&2 Simple Comparative Experiments

 Statistical Plots

 Sampling and Sampling Distributions

 Hypothesis Testing

 Confidence Interval

&Two DOE Class 90a 1

~

點圖 (Dot Diagram)

&Two DOE Class 90a 2

 直方圖 (Histogram)

&Two DOE Class 90a 3

 盒形圖 (Box Plot)

&Two DOE Class 90a 4

 時間序列圖 (Time Series Plot)

&Two DOE Class 90a 5

 期望值與變異數之公式
 母體平均數(m ) = 隨機變數之期望值 E(X)
 母體變異數(s 2) = 隨機變數之變異數 V(X)

  xf ( x)dx, continuous
 
E( X )  
x xp(x), discrete
 All


V ( X )  E x  m 
2

&Two DOE Class 90a 6
 期望值與變異數之公式
E (ax1  bx2 )  aE( x1 )  bE( x2 )
V (ax1 )  a 2V ( x1 )
V ( x1  x2 )  V ( x1 )  V ( x2 )  2Cov( x1 , x2 ),
其中 Cov( x1 , x2 )  E  x1  m1  x2  m 2 
若x1 , x2 獨立 , 則Cov( x1 , x2 )  0, 且V ( x1  x2 )  V ( x1 )  V ( x2 )
E ( x1  x2 )  E ( x1 )  E ( x2 )
x1 E ( x1 )
E( )
x2 E ( x2 )
&Two DOE Class 90a 7
 Sample and Sampling

&Two DOE Class 90a 8

 點估計(Point Estimation)
 以抽樣得來之樣本資料, 依循某一公式計算出單一數值,
來估計母體參數, 稱為點估計.
 好的點估計公式之條件:
 不偏性
 最小變異
 常用之點估計:
 母體平均數(m)
X 
 X i

 X 
n
2
母體變異數(s ) i  X
 2

S2  i 1

n 1
&Two DOE Class 90a 9
 Central Limit Theorem

&Two DOE Class 90a 10

 假設檢定(Hypothesis Testing)
 “A person is innocent until proven guilty beyond a
reasonable doubt.” 在沒有充分證據證明其犯罪之前,
任何人皆是清白的.
 假設檢定
H0: m = 50 cm/s
H1: m  50 cm/s
 Null Hypothesis (H0) Vs. Alternative Hypothesis (H1)
 One-sided and two-sided Hypotheses
 A statistical hypothesis is a statement about the
parameters of one or more populations.

&Two DOE Class 90a 11

 About Testing
 Critical Region
 Acceptance Region
 Critical Values

&Two DOE Class 90a 12

 Errors in Hypothesis Testing
 檢定結果可能為

 Type I Error(a): Reject H0 while H0 is true.

 Type II Error(b): Fail to reject H0 while H0 is false.

&Two DOE Class 90a 13

  H 0 : 無辜( Innocent)

 H1 : 有罪(Guilty)

The Jury finds the The Defendant is

person Innocent Guilty
Innocent Type II Error
Guilty Type I Error

&Two DOE Class 90a 14

 Making Conclusions
 We always know the risk of rejecting H0, i.e., a, the
significant level or the risk.
 We therefore do not know the probability of
committing a type II error (b).

 Two ways of making conclusion:

1. Reject H0
2. Fail to reject H0, (Do not say accept H0)
or there is not enough evidence to reject H0.

&Two DOE Class 90a 15

 Significant Level (a)
 a = P(type I error) = P(reject H0 while H0 is true)

n = 10, s = 2.5
s/n = 0.79

&Two DOE Class 90a 16

&Two DOE Class 90a 17
&Two DOE Class 90a 18
&Two DOE Class 90a 19
&Two DOE Class 90a 20
 The Power of a Statistical Test

 Power = 1 - b
 Power = the sensitivity of a statistical test

&Two DOE Class 90a 21

 General Procedure for Hypothesis Testing
1. From the problem context, identify the parameter of
interest.
2. State the null hypothesis, H0.
3. Specify an appropriate alternative hypothesis, H1.
4. Choose a significance level a.
5. State an appropriate test statistic.
6. State the rejection region for the statistic.
7. Compute any necessary sample quantities, substitute
these into the equation for the test statistic, and compute
that value.
8. Decide whether or not H0 should be rejected and report
that in the problem context.

&Two DOE Class 90a 22

 Inference on the Mean of a Population
-Variance Known

 H0: m = m0
H1: m  m0 , where m0 is a specified constant.
 Sample mean is the unbiased point estimator for population
mean.

If X 1 , X 2 ,, X n are samples drawn from a distribution


with mean m and variance s 2 , then X ~ N m , s
2

n
.
Therefore, if H 0 is true ( m  m 0 ), then
X  m0
Z0  ~ N 0,1
s n

&Two DOE Class 90a 23

 Example 8-2
Aircrew escape systems are powered by a solid
propellant. The burning rate of this propellant is an
important product characteristic. Specifications require
that the mean burning rate must be 50 cm/s. We know
that the standard deviation of burning rate is 2 cm/s.
The experimenter decides to specify a type I error
probability or significance level of α = 0.05. He selects
a random sample of n = 25 and obtains a sample average
of the burning rate of x = 51.3 cm/s. What conclusions
should be drawn?

&Two DOE Class 90a 24

 1. The parameter of interest is m, the meaning burning rate.
2. H0: m = 50 cm/s
3. H1: m  50 cm/s
4. a = 0.05
5. The test statistics is:
x  m0
Z0 
s/ n
6. Reject H0 if Z0 > 1.96 or Z0 < -1.96 (because Za/2 = Z0.025 = 1.96)
7. Computations:
51.3  50
Z0   3.25
2 / 25
8. Conclusions: Since Z0 = 3.25 > 1.96, we reject H0: m = 50 at the 0.05
level of significance. We conclude that the mean burning rate differs
from 50 cm/s, based on a sample of 25 measurements. In fact, there is
string evidence that the mean burning rate exceeds 50 cm/s.
&Two DOE Class 90a 25
 P-Values in Hypothesis Tests

 Where Z0 is the test statistic, and (z) is the standard normal

cumulative function.

&Two DOE Class 90a 26

 The Sample Size (I)
 Given values of a and d, find the required sample
size n to achieve a particular level of b..
 d   d 
Since b   Za / 2     
 a /2
Z  
 s/ n  s/ n
 d 
  Za / 2   when d  0
 s/ n
Let b    Z b 
d
Then,  Z b  Za / 2 
s/ n

n
Za / 2  Z b  s 2
2

whe re d  m  m 0
d2
&Two DOE Class 90a 27
 The Operating Characteristic Curves
- Normal test (z-test)

 Use to performing sample size or type II error

calculations.
 The parameter d is defined as:
| m  m0 | |d |
d 
s s
so that it can be used for all problems regardless of
the values of m0 and s.
 課本41頁之公式為兩平均數差之假設檢定所需之樣本
數公式。

&Two DOE Class 90a 28

&Two DOE Class 90a 29
 Construction of the C.I.
 From Central Limit Theory,

If X ~ m , s  and n  25, X ~ N m ,
2 s2

n
. 
 Use standardization and the properties of Z,
X m
Z  and P  za / 2  Z  za / 2   1  a
s n
 X m 

 P  za / 2   za / 2   1  a
 s n 
 
 P X  za / 2s / n  m  X  za / 2s / n  1  a

&Two DOE Class 90a 30

 Inference on the Mean of a Population
-Variance Unknown
 Let X1, X2, …, Xn be a random sample for a normal
distribution with unknown mean m and unknown
variance s2. The quantity

has a t distribution with n - 1 degrees of freedom.

&Two DOE Class 90a 31

 Inference on the Mean of a Population
-Variance Unknown
 H0: m = m0
H1: m  m0 , where m0 is a specified constant.
 Variance unknown, therefore, use s instead of s in
the test statistic.

 If n is large enough ( 30), we can use Z-test.

However, n is usually small. In this case, T0 will not
follow the standard normal distribution.

&Two DOE Class 90a 32

 Inference for the Difference in Means
-Two Normal Distributions and Variance Unknown

 Why?

&Two DOE Class 90a 33

&Two DOE Class 90a 34
 Re placing σ 12 and σ 22 with S12 and S22 , we still have

is distributed approximately as t with degrees of freedom

given by

&Two DOE Class 90a 35

 C.I. on the Difference in Means

&Two DOE Class 90a 36

 C.I. on the Difference in Means

&Two DOE Class 90a 37

 Paired t-Test

 When the observations on the two populations of

interest are collected in pairs.

 Let (X11, X21), (X12, X22), …, (X1n, X2n) be a set of n

paired observations, in which X1j~(m1, s12) and
X2j~(m2, s22) and Dj = X1j – X2j, j = 1, 2, …, n. Then,
to test H0: m1= m2 is the same as performing a one-
sample t-test H0: mD = 0 since
mD = E(X1-X2) = E(X1)-E(X2) = m1 - m2

&Two DOE Class 90a 38

&Two DOE Class 90a 39
 Inference on the Variance of a Normal
Population (I)
 H0: s2 = s02
H1: s2  s02 , where s02 is a specified constant.
 Sampling from a normal distribution with unknown
mean m and unknown variance s2, the quantity

2 
n  1S 2

s2
has a Chi-square distribution with n-1 degrees of
freedom. That is,
 
2 n  1S 2
~  2
n 1
2
s

&Two DOE Class 90a 40

 Inference on the Variance of a Normal
Population (II)
 Let X1, X2, …, Xn be a random sample for a normal
distribution with unknown mean m and unknown
variance s2. To test the hypothesis
H0: s2 = s02
H1: s2  s02 , where s02 is a specified constant.
We use the statistic

 If H0 is true, then the statistic has a chi-square

distribution with n-1 d.f..

&Two DOE Class 90a 41

 PDF of chi - square distribution :

f x  
1  k / 2 1  x / 2
x e x0
2 k / 2 
k /2

k is the number of degrees of freedom.

In addition,
mk
s 2  2k

&Two DOE Class 90a 42

 The Reasoning
 For H0 to be true, the value of 02 can not be too
large or too small.

 What values of 02 should we reject H0? (based on a

value)
What values of 02 should we conclude that there is
not enough evidence to reject H0?

&Two DOE Class 90a 43

&Two DOE Class 90a 44
 Example 8-11
An automatic filling machine is used to fill bottles
with liquid detergent. A random sample of 20 bottles
results in a sample variance of fill volume of s2 =
0.0153 (fluid ounces)2. If the variance of fill volume
exceeds 0.01 (fluid ounces)2, an unacceptable
proportion of bottles will be underfilled and overfilled.
Is there evidence in the sample data to suggest that
the manufacturer has a problem with underfilled and
overfilled bottles? Use a = 0.05, and assume that fill
volume has a normal distribution.

&Two DOE Class 90a 45

 1. The parameter of interest is the population variance s2.
2. H0: s2 = 0.01
3. H1: s2  0.01
4. a = 0.05
5. The test statistics is 2 n
0 
 1 s 2
 
s 02
6. Reject H0 if  02   02.05,19  30.14
7. Computations:
190.0153
 02   29.07
0.01
8. Conclusions: Since  0  29.07   0.05,19  30.14 , we conclude that
2 2

there is no strong evidence that the variance of fill volume exceeds

0.01 (fluid ounces)2.

&Two DOE Class 90a 46

 Hypothesis Testing on Variance
- Normal Population

H1 Test Statistic Reject H0 if

s2  s02  02   a2 / 2 , n 1 or  02   12a / 2 , n 1

 
n  1S 2
 02   a2 , n 1
2
s2 > s02 0
s02

s2 < s02  02   12a , n 1

&Two DOE Class 90a 47

 The Test Procedure for Two Variances
Comparison

&Two DOE Class 90a 48

&Two DOE Class 90a 49
&Two DOE Class 90a 50
&Two DOE Class 90a 51
 Hypothesis Testing on the Ratio of Two
Variances

&Two DOE Class 90a 52