Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture...
Transcript of Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture...
![Page 1: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/1.jpg)
1
Lecture 17
! The Chi-Square Distribution
! Joint Distribution of the Sample Mean and Sample Variance
! The t Distribution
![Page 2: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/2.jpg)
22
小小的复习:
很多时候,我们通过样本来了解总体。
总体样本
![Page 3: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/3.jpg)
33
基本概念
n 总体与样本n 总体:所有个体的全体
n 样本:观测到的个体
![Page 4: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/4.jpg)
4
为什么需要抽样?
1)总体无法得到。例:光临麦当劳的所有顾客(无限总体)。
2)时间和成本不允许。例:美国总统选举的民意测验。
3)实验具有破坏性。例:测量产品的寿命。
![Page 5: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/5.jpg)
5
统计分析的任务: 通过样本的统计量来了解总体的参数。
n 参数与统计量n 参数:关于总体的度量,如:µ,p,sn 统计量:关于样本的度量,如: , , x p s
样本统计量 总体参数
![Page 6: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/6.jpg)
6
参数的点估计
( )
1
22 2
1
1
1
1 1
1 , =1 0.
n
ii
n
ii
n
i ii
x xn
s x xn
p x p xn
µ
s
=
=
=
= ®
= - ®-
= ®
å
å
å 这里 或
![Page 7: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/7.jpg)
7
考虑到所有可能的样本...
1
2
3
4
M
xxx xx
xL
对不同
的样本,
取值
通常也
不同
![Page 8: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/8.jpg)
8
Estimation Error
n 如:样本均值
( ) ( )
( ) ( )
2 2
2
/
E( ) /
MSE x E x n
SE x E x x n
µ s
s
= - =
= - =
( ) ( )2( ) ( )MSE x E xd d q= -
![Page 9: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/9.jpg)
9
The Chi-square Distribution
! For any given positive integer n, the distribution with n degrees of freedom.
where G is the gamma function defined as
2c
/2 1 /2n/2
1( ) 02 ( / 2)
n xf x x e xn
- -= >G
( ) ( ) 2E X n Var X n= =
0for)(0
1 >=G ò¥ -- aa a dxex x
![Page 10: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/10.jpg)
10
n=2
n=3
n=5n=10
![Page 11: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/11.jpg)
11
Theorem
! If the random variables X1,…,Xk areindependent and if Xi has a distributionwith ni degrees of freedom (i=1,…,k), then thesum X1+…+Xk has a distribution withn1+…+nk degrees of freedom.
2c
2c
![Page 12: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/12.jpg)
12
Relation of the distribution with the Normal Distribution
! If a random variable X has a standard normal distribution, then the random variablewill have a distribution with one degree of freedom.
2c
2XY =
2c
![Page 13: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/13.jpg)
13
Relation of the distribution with the Normal Distribution
! If a random variable X has a standard normal distribution, then the random variablewill have a distribution with one degree of freedom.Proof. For any y>0,
2c
2XY =
21
221
2
21)(
21)()(
21)(
21)()()(
)()()()()(
cp
pff
ff
FF
~eyyf
eyy
yy
yyy'Fyf
yyyXyPryYPryF
/y/
/y
--
-
=Þ
=-=
-+==
--=££-=£=
2c
![Page 14: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/14.jpg)
14
Theorem
! If the random variables X1,…,Xn are i.i.d., and if each has a standard normal distribution, then the sum of squareshas a distribution with n degrees of freedom.
2c
2 21 nX X+ +L
![Page 15: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/15.jpg)
15
Example. Acid Concentration in Cheese
! The variation in concentrations of chemicals like lactic acid can lead to variation in the taste of cheese. Suppose that we model the concentration of lactic acid in several chunks of cheese as independent normal random variables with mean µ and variance .
! Let X1,…,Xn be the concentrations in n chunks, and let Zi=(Xi -µ)/s, then
is one measure of how much the n concentrations differ from µ. Suppose that a difference of u or more in lactic acid concentration is enough to cause a noticeable difference in taste. We wish to calculate
2s
åå==
=-=n
ii
n
ii Z
nX
nY
1
22
1
2||1 sµ
)Pr( 2uY £
![Page 16: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/16.jpg)
16
! Suppose , and n=10, u=0.3,09.02 =s
2
1
22 ~/ n
n
iiZnYW cs å
=
==
56.0)10Pr(09.03.010Pr)3.0Pr(2
2 =£=÷÷ø
öççè
æ ´£=£ WWY
![Page 17: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/17.jpg)
17
Joint Distribution of the Sample Mean and Sample Variances
! Theorem. Suppose that X1,…,Xn form a random sample from a normal distribution with mean µ and variance . Then the sample mean and the sample variance
are independent random variables, and
2s
nX( ) ( ) Unbiasedfrom
11or MLE from 1
1
2
1
2 åå ==-
--
n
i nin
i ni XXn
XXn
( ) 21
21
2
2
-=å -
÷÷ø
öççè
æ
nn
i ni
n
~/XX
n,N~X
cs
sµ
![Page 18: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/18.jpg)
18
The t Distribution! Consider two independent random variables
Y and Z, such that Z has a standard normal distribution and Y has a distribution with ndegrees of freedom. Suppose a random variable X is defined by
Then the distribution of X is called the t distribution with n degrees of freedom.
nYZX/
=
2c
![Page 19: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/19.jpg)
19
The p.d.f.! Let W=Y, then
! Since
WYnWXZ ==
¥<<¥-÷÷ø
öççè
æ+
÷øö
çèæG
÷øö
çèæ +
G==
úû
ùêë
é÷÷ø
öççè
æ+-
G=
•÷÷ø
öççè
æ-
G=
÷÷ø
öççè
æ-
G==
+-¥
-++
--+
--
ò xnx
nn
n
dwwxfxg
wnxw
nn
nww
nxew
nwxf
zeyn
zfyfzyf
n
nn
wnn
ynn
2/)1(2
0
212/)1(
2/)1(
22/12/
2/)1(
22/12/
2/
1
2
21
),()(
121exp
)2/(21
21exp
)2/(21),(
2exp
21
)2/(21)()(),(
p
p
p
p
![Page 20: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/20.jpg)
20
Relation to the Normal Distribution
! When , g(x) converges to the p.d.f. f(x) of the standard normal distribution for every value of x.
! When n is large, the t distribution with n degrees of freedom can be approximated by the standard normal distribution.
¥®n
![Page 21: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/21.jpg)
21
n=1 (Cauchy)n=2n=5n=10
¥=n (normal)
![Page 22: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/22.jpg)
22
Mean and Variance of the t Distribution
! The mean does not exist when n=1. It exists and is equal to 0 for any value of n>1.
! If X has a t distribution with n degrees of freedom (n>2), then Var(X)=n/(n-2).
![Page 23: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/23.jpg)
23
Relation to Random Samples from a Normal Distribution
! Suppose that X1,…,Xn form a random sample from a normal distribution with mean µ and variance . Since
and they are independent of each other,
2s
( ) ( )
( ) 21
21
2
2
~/
)10(~/,~
-=å -
-Þ
nn
i ni
nn
XX
,NXnnNX
css
µsµ
( )
( )( )( ) 1
1
221
2
11
-
==
-
-
-=
-
-
-
=åå
nn
i ni
nn
i ni
n
t~
nXX
Xn
n/XX
Xn
U µ
ss
µ
![Page 24: Lecture 17 - PKUmwfy.gsm.pku.edu.cn/miao_files/ProbStat/lecture17.pdf · 2020-04-22 · Lecture 17!The Chi-Square Distribution!Joint Distribution of the Sample Mean and Sample Variance!The](https://reader035.fdocuments.net/reader035/viewer/2022070822/5f28b0c2e95d38765f37aa69/html5/thumbnails/24.jpg)
24
! Define
They are often referred to as sample variance and sample standard deviation.
! When the variance is known,
when it is unknown, we can replace by S,
They can be used to make statistical inferenceabout µ.
( )221
11
ni ni
S X Xn =
= -- å
2s
( ) )10( ,N~Xn n
sµ-
s
( )1~n
n
n Xt
Sµ
-
-