Download Sampling distributions - Laboratory for Remote Sensing Hydrology

Document related concepts
no text concepts found
Transcript
STATISTICS
Sampling and Sampling
Distributions
Professor Ke-Sheng Cheng
Department of Bioenvironmental Systems Engineering
National Taiwan University
Random sample
• Let the random variables X1, X2, …, Xn have a
joint density f X1 , X 2 ,, X n (,, ,) that factors as
follows:
f X1 , X 2 ,, X n ( x1 , x 2 , x n )  f ( x1 ) f ( x 2 ) f ( x n )
where f () is the common density of each Xi .
Then (X1, X2, …, Xn) is defined to be a random
sample of size n from a population with
density f () .
• If X1, X2, …, Xn is a random sample of size n
from f () , then X1, X2, …, Xn are stochastically
independent.
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
2
Statistic
• A statistic is a function of observable random
variables, which is itself an observable random
variable and does not contain any unknown
parameters.
• A statistic must be observable because we
intend to use it to make inferences about the
density functions of the random variables.
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
3
• For example, if a random variable has a
probability density function N (,  2 ) where  and
 are unknown, then  X   is not a statistic.
• If a statistic is not observable, then it can not be
used to inference the parameters of the density
function.
n
i 1
5/9/2017
2
i
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
4
• An observation of random sample of size n can
be regarded as n independent observations of a
random variable.
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
5
• One of the central problems in statistics is to
find suitable statistics to represent parameters
of the probability distribution function of a
random variable.
Sample {x1 ,, xn }
Population N (  , 2 )
Statistics ( x , s 2 )
2
Parameters (  ,  )
Observable
5/9/2017
Unknown
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
6
Sample moments
• Let X1, X2, …, Xn be a random sample from the
density f () . Then the rth sample moment about
0 is defined as
n
1
r
'
Mr   Xi
n i 1
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
7
• In particular, if r = 1, we have the sample
mean X n ; that is,
1 n
Xn   Xi
n i 1
• Also, the rth sample moment about the sample
mean is defined as
1 n
r
Mr   (Xi  Xn)
n i 1
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
8
• Theorem – Let X1, X2, …, Xn be a random
sample from the density f () . The expected
value of the rth sample moment about 0 is equal
'
'
th
to the r population moment; i.e., E[ M r ]   r
Also,
Var[ M r' ]
1
1 '
2r
r 2
 {E[ X ]  ( E[ X ]) }  [  2 r  (  r' ) 2 ]
n
n
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
9
• Special case: r=1
1
2
2
Var[ X ]  {E[ X ]  ( E[ X ]) }
n
1 '
Var
(
X
)
' 2
2
 [  2  ( 1 ) ] 
X /n
n
n
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
10
Sample statistics
• Let X1, X2, …, Xn be a random sample from the
distribution of a random variable X. Sample
mean and sample variance of the distribution are
respectively defined to be
n
1
X   Xi
n i 1
5/9/2017
n
1
2
2
S 
(Xi  X )

n  1 i 1
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
11
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
12
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
13
Estimating the mean
• Given a random sample x1 , x2 , xn from a probability
density function f(.) with unknown mean μ and finite
variance σ2, we want to estimate the mean using the
random sample.
• Using only a finite number of values of X (a random
sample of size n), can any reliable inferences be made
about E(X), the average of an infinite number of values
of X?
• Will the estimate be more reliable if the size of the
random sample is larger?
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
14
R-program demonstration
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
15
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
16
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
17
Standard deviation of sample means w.r.t. sample size
5
4.5
4
3.5
What is the theoretical basis?
3
y = 19.938x-0.4998
Y=f(x)=?
R = 0.9995
2.5
2
2
1.5
1
0.5
0
0
5/9/2017
1000
2000
3000
4000
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
5000
18
Histograms of sample mean and sample standard deviation
ns=30
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
19
Histograms of sample mean and sample standard deviation
ns=5000
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
20
Weak Law of Large Numbers
(WLLN)
• Let f(.) be a density with mean μ and variance
σ2, and let X n be the sample mean of a random
sample of size n from f(.). Let ε and δ be any
two specified numbers satisfying ε>0 and 0<δ<1.
2

If n is any integer greater than
, then
2
 
P[  X n     ]  1  
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
21
Recall the theorem
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
22
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
23
• (Example 1) Suppose that some distribution
with an unknown mean has its variance equal
to 1. How large a random sample must be
taken such that the probability will be at least
0.95 that the sample mean X n will lie within
0.5 of the population mean?
  1   0.5
2
  1  0.95  0.05
1
n
 80
2
(0.05)(0.5)
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
24
(Example 2) How large a random sample must
be taken in order that you are at least 99%
certain that X n is within 0.5σ from μ?
  0.5
  1  0.992  0.01
n

(0.01)(0.5 )
2
 400
What if we know in advance that the random sample
is to be drawn from a normal distribution?
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
25
A much smaller sample size is required if the distribution
is known in advance.
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
26
The Central Limit Theorem
• Let f(.) be a density with mean μ and finite
variance σ2. Let X n be the sample mean of a
random sample of size n from f(.). Then
Zn 
Xn  

n
approaches the standard normal distribution as
n approaches infinity.
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
27
• The importance of the CLT is the fact that the
mean X n of a random sample from any
distribution with finite variance σ2 and mean μ is
approximately distributed as a normal2 random
variable with mean μ and variance  n .






X n    Zn 
 ~ N  ,

n
n


5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
28
R-program demonstration
- Central Limit Theorem
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
29
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
30
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
31
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
32
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
33
Sampling distributions
• Given random samples of certain probability densities,
we often are interested in knowing the probability
densities of sampling statistics.
–
–
–
–
–
–
Poisson distribution
Exponential distribution
Normal distribution
Chi-square distribution
Standard normal and chi-square distributions
Student’s t-distribution
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
34
Poisson distribution
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
35
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
36
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
37
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
38
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
39
Exponential distribution
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
40
Chi-squared distribution
1
 x
f X ( x; k ) 
 
2(k / 2)  2 
( k / 2 ) 1
e  x / 2 I[ 0, ) ( x) , k  1,2, .
E[ X ]  k Var[ X ]  2k
m X (t )  (1  2t )  k / 2 for t  1 / 2.
• The chi-squared distribution is a special
case of the gamma distribution with
  k / 2 and   2.
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
41
Normal distribution
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
42
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
43
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
44
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
45
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
46
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
47
The sample mean and sample
standard deviation are independently
distributed. [Only valid for the
normal distribution.]
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
48
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
49
Chi-square distribution
F distribution with degrees of freedom m and n
[( m  n) / 2]  m 
f X ( x) 
 
(m / 2)(n / 2)  n 
5/9/2017
m/2
x ( m2) / 2
I
( x)
( m  n ) / 2 ( 0 , )
[1  (m / n) x]
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
50
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
51
Standard normal and chi-square
distributions
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
52
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
53
Student’s t-distribution
Student’s t distribution with k degrees of freedom
As the number of degrees of freedom increases, the
Student’s t distribution approaches the standard normal
distribution.
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
54
• The "student's" distribution was published in
1908 by W. S. Gosset. Gosset, however, was
employed at a brewery that forbade the
publication of research by its staff members. To
circumvent this restriction, Gosset used the
name "Student", and consequently the
distribution was named "Student t-distribution.
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
55
Order statistics
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
56
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
57
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
58
Extremal Types Theorem
• Reference source
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
59
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
60
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
61
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
62
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
63
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
64
Gumbel Distribution
(Extreme Value Type I)
5/9/2017
Laboratory for Remote Sensing Hydrology and Spatial Modeling,
Dept of Bioenvironmental Systems Engineering, National Taiwan Univ.
65
Related documents