CHAPTER 3

Randomized Blocks, Latin Squares and

Related Designs
§ 3.1 The Randomized Complete Blocked Design (RCBD)
§ 3.2 The Latin Square Design
§ 3.3 The Graeco-Latin Square Design
§ 3.4 Balanced Incomplete Block Designs (BIBD)

1
3.2 The Latin Square Design

2
Latin Square Design
• Latin square design is used to eliminate two nuisance
sources of variability; that is, it systematically allows blocking
in two directions.
• Latin square design is a square arrangement and the
treatments are denoted by the Latin letters for example A,
B, C, D, and E if the experiment have five treatments.
• In general, a Latin square for 𝑝 factors is a square containing
𝑝 rows and 𝑝 columns.
• Each treatments, and each letter occurs once and only once
in each row and column.
• Each row contains every treatment.
• Each column contains every treatment.

3
Latin Square Design
• Example of Latin square design

4
Example 3.4 (Latin square)
An experiment to investigate the effects of various dietary starch levels on milk production was conducted
on four cows. The four diets (A, B, C, and D) were fed for three weeks to each cow and the total yield of
milk in the third week of each period was recorded in table below. The trial lasted 12 weeks since each cow
received each treatment, and each treatment required three weeks. The investigator felt strongly that time
period effects might be important (i.e earlier periods in the experiment might influence milk yields differently
compared to later periods). Hence, the investigator wanted to block on both cow and period. The
investigator decides to run the experiment as a Latin square so that cow and period effects may be
systematically controlled.
Cow
Period
1 2 3 4
1 D (192) A (195) C (292) B (249)
2 A (190) D (203) B (218) C (210)
3 C (214) B (139) A (245) D (163)
4 B (221) C (152) D (204) A (134)
5
 Statistical Analysis of the
Latin square Design
Fixed Effect Model
𝑥"#! = 𝜇 + 𝛼" + 𝛽# + 𝛾! + 𝜀"#! ,
𝑖 = 1,2, … , 𝑝 𝑗 = 1,2, … , 𝑝 𝑘 = 1,2, … , 𝑝
where
!" !"
𝑥"#! : the observation in the 𝑗 row and 𝑘 column
for the 𝑖 !" treatment
𝜇 : the overall mean
𝛼" : the 𝑖 !" treatment (level) effect
𝛽# : the 𝑗!" row effect
𝛾! : the 𝑘 !" column effect
𝜀"#! : the random error
𝑝 : number of treatments
**The model is completely additive; that is, there is no interaction
between rows, columns, and treatments 6
Statistical Analysis of the
Latin square Design
In an experiment involving the Latin square Design, we
are interested in testing the equality of the treatment
means.Thus, the hypotheses of interest are

𝐻! : 𝜇" = 𝜇# = ⋯ = 𝜇$
𝐻# : 𝜇$ ≠ 𝜇% for at least one 𝑖 ≠ 𝑗
where 𝑖, 𝑗 = 1,2, … , 𝑘

7
 Statistical
Statistical Analysis
Analysis of the
of the RCBD
Latin square Design
Example 3.4 (Latin square)
An experiment to investigate the effects of various dietary starch levels on milk production was conducted
on four cows. The four diets (A, B, C, and D) were fed for three weeks to each cow and the total yield of
milk in the third week of each period was recorded in table below. The trial lasted 12 weeks since each cow
received each treatment, and each treatment required three weeks. The investigator felt strongly that time
period effects might be important (i.e earlier periods in the experiment might influence milk yields differently
compared to later periods). Hence, the investigator wanted to block on both cow and period. The
investigator decides to run the experiment as a Latin square so that cow and period effects may be
systematically controlled.
Cow
Period
1 2 3 4
1 D (192) A (195) C (292) B (249)
2 A (190) D (203) B (218) C (210)
3 C (214) B (139) A (245) D (163)
4 B (221) C (152) D (204) A (134)
9
Example 3.4 (Latin square)
a) Determine the nuisance factor in this experiment.

b) What are the control and response variable?

c) How many treatments and list all the treatments.

d) Write down the model for this experiment.

10
Example 3.4 (Latin square)
a) Figure 1 shows the statistical analysis output from R.

Figure 1

i) State the value of mean square error and f-test treatment. Then, shows the manual calculation for the
values.
ii) Perform a complete statistical analysis to determine is there any significant difference between the
treatments.
iii) Based on your answer in iv), is it appropriate to conduct post-hoc test? Give a reason.

11
Example 3.4 (Latin square)
b) Figure 2 shows the normality test results of the residuals
using Shapiro-Wilk test. Conduct a necessary statistical test
for the residuals and draw your conclusion.

Figure 2

12
Exercise 3.4 (Latin square)
The effect of four different methods (A, B, C, D) on the time to assemble a device is being studied. The
experimenter wants to investigate how long each method takes with each of 4 workers. Also, since workers
might get tired or get more practiced, they might slow down or speed up near the end. Therefore, the order
that each worker tries a method also matters. Then, the experimenter decides to run the experiment as a
Latin square so that worker and order effects may be systematically controlled. The data was recorded
below.
Worker
Order
1 2 3 4
1 C=10 D=14 A=7 B=8
2 B=7 C=18 D=11 A=8
3 A=5 B=10 C=11 D=9
4 D=10 A=10 B=12 C=14

13
Exercise 3.4 (Latin square)
a) Determine the nuisance factor in this experiment.

b) What are the control and response variable?

c) How many treatments and list all the treatments.

d) Write down the model for this experiment.

14
Exercise 3.4 (Latin square)
a) Figure 1 shows the statistical analysis output from R.

P R

Figure 1

i) What kind of statistical analysis in Figure 1?

ii) State the value of mean square error.
iii) Determine the value of P, Q, and R.
iv) Perform a complete statistical analysis to determine is there any significant difference between the
treatments.
v) Based on your answer in iv), is it appropriate to conduct post-hoc test? Give a reason.

15
Exercise 3.4 (Latin square)
b) Figure 2 shows the Tukey’s confidence interval plot for
the experiment. Based on Figure 2,determine the significant
and the most significant comparison among treatments.
Justify your answer.

Figure 2

16
3.3 The Graeco-Latin Square
Design

17
Graeco-Latin Square
Design
• Graeco-Latin square design can be used to control
systematically three sources of extraneous variability (three
nuisance factor), that is, to block in three directions.
• The design allows investigation of four factors (rows,
columns, Latin letters (treatments), and Greek letters), each
at 𝑝 levels in only 𝑝# runs. Graeco-Latin squares exist for all
𝑝 ≥ 3 except 𝑝 = 6.

18
Graeco-Latin Square
Design
• The Greek letters appear exactly once in each row and
column and exactly once with each Latin letter.
• The factor represented by the Greek letters is orthogonal
to rows, columns, and Latin letter treatments.

19
Graeco-Latin Square
Design

20
Example 3.5 (Graeco-Latin Square Design)
An industrial engineer is investigating the effect of four assembly methods (A, B, C, D) on the assembly time
for a color television component. Four operators are selected for the study. Furthermore, the engineer
knows that each assembly method produces such fatigue that the time required for the last assembly may be
greater than the time required for the first, regardless of the method. That is, a trend develops in the required
assembly time. In addition, the engineer suspects that the workplaces used by the four operators may
represent an additional source of variation. A fourth factor, workplace (𝛼, 𝛽, 𝛾, 𝛿) may be introduced and
experiment was conducted, yielding the Graeco-Latin square as in table below .
Operator
Order of assembly
1 2 3 4
1 C𝛽 = 11 B𝛾 = 10 D𝛿 = 14 A𝛼 = 8
2 B𝛼 = 8 C𝛿 = 12 A𝛾 = 10 D𝛽 = 12
3 A𝛿 = 9 D𝛼 = 11 B𝛽 = 7 C𝛾 = 15
4 D𝛾 = 9 A𝛽 = 8 C𝛼 = 18 B𝛿 = 6

21
 Statistical Analysis of the
Graeco-Latin Square Design
Fixed Effect Model
𝑥"#! = 𝜇 + 𝛼" + 𝛽# + 𝛾! + 𝛿$ + 𝜀"#!$ ,
𝑖 = 1,2, … , 𝑝 𝑗 = 1,2, … , 𝑝 𝑘 = 1,2, … , 𝑝 𝑙 = 1,2, … , 𝑝
where
!" !"
𝑥"#! : the observation in the 𝑗 row and 𝑘 column
for the 𝑖 !" treatment
𝜇 : the overall mean
𝛼" : the 𝑖 !" treatment (level) effect
𝛽# : the 𝑗!" row effect
𝛾! : the 𝑘 !" column effect
𝜀"#!$: the random error
𝑝 : number of treatments
**The model is completely additive; that is, there is no interaction
between rows, columns, and treatments 22
Statistical Analysis of the
Graeco-Latin Square Design
In an experiment involving the Latin square Design, we
are interested in testing the equality of the treatment
means.Thus, the hypotheses of interest are

𝐻! : 𝜇" = 𝜇# = ⋯ = 𝜇$
𝐻# : 𝜇$ ≠ 𝜇% for at least one 𝑖 ≠ 𝑗
where 𝑖, 𝑗 = 1,2, … , 𝑘

23
Graeco-Latin
Square Design

24
Example 3.5 (Graeco-Latin Square Design)
An industrial engineer is investigating the effect of four assembly methods (A, B, C, D) on the assembly time
for a color television component. Four operators are selected for the study. Furthermore, the engineer
knows that each assembly method produces such fatigue that the time required for the last assembly may be
greater than the time required for the first, regardless of the method. That is, a trend develops in the required
assembly time. In addition, the engineer suspects that the workplaces used by the four operators may
represent an additional source of variation. A fourth factor, workplace (𝛼, 𝛽, 𝛾, 𝛿) may be introduced and
experiment was conducted, yielding the Graeco-Latin square as in table below.

Operator
Order of assembly
1 2 3 4
1 C𝛽 = 11 B𝛾 = 10 D𝛿 = 14 A𝛼 = 8
2 B𝛼 = 8 C𝛿 = 12 A𝛾 = 10 D𝛽 = 12
3 A𝛿 = 9 D𝛼 = 11 B𝛽 = 7 C𝛾 = 15
4 D𝛾 = 9 A𝛽 = 8 C𝛼 = 18 B𝛿 = 6
25
Example 3.5 (Graeco-Latin square)
a) Determine the nuisance factor in this experiment.

b) What are the control and response variable?

c) How many treatments and list all the treatments.

d) State TWO conditions of Graeco-Latin square design.

26
Example 3.5 (Graeco-Latin square)
e) Figure 1 shows the statistical analysis output from R.

Figure 1

i) Perform a complete statistical analysis to determine is there any significant difference between the
treatments.
ii) Based on your answer in iv), is it appropriate to conduct post-hoc test? Give a reason.

27
Example 3.5 (Graeco-Latin square)

Figure 2

f) Figure 2 shows the independence test results of the residuals using Durbin-Watson test. Conduct a necessary
statistical test for the residuals and draw your conclusion.

28
Exercise 3.5 (Graeco-Latin Square Design)
The yield of a chemical process was measured using five batches of raw material, five acid concentrations, five
catalyst concentrations (A, B, C, D, E), and five standing times (𝛼, 𝛽, 𝛾, 𝛿, 𝜖). The Graeco-Latin square that as in
table below.

Acid Concentration
Batch
1 2 3 4 5
1 A𝛼 = 26 B𝛽 = 16 C𝛾 = 19 D = 16 E = 13
2 B𝛾 = 18 C𝛿 = 21 D𝜖 = 18 E = 11 A = 21
3 C𝜖 = 20 D𝛼 = 12 E𝛽 = 16 A = 25 B = 13
4 D𝛽 = 15 E𝛾 = 15 A𝛿 = 22 B = 14 C = 17
5 E𝛿 = 10 A𝜖 = 24 B𝛼 = 17 C = 17 D = 14

29
Exercise 3.5 (Graeco-Latin square)
a) Determine the nuisance factor in this experiment.

b) What are the control and response variable?

c) How many treatments and list all the treatments.

d) What is the purpose of Graeco-Latin square design?

e) Fill in the Greek Letter in the table.

30
Exercise 3.5 (Graeco-Latin square)
f) Figure 1 shows the statistical analysis output from R.

P R

Figure 1

i) Determine the degree of freedom for the treatments.

ii) Perform a complete statistical analysis to determine is there any significant difference between the
treatments.
iii) Based on your answer in iv), is it appropriate to conduct post-hoc test? Give a reason.

31
Exercise 3.5 (Graeco-Latin square)

f) Figure 2 shows the output

for the Tukey’s test. Based on
Figure 2, determine the
significant and the most
significant comparison among
treatments. Justify your
answer.

Figure 2 32