001/*
002 * Licensed to the Apache Software Foundation (ASF) under one or more
003 * contributor license agreements.  See the NOTICE file distributed with
004 * this work for additional information regarding copyright ownership.
005 * The ASF licenses this file to You under the Apache License, Version 2.0
006 * (the "License"); you may not use this file except in compliance with
007 * the License.  You may obtain a copy of the License at
008 *
009 *      http://www.apache.org/licenses/LICENSE-2.0
010 *
011 * Unless required by applicable law or agreed to in writing, software
012 * distributed under the License is distributed on an "AS IS" BASIS,
013 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
014 * See the License for the specific language governing permissions and
015 * limitations under the License.
016 */
017package org.apache.commons.math4.legacy.stat.inference;
018
019import org.apache.commons.statistics.distribution.ChiSquaredDistribution;
020import org.apache.commons.math4.legacy.exception.DimensionMismatchException;
021import org.apache.commons.math4.legacy.exception.MaxCountExceededException;
022import org.apache.commons.math4.legacy.exception.NotPositiveException;
023import org.apache.commons.math4.legacy.exception.NotStrictlyPositiveException;
024import org.apache.commons.math4.legacy.exception.NullArgumentException;
025import org.apache.commons.math4.legacy.exception.OutOfRangeException;
026import org.apache.commons.math4.legacy.exception.ZeroException;
027import org.apache.commons.math4.legacy.exception.util.LocalizedFormats;
028import org.apache.commons.math4.core.jdkmath.JdkMath;
029import org.apache.commons.math4.legacy.core.MathArrays;
030
031/**
032 * Implements Chi-Square test statistics.
033 *
034 * <p>This implementation handles both known and unknown distributions.</p>
035 *
036 * <p>Two samples tests can be used when the distribution is unknown <i>a priori</i>
037 * but provided by one sample, or when the hypothesis under test is that the two
038 * samples come from the same underlying distribution.</p>
039 *
040 */
041public class ChiSquareTest {
042
043    /**
044     * Construct a ChiSquareTest.
045     */
046    public ChiSquareTest() {
047        super();
048    }
049
050    /**
051     * Computes the <a href="http://www.itl.nist.gov/div898/handbook/eda/section3/eda35f.htm">
052     * Chi-Square statistic</a> comparing <code>observed</code> and <code>expected</code>
053     * frequency counts.
054     * <p>
055     * This statistic can be used to perform a Chi-Square test evaluating the null
056     * hypothesis that the observed counts follow the expected distribution.</p>
057     * <p>
058     * <strong>Preconditions</strong>: <ul>
059     * <li>Expected counts must all be positive.
060     * </li>
061     * <li>Observed counts must all be &ge; 0.
062     * </li>
063     * <li>The observed and expected arrays must have the same length and
064     * their common length must be at least 2.
065     * </li></ul><p>
066     * If any of the preconditions are not met, an
067     * <code>IllegalArgumentException</code> is thrown.</p>
068     * <p><strong>Note: </strong>This implementation rescales the
069     * <code>expected</code> array if necessary to ensure that the sum of the
070     * expected and observed counts are equal.</p>
071     *
072     * @param observed array of observed frequency counts
073     * @param expected array of expected frequency counts
074     * @return chiSquare test statistic
075     * @throws NotPositiveException if <code>observed</code> has negative entries
076     * @throws NotStrictlyPositiveException if <code>expected</code> has entries that are
077     * not strictly positive
078     * @throws DimensionMismatchException if the arrays length is less than 2
079     */
080    public double chiSquare(final double[] expected, final long[] observed)
081        throws NotPositiveException, NotStrictlyPositiveException,
082        DimensionMismatchException {
083
084        if (expected.length < 2) {
085            throw new DimensionMismatchException(expected.length, 2);
086        }
087        if (expected.length != observed.length) {
088            throw new DimensionMismatchException(expected.length, observed.length);
089        }
090        MathArrays.checkPositive(expected);
091        MathArrays.checkNonNegative(observed);
092
093        double sumExpected = 0d;
094        double sumObserved = 0d;
095        for (int i = 0; i < observed.length; i++) {
096            sumExpected += expected[i];
097            sumObserved += observed[i];
098        }
099        double ratio = 1.0d;
100        boolean rescale = false;
101        if (JdkMath.abs(sumExpected - sumObserved) > 10E-6) {
102            ratio = sumObserved / sumExpected;
103            rescale = true;
104        }
105        double sumSq = 0.0d;
106        for (int i = 0; i < observed.length; i++) {
107            if (rescale) {
108                final double dev = observed[i] - ratio * expected[i];
109                sumSq += dev * dev / (ratio * expected[i]);
110            } else {
111                final double dev = observed[i] - expected[i];
112                sumSq += dev * dev / expected[i];
113            }
114        }
115        return sumSq;
116    }
117
118    /**
119     * Returns the <i>observed significance level</i>, or <a href=
120     * "http://www.cas.lancs.ac.uk/glossary_v1.1/hyptest.html#pvalue">
121     * p-value</a>, associated with a
122     * <a href="http://www.itl.nist.gov/div898/handbook/eda/section3/eda35f.htm">
123     * Chi-square goodness of fit test</a> comparing the <code>observed</code>
124     * frequency counts to those in the <code>expected</code> array.
125     * <p>
126     * The number returned is the smallest significance level at which one can reject
127     * the null hypothesis that the observed counts conform to the frequency distribution
128     * described by the expected counts.</p>
129     * <p>
130     * <strong>Preconditions</strong>: <ul>
131     * <li>Expected counts must all be positive.
132     * </li>
133     * <li>Observed counts must all be &ge; 0.
134     * </li>
135     * <li>The observed and expected arrays must have the same length and
136     * their common length must be at least 2.
137     * </li></ul><p>
138     * If any of the preconditions are not met, an
139     * <code>IllegalArgumentException</code> is thrown.</p>
140     * <p><strong>Note: </strong>This implementation rescales the
141     * <code>expected</code> array if necessary to ensure that the sum of the
142     * expected and observed counts are equal.</p>
143     *
144     * @param observed array of observed frequency counts
145     * @param expected array of expected frequency counts
146     * @return p-value
147     * @throws NotPositiveException if <code>observed</code> has negative entries
148     * @throws NotStrictlyPositiveException if <code>expected</code> has entries that are
149     * not strictly positive
150     * @throws DimensionMismatchException if the arrays length is less than 2
151     * @throws MaxCountExceededException if an error occurs computing the p-value
152     */
153    public double chiSquareTest(final double[] expected, final long[] observed)
154        throws NotPositiveException, NotStrictlyPositiveException,
155        DimensionMismatchException, MaxCountExceededException {
156
157        // pass a null rng to avoid unneeded overhead as we will not sample from this distribution
158        final ChiSquaredDistribution distribution =
159            ChiSquaredDistribution.of(expected.length - 1.0);
160        return distribution.survivalProbability(chiSquare(expected, observed));
161    }
162
163    /**
164     * Performs a <a href="http://www.itl.nist.gov/div898/handbook/eda/section3/eda35f.htm">
165     * Chi-square goodness of fit test</a> evaluating the null hypothesis that the
166     * observed counts conform to the frequency distribution described by the expected
167     * counts, with significance level <code>alpha</code>.  Returns true iff the null
168     * hypothesis can be rejected with 100 * (1 - alpha) percent confidence.
169     * <p>
170     * <strong>Example:</strong><br>
171     * To test the hypothesis that <code>observed</code> follows
172     * <code>expected</code> at the 99% level, use </p><p>
173     * <code>chiSquareTest(expected, observed, 0.01) </code></p>
174     * <p>
175     * <strong>Preconditions</strong>: <ul>
176     * <li>Expected counts must all be positive.
177     * </li>
178     * <li>Observed counts must all be &ge; 0.
179     * </li>
180     * <li>The observed and expected arrays must have the same length and
181     * their common length must be at least 2.
182     * <li> <code> 0 &lt; alpha &lt; 0.5 </code>
183     * </li></ul><p>
184     * If any of the preconditions are not met, an
185     * <code>IllegalArgumentException</code> is thrown.</p>
186     * <p><strong>Note: </strong>This implementation rescales the
187     * <code>expected</code> array if necessary to ensure that the sum of the
188     * expected and observed counts are equal.</p>
189     *
190     * @param observed array of observed frequency counts
191     * @param expected array of expected frequency counts
192     * @param alpha significance level of the test
193     * @return true iff null hypothesis can be rejected with confidence
194     * 1 - alpha
195     * @throws NotPositiveException if <code>observed</code> has negative entries
196     * @throws NotStrictlyPositiveException if <code>expected</code> has entries that are
197     * not strictly positive
198     * @throws DimensionMismatchException if the arrays length is less than 2
199     * @throws OutOfRangeException if <code>alpha</code> is not in the range (0, 0.5]
200     * @throws MaxCountExceededException if an error occurs computing the p-value
201     */
202    public boolean chiSquareTest(final double[] expected, final long[] observed,
203                                 final double alpha)
204        throws NotPositiveException, NotStrictlyPositiveException,
205        DimensionMismatchException, OutOfRangeException, MaxCountExceededException {
206
207        if (alpha <= 0 || alpha > 0.5) {
208            throw new OutOfRangeException(LocalizedFormats.OUT_OF_BOUND_SIGNIFICANCE_LEVEL,
209                                          alpha, 0, 0.5);
210        }
211        return chiSquareTest(expected, observed) < alpha;
212    }
213
214    /**
215     *  Computes the Chi-Square statistic associated with a
216     * <a href="http://www.itl.nist.gov/div898/handbook/prc/section4/prc45.htm">
217     *  chi-square test of independence</a> based on the input <code>counts</code>
218     *  array, viewed as a two-way table.
219     * <p>
220     * The rows of the 2-way table are
221     * <code>count[0], ... , count[count.length - 1] </code></p>
222     * <p>
223     * <strong>Preconditions</strong>: <ul>
224     * <li>All counts must be &ge; 0.
225     * </li>
226     * <li>The sum of each row and column must be &gt; 0.
227     * </li>
228     * <li>The count array must be rectangular (i.e. all count[i] subarrays
229     *  must have the same length).
230     * </li>
231     * <li>The 2-way table represented by <code>counts</code> must have at
232     *  least 2 columns and at least 2 rows.
233     * </li>
234     * </ul><p>
235     * If any of the preconditions are not met, an
236     * <code>IllegalArgumentException</code> is thrown.</p>
237     * <p>
238     * If a column or row contains only zeros this is invalid input and a
239     * <code>ZeroException</code> is thrown. The empty column/row should
240     * be removed from the input counts.</p>
241     *
242     * @param counts array representation of 2-way table
243     * @return chiSquare test statistic
244     * @throws NullArgumentException if the array is null
245     * @throws DimensionMismatchException if the array is not rectangular
246     * @throws NotPositiveException if {@code counts} has negative entries
247     * @throws ZeroException if the sum of a row or column is zero
248     */
249    public double chiSquare(final long[][] counts)
250        throws NullArgumentException, NotPositiveException,
251        DimensionMismatchException {
252
253        checkArray(counts);
254        int nRows = counts.length;
255        int nCols = counts[0].length;
256
257        // compute row, column and total sums
258        double[] rowSum = new double[nRows];
259        double[] colSum = new double[nCols];
260        double total = 0.0d;
261        for (int row = 0; row < nRows; row++) {
262            for (int col = 0; col < nCols; col++) {
263                rowSum[row] += counts[row][col];
264                colSum[col] += counts[row][col];
265                total += counts[row][col];
266            }
267            checkNonZero(rowSum[row], "row", row);
268        }
269
270        for (int col = 0; col < nCols; col++) {
271            checkNonZero(colSum[col], "column", col);
272        }
273
274        // compute expected counts and chi-square
275        double sumSq = 0.0d;
276        double expected = 0.0d;
277        for (int row = 0; row < nRows; row++) {
278            for (int col = 0; col < nCols; col++) {
279                expected = (rowSum[row] * colSum[col]) / total;
280                sumSq += ((counts[row][col] - expected) *
281                        (counts[row][col] - expected)) / expected;
282            }
283        }
284        return sumSq;
285    }
286
287    /**
288     * Returns the <i>observed significance level</i>, or <a href=
289     * "http://www.cas.lancs.ac.uk/glossary_v1.1/hyptest.html#pvalue">
290     * p-value</a>, associated with a
291     * <a href="http://www.itl.nist.gov/div898/handbook/prc/section4/prc45.htm">
292     * chi-square test of independence</a> based on the input <code>counts</code>
293     * array, viewed as a two-way table.
294     * <p>
295     * The rows of the 2-way table are
296     * <code>count[0], ... , count[count.length - 1] </code></p>
297     * <p>
298     * <strong>Preconditions</strong>: <ul>
299     * <li>All counts must be &ge; 0.
300     * </li>
301     * <li>The sum of each row and column must be &gt; 0.
302     * </li>
303     * <li>The count array must be rectangular (i.e. all count[i] subarrays must have
304     *     the same length).
305     * </li>
306     * <li>The 2-way table represented by <code>counts</code> must have at least 2
307     *     columns and at least 2 rows.
308     * </li>
309     * </ul><p>
310     * If any of the preconditions are not met, an
311     * <code>IllegalArgumentException</code> is thrown.</p>
312     * <p>
313     * If a column or row contains only zeros this is invalid input and a
314     * <code>ZeroException</code> is thrown. The empty column/row should
315     * be removed from the input counts.</p>
316     *
317     * @param counts array representation of 2-way table
318     * @return p-value
319     * @throws NullArgumentException if the array is null
320     * @throws DimensionMismatchException if the array is not rectangular
321     * @throws NotPositiveException if {@code counts} has negative entries
322     * @throws MaxCountExceededException if an error occurs computing the p-value
323     * @throws ZeroException if the sum of a row or column is zero
324     */
325    public double chiSquareTest(final long[][] counts)
326        throws NullArgumentException, DimensionMismatchException,
327        NotPositiveException, MaxCountExceededException {
328
329        checkArray(counts);
330        double df = ((double) counts.length -1) * ((double) counts[0].length - 1);
331        // pass a null rng to avoid unneeded overhead as we will not sample from this distribution
332        final ChiSquaredDistribution distribution = ChiSquaredDistribution.of(df);
333        return distribution.survivalProbability(chiSquare(counts));
334    }
335
336    /**
337     * Performs a <a href="http://www.itl.nist.gov/div898/handbook/prc/section4/prc45.htm">
338     * chi-square test of independence</a> evaluating the null hypothesis that the
339     * classifications represented by the counts in the columns of the input 2-way table
340     * are independent of the rows, with significance level <code>alpha</code>.
341     * Returns true iff the null hypothesis can be rejected with 100 * (1 - alpha) percent
342     * confidence.
343     * <p>
344     * The rows of the 2-way table are
345     * <code>count[0], ... , count[count.length - 1] </code></p>
346     * <p>
347     * <strong>Example:</strong><br>
348     * To test the null hypothesis that the counts in
349     * <code>count[0], ... , count[count.length - 1] </code>
350     *  all correspond to the same underlying probability distribution at the 99% level, use</p>
351     * <p><code>chiSquareTest(counts, 0.01)</code></p>
352     * <p>
353     * <strong>Preconditions</strong>: <ul>
354     * <li>All counts must be &ge; 0.
355     * </li>
356     * <li>The sum of each row and column must be &gt; 0.
357     * </li>
358     * <li>The count array must be rectangular (i.e. all count[i] subarrays must have the
359     *     same length).</li>
360     * <li>The 2-way table represented by <code>counts</code> must have at least 2 columns and
361     *     at least 2 rows.</li>
362     * </ul><p>
363     * If any of the preconditions are not met, an
364     * <code>IllegalArgumentException</code> is thrown.</p>
365     * <p>
366     * If a column or row contains only zeros this is invalid input and a
367     * <code>ZeroException</code> is thrown. The empty column/row should
368     * be removed from the input counts.</p>
369     *
370     * @param counts array representation of 2-way table
371     * @param alpha significance level of the test
372     * @return true iff null hypothesis can be rejected with confidence
373     * 1 - alpha
374     * @throws NullArgumentException if the array is null
375     * @throws DimensionMismatchException if the array is not rectangular
376     * @throws NotPositiveException if {@code counts} has any negative entries
377     * @throws OutOfRangeException if <code>alpha</code> is not in the range (0, 0.5]
378     * @throws MaxCountExceededException if an error occurs computing the p-value
379     * @throws ZeroException if the sum of a row or column is zero
380     */
381    public boolean chiSquareTest(final long[][] counts, final double alpha)
382        throws NullArgumentException, DimensionMismatchException,
383        NotPositiveException, OutOfRangeException, MaxCountExceededException {
384
385        if (alpha <= 0 || alpha > 0.5) {
386            throw new OutOfRangeException(LocalizedFormats.OUT_OF_BOUND_SIGNIFICANCE_LEVEL,
387                                          alpha, 0, 0.5);
388        }
389        return chiSquareTest(counts) < alpha;
390    }
391
392    /**
393     * <p>Computes a
394     * <a href="http://www.itl.nist.gov/div898/software/dataplot/refman1/auxillar/chi2samp.htm">
395     * Chi-Square two sample test statistic</a> comparing bin frequency counts
396     * in <code>observed1</code> and <code>observed2</code>.  The
397     * sums of frequency counts in the two samples are not required to be the
398     * same.  The formula used to compute the test statistic is</p>
399     * <code>
400     * &sum;[(K * observed1[i] - observed2[i]/K)<sup>2</sup> / (observed1[i] + observed2[i])]
401     * </code> where
402     * <br><code>K = &radic;[&sum;(observed2 / &sum;(observed1)]</code>
403     *
404     * <p>This statistic can be used to perform a Chi-Square test evaluating the
405     * null hypothesis that both observed counts follow the same distribution.</p>
406     * <p>
407     * <strong>Preconditions</strong>: <ul>
408     * <li>Observed counts must be non-negative.
409     * </li>
410     * <li>Observed counts for a specific bin must not both be zero.
411     * </li>
412     * <li>Observed counts for a specific sample must not all be 0.
413     * </li>
414     * <li>The arrays <code>observed1</code> and <code>observed2</code> must have
415     * the same length and their common length must be at least 2.
416     * </li></ul><p>
417     * If any of the preconditions are not met, an
418     * <code>IllegalArgumentException</code> is thrown.</p>
419     *
420     * @param observed1 array of observed frequency counts of the first data set
421     * @param observed2 array of observed frequency counts of the second data set
422     * @return chiSquare test statistic
423     * @throws DimensionMismatchException the length of the arrays does not match
424     * @throws NotPositiveException if any entries in <code>observed1</code> or
425     * <code>observed2</code> are negative
426     * @throws ZeroException if either all counts of <code>observed1</code> or
427     * <code>observed2</code> are zero, or if the count at some index is zero
428     * for both arrays
429     * @since 1.2
430     */
431    public double chiSquareDataSetsComparison(long[] observed1, long[] observed2)
432        throws DimensionMismatchException, NotPositiveException, ZeroException {
433
434        // Make sure lengths are same
435        if (observed1.length < 2) {
436            throw new DimensionMismatchException(observed1.length, 2);
437        }
438        if (observed1.length != observed2.length) {
439            throw new DimensionMismatchException(observed1.length, observed2.length);
440        }
441
442        // Ensure non-negative counts
443        MathArrays.checkNonNegative(observed1);
444        MathArrays.checkNonNegative(observed2);
445
446        // Compute and compare count sums
447        long countSum1 = 0;
448        long countSum2 = 0;
449        boolean unequalCounts = false;
450        double weight = 0.0;
451        for (int i = 0; i < observed1.length; i++) {
452            countSum1 += observed1[i];
453            countSum2 += observed2[i];
454        }
455        // Ensure neither sample is uniformly 0
456        if (countSum1 == 0 || countSum2 == 0) {
457            throw new ZeroException();
458        }
459        // Compare and compute weight only if different
460        unequalCounts = countSum1 != countSum2;
461        if (unequalCounts) {
462            weight = JdkMath.sqrt((double) countSum1 / (double) countSum2);
463        }
464        // Compute ChiSquare statistic
465        double sumSq = 0.0d;
466        double dev = 0.0d;
467        double obs1 = 0.0d;
468        double obs2 = 0.0d;
469        for (int i = 0; i < observed1.length; i++) {
470            if (observed1[i] == 0 && observed2[i] == 0) {
471                throw new ZeroException(LocalizedFormats.OBSERVED_COUNTS_BOTTH_ZERO_FOR_ENTRY, i);
472            } else {
473                obs1 = observed1[i];
474                obs2 = observed2[i];
475                if (unequalCounts) { // apply weights
476                    dev = obs1/weight - obs2 * weight;
477                } else {
478                    dev = obs1 - obs2;
479                }
480                sumSq += (dev * dev) / (obs1 + obs2);
481            }
482        }
483        return sumSq;
484    }
485
486    /**
487     * <p>Returns the <i>observed significance level</i>, or <a href=
488     * "http://www.cas.lancs.ac.uk/glossary_v1.1/hyptest.html#pvalue">
489     * p-value</a>, associated with a Chi-Square two sample test comparing
490     * bin frequency counts in <code>observed1</code> and
491     * <code>observed2</code>.
492     * </p>
493     * <p>The number returned is the smallest significance level at which one
494     * can reject the null hypothesis that the observed counts conform to the
495     * same distribution.
496     * </p>
497     * <p>See {@link #chiSquareDataSetsComparison(long[], long[])} for details
498     * on the formula used to compute the test statistic. The degrees of
499     * of freedom used to perform the test is one less than the common length
500     * of the input observed count arrays.
501     * </p>
502     * <strong>Preconditions</strong>: <ul>
503     * <li>Observed counts must be non-negative.
504     * </li>
505     * <li>Observed counts for a specific bin must not both be zero.
506     * </li>
507     * <li>Observed counts for a specific sample must not all be 0.
508     * </li>
509     * <li>The arrays <code>observed1</code> and <code>observed2</code> must
510     * have the same length and
511     * their common length must be at least 2.
512     * </li></ul><p>
513     * If any of the preconditions are not met, an
514     * <code>IllegalArgumentException</code> is thrown.</p>
515     *
516     * @param observed1 array of observed frequency counts of the first data set
517     * @param observed2 array of observed frequency counts of the second data set
518     * @return p-value
519     * @throws DimensionMismatchException the length of the arrays does not match
520     * @throws NotPositiveException if any entries in <code>observed1</code> or
521     * <code>observed2</code> are negative
522     * @throws ZeroException if either all counts of <code>observed1</code> or
523     * <code>observed2</code> are zero, or if the count at the same index is zero
524     * for both arrays
525     * @throws MaxCountExceededException if an error occurs computing the p-value
526     * @since 1.2
527     */
528    public double chiSquareTestDataSetsComparison(long[] observed1, long[] observed2)
529        throws DimensionMismatchException, NotPositiveException, ZeroException,
530        MaxCountExceededException {
531
532        // pass a null rng to avoid unneeded overhead as we will not sample from this distribution
533        final ChiSquaredDistribution distribution =
534                ChiSquaredDistribution.of((double) observed1.length - 1);
535        return distribution.survivalProbability(
536                chiSquareDataSetsComparison(observed1, observed2));
537    }
538
539    /**
540     * <p>Performs a Chi-Square two sample test comparing two binned data
541     * sets. The test evaluates the null hypothesis that the two lists of
542     * observed counts conform to the same frequency distribution, with
543     * significance level <code>alpha</code>.  Returns true iff the null
544     * hypothesis can be rejected with 100 * (1 - alpha) percent confidence.
545     * </p>
546     * <p>See {@link #chiSquareDataSetsComparison(long[], long[])} for
547     * details on the formula used to compute the Chisquare statistic used
548     * in the test. The degrees of of freedom used to perform the test is
549     * one less than the common length of the input observed count arrays.
550     * </p>
551     * <strong>Preconditions</strong>: <ul>
552     * <li>Observed counts must be non-negative.
553     * </li>
554     * <li>Observed counts for a specific bin must not both be zero.
555     * </li>
556     * <li>Observed counts for a specific sample must not all be 0.
557     * </li>
558     * <li>The arrays <code>observed1</code> and <code>observed2</code> must
559     * have the same length and their common length must be at least 2.
560     * </li>
561     * <li> <code> 0 &lt; alpha &lt; 0.5 </code>
562     * </li></ul><p>
563     * If any of the preconditions are not met, an
564     * <code>IllegalArgumentException</code> is thrown.</p>
565     *
566     * @param observed1 array of observed frequency counts of the first data set
567     * @param observed2 array of observed frequency counts of the second data set
568     * @param alpha significance level of the test
569     * @return true iff null hypothesis can be rejected with confidence
570     * 1 - alpha
571     * @throws DimensionMismatchException the length of the arrays does not match
572     * @throws NotPositiveException if any entries in <code>observed1</code> or
573     * <code>observed2</code> are negative
574     * @throws ZeroException if either all counts of <code>observed1</code> or
575     * <code>observed2</code> are zero, or if the count at the same index is zero
576     * for both arrays
577     * @throws OutOfRangeException if <code>alpha</code> is not in the range (0, 0.5]
578     * @throws MaxCountExceededException if an error occurs performing the test
579     * @since 1.2
580     */
581    public boolean chiSquareTestDataSetsComparison(final long[] observed1,
582                                                   final long[] observed2,
583                                                   final double alpha)
584        throws DimensionMismatchException, NotPositiveException,
585        ZeroException, OutOfRangeException, MaxCountExceededException {
586
587        if (alpha <= 0 ||
588            alpha > 0.5) {
589            throw new OutOfRangeException(LocalizedFormats.OUT_OF_BOUND_SIGNIFICANCE_LEVEL,
590                                          alpha, 0, 0.5);
591        }
592        return chiSquareTestDataSetsComparison(observed1, observed2) < alpha;
593    }
594
595    /**
596     * Checks to make sure that the input long[][] array is rectangular,
597     * has at least 2 rows and 2 columns, and has all non-negative entries.
598     *
599     * @param in input 2-way table to check
600     * @throws NullArgumentException if the array is null
601     * @throws DimensionMismatchException if the array is not valid
602     * @throws NotPositiveException if the array contains any negative entries
603     */
604    private void checkArray(final long[][] in)
605        throws NullArgumentException, DimensionMismatchException,
606        NotPositiveException {
607
608        if (in.length < 2) {
609            throw new DimensionMismatchException(in.length, 2);
610        }
611
612        if (in[0].length < 2) {
613            throw new DimensionMismatchException(in[0].length, 2);
614        }
615
616        MathArrays.checkRectangular(in);
617        MathArrays.checkNonNegative(in);
618    }
619
620    /**
621     * Check the array value is non-zero.
622     *
623     * @param value Value
624     * @param name Name of the array
625     * @param index Index in the array
626     * @throws ZeroException if the value is zero
627     */
628    private static void checkNonZero(double value, String name, int index) {
629        if (value == 0) {
630            throw new ZeroException(LocalizedFormats.OBSERVED_COUNTS_ALL_ZERO,
631                name + " " + index);
632        }
633    }
634}