An unadulterated confidence range , also an unadulterated range estimator or an undistorted confidence range is a special confidence range in mathematical statistics . The authenticity itself is not a concept of optimality, but enables the construction of optimal confidence ranges as well as confidence ranges with minimal volume. If the confidence range is one-dimensional, one speaks accordingly of an undistorted / undistorted confidence interval or of an undistorted interval estimation function .
definition
A statistical model as well as a decision space and a function to be estimated are given
(
X
,
A.
,
(
P
ϑ
)
ϑ
∈
Θ
)
{\ displaystyle (X, {\ mathcal {A}}, (P _ {\ vartheta}) _ {\ vartheta \ in \ Theta})}
(
Γ
,
A.
Γ
)
{\ displaystyle (\ Gamma, {\ mathcal {A}} _ {\ Gamma})}
G
:
Θ
→
Γ
{\ displaystyle g \ colon \ Theta \ to \ Gamma}
,
which in the parametric case is also referred to as a parameter function.
A confidence range
C.
:
X
→
P
(
Γ
)
{\ displaystyle C \ colon X \ to {\ mathcal {P}} (\ Gamma)}
for the confidence level is called an unadulterated confidence range, if for all
1
-
α
{\ displaystyle 1- \ alpha}
ϑ
,
ϑ
′
∈
Θ
{\ displaystyle \ vartheta, \ vartheta '\ in \ Theta}
P
ϑ
(
{
G
(
ϑ
)
∈
C.
}
)
≥
P
ϑ
(
{
G
(
ϑ
′
)
∈
C.
}
)
{\ displaystyle P _ {\ vartheta} (\ {g (\ vartheta) \ in C \}) \ geq P _ {\ vartheta} (\ {g (\ vartheta ') \ in C \})}
applies. For each , the probability of covering the correct parameter is greater than the probability of covering any other parameter .
ϑ
{\ displaystyle \ vartheta}
G
(
ϑ
)
{\ displaystyle g (\ vartheta)}
G
(
ϑ
′
)
{\ displaystyle g (\ vartheta ')}
example
Given is the normal distribution model with known variance and unknown expected value
σ
0
2
{\ displaystyle \ sigma _ {0} ^ {2}}
, i.e. the statistical model , provided with the distribution class . The expected value should be covered , the parameter function is therefore given by
(
R.
n
,
B.
(
R.
n
)
,
P
)
{\ displaystyle (\ mathbb {R} ^ {n}, {\ mathcal {B}} (\ mathbb {R} ^ {n}), {\ mathcal {P}})}
P
=
{
N
⊗
n
(
μ
,
σ
0
2
)
|
μ
∈
R.
}
{\ displaystyle {\ mathcal {P}} = \ {{\ mathcal {N}} ^ {\ otimes n} (\ mu, {\ sigma _ {0}} ^ {2}) \; | \; \ mu \ in \ mathbb {R} \}}
μ
{\ displaystyle \ mu}
G
(
μ
)
=
μ
{\ displaystyle g (\ mu) = \ mu}
.
A mutual confidence range for the expected value is given by, for example
μ
{\ displaystyle \ mu}
C.
(
x
)
=
[
x
¯
-
σ
0
n
u
1
-
α
/
2
;
x
¯
+
σ
0
n
u
1
-
α
/
2
]
{\ displaystyle C (x) = \ left [{\ overline {x}} - {\ tfrac {\ sigma _ {0}} {\ sqrt {n}}} u_ {1- \ alpha / 2}; {\ overline {x}} + {\ tfrac {\ sigma _ {0}} {\ sqrt {n}}} u_ {1- \ alpha / 2} \ right]}
.
The - is the quantile of the standard normal distribution and the sample mean .
u
α
{\ displaystyle u _ {\ alpha}}
α
{\ displaystyle \ alpha}
x
¯
{\ displaystyle {\ overline {x}}}
The confidence range is unadulterated because it is forever
μ
≠
μ
′
{\ displaystyle \ mu \ neq \ mu '}
P
μ
(
{
μ
′
∈
C.
}
)
=
P
μ
(
x
¯
-
σ
0
n
u
1
-
α
/
2
≤
μ
′
≤
x
¯
+
σ
0
n
u
1
-
α
/
2
)
=
P
μ
(
μ
′
-
μ
σ
0
/
n
-
u
1
-
α
/
2
≤
x
¯
-
μ
σ
0
/
n
≤
μ
′
-
μ
σ
0
/
n
+
u
1
-
α
/
2
)
=
Φ
(
μ
′
-
μ
σ
0
/
n
+
u
1
-
α
/
2
)
-
Φ
(
μ
′
-
μ
σ
0
/
n
-
u
1
-
α
/
2
)
{\ displaystyle {\ begin {aligned} P _ {\ mu} \ left (\ {\ mu '\ in C \} \ right) & = P _ {\ mu} \ left ({\ overline {x}} - {\ tfrac {\ sigma _ {0}} {\ sqrt {n}}} u_ {1- \ alpha / 2} \ leq \ mu '\ leq {\ overline {x}} + {\ tfrac {\ sigma _ {0 }} {\ sqrt {n}}} u_ {1- \ alpha / 2} \ right) \\ & = P _ {\ mu} \ left ({\ frac {\ mu '- \ mu} {\ sigma _ { 0} / {\ sqrt {n}}}} - u_ {1- \ alpha / 2} \ leq {\ frac {{\ overline {x}} - \ mu} {\ sigma _ {0} / {\ sqrt {n}}}} \ leq {\ frac {\ mu '- \ mu} {\ sigma _ {0} / {\ sqrt {n}}}} + u_ {1- \ alpha / 2} \ right) \ \ & = \ Phi \ left ({\ frac {\ mu '- \ mu} {\ sigma _ {0} / {\ sqrt {n}}}} + u_ {1- \ alpha / 2} \ right) - \ Phi \ left ({\ frac {\ mu '- \ mu} {\ sigma _ {0} / {\ sqrt {n}}}} - u_ {1- \ alpha / 2} \ right) \ end {aligned }}}
,
where denotes the distribution function of the standard normal distribution. But the last expression is maximal for , so the confidence range is unadulterated.
Φ
(
⋅
)
{\ displaystyle \ Phi (\ cdot)}
μ
=
μ
′
{\ displaystyle \ mu = \ mu '}
General definition via form hypotheses
Under the same general conditions as above, a confidence range for the form hypotheses and for the confidence level is called an unadulterated confidence range, if for all
C.
{\ displaystyle C}
(
H
~
ϑ
,
K
~
ϑ
)
ϑ
∈
Θ
{\ displaystyle ({\ tilde {H}} _ {\ vartheta}, {\ tilde {K}} _ {\ vartheta}) _ {\ vartheta \ in \ Theta}}
1
-
α
{\ displaystyle 1- \ alpha}
ϑ
∈
Θ
{\ displaystyle \ vartheta \ in \ Theta}
P
ϑ
(
{
γ
∈
C.
}
)
≤
1
-
α
{\ displaystyle P _ {\ vartheta} (\ {\ gamma \ in C \}) \ leq 1- \ alpha}
for all
γ
∈
K
~
ϑ
{\ displaystyle \ gamma \ in {\ tilde {K}} _ {\ vartheta}}
is.
Each value from the "set to be avoided" is therefore covered less often than each value from the "set to be covered" (see shape hypotheses # confidence ranges for shape hypotheses )
K
~
ϑ
{\ displaystyle {\ tilde {K}} _ {\ vartheta}}
H
~
ϑ
{\ displaystyle {\ tilde {H}} _ {\ vartheta}}
The first formulation results from using the form hypotheses
H
~
ϑ
=
{
G
(
ϑ
)
}
{\ displaystyle {\ tilde {H}} _ {\ vartheta} = \ {g (\ vartheta) \}}
and
K
~
ϑ
=
{
G
(
ϑ
′
)
∣
ϑ
′
≠
ϑ
}
{\ displaystyle {\ tilde {K}} _ {\ vartheta} = \ {g (\ vartheta ') \ mid \ vartheta' \ neq \ vartheta \}}
and assuming that is injective .
G
{\ displaystyle g}
Related terms
The corresponding term for statistical tests in the sense of the duality of tests and confidence ranges are the unadulterated tests .
literature
Individual evidence
↑ Claudia Czado, Thorsten Schmidt: Mathematical Statistics . Springer-Verlag, Berlin Heidelberg 2011, ISBN 978-3-642-17260-1 , p. 142 , doi : 10.1007 / 978-3-642-17261-8 .
^ Ludger Rüschendorf: Mathematical Statistics . Springer Verlag, Berlin Heidelberg 2014, ISBN 978-3-642-41996-6 , p. 241 , doi : 10.1007 / 978-3-642-41997-3 .
<img src="https://de.wikipedia.org/wiki/Special:CentralAutoLogin/start?type=1x1" alt="" title="" width="1" height="1" style="border: none; position: absolute;">