoid precision loss)
*/
final long n1n2prod = (long) n1 * n2;
// http://en.wikipedia.org/wiki/Mann%E2%80%93Whitney_U#Normal_approximation
final double EU = n1n2prod / 2.0;
final double VarU = n1n2prod * (n1 + n2 + 1) / 12.0;
final double z = (Umin - EU) / FastMath.sqrt(VarU);
// No try-catch or advertised exception because args are valid
final NormalDistribution standardNormal = new NormalDistribution(0, 1);
return 2 * standardNormal.cumulativeProbability(z);
}
/**
* Returns the asymptotic observed significance level, or
*
* p-value, associated with a
* Mann-Whitney
* U statistic comparing mean for two independent samples.
*
* Let Xi denote the i'th individual of the first sample and
* Yj the j'th individual in the second sample. Note that the
* samples would often have different length.
*
* Preconditions:
*
* - All observations in the two samples are independent.
* - The observations are at least ordinal (continuous are also ordinal).
*
*
* Ties give rise to biased variance at the moment. See e.g.
*
* http://mlsc.lboro.ac.uk/resources/statistics/Mannwhitney.pdf.
*
* @param x the first sample
* @param y the second sample
* @return asymptotic p-value
* @throws NullArgumentException if {@code x} or {@code y} are {@code null}.
* @throws MathIllegalArgumentException if {@code x} or {@code y} are zero-length.
* @throws MathIllegalStateException if the p-value can not be computed due to a
* convergence error
* @throws MathIllegalStateException if the maximum number of iterations
* is exceeded
*/
public double mannWhitneyUTest(final double[] x, final double[] y)
throws MathIllegalArgumentException, NullArgumentException,
MathIllegalStateException {
ensureDataConformance(x, y);
final double Umax = mannWhitneyU(x, y);
/*
* It can be shown that U1 + U2 = n1 * n2
*/
final double Umin = (long) x.length * y.length - Umax;
return calculateAsymptoticPValue(Umin, x.length, y.length);
}
}