-
-
Notifications
You must be signed in to change notification settings - Fork 5.1k
/
_binomtest.py
375 lines (324 loc) · 12.7 KB
/
_binomtest.py
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
from math import sqrt
import numpy as np
from scipy._lib._util import _validate_int
from scipy.optimize import brentq
from scipy.special import ndtri
from ._discrete_distns import binom
from ._common import ConfidenceInterval
class BinomTestResult:
"""
Result of `scipy.stats.binomtest`.
Attributes
----------
k : int
The number of successes (copied from `binomtest` input).
n : int
The number of trials (copied from `binomtest` input).
alternative : str
Indicates the alternative hypothesis specified in the input
to `binomtest`. It will be one of ``'two-sided'``, ``'greater'``,
or ``'less'``.
statistic: float
The estimate of the proportion of successes.
pvalue : float
The p-value of the hypothesis test.
"""
def __init__(self, k, n, alternative, statistic, pvalue):
self.k = k
self.n = n
self.alternative = alternative
self.statistic = statistic
self.pvalue = pvalue
# add alias for backward compatibility
self.proportion_estimate = statistic
def __repr__(self):
s = ("BinomTestResult("
f"k={self.k}, "
f"n={self.n}, "
f"alternative={self.alternative!r}, "
f"statistic={self.statistic}, "
f"pvalue={self.pvalue})")
return s
def proportion_ci(self, confidence_level=0.95, method='exact'):
"""
Compute the confidence interval for ``statistic``.
Parameters
----------
confidence_level : float, optional
Confidence level for the computed confidence interval
of the estimated proportion. Default is 0.95.
method : {'exact', 'wilson', 'wilsoncc'}, optional
Selects the method used to compute the confidence interval
for the estimate of the proportion:
'exact' :
Use the Clopper-Pearson exact method [1]_.
'wilson' :
Wilson's method, without continuity correction ([2]_, [3]_).
'wilsoncc' :
Wilson's method, with continuity correction ([2]_, [3]_).
Default is ``'exact'``.
Returns
-------
ci : ``ConfidenceInterval`` object
The object has attributes ``low`` and ``high`` that hold the
lower and upper bounds of the confidence interval.
References
----------
.. [1] C. J. Clopper and E. S. Pearson, The use of confidence or
fiducial limits illustrated in the case of the binomial,
Biometrika, Vol. 26, No. 4, pp 404-413 (Dec. 1934).
.. [2] E. B. Wilson, Probable inference, the law of succession, and
statistical inference, J. Amer. Stat. Assoc., 22, pp 209-212
(1927).
.. [3] Robert G. Newcombe, Two-sided confidence intervals for the
single proportion: comparison of seven methods, Statistics
in Medicine, 17, pp 857-872 (1998).
Examples
--------
>>> from scipy.stats import binomtest
>>> result = binomtest(k=7, n=50, p=0.1)
>>> result.statistic
0.14
>>> result.proportion_ci()
ConfidenceInterval(low=0.05819170033997342, high=0.26739600249700846)
"""
if method not in ('exact', 'wilson', 'wilsoncc'):
raise ValueError("method must be one of 'exact', 'wilson' or "
"'wilsoncc'.")
if not (0 <= confidence_level <= 1):
raise ValueError('confidence_level must be in the interval '
'[0, 1].')
if method == 'exact':
low, high = _binom_exact_conf_int(self.k, self.n,
confidence_level,
self.alternative)
else:
# method is 'wilson' or 'wilsoncc'
low, high = _binom_wilson_conf_int(self.k, self.n,
confidence_level,
self.alternative,
correction=method == 'wilsoncc')
return ConfidenceInterval(low=low, high=high)
def _findp(func):
try:
p = brentq(func, 0, 1)
except RuntimeError:
raise RuntimeError('numerical solver failed to converge when '
'computing the confidence limits') from None
except ValueError as exc:
raise ValueError('brentq raised a ValueError; report this to the '
'SciPy developers') from exc
return p
def _binom_exact_conf_int(k, n, confidence_level, alternative):
"""
Compute the estimate and confidence interval for the binomial test.
Returns proportion, prop_low, prop_high
"""
if alternative == 'two-sided':
alpha = (1 - confidence_level) / 2
if k == 0:
plow = 0.0
else:
plow = _findp(lambda p: binom.sf(k-1, n, p) - alpha)
if k == n:
phigh = 1.0
else:
phigh = _findp(lambda p: binom.cdf(k, n, p) - alpha)
elif alternative == 'less':
alpha = 1 - confidence_level
plow = 0.0
if k == n:
phigh = 1.0
else:
phigh = _findp(lambda p: binom.cdf(k, n, p) - alpha)
elif alternative == 'greater':
alpha = 1 - confidence_level
if k == 0:
plow = 0.0
else:
plow = _findp(lambda p: binom.sf(k-1, n, p) - alpha)
phigh = 1.0
return plow, phigh
def _binom_wilson_conf_int(k, n, confidence_level, alternative, correction):
# This function assumes that the arguments have already been validated.
# In particular, `alternative` must be one of 'two-sided', 'less' or
# 'greater'.
p = k / n
if alternative == 'two-sided':
z = ndtri(0.5 + 0.5*confidence_level)
else:
z = ndtri(confidence_level)
# For reference, the formulas implemented here are from
# Newcombe (1998) (ref. [3] in the proportion_ci docstring).
denom = 2*(n + z**2)
center = (2*n*p + z**2)/denom
q = 1 - p
if correction:
if alternative == 'less' or k == 0:
lo = 0.0
else:
dlo = (1 + z*sqrt(z**2 - 2 - 1/n + 4*p*(n*q + 1))) / denom
lo = center - dlo
if alternative == 'greater' or k == n:
hi = 1.0
else:
dhi = (1 + z*sqrt(z**2 + 2 - 1/n + 4*p*(n*q - 1))) / denom
hi = center + dhi
else:
delta = z/denom * sqrt(4*n*p*q + z**2)
if alternative == 'less' or k == 0:
lo = 0.0
else:
lo = center - delta
if alternative == 'greater' or k == n:
hi = 1.0
else:
hi = center + delta
return lo, hi
def binomtest(k, n, p=0.5, alternative='two-sided'):
"""
Perform a test that the probability of success is p.
The binomial test [1]_ is a test of the null hypothesis that the
probability of success in a Bernoulli experiment is `p`.
Details of the test can be found in many texts on statistics, such
as section 24.5 of [2]_.
Parameters
----------
k : int
The number of successes.
n : int
The number of trials.
p : float, optional
The hypothesized probability of success, i.e. the expected
proportion of successes. The value must be in the interval
``0 <= p <= 1``. The default value is ``p = 0.5``.
alternative : {'two-sided', 'greater', 'less'}, optional
Indicates the alternative hypothesis. The default value is
'two-sided'.
Returns
-------
result : `~scipy.stats._result_classes.BinomTestResult` instance
The return value is an object with the following attributes:
k : int
The number of successes (copied from `binomtest` input).
n : int
The number of trials (copied from `binomtest` input).
alternative : str
Indicates the alternative hypothesis specified in the input
to `binomtest`. It will be one of ``'two-sided'``, ``'greater'``,
or ``'less'``.
statistic : float
The estimate of the proportion of successes.
pvalue : float
The p-value of the hypothesis test.
The object has the following methods:
proportion_ci(confidence_level=0.95, method='exact') :
Compute the confidence interval for ``statistic``.
Notes
-----
.. versionadded:: 1.7.0
References
----------
.. [1] Binomial test, https://en.wikipedia.org/wiki/Binomial_test
.. [2] Jerrold H. Zar, Biostatistical Analysis (fifth edition),
Prentice Hall, Upper Saddle River, New Jersey USA (2010)
Examples
--------
>>> from scipy.stats import binomtest
A car manufacturer claims that no more than 10% of their cars are unsafe.
15 cars are inspected for safety, 3 were found to be unsafe. Test the
manufacturer's claim:
>>> result = binomtest(3, n=15, p=0.1, alternative='greater')
>>> result.pvalue
0.18406106910639114
The null hypothesis cannot be rejected at the 5% level of significance
because the returned p-value is greater than the critical value of 5%.
The test statistic is equal to the estimated proportion, which is simply
``3/15``:
>>> result.statistic
0.2
We can use the `proportion_ci()` method of the result to compute the
confidence interval of the estimate:
>>> result.proportion_ci(confidence_level=0.95)
ConfidenceInterval(low=0.05684686759024681, high=1.0)
"""
k = _validate_int(k, 'k', minimum=0)
n = _validate_int(n, 'n', minimum=1)
if k > n:
raise ValueError('k must not be greater than n.')
if not (0 <= p <= 1):
raise ValueError("p must be in range [0,1]")
if alternative not in ('two-sided', 'less', 'greater'):
raise ValueError("alternative not recognized; \n"
"must be 'two-sided', 'less' or 'greater'")
if alternative == 'less':
pval = binom.cdf(k, n, p)
elif alternative == 'greater':
pval = binom.sf(k-1, n, p)
else:
# alternative is 'two-sided'
d = binom.pmf(k, n, p)
rerr = 1 + 1e-7
if k == p * n:
# special case as shortcut, would also be handled by `else` below
pval = 1.
elif k < p * n:
ix = _binary_search_for_binom_tst(lambda x1: -binom.pmf(x1, n, p),
-d*rerr, np.ceil(p * n), n)
# y is the number of terms between mode and n that are <= d*rerr.
# ix gave us the first term where a(ix) <= d*rerr < a(ix-1)
# if the first equality doesn't hold, y=n-ix. Otherwise, we
# need to include ix as well as the equality holds. Note that
# the equality will hold in very very rare situations due to rerr.
y = n - ix + int(d*rerr == binom.pmf(ix, n, p))
pval = binom.cdf(k, n, p) + binom.sf(n - y, n, p)
else:
ix = _binary_search_for_binom_tst(lambda x1: binom.pmf(x1, n, p),
d*rerr, 0, np.floor(p * n))
# y is the number of terms between 0 and mode that are <= d*rerr.
# we need to add a 1 to account for the 0 index.
# For comparing this with old behavior, see
# tst_binary_srch_for_binom_tst method in test_morestats.
y = ix + 1
pval = binom.cdf(y-1, n, p) + binom.sf(k-1, n, p)
pval = min(1.0, pval)
result = BinomTestResult(k=k, n=n, alternative=alternative,
statistic=k/n, pvalue=pval)
return result
def _binary_search_for_binom_tst(a, d, lo, hi):
"""
Conducts an implicit binary search on a function specified by `a`.
Meant to be used on the binomial PMF for the case of two-sided tests
to obtain the value on the other side of the mode where the tail
probability should be computed. The values on either side of
the mode are always in order, meaning binary search is applicable.
Parameters
----------
a : callable
The function over which to perform binary search. Its values
for inputs lo and hi should be in ascending order.
d : float
The value to search.
lo : int
The lower end of range to search.
hi : int
The higher end of the range to search.
Returns
-------
int
The index, i between lo and hi
such that a(i)<=d<a(i+1)
"""
while lo < hi:
mid = lo + (hi-lo)//2
midval = a(mid)
if midval < d:
lo = mid+1
elif midval > d:
hi = mid-1
else:
return mid
if a(lo) <= d:
return lo
else:
return lo-1