Scribd
Upload
226 views10 pages

Convergence and Large Sample Approximations - Part 1 Definition: Convergence in Distribution

< ww < ∞. Then WWnn does not converge. 1) The document defines convergence in distribution and convergence in probability for sequences of random variables. Convergence in distribution means the CDFs converge at all points of continuity, while convergence in probability means the probability of the variables being far apart goes to 0. 2) Several examples are provided to illustrate the definitions and differences between the types of convergence. Example 1 shows a sequence converging in distribution to a constant, while Example 3 shows convergence in distribution but not probability. 3) Large sample approximations are discussed, along with conditions for convergence. The document provides examples applying the definitions and theorems to establish various convergence results for sequences of random variables

Uploaded by

d
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
226 views10 pages

Convergence and Large Sample Approximations - Part 1 Definition: Convergence in Distribution

< ww < ∞. Then WWnn does not converge. 1) The document defines convergence in distribution and convergence in probability for sequences of random variables. Convergence in distribution means the CDFs converge at all points of continuity, while convergence in probability means the probability of the variables being far apart goes to 0. 2) Several examples are provided to illustrate the definitions and differences between the types of convergence. Example 1 shows a sequence converging in distribution to a constant, while Example 3 shows convergence in distribution but not probability. 3) Large sample approximations are discussed, along with conditions for convergence. The document provides examples applying the definitions and theorems to establish various convergence results for sequences of random variables

Uploaded by

d
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/ 10

STAT 410

Convergence and Large Sample


Approximations Part 1

Fall 2016

Definition: Convergence in Distribution


Let { } be a sequence of random variables and let be a random variable. Let

and be, respectively, the cdfs of and . Let ( ) denote the set of all

points where is continuous. We say that converges in distribution to if


lim () = () , ( )

We denote this convergence by

Definition: Convergence in Probability

Let 1 , 2 , be an infinite sequence of random variables, and let be another


random variable. Then the sequence { } converges in probability to , if for all
> 0,
lim P(| | ) = 0, lim P(| | < ) = 1

and write .

Theorem 1

Theorem 2

, constant

Example 1 Consider a sequence of discrete random variables where


( = 0) =

3
1
1
and = = , = 1,2,3,
4
4

0, < 0
1
,
0

<
For each n the cdf is () = 4

1
1,
1

STAT 410

We see that

Convergence and Large Sample


Approximations Part 1

Fall 2016

< 0 implies () 0 as
> 0 implies () 1 as

Therefore () () =

0, < 0
for every 0, i.e. for every point of
1, 0

continuity for F. This is the cdf of the constant 0, i.e. ( = 0) = 1. Thus 0.

By Theorem 2 we would conclude that 0 as well. For a direct proof note that
for any fixed > 0 we have

3
,
4
(| | ) =
0,

1
<

So the probability = 0 eventually, for all > . Since > 0 was arbitrary, we

conclude by definition that 0.

STAT 410

Fall 2016

Convergence and Large Sample


Approximations Part 1
1

Example 2. Let = + where Y and Z are independent (0,1) random

variables. It follows that ~(0, 1 +

). To establish its limiting distribution

consider the limiting moment generating function:


1
1
1+ 2 2

() = 2

1 2

2 = () as

By a result stated below (Theorem 7) this implies that .


We could have also proven this directly via the cdf. Note that

() =

1 +

for all t. Hence, by definition .

() = ()

1
2

In fact we can show a stronger convergence. For any > 0 we have


(| | ) =

= (|| )

= 21 () 0

Therefore by definition we have that , which also imples .

STAT 410

Fall 2016

Convergence and Large Sample


Approximations Part 1

Example 3. Heres an example where convergence in distribution holds, but not

convergence in probability: Let = for all n where ~(0,1). Then

but (| | ) = (|| ) = || = 2 1 > 0 for all n.


2
2
It follows that does not converge in probability to Z.

Example 4.
Let 1 , 2 , be i.i.d. Uniform(0, ). Let = max(1 , 2 , , ).

First show that . This follows because, given any > 0 and less than ,

(| | ) = ( ) =
,

which converges to 0 as n increases, because |( )/| < 1.


Next find the limiting distribution of = ( ).


() = max () = , 0 < < .

() = [( ) ] = > = 1 1 , 0 < < .

() 1 , > 0, .

, where ~().

STAT 410

Fall 2016

Convergence and Large Sample


Approximations Part 1

Example 5. Let 1 , , be a random sample from the distribution with


probability density function
1 1
(;
) = , 0 < < 1,0 < <

Let 1 < 2 < denote the corresponding order statistics.


a)

For which values of does = (1 ) converge in distribution? Find

the limiting distribution of .

() = ( ) = , 0 < < 1

() = (1 ) = 1
= 1 1

/
, 0 < < .

If = 1, lim () = 1 , 0 < < , Then ~().

If < 1, lim () = 1 , 0 < < ,


If > 1, lim () = 0 , 0 < < ,
b)

Then does not have a limiting


distribution.

For which values of does = 1 converge in distribution? Find the

limiting distribution of .

1 () = (1 ) = 1 1

() = 1
If = ,

0 < < ,

1

1

, 0 < < 1

= 1 1 , 0 < < .

lim () = 1

1/

Then 0 and thus 0.

Then ~().

STAT 410

If < ,
If > ,

Convergence and Large Sample


Approximations Part 1

Fall 2016

lim () = 1 , 0 < < , Then


0, and thus 0.

lim () = 0 , 0 < < , Then does not have a limiting


distribution.

, is continuous on the support of

Theorem 3

( ) ()

Theorem 4

, 0 +

Theorem 5

Slutskys Theorem

, ,

+ +

() () for || < .

Theorem 6

Example 6. Let ~ , = . Find the limiting distribution of .

Let ~ , = . Then

() =

() = 1 + 1 as .

, where ~() (Poisson approximation to

Binomial distribution ).

STAT 410

Fall 2016

Convergence and Large Sample


Approximations Part 1

Example 7. Let ~ 2 (). Recall ( ) = and ( ) = 2.


a)

Let = /. Find the limiting distribution of .

Let ~ 2 () and = /. Then,


() =

= = 1 2 as .

Note () = , where ( = 1) = 1 1 1.
b)

Let = ( )/2. Find the limiting distribution of .


() =

2 2

1 2

, < .
2

By Taylor approximation,

2
1
1
= 1 + + 2 + .

So for < ,
2

STAT 410

Fall 2016

Convergence and Large Sample


Approximations Part 1

2
2
1
1

() = 1 + + 2 + 1

= 1
=

1 2

1
+

1
2

2
1
1 +
1

1 2
2

1 2

= 2

as

As , () 2 = (), where ~(0,1) .

STAT 410

Convergence and Large Sample


Approximations Part 1

Fall 2016

Distribution-free convergence of sample averages


Weak Law of Large Numbers
1 , 2 , , are i.i.d. with mean and variance 2 . Then

1
=

=1

Proof: For every fixed > 0 we have, using Markovs inequality,


(| | > ) = (( )2 > 2 )

2
( )2
= 20

as . Therefore by definition of convergence of probability.

Example 8. Let 1 , 2 , , are i.i.d. with mean and variance 2 and finite
fourth moment 4 = ( 4 ). Then, by the weak law of large numbers,

1
2 (12 ) = 2 + 2

=1

Furthermore, using our previous results we can show convergence of the sample
variance:

1
2 =
( )2
1
=

=1

2 2 (1){( 2 + 2 ) 2 } = 2

1
=1

Example 9. Let 1 , , be iid (0,1). Show the following:


9

STAT 410

Convergence and Large Sample


Approximations Part 1

1
a.

b.

c.

d.

e.

f.

2
1 2 1


=1
2
12
2

=1

=1 ln( ) 1

=1

+1
1

>
=1
2
2

Central Limit Theorem


1 , 2 , , are i.i.d. with mean and variance 2 .

( ) (=1 )
=
~(0,1).

10

Fall 2016

You might also like