Goodness of Fit and Contingency

Tables

主讲老师：刘子真
 Chi-square Distribution
𝒁𝟏 , 𝒁𝟐 , … , 𝒁 𝒌 are independent variables following
standard normal distribution, then
𝑸 = 𝒁𝟏 𝟐 + 𝒁𝟐 𝟐 + ⋯ + 𝒁𝒌 𝟐 ~𝝌𝟐𝒌 𝒅𝒆𝒈𝒓𝒆𝒆 𝒐𝒇 𝒇𝒓𝒆𝒆𝒅𝒐𝒎 = 𝒌
 Number on die 1 2 3 4 5 6
Observed 23 15 25 18 21 18
Expected 20 20 20 20 20 20

𝑯𝟎 : there is no difference between the observed

and the theoretical distribution.

𝑯𝟏 : there is a difference between the observed and

the theoretical distribution.
 Number on die 1 2 3 4 5 6
Observed 23 15 25 18 21 18
Expected 20 20 20 20 20 20

The measure of goodness of fit:

(𝑶𝒊 − 𝑬𝒊 )𝟐
𝑿𝟐 = ~𝝌𝟐𝒏−𝟏
𝑬𝒊

Degree of Freedom= # cells − # constraints

Number on die 1 2 3 4 5 6
Observed 23 15 25 18 21 18
Expected 20 20 20 20 20 20
(𝑶𝒊 − 𝑬𝒊 )𝟐
𝟎. 𝟒𝟓 𝟏. 𝟐𝟓 𝟏. 𝟐𝟓 𝟎. 𝟐 𝟎. 𝟎𝟓 𝟎. 𝟐
𝑬𝒊

(𝑶𝒊 − 𝑬𝒊 )𝟐
𝑿𝟐 = = 𝟑. 𝟒
𝑬𝒊

Degree of Freedom= 6 − 1= 5
𝝂=𝟓
 (𝑶𝒊 − 𝑬𝒊 )𝟐
𝑿𝟐 = = 𝟑. 𝟒
𝑬𝒊

𝝌𝟐𝟓 𝟓% = 𝟏𝟏. 𝟎𝟕𝟎

Since 𝟑. 𝟒 < 𝟏𝟏. 𝟎𝟕𝟎 , do not reject the null

hypothesis. There is no evidence that the die
is biased.
1. Determine which distribution is likely to be a
good model.

2. Set the significance level, for example, 5%.

3. Estimate the parameter (if necessary).

4. Form your hypothesis.

5. Calculate expected value.

6. Combine any expected values so that none are

less than 5.
7. Find 𝝂 = #𝐜𝐞𝐥𝐥𝐬 𝐚𝐟𝐭𝐞𝐫 𝐜𝐨𝐦𝐛𝐢𝐧𝐞 − #𝐜𝐨𝐧𝐬𝐭𝐫𝐚𝐢𝐧𝐭𝐬

8. Find the critical value of 𝝌𝟐 .

(𝑶𝒊 − 𝑬𝒊 )𝟐
9. Calculate .
𝑬𝒊

10. See if your value is significant.

11. Draw the conclusion and interpret.

 Digit 0 1 2 3 4 5 6 7 8 9
OB 11 8 8 7 8 9 12 9 13 15

Could the digits be from a random number table?

Test at the 0.05 level.

𝑯𝟎 : A discrete uniform distribution is a suitable

model.

𝑯𝟏 : A discrete uniform distribution is not a

suitable model.
 Digit 0 1 2 3 4 5 6 7 8 9
OB 11 8 8 7 8 9 12 9 13 15
EX 10 10 10 10 10 10 10 10 10 10
𝑿𝟐 0.1 0.4 0.4 0.9 0.4 0.1 0.4 0.1 0.9 2.5

𝝂 = 𝟏𝟎 − 𝟏 = 𝟗 𝝌𝟐𝟗 𝟓% = 𝟏𝟔. 𝟗𝟏𝟗

(𝑶𝒊 − 𝑬𝒊 )𝟐
𝑿𝟐 = = 𝟔. 𝟐
𝑬𝒊
Since 𝟔. 𝟐 < 𝟏𝟔. 𝟗𝟏𝟗, do not reject the null hypothesis.
There is no evidence that the digits are not random.
 𝒙 0 1 2 3 4 5 6 7 8
Frequency 12 28 28 17 7 4 2 2 0

The data are thought to be modelled by a binomial

B(10,0.2). Use the table for the binomial cumulative
distribution function to find expected values, and
conduct a test to see if this is a good model. Use a 5%
significance level.
 𝒙 0 1 2 3 4 5 6 7 8
Frequency 12 28 28 17 7 4 2 2 0

𝑯𝟎 : A B(10,0.2) distribution is a suitable model.

𝑯𝟏 : A B(10,0.2) distribution is not a suitable model.

𝒙 0 1 2 3
Probability 0.1074 0.2684 0.3020 0.2013
Expected 10.74 26.84 30.2 20.13
 𝒙 4 5 6 7 8
Probability 0.0881 0.0264 0.0055 0.0008 0.0001
Expected 8.81 2.64 0.55 0.08 0.01

𝒙 0 1 2 3 ≥4
Observed 12 28 28 17 15
Expected 10.74 26.84 30.2 20.13 12.09
(𝑶𝒊 − 𝑬𝒊 )𝟐
𝑬𝒊 0.1478 0.0501 0.1603 0.4867 0.7004

𝝂=𝟓−𝟏=𝟒 𝝌𝟐𝟒 𝟓% = 𝟗. 𝟒𝟖𝟖

 (𝑶𝒊 − 𝑬𝒊 )𝟐
𝑿𝟐 = = 𝟏. 𝟓𝟒𝟓𝟑
𝑬𝒊

Since 𝟏. 𝟓𝟒𝟓𝟑 < 𝟗. 𝟒𝟖𝟖 , do not reject the null

hypothesis. There is no evidence that a B(10,0.2)
distribution is not a suitable model.
 Girls 0 1 2 3 4 5
Observed 13 18 38 20 10 1

A study of the number of girls in families with five

children was done on 100 such families.

a. Test, at the 5% significance level, whether a

B(5,0.5) distribution is a good model.

𝑯𝟎 : A B(5,0.5) distribution is a suitable model.

𝑯𝟏 : A B(5,0.5) distribution is not a suitable model.

 Girls 0 1 2 3 4 5
Observed 13 18 38 20 10 1
Expected 3.12 15.63 31.25 31.25 15.63 3.12

Girls 0 or 1 2 3 4 or 5
Observed 31 38 20 11
Expected 18.75 31.25 31.25 18.75

𝝂=𝟒−𝟏=𝟑 𝝌𝟐𝟑 𝟓% = 𝟕. 𝟖𝟏𝟓

 (𝑶𝒊 − 𝑬𝒊 )𝟐
𝑿𝟐 = = 𝟏𝟔. 𝟕𝟏𝟒
𝑬𝒊

Since 𝟏𝟔. 𝟕𝟏𝟒 > 𝟕. 𝟖𝟏𝟓, reject the null hypothesis.

There is evidence that a B(5,0.5) distribution is not
a suitable model.
b. Test, at the 5% significance level, whether a
binomial distribution is a good model.

𝟏𝟗𝟗
𝒑= = 𝟎. 𝟑𝟗𝟖
𝟏𝟎𝟎 × 𝟓

𝑯𝟎 : A B(100,0.398) distribution is a suitable model.

𝑯𝟏 : A B(100,0.398) distribution is not a suitable

model.
 Girls 0 1 2 3 4 5
Observed 13 18 38 20 10 1
Expected 7.91 26.14 34.56 22.85 7.55 0.99

Girls 0 1 2 3 4 or 5
Observed 13 18 38 20 11
Expected 7.91 26.14 34.56 22.85 8.54

𝝂=𝟓−𝟐=𝟑 𝝌𝟐𝟑 𝟓% = 𝟕. 𝟖𝟏𝟓

 (𝑶𝒊 − 𝑬𝒊 )𝟐
𝑿𝟐 = = 𝟕. 𝟐𝟐
𝑬𝒊

Since 𝟕. 𝟐𝟐 < 𝟕. 𝟖𝟏𝟓 , do not reject the null

hypothesis. There is no evidence that a
B(100,0.398) distribution is not a suitable model.
 𝒙 0 1 2 3 4 5 6 7 8
Frequency 8 9 26 13 7 5 1 1 0

The number of telephone calls arriving at an

exchange in six-minute periods were recorded over a
period of 8 hours. Can these results be modelled by a
Poisson distribution? Test at the 5% significance
level.
𝟏𝟕𝟔
𝝀= = 𝟐. 𝟐
𝟖 × 𝟔𝟎 ÷ 𝟔
𝑯𝟎 : A Po(2.2) distribution is a suitable model.

𝑯𝟏 : A Po(2.2) distribution is not a suitable model.

𝟐. 𝟐𝒏 −𝟐.𝟐
𝑷 𝑿=𝒏 = 𝒆
𝒏!

𝒙 0 1 2 3 4 5 ≥6
E 8.864 19.504 21.448 15.728 8.656 3.808 1.692
𝒙 0 1 2 3 4 ≥𝟓
O 8 19 26 13 7 7
E 8.864 19.504 21.448 15.728 8.656 5.8

𝝂=𝟔−𝟐=𝟒 𝝌𝟐𝟒 𝟓% = 𝟗. 𝟒𝟖𝟖

(𝑶 − 𝑬 ) 𝟐
𝒊 𝒊
𝑿𝟐 = = 𝟐. 𝟏𝟎𝟏𝟔
𝑬𝒊
Since 𝟐. 𝟏𝟎𝟏𝟔 < 𝟗. 𝟒𝟖𝟖 , do not reject the null
hypothesis. There is no evidence that a Po(2.2)
distribution is not a suitable model.
Height 150- 155- 160- 170- 175- 180- 185- 190-
(cm) 154 159 169 174 179 184 189 194
OB 11 8 8 7 8 9 12 9

a. Test at the 5% significance level to see if the height

could be modelled by a normal distribution with mean 172
and standard deviation 6.

𝑯𝟎 : The N(172,36) distribution is a suitable model.

𝑯𝟏 : The N(172,36) distribution is not a suitable model.

 𝝂=𝟓−𝟏=𝟒

𝝌𝟐𝟒 𝟓% = 𝟗. 𝟒𝟖𝟖

(𝑶𝒊 − 𝑬𝒊 )𝟐
𝑬𝒊
= 𝟏𝟐. 𝟏𝟎𝟕𝟒

Since 𝟏𝟐. 𝟏𝟎𝟕𝟒 > 𝟗. 𝟒𝟖𝟖, reject the null hypothesis.

There is evidence that the N(172,36) distribution is
not a suitable model.
b. Describe how you would modify this test if the
mean and variance were unknown.

1. Estimate the parameter

𝒙 = 𝟏𝟕𝟑. 𝟏𝟓 𝑺𝟐 = 𝟓𝟖. 𝟐𝟐

2. Calculate the degrees of freedom

𝝂 = #𝐜𝐞𝐥𝐥𝐬 𝐚𝐟𝐭𝐞𝐫 𝐜𝐨𝐧𝐦𝐢𝐧𝐠 − 𝟑

Chi-square test of Independence

Observed Employed Unemployed Total

Middle 1153 110 1263
High 3585 206 3791
College 3297 146 3443
Total 8035 462 8497
Chi-square test of Independence

Expected Employed Unemployed Total

Middle 8035∙(1263/8497) 462∙ (1263/8497) 1263
High 8035 ∙ (3791/8497) 462∙ (3791/8497) 3791
College 8035 ∙ (3443/8497) 462∙ (3443/8497) 3443
Total 8035 462 8497
𝑯𝟎 : There is no association between employment and
education.
𝑯𝟏 : There is an association between employment and
education.
Chi-square test of Independence

Expected Employed Unemployed Total

Middle 8035∙(1263/8497) 462∙ (1263/8497) 1263
High 8035 ∙ (3791/8497) 462∙ (3791/8497) 3791
College 8035 ∙ (3443/8497) 462∙ (3443/8497) 3443
Total 8035 462 8497

𝝂 = (𝟑 − 𝟏) × (𝟐 − 𝟏) = 𝟐 𝝌𝟐𝟐 𝟓% = 𝟓. 𝟗𝟗𝟏
(𝑶 − 𝑬)𝟐
𝝌𝟐 = = 𝟑𝟓. 𝟖𝟗𝟐 > 𝟓. 𝟗𝟗𝟏 Reject 𝑯𝟎 .
𝑬