|
1
|
|
|
import random |
|
2
|
|
|
|
|
3
|
|
|
from pysie.stats.distributions import DistributionFamily |
|
4
|
|
|
from scipy.stats import norm, t |
|
5
|
|
|
import math |
|
6
|
|
|
|
|
7
|
|
|
|
|
8
|
|
|
class MeanDiffTesting(object): |
|
9
|
|
|
sampling_distribution = None |
|
10
|
|
|
p_value_one_tail = None |
|
11
|
|
|
p_value_two_tail = None |
|
12
|
|
|
test_statistic = None |
|
13
|
|
|
significance_level = None |
|
14
|
|
View Code Duplication |
reject_mean_same = None |
|
|
|
|
|
|
15
|
|
|
|
|
16
|
|
|
def __init__(self, sampling_distribution, significance_level=None): |
|
17
|
|
|
self.sampling_distribution = sampling_distribution |
|
18
|
|
|
if significance_level is not None: |
|
19
|
|
|
self.significance_level = significance_level |
|
20
|
|
|
|
|
21
|
|
|
if self.sampling_distribution.distribution_family == DistributionFamily.normal: |
|
22
|
|
|
Z = sampling_distribution.point_estimate / sampling_distribution.standard_error |
|
23
|
|
|
self.test_statistic = Z |
|
24
|
|
|
pf = norm.cdf(Z) |
|
25
|
|
|
if Z < 0: |
|
26
|
|
|
pf = 1 - pf |
|
27
|
|
|
self.p_value_one_tail = 1 - pf |
|
28
|
|
|
self.p_value_two_tail = self.p_value_one_tail * 2 |
|
29
|
|
|
else: |
|
30
|
|
|
td_df = sampling_distribution.point_estimate / sampling_distribution.standard_error |
|
31
|
|
|
self.test_statistic = td_df |
|
32
|
|
|
pf = t.cdf(td_df, sampling_distribution.df) |
|
33
|
|
|
if td_df < 0: |
|
34
|
|
|
pf = 1 - pf |
|
35
|
|
|
self.p_value_one_tail = 1 - pf |
|
36
|
|
|
self.p_value_two_tail = self.p_value_one_tail * 2 |
|
37
|
|
|
|
|
38
|
|
|
if significance_level is not None: |
|
39
|
|
|
self.reject_mean_same = (self.p_value_one_tail < significance_level, |
|
40
|
|
|
self.p_value_two_tail < significance_level) |
|
41
|
|
|
|
|
42
|
|
|
def will_reject(self, significance_level): |
|
43
|
|
|
|
|
44
|
|
|
return self.p_value_one_tail < significance_level, self.p_value_two_tail < significance_level |
|
45
|
|
|
|
|
46
|
|
|
|
|
47
|
|
|
class ProportionDiffTesting(object): |
|
48
|
|
|
sampling_distribution = None |
|
49
|
|
|
p_value_one_tail = None |
|
50
|
|
|
p_value_two_tail = None |
|
51
|
|
|
p_null = None |
|
52
|
|
|
test_statistic = None |
|
53
|
|
|
significance_level = None |
|
54
|
|
|
reject_proportion_same = None |
|
55
|
|
|
|
|
56
|
|
|
def __init__(self, sampling_distribution, significance_level=None): |
|
57
|
|
|
self.sampling_distribution = sampling_distribution |
|
58
|
|
|
p_null = (sampling_distribution.grp1_point_estimate + sampling_distribution.grp2_point_estimate) / 2 |
|
59
|
|
|
self.p_null = p_null |
|
60
|
|
|
if significance_level is not None: |
|
61
|
|
|
self.significance_level = significance_level |
|
62
|
|
|
|
|
63
|
|
|
if self.sampling_distribution.distribution_family == DistributionFamily.normal: |
|
64
|
|
|
standard_error_null = math.sqrt(p_null * (1 - p_null) / sampling_distribution.grp1_sample_size + p_null * (1-p_null) / sampling_distribution.grp2_sample_size) |
|
65
|
|
|
Z = sampling_distribution.point_estimate / standard_error_null |
|
66
|
|
|
self.test_statistic = Z |
|
67
|
|
|
pf = norm.cdf(Z) |
|
68
|
|
|
if Z < 0: |
|
69
|
|
|
pf = 1 - pf |
|
70
|
|
|
self.p_value_one_tail = 1 - pf |
|
71
|
|
|
self.p_value_two_tail = self.p_value_one_tail * 2 |
|
72
|
|
|
else: |
|
73
|
|
|
simulated_proportions = self.simulate() |
|
74
|
|
|
diff = sampling_distribution.grp1_point_estimate - sampling_distribution.grp2_point_estimate |
|
75
|
|
|
pf = sum(1.0 for x in simulated_proportions if x > diff) / 1000.0 |
|
76
|
|
|
self.p_value_one_tail = pf |
|
77
|
|
|
self.p_value_two_tail = sum(1.0 for x in simulated_proportions if x > diff or x < -diff) / 1000.0 |
|
78
|
|
|
|
|
79
|
|
|
if significance_level is not None: |
|
80
|
|
|
self.reject_proportion_same = (self.p_value_one_tail < significance_level, |
|
81
|
|
|
self.p_value_two_tail < significance_level) |
|
82
|
|
|
|
|
83
|
|
|
def simulate(self): |
|
84
|
|
|
simulated_proportions = [0] * 1000 |
|
85
|
|
|
for i in range(1000): |
|
86
|
|
|
count1 = 0 |
|
87
|
|
|
for trials in range(self.sampling_distribution.grp1_sample_size): |
|
88
|
|
|
if random.random() <= self.p_null: |
|
89
|
|
|
count1 += 1 |
|
90
|
|
|
count2 = 0 |
|
91
|
|
|
for trials in range(self.sampling_distribution.grp2_sample_size): |
|
92
|
|
|
if random.random() <= self.p_null: |
|
93
|
|
|
count2 += 1 |
|
94
|
|
|
|
|
95
|
|
|
simulated_proportions[i] = float(count1) / self.sampling_distribution.grp1_sample_size - float(count2) / self.sampling_distribution.grp2_sample_size |
|
96
|
|
|
return sorted(simulated_proportions) |
|
97
|
|
|
|
|
98
|
|
|
def will_reject(self, significance_level): |
|
99
|
|
|
|
|
100
|
|
|
return self.p_value_one_tail < significance_level, self.p_value_two_tail < significance_level |
chen0040 /
pysie
Loading history...