Chapter 8

Analysis of Variance
and Design of
Experiments
University of Economics
Ho Chi Minh City

Dang Van Thac 1/15

Outline
• Introduction to ANOVA
• The Completely Randomized Design (One-Way
ANOVA)
• Multiple Comparison Tests
• A Factorial Design (Two-Way ANOVA)

Dang Van Thac 2/15

 Introduction to ANOVA
• In an experimental design, an independent variable
(treatment variable) is a variable the experimenter
controls or modifies in the experiment. Independent
variables are sometimes also referred to as factors.
Each independent variable has two or more levels, or
classifications
• The other type of variable in an experimental design is
a dependent variable. A dependent variable is the
response to the different levels of the independent
variables. Dang Van Thac 3/15
Introduction to ANOVA
Example: Wal-Mart executives might perform a study
to compare daily sales volumes for a given size store
in four different demographic settings: (1) inner-city
stores (large city), (2) suburban stores, (3) stores in a
medium-sized city, and (4) stores in a small town.
Managers might also decide to compare sales on
the five different weekdays (Monday through Friday).
In this study, the independent variables are store
demographics and day of the week. The dependent
variable is the dollar amount of daily total sales.
Dang Van Thac 4/15
One-Way ANOVA
• In One-way ANOVA, subjects are assigned randomly to treatments
(Independent variable). It contains only one independent variable, with
more than two treatment levels, or classifications.
• In general, if k samples are being analyzed, the following hypotheses
are being tested in a one-way ANOVA:
𝐻0 : 𝜇1 = 𝜇2 = ⋯ = 𝜇𝑘
𝐻𝑎 : 𝐴𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑚𝑒𝑎𝑛𝑠 𝑖𝑠 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑡ℎ𝑒 𝑜𝑡ℎ𝑒𝑟𝑠
• 𝑆𝑆𝑇 = 𝑆𝑆𝐶 + 𝑆𝑆𝐸
SST=total sum of square; SSC=sum of squares column (treatment);
SSE=sum of square error; i=particular member of a treatment level; j=a
treatment level.

Dang Van Thac 5/15

The Completely Randomized Design (One-Way ANOVA)
𝑛𝑗 𝐶 2
𝑆𝑆𝑇 = 𝑖=1 𝑗=1 𝑥𝑖𝑗 − 𝑥 𝑀𝑆𝐶 = 𝑆𝑆𝐶/𝑑𝑓𝐶
𝐶
𝑆𝑆𝐶 = 𝑗=1 𝑛𝑗 (𝑥𝑗 − 𝑥 )2 𝑀𝑆𝐸 = 𝑆𝑆𝐸/𝑑𝑓𝐸

𝑛𝑗 2
𝑆𝑆𝐸 = 𝐶
𝑗=1 𝑥𝑖𝑗 − 𝑥𝑗 𝐹 = 𝑀𝑆𝐶/𝑀𝑆𝐸
𝑖=1

𝑑𝑓𝐶 = 𝐶 − 1 i = a particular member of a treatment level

j = a treatment level
𝑑𝑓𝐸 = 𝑁 − 𝐶 C = number of treatment levels
𝑛𝑗 = number of observations in a given treatment level
𝑑𝑓𝑇 = 𝑁 − 1 𝑥 = grand mean
𝑥𝑗 = column mean
𝑥𝑖𝑗 = individual value
Dang Van Thac 6/15
The Completely Randomized Design (One-Way ANOVA)

Example: A company has three

manufacturing plants, and company
officials want to determine whether there is
a difference in the average age of workers
at the three locations. The following data
are the ages of five randomly selected
workers at each plant. Perform a one-way
ANOVA to determine whether there is a
significant difference in the mean ages of
the workers at the three plants. Use α= .01
and note that the sample sizes are equal.
Dang Van Thac 7/15
The Completely Randomized Design (One-Way ANOVA)
Step 1: 𝐻0 : 𝜇1 = 𝜇2 = 𝜇3
𝐻𝑎 : 𝐴𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑚𝑒𝑎𝑛𝑠 𝑖𝑠 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑡ℎ𝑒 𝑜𝑡ℎ𝑒𝑟𝑠
Step 2: The statistical test is ANOVA F test
Step 3: The value of alpha is .01
Step 4: 𝑑𝑓𝐶 = 3 − 1 = 2; 𝑑𝑓𝐸 = 15 − 3 = 12. Check the F table: 𝐹.01,𝟐,𝟏𝟐 = 6.93
=> the decision is to reject the null hypotheses if the observed
F value is greater than the critical F value.
Step 5: Sample data
Step 6: Calculate the sample statistics:

𝑛1 = 5, 𝑛2 = 5, 𝑛3 = 5
𝑥1 = 28.2, 𝑥2 = 32.0, 𝑥3 = 24.8, 𝑥 = 28.33
𝑑𝑓𝐶 = 2, 𝑑𝑓𝐸 = 12, 𝑑𝑓𝑇 = 15 − 1 = 14

Dang Van Thac 8/15

The Completely Randomized Design (One-Way ANOVA)
2
𝑆𝑆𝐶 = 5 ∗ 28.2 − 28.33 + 5 ∗ 32.0 − 28.33 2 + 5 ∗ 24.8 − 28.33 2 =129.73
𝑆𝑆𝐸 = (29 − 28.2)2 + (27 − 28.2)2 +…+ (26 − 24.8)2 = 19.60
𝑆𝑆𝑇 = (29 − 28.33)2 +(27 − 28.33)2 +…+(26 − 28.33)2 = 149.33
𝑀𝑆𝐶 = 𝑆𝑆𝐶/𝑑𝑓𝐶 =129.73/2=64.87
𝑀𝑆E = 𝑆𝑆E/𝑑𝑓𝐸 =19.60/12=1.63
𝐹 = 𝑀𝑆𝐶 𝑀𝑆𝐸 = 64.87 1.63 = 39.80

Step 7: Because 39.80>6.93 => reject H0

Dang Van Thac 9/15
Multiple Comparison Tests
• Multiple comparisons (Post Hoc multiple comparisons)
are to be used only when an overall significant difference
between groups has been obtained by using the F value
of the analysis of variance.
• Post hoc pairwise comparisons are made after the
experiment when the researcher decides to test for any
significant differences in the samples based on a
significant overall F value.
• LSD test, Bonferroni test, Scheffe test, Tukey test, Waller-
Duncan test...
Dang Van Thac 10/15
Two-Way ANOVA
• Some experiments are designed so that two or more treatments
(independent variables) are explored simultaneously. Such experimental
designs are referred to as factorial designs (Two-Way ANOVA).

Dang Van Thac 11/15

 A Factorial Design (Two-Way ANOVA)

• Hypotheses:
Row effects: 𝐻0 : 𝑅𝑜𝑤 𝑚𝑒𝑎𝑛𝑠 𝑎𝑙𝑙 𝑎𝑟𝑒 𝑒𝑞𝑢𝑎𝑙
𝐻𝑎 : 𝐴𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑟𝑜𝑤 𝑚𝑒𝑎𝑛 𝑖𝑠 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑡ℎ𝑒 𝑜𝑡ℎ𝑒𝑟𝑠
Column effects: 𝐻0 : 𝐶𝑜𝑙𝑢𝑚𝑛 𝑚𝑒𝑎𝑛𝑠 𝑎𝑙𝑙 𝑎𝑟𝑒 𝑒𝑞𝑢𝑎𝑙
𝐻𝑎 : 𝐴𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝑐𝑜𝑙𝑢𝑚𝑛 𝑚𝑒𝑎𝑛 𝑖𝑠 𝑑𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑡 𝑓𝑟𝑜𝑚 𝑡ℎ𝑒 𝑜𝑡ℎ𝑒𝑟𝑠
Interaction effects: 𝐻0 : 𝑇ℎ𝑒 𝑖𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛 𝑒𝑓𝑓𝑒𝑐𝑡𝑠 𝑎𝑟𝑒 𝑧𝑒𝑟𝑜
𝐻𝑎 : 𝐴𝑛 𝑖𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛 𝑒𝑓𝑓𝑒𝑐𝑡 𝑖𝑠 𝑝𝑟𝑒𝑠𝑒𝑛𝑡

Dang Van Thac 12/15

 A Factorial Design (Two-Way ANOVA)
𝑅 𝑅 𝐶

𝑆𝑆𝑅 = 𝑛𝐶 (𝑥𝑖 − 𝑥)2 𝑆𝑆𝐼 = 𝑛 (𝑥𝑖𝑗 − 𝑥𝑖 − 𝑥𝑗 + 𝑥 )2

𝑖=1 𝑖=1 𝑗=1
𝐶 𝑅 𝐶 𝑛 𝑅 𝐶 𝑛

𝑆𝑆𝐶 = 𝑛𝑅 (𝑥𝑗 − 𝑥)2 𝑆𝑆𝐸 = (𝑥𝑖𝑗𝑘 − 𝑥𝑖𝑗 )2 𝑆𝑆𝑇 = (𝑥𝑖𝑗𝑘 − 𝑥)2

𝑗=1 𝑖=1 𝑗=1 𝑘=1 𝑖=1 𝑗=1 𝑘=1

𝑀𝑆𝑅 = 𝑆𝑆𝑅/(𝑅 − 1) 𝐹𝑅 = 𝑀𝑆𝑅/𝑀𝑆𝐸

𝑀𝑆𝐶 = 𝑆𝑆𝐶/(𝐶 − 1) 𝐹𝐶 = 𝑀𝑆𝐶/𝑀𝑆𝐸
𝑀𝑆𝐼 = 𝑆𝑆𝑅/(𝑅 − 1)(𝐶 − 1) 𝐹𝐼 = 𝑀𝑆𝐼/𝑀𝑆𝐸
𝑀𝑆𝐸 = 𝑆𝑆𝐸/𝑅𝐶(𝑛 − 1)
Dang Van Thac 13/15
A Factorial Design (Two-Way ANOVA)
• Interaction occurs when the effects of one treatment vary according to the
levels of treatment of the other effect.

Dang Van Thac 14/15

A Factorial Design (Two-Way ANOVA)

Dang Van Thac 15/15