• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1[section:students_t_dist Students t Distribution]
2
3``#include <boost/math/distributions/students_t.hpp>``
4
5   namespace boost{ namespace math{
6
7   template <class RealType = double,
8             class ``__Policy``   = ``__policy_class`` >
9   class students_t_distribution;
10
11   typedef students_t_distribution<> students_t;
12
13   template <class RealType, class ``__Policy``>
14   class students_t_distribution
15   {
16      typedef RealType value_type;
17      typedef Policy   policy_type;
18
19      // Constructor:
20      students_t_distribution(const RealType& v);
21
22      // Accessor:
23      RealType degrees_of_freedom()const;
24
25      // degrees of freedom estimation:
26      static RealType find_degrees_of_freedom(
27         RealType difference_from_mean,
28         RealType alpha,
29         RealType beta,
30         RealType sd,
31         RealType hint = 100);
32   };
33
34   }} // namespaces
35
36Student's t-distribution is a statistical distribution published by William Gosset in 1908.
37His employer, Guinness Breweries, required him to publish under a
38pseudonym (possibly to hide that they were using statistics to improve beer quality),
39so he chose "Student".
40
41Given N independent measurements, let
42
43[equation students_t_dist]
44
45where /M/ is the population mean, [mu] is the sample mean, and /s/ is the sample variance.
46
47[@https://en.wikipedia.org/wiki/Student%27s_t-distribution Student's t-distribution]
48is defined as the distribution of the random
49variable t which is  - very loosely - the "best" that we can do while not
50knowing the true standard deviation of the sample.  It has the PDF:
51
52[equation students_t_ref1]
53
54The Student's t-distribution takes a single parameter: the number of
55degrees of freedom of the sample. When the degrees of freedom is
56/one/ then this distribution is the same as the Cauchy-distribution.
57As the number of degrees of freedom tends towards infinity, then this
58distribution approaches the normal-distribution.  The following graph
59illustrates how the PDF varies with the degrees of freedom [nu]:
60
61[graph students_t_pdf]
62
63[h4 Member Functions]
64
65   students_t_distribution(const RealType& v);
66
67Constructs a Student's t-distribution with /v/ degrees of freedom.
68
69Requires /v/ > 0, including infinity (if RealType permits),
70otherwise calls __domain_error.  Note that
71non-integral degrees of freedom are supported,
72and are meaningful under certain circumstances.
73
74   RealType degrees_of_freedom()const;
75
76returns the number of degrees of freedom of this distribution.
77
78   static RealType find_degrees_of_freedom(
79      RealType difference_from_mean,
80      RealType alpha,
81      RealType beta,
82      RealType sd,
83      RealType hint = 100);
84
85returns the number of degrees of freedom required to observe a significant
86result in the Student's t test when the mean differs from the "true"
87mean by /difference_from_mean/.
88
89[variablelist
90[[difference_from_mean][The difference between the true mean and the sample mean
91                        that we wish to show is significant.]]
92[[alpha][The maximum acceptable probability of rejecting the null hypothesis
93        when it is in fact true.]]
94[[beta][The maximum acceptable probability of failing to reject the null hypothesis
95        when it is in fact false.]]
96[[sd][The sample standard deviation.]]
97[[hint][A hint for the location to start looking for the result, a good choice for this
98      would be the sample size of a previous borderline Student's t test.]]
99]
100
101[note
102Remember that for a two-sided test, you must divide alpha by two
103before calling this function.]
104
105For more information on this function see the
106[@http://www.itl.nist.gov/div898/handbook/prc/section2/prc222.htm
107NIST Engineering Statistics Handbook].
108
109[h4 Non-member Accessors]
110
111All the [link math_toolkit.dist_ref.nmp usual non-member accessor functions] that are generic to all
112distributions are supported: __usual_accessors.
113
114The domain of the random variable is \[-[infin], +[infin]\].
115
116[h4 Examples]
117
118Various [link math_toolkit.stat_tut.weg.st_eg worked examples] are available illustrating the use of the Student's t
119distribution.
120
121[h4 Accuracy]
122
123The normal distribution is implemented in terms of the
124[link math_toolkit.sf_beta.ibeta_function incomplete beta function]
125and [link math_toolkit.sf_beta.ibeta_inv_function its inverses],
126refer to accuracy data on those functions for more information.
127
128[h4 Implementation]
129
130In the following table /v/ is the degrees of freedom of the distribution,
131/t/ is the random variate, /p/ is the probability and /q = 1-p/.
132
133[table
134[[Function][Implementation Notes]]
135[[pdf][Using the relation: [role serif_italic pdf = (v \/ (v + t[super 2]))[super (1+v)\/2 ] / (sqrt(v) * __beta(v\/2, 0.5))] ]]
136[[cdf][Using the relations:
137
138[role serif_italic p = 1 - z /iff t > 0/]
139
140[role serif_italic p = z     /otherwise/]
141
142where z is given by:
143
144__ibeta(v \/ 2, 0.5, v \/ (v + t[super 2])) \/ 2 ['iff v < 2t[super 2]]
145
146__ibetac(0.5, v \/ 2, t[super 2 ] / (v + t[super 2]) \/ 2   /otherwise/]]
147[[cdf complement][Using the relation: q = cdf(-t) ]]
148[[quantile][Using the relation: [role serif_italic t = sign(p - 0.5) * sqrt(v * y \/ x)]
149
150where:
151
152[role serif_italic x = __ibeta_inv(v \/ 2, 0.5, 2 * min(p, q)) ]
153
154[role serif_italic y = 1 - x]
155
156The quantities /x/ and /y/ are both returned by __ibeta_inv
157without the subtraction implied above.]]
158[[quantile from the complement][Using the relation: t = -quantile(q)]]
159[[mode][0]]
160[[mean][0]]
161[[variance][if (v > 2) v \/ (v - 2) else NaN]]
162[[skewness][if (v > 3) 0 else NaN ]]
163[[kurtosis][if (v > 4) 3 * (v - 2) \/ (v - 4) else NaN]]
164[[kurtosis excess][if (v > 4) 6 \/ (df - 4) else NaN]]
165]
166
167If the moment index /k/ is less than /v/, then the moment is undefined.
168Evaluating the moment will throw a __domain_error unless ignored by a policy,
169when it will return `std::numeric_limits<>::quiet_NaN();`
170
171[h5:implementation Implementation]
172
173(By popular demand, we now support infinite argument and random deviate.
174But we have not implemented the return of infinity
175as suggested by [@http://en.wikipedia.org/wiki/Student%27s_t-distribution Wikipedia Student's t],
176instead throwing a domain error or return NaN.
177See also [@https://svn.boost.org/trac/boost/ticket/7177].)
178
179[endsect] [/section:students_t_dist Students t]
180
181[/ students_t.qbk
182  Copyright 2006, 2012, 2017 John Maddock and Paul A. Bristow.
183  Distributed under the Boost Software License, Version 1.0.
184  (See accompanying file LICENSE_1_0.txt or copy at
185  http://www.boost.org/LICENSE_1_0.txt).
186]
187
188