Scribd
Upload
33 views1 page

Notes Confidence Intervalsintercvacnfidence

The document discusses symmetric confidence intervals for parameters based on data. It provides three examples: 1) A confidence interval for the mean of bounded random variables using sample average and Hoeffding's inequality. 2) A confidence interval derived from an asymptotically normal estimator, using a data-based variance estimator. 3) A t-distribution based confidence interval for the mean of a normal distribution using sample mean and variance. For large samples, it can be approximated by a normal-based interval.

Uploaded by

Itha Oppa'minoz Cii'befoosee
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
33 views1 page

Notes Confidence Intervalsintercvacnfidence

The document discusses symmetric confidence intervals for parameters based on data. It provides three examples: 1) A confidence interval for the mean of bounded random variables using sample average and Hoeffding's inequality. 2) A confidence interval derived from an asymptotically normal estimator, using a data-based variance estimator. 3) A t-distribution based confidence interval for the mean of a normal distribution using sample mean and variance. For large samples, it can be approximated by a normal-based interval.

Uploaded by

Itha Oppa'minoz Cii'befoosee
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 1

E0 232 Probability and Statistics Oct 30, 2012

Brief Note on Symmetric Condence Intervals


Shivani Agarwal
Recall that a (1 ) condence interval for a parameter based on data X
1
, . . . , X
n
is a (random) interval
[

n
,

+
n
], where

n
= g

(X
1
, . . . , X
n
) and

+
n
= g
+
(X
1
, . . . , X
n
) are some functions of the data satisfying
P

+
n

1 .
Often, it is natural to construct a symmetric condence interval [

n
,

n
+] around some estimator

n
.
This reduces the problem to nding > 0 such that
P

.
Example 1 (Condence interval for mean of a bounded random variable). Suppose X
1
, . . . , X
n
are iid
random variables in some bounded range [a, b] with unknown mean . Then a condence interval for using
the sample average estimator

M
n
=
1
n

n
i=1
X
i
can be obtained using Hoedings inequality. In particular,
taking = (b a)

1
2n
ln(
2

) yields a (1 ) condence interval [

M
n
,

M
n
+] for .
Example 2 (Condence interval derived from an asymptotically normal estimator). Suppose X
1
, . . . , X
n
are
iid random variables whose distribution is governed by some unknown parameter , and

n
= g(X
1
, . . . , X
n
)
is an asymptotically normal estimator of , i.e. that

Var

n
)
D
N(0, 1) .
Unless Var

n
) is known, this does not directly yield a condence interval for . However, if for some
data-based estimator

V
n
= h(X
1
, . . . , X
n
) of Var

n
) we have

V
n
D
N(0, 1) ,
then for large n, this can be used to obtain a normal-based (approximate) condence interval for . In
particular, taking =

V
n
z
/2
, where z
/2
=
1
(1

2
), yields a (1) condence interval [

n
,

n
+]
for .
Example 3 (Condence interval for mean of a normal random variable). Suppose X
1
, . . . , X
n
are iid normal
random variables with unknown mean and unknown variance
2
, and let

M
n
=
1
n

n
i=1
X
i
and

S
2
n
=
1
n1

n
i=1
(X
i


M
n
)
2
. Then

M
n

S
n

n
t
n1
,
where t
n1
denotes the t distribution with n 1 degrees of freedom, and whose CDF
n1
is available in
tables similarly to the standard normal CDF tables. In this case, one can obtain a t distribution based
condence interval for ; in particular, taking =

S
n

n
z

/2
, where z

/2
=
1
n1
(1

2
), yields a (1 )
condence interval [

M
n
,

M
n
+ ] for . For large n, t
n1
is well-approximated by the standard normal
N(0, 1), and so one can again fall back on a normal-based approximation, but for small n, the t
n1
-based
condence interval should be used.
1

You might also like