File size: 17,826 Bytes
7885a28
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
import copy

import numpy as np
import pytest
from numpy.testing import assert_allclose

from scipy import stats
from scipy.stats._multicomp import _pvalue_dunnett, DunnettResult


class TestDunnett:
    # For the following tests, p-values were computed using Matlab, e.g.
    #     sample = [18.  15.  18.  16.  17.  15.  14.  14.  14.  15.  15....
    #               14.  15.  14.  22.  18.  21.  21.  10.  10.  11.  9....
    #               25.  26.  17.5 16.  15.5 14.5 22.  22.  24.  22.5 29....
    #               24.5 20.  18.  18.5 17.5 26.5 13.  16.5 13.  13.  13....
    #               28.  27.  34.  31.  29.  27.  24.  23.  38.  36.  25....
    #               38. 26.  22.  36.  27.  27.  32.  28.  31....
    #               24.  27.  33.  32.  28.  19. 37.  31.  36.  36....
    #               34.  38.  32.  38.  32....
    #               26.  24.  26.  25.  29. 29.5 16.5 36.  44....
    #               25.  27.  19....
    #               25.  20....
    #               28.];
    #     j = [0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
    #          0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ...
    #          0 0 0 0...
    #          1 1 1 1 1 1 1 1 1 1 1 1 1 1 1...
    #          2 2 2 2 2 2 2 2 2...
    #          3 3 3...
    #          4 4...
    #          5];
    #     [~, ~, stats] = anova1(sample, j, "off");
    #     [results, ~, ~, gnames] = multcompare(stats, ...
    #     "CriticalValueType", "dunnett", ...
    #     "Approximate", false);
    #     tbl = array2table(results, "VariableNames", ...
    #     ["Group", "Control Group", "Lower Limit", ...
    #     "Difference", "Upper Limit", "P-value"]);
    #     tbl.("Group") = gnames(tbl.("Group"));
    #     tbl.("Control Group") = gnames(tbl.("Control Group"))

    # Matlab doesn't report the statistic, so the statistics were
    # computed using R multcomp `glht`, e.g.:
    #     library(multcomp)
    #     options(digits=16)
    #     control < - c(18.0, 15.0, 18.0, 16.0, 17.0, 15.0, 14.0, 14.0, 14.0,
    #                   15.0, 15.0, 14.0, 15.0, 14.0, 22.0, 18.0, 21.0, 21.0,
    #                   10.0, 10.0, 11.0, 9.0, 25.0, 26.0, 17.5, 16.0, 15.5,
    #                   14.5, 22.0, 22.0, 24.0, 22.5, 29.0, 24.5, 20.0, 18.0,
    #                   18.5, 17.5, 26.5, 13.0, 16.5, 13.0, 13.0, 13.0, 28.0,
    #                   27.0, 34.0, 31.0, 29.0, 27.0, 24.0, 23.0, 38.0, 36.0,
    #                   25.0, 38.0, 26.0, 22.0, 36.0, 27.0, 27.0, 32.0, 28.0,
    #                   31.0)
    #     t < - c(24.0, 27.0, 33.0, 32.0, 28.0, 19.0, 37.0, 31.0, 36.0, 36.0,
    #             34.0, 38.0, 32.0, 38.0, 32.0)
    #     w < - c(26.0, 24.0, 26.0, 25.0, 29.0, 29.5, 16.5, 36.0, 44.0)
    #     x < - c(25.0, 27.0, 19.0)
    #     y < - c(25.0, 20.0)
    #     z < - c(28.0)
    #
    #     groups = factor(rep(c("control", "t", "w", "x", "y", "z"),
    #                         times=c(length(control), length(t), length(w),
    #                                 length(x), length(y), length(z))))
    #     df < - data.frame(response=c(control, t, w, x, y, z),
    #                       group=groups)
    #     model < - aov(response
    #     ~group, data = df)
    #     test < - glht(model=model,
    #                   linfct=mcp(group="Dunnett"),
    #                   alternative="g")
    #     summary(test)
    #     confint(test)
    # p-values agreed with those produced by Matlab to at least atol=1e-3

    # From Matlab's documentation on multcompare
    samples_1 = [
        [
            24.0, 27.0, 33.0, 32.0, 28.0, 19.0, 37.0, 31.0, 36.0, 36.0,
            34.0, 38.0, 32.0, 38.0, 32.0
        ],
        [26.0, 24.0, 26.0, 25.0, 29.0, 29.5, 16.5, 36.0, 44.0],
        [25.0, 27.0, 19.0],
        [25.0, 20.0],
        [28.0]
    ]
    control_1 = [
        18.0, 15.0, 18.0, 16.0, 17.0, 15.0, 14.0, 14.0, 14.0, 15.0, 15.0,
        14.0, 15.0, 14.0, 22.0, 18.0, 21.0, 21.0, 10.0, 10.0, 11.0, 9.0,
        25.0, 26.0, 17.5, 16.0, 15.5, 14.5, 22.0, 22.0, 24.0, 22.5, 29.0,
        24.5, 20.0, 18.0, 18.5, 17.5, 26.5, 13.0, 16.5, 13.0, 13.0, 13.0,
        28.0, 27.0, 34.0, 31.0, 29.0, 27.0, 24.0, 23.0, 38.0, 36.0, 25.0,
        38.0, 26.0, 22.0, 36.0, 27.0, 27.0, 32.0, 28.0, 31.0
    ]
    pvalue_1 = [4.727e-06, 0.022346, 0.97912, 0.99953, 0.86579]  # Matlab
    # Statistic, alternative p-values, and CIs computed with R multcomp `glht`
    p_1_twosided = [1e-4, 0.02237, 0.97913, 0.99953, 0.86583]
    p_1_greater = [1e-4, 0.011217, 0.768500, 0.896991, 0.577211]
    p_1_less = [1, 1, 0.99660, 0.98398, .99953]
    statistic_1 = [5.27356, 2.91270, 0.60831, 0.27002, 0.96637]
    ci_1_twosided = [[5.3633917835622, 0.7296142201217, -8.3879817106607,
                      -11.9090753452911, -11.7655021543469],
                     [15.9709832164378, 13.8936496687672, 13.4556900439941,
                      14.6434503452911, 25.4998771543469]]
    ci_1_greater = [5.9036402398526, 1.4000632918725, -7.2754756323636,
                    -10.5567456382391, -9.8675629499576]
    ci_1_less = [15.4306165948619, 13.2230539537359, 12.3429406339544,
                 13.2908248513211, 23.6015228251660]
    pvalues_1 = dict(twosided=p_1_twosided, less=p_1_less, greater=p_1_greater)
    cis_1 = dict(twosided=ci_1_twosided, less=ci_1_less, greater=ci_1_greater)
    case_1 = dict(samples=samples_1, control=control_1, statistic=statistic_1,
                  pvalues=pvalues_1, cis=cis_1)

    # From Dunnett1955 comparing with R's DescTools: DunnettTest
    samples_2 = [[9.76, 8.80, 7.68, 9.36], [12.80, 9.68, 12.16, 9.20, 10.55]]
    control_2 = [7.40, 8.50, 7.20, 8.24, 9.84, 8.32]
    pvalue_2 = [0.6201, 0.0058]
    # Statistic, alternative p-values, and CIs computed with R multcomp `glht`
    p_2_twosided = [0.6201020, 0.0058254]
    p_2_greater = [0.3249776, 0.0029139]
    p_2_less = [0.91676, 0.99984]
    statistic_2 = [0.85703, 3.69375]
    ci_2_twosided = [[-1.2564116462124, 0.8396273539789],
                     [2.5564116462124, 4.4163726460211]]
    ci_2_greater = [-0.9588591188156, 1.1187563667543]
    ci_2_less = [2.2588591188156, 4.1372436332457]
    pvalues_2 = dict(twosided=p_2_twosided, less=p_2_less, greater=p_2_greater)
    cis_2 = dict(twosided=ci_2_twosided, less=ci_2_less, greater=ci_2_greater)
    case_2 = dict(samples=samples_2, control=control_2, statistic=statistic_2,
                  pvalues=pvalues_2, cis=cis_2)

    samples_3 = [[55, 64, 64], [55, 49, 52], [50, 44, 41]]
    control_3 = [55, 47, 48]
    pvalue_3 = [0.0364, 0.8966, 0.4091]
    # Statistic, alternative p-values, and CIs computed with R multcomp `glht`
    p_3_twosided = [0.036407, 0.896539, 0.409295]
    p_3_greater = [0.018277, 0.521109, 0.981892]
    p_3_less = [0.99944, 0.90054, 0.20974]
    statistic_3 = [3.09073, 0.56195, -1.40488]
    ci_3_twosided = [[0.7529028025053, -8.2470971974947, -15.2470971974947],
                     [21.2470971974947, 12.2470971974947, 5.2470971974947]]
    ci_3_greater = [2.4023682323149, -6.5976317676851, -13.5976317676851]
    ci_3_less = [19.5984402363662, 10.5984402363662, 3.5984402363662]
    pvalues_3 = dict(twosided=p_3_twosided, less=p_3_less, greater=p_3_greater)
    cis_3 = dict(twosided=ci_3_twosided, less=ci_3_less, greater=ci_3_greater)
    case_3 = dict(samples=samples_3, control=control_3, statistic=statistic_3,
                  pvalues=pvalues_3, cis=cis_3)

    # From Thomson and Short,
    # Mucociliary function in health, chronic obstructive airway disease,
    # and asbestosis, Journal of Applied Physiology, 1969. Table 1
    # Comparing with R's DescTools: DunnettTest
    samples_4 = [[3.8, 2.7, 4.0, 2.4], [2.8, 3.4, 3.7, 2.2, 2.0]]
    control_4 = [2.9, 3.0, 2.5, 2.6, 3.2]
    pvalue_4 = [0.5832, 0.9982]
    # Statistic, alternative p-values, and CIs computed with R multcomp `glht`
    p_4_twosided = [0.58317, 0.99819]
    p_4_greater = [0.30225, 0.69115]
    p_4_less = [0.91929, 0.65212]
    statistic_4 = [0.90875, -0.05007]
    ci_4_twosided = [[-0.6898153448579, -1.0333456251632],
                     [1.4598153448579, 0.9933456251632]]
    ci_4_greater = [-0.5186459268412, -0.8719655502147 ]
    ci_4_less = [1.2886459268412, 0.8319655502147]
    pvalues_4 = dict(twosided=p_4_twosided, less=p_4_less, greater=p_4_greater)
    cis_4 = dict(twosided=ci_4_twosided, less=ci_4_less, greater=ci_4_greater)
    case_4 = dict(samples=samples_4, control=control_4, statistic=statistic_4,
                  pvalues=pvalues_4, cis=cis_4)

    @pytest.mark.parametrize(
        'rho, n_groups, df, statistic, pvalue, alternative',
        [
            # From Dunnett1955
            # Tables 1a and 1b pages 1117-1118
            (0.5, 1, 10, 1.81, 0.05, "greater"),  # different than two-sided
            (0.5, 3, 10, 2.34, 0.05, "greater"),
            (0.5, 2, 30, 1.99, 0.05, "greater"),
            (0.5, 5, 30, 2.33, 0.05, "greater"),
            (0.5, 4, 12, 3.32, 0.01, "greater"),
            (0.5, 7, 12, 3.56, 0.01, "greater"),
            (0.5, 2, 60, 2.64, 0.01, "greater"),
            (0.5, 4, 60, 2.87, 0.01, "greater"),
            (0.5, 4, 60, [2.87, 2.21], [0.01, 0.05], "greater"),
            # Tables 2a and 2b pages 1119-1120
            (0.5, 1, 10, 2.23, 0.05, "two-sided"),  # two-sided
            (0.5, 3, 10, 2.81, 0.05, "two-sided"),
            (0.5, 2, 30, 2.32, 0.05, "two-sided"),
            (0.5, 3, 20, 2.57, 0.05, "two-sided"),
            (0.5, 4, 12, 3.76, 0.01, "two-sided"),
            (0.5, 7, 12, 4.08, 0.01, "two-sided"),
            (0.5, 2, 60, 2.90, 0.01, "two-sided"),
            (0.5, 4, 60, 3.14, 0.01, "two-sided"),
            (0.5, 4, 60, [3.14, 2.55], [0.01, 0.05], "two-sided"),
        ],
    )
    def test_critical_values(
        self, rho, n_groups, df, statistic, pvalue, alternative
    ):
        rng = np.random.default_rng(165250594791731684851746311027739134893)
        rho = np.full((n_groups, n_groups), rho)
        np.fill_diagonal(rho, 1)

        statistic = np.array(statistic)
        res = _pvalue_dunnett(
            rho=rho, df=df, statistic=statistic,
            alternative=alternative,
            rng=rng
        )
        assert_allclose(res, pvalue, atol=5e-3)

    @pytest.mark.parametrize(
        'samples, control, pvalue, statistic',
        [
            (samples_1, control_1, pvalue_1, statistic_1),
            (samples_2, control_2, pvalue_2, statistic_2),
            (samples_3, control_3, pvalue_3, statistic_3),
            (samples_4, control_4, pvalue_4, statistic_4),
        ]
    )
    def test_basic(self, samples, control, pvalue, statistic):
        rng = np.random.default_rng(11681140010308601919115036826969764808)

        res = stats.dunnett(*samples, control=control, rng=rng)

        assert isinstance(res, DunnettResult)
        assert_allclose(res.statistic, statistic, rtol=5e-5)
        assert_allclose(res.pvalue, pvalue, rtol=1e-2, atol=1e-4)

    @pytest.mark.parametrize(
        'alternative',
        ['two-sided', 'less', 'greater']
    )
    def test_ttest_ind(self, alternative):
        # check that `dunnett` agrees with `ttest_ind`
        # when there are only two groups
        rng = np.random.default_rng(114184017807316971636137493526995620351)

        for _ in range(10):
            sample = rng.integers(-100, 100, size=(10,))
            control = rng.integers(-100, 100, size=(10,))

            # preserve use of old random_state during SPEC 7 transition
            res = stats.dunnett(
                sample, control=control,
                alternative=alternative, random_state=rng
            )
            ref = stats.ttest_ind(
                sample, control,
                alternative=alternative
            )

            assert_allclose(res.statistic, ref.statistic, rtol=1e-3, atol=1e-5)
            assert_allclose(res.pvalue, ref.pvalue, rtol=1e-3, atol=1e-5)

    @pytest.mark.parametrize(
        'alternative, pvalue',
        [
            ('less', [0, 1]),
            ('greater', [1, 0]),
            ('two-sided', [0, 0]),
        ]
    )
    def test_alternatives(self, alternative, pvalue):
        rng = np.random.default_rng(114184017807316971636137493526995620351)

        # width of 20 and min diff between samples/control is 60
        # and maximal diff would be 100
        sample_less = rng.integers(0, 20, size=(10,))
        control = rng.integers(80, 100, size=(10,))
        sample_greater = rng.integers(160, 180, size=(10,))

        res = stats.dunnett(
            sample_less, sample_greater, control=control,
            alternative=alternative, rng=rng
        )
        assert_allclose(res.pvalue, pvalue, atol=1e-7)

        ci = res.confidence_interval()
        # two-sided is comparable for high/low
        if alternative == 'less':
            assert np.isneginf(ci.low).all()
            assert -100 < ci.high[0] < -60
            assert 60 < ci.high[1] < 100
        elif alternative == 'greater':
            assert -100 < ci.low[0] < -60
            assert 60 < ci.low[1] < 100
            assert np.isposinf(ci.high).all()
        elif alternative == 'two-sided':
            assert -100 < ci.low[0] < -60
            assert 60 < ci.low[1] < 100
            assert -100 < ci.high[0] < -60
            assert 60 < ci.high[1] < 100

    @pytest.mark.parametrize("case", [case_1, case_2, case_3, case_4])
    @pytest.mark.parametrize("alternative", ['less', 'greater', 'two-sided'])
    def test_against_R_multicomp_glht(self, case, alternative):
        rng = np.random.default_rng(189117774084579816190295271136455278291)
        samples = case['samples']
        control = case['control']
        alternatives = {'less': 'less', 'greater': 'greater',
                        'two-sided': 'twosided'}
        p_ref = case['pvalues'][alternative.replace('-', '')]

        res = stats.dunnett(*samples, control=control, alternative=alternative,
                            rng=rng)
        # atol can't be tighter because R reports some pvalues as "< 1e-4"
        assert_allclose(res.pvalue, p_ref, rtol=5e-3, atol=1e-4)

        ci_ref = case['cis'][alternatives[alternative]]
        if alternative == "greater":
            ci_ref = [ci_ref, np.inf]
        elif alternative == "less":
            ci_ref = [-np.inf, ci_ref]
        assert res._ci is None
        assert res._ci_cl is None
        ci = res.confidence_interval(confidence_level=0.95)
        assert_allclose(ci.low, ci_ref[0], rtol=5e-3, atol=1e-5)
        assert_allclose(ci.high, ci_ref[1], rtol=5e-3, atol=1e-5)

        # re-run to use the cached value "is" to check id as same object
        assert res._ci is ci
        assert res._ci_cl == 0.95
        ci_ = res.confidence_interval(confidence_level=0.95)
        assert ci_ is ci

    @pytest.mark.parametrize('alternative', ["two-sided", "less", "greater"])
    def test_str(self, alternative):
        rng = np.random.default_rng(189117774084579816190295271136455278291)

        res = stats.dunnett(
            *self.samples_3, control=self.control_3, alternative=alternative,
            rng=rng
        )

        # check some str output
        res_str = str(res)
        assert '(Sample 2 - Control)' in res_str
        assert '95.0%' in res_str

        if alternative == 'less':
            assert '-inf' in res_str
            assert '19.' in res_str
        elif alternative == 'greater':
            assert 'inf' in res_str
            assert '-13.' in res_str
        else:
            assert 'inf' not in res_str
            assert '21.' in res_str

    def test_warnings(self):
        rng = np.random.default_rng(189117774084579816190295271136455278291)

        res = stats.dunnett(
            *self.samples_3, control=self.control_3, rng=rng
        )
        msg = r"Computation of the confidence interval did not converge"
        with pytest.warns(UserWarning, match=msg):
            res._allowance(tol=1e-5)

    def test_raises(self):
        samples, control = self.samples_3, self.control_3

        # alternative
        with pytest.raises(ValueError, match="alternative must be"):
            stats.dunnett(*samples, control=control, alternative='bob')

        # 2D for a sample
        samples_ = copy.deepcopy(samples)
        samples_[0] = [samples_[0]]
        with pytest.raises(ValueError, match="must be 1D arrays"):
            stats.dunnett(*samples_, control=control)

        # 2D for control
        control_ = copy.deepcopy(control)
        control_ = [control_]
        with pytest.raises(ValueError, match="must be 1D arrays"):
            stats.dunnett(*samples, control=control_)

        # No obs in a sample
        samples_ = copy.deepcopy(samples)
        samples_[1] = []
        with pytest.raises(ValueError, match="at least 1 observation"):
            stats.dunnett(*samples_, control=control)

        # No obs in control
        control_ = []
        with pytest.raises(ValueError, match="at least 1 observation"):
            stats.dunnett(*samples, control=control_)

        res = stats.dunnett(*samples, control=control)
        with pytest.raises(ValueError, match="Confidence level must"):
            res.confidence_interval(confidence_level=3)

    @pytest.mark.filterwarnings("ignore:Computation of the confidence")
    @pytest.mark.parametrize('n_samples', [1, 2, 3])
    def test_shapes(self, n_samples):
        rng = np.random.default_rng(689448934110805334)
        samples = rng.normal(size=(n_samples, 10))
        control = rng.normal(size=10)
        res = stats.dunnett(*samples, control=control, rng=rng)
        assert res.statistic.shape == (n_samples,)
        assert res.pvalue.shape == (n_samples,)
        ci = res.confidence_interval()
        assert ci.low.shape == (n_samples,)
        assert ci.high.shape == (n_samples,)