4.

3 THE GRAECO-LATIN SQUARE DESIGN

INTRODUCTION

In many experimental situations, researchers must control several sources of nuisance

variability to obtain accurate results. When two blocking factors are not enough, the Graeco-
Latin Square Design becomes a powerful tool. It extends the Latin Square by incorporating a
second set of treatment symbols (Greek letters), allowing control of three nuisance factors
while studying one primary treatment factor. This design uses two orthogonal Latin squares of
the same order, ensuring that every Latin letter occurs exactly once with every Greek letter.

OBJECTIVES

By the end of this section, you should be able to:

1. Explain the concept and structure of a Graeco-Latin Square Design.

2. Identify conditions where this design is appropriate.

3. Describe the role of orthogonality and how it ensures balance.

4. Understand the statistical model associated with the design.

5. Interpret the ANOVA results produced by a Graeco-Latin Square Design.

TRY THIS!

“Graeco-Latin Square Designs - Controlling Two Blocking Factors”

You’re designing an experiment to test four different formulations of a new beverage.

However, two nuisance factors could influence the results:

• Day (4 levels)
• Taster (4 different panelists)

To control both nuisance factors at the same time, the researcher chooses a Graeco–Latin
Square Design, where:

• Each treatment appears once in every row and column,

• And each treatment appears once with each Greek-letter symbol, representing the
second blocking factor.
Your task is to understand how the design works and identify its key components.

1. What are the variables in the experiment?

2. Formulate the statistical model assumptions for a Graeco-Latin square design.

3. Why is a Graeco-Latin square design preferred when two nuisance factors must be
controlled?

4. What would be an appropriate research question for this design?

THINK AHEAD!

“Digging Deeper – Why this design works?”

You used a Graeco–Latin Square ANOVA, which controls three nuisance factors (rows,
columns, and Greek letters). Reflect on your choices and understanding using the questions
below.

1. Choosing the Method:

• Why is a Graeco-Latin Square the appropriate analysis for this experiment?

• How did orthogonality between the Latin and Greek squares ensure valid
comparisons among treatments?
 2. Facing Uncertainty:

• Was there a moment when you were unsure whether to use a Latin Square,
RCBD, or Graeco-Latin Square?

• What helped you confirm that three nuisance variables were present and
needed to be controlled?

3. Interpreting Results:

• How confident were you in understanding the ANOVA table for a Graeco-
Latin Square?

• Did the results match your expectations? Why or why not?

4. If Assumptions Fail:

• If normality or equal variances were violated, what adjustments could you

make (e.g., transformation, alternative design, or re-checking randomization)?

READ AND PONDER!

A Graeco-Latin Square of order p uses:

• Rows (nuisance factor 1)
• Columns (nuisance factor 2)
• Latin letters (primary treatments)
• Greek letters (nuisance factor 3)
 Two Latin squares are orthogonal if, when superimposed, each Latin symbol appears
exactly once with each Greek symbol. This guarantees balanced treatment comparisons.

Statistical Model:

𝑌𝑖𝑗𝑘𝑙 = 𝜇 + 𝛼𝑖 + 𝛽𝑗 + γk + 𝛿𝑙 + 𝜀𝑖𝑗𝑘𝑙
Where:
𝛼𝑖 = row effect
𝛽𝑗 = column effect
γk = Latin-letter treatment effect
𝛿𝑙 = Greek-letter treatment effect
𝜀𝑖𝑗𝑘𝑙 = error term (normally distributed)

ANOVA partitions variation into rows, columns, Latin letters, Greek letters, and
error. Each factor contributes (p-1) degrees of freedom, while error contributes (p-1)(p-3).

STEP-BY-STEP PROCEDURE

1. Identify three nuisance factors and one primary treatment factor.

2. Check if the order p of the design allows a Graeco-Latin Square (p ≠ 6).
3. Construct two orthogonal Latin Squares.
4. Superimpose them to obtain paired treatments.
5. Randomize rows, columns, and treatment labels.
6. Conduct experiments according to the layout.
7. Perform ANOVA to test treatment and nuisance factor effects.

CONCLUSION
Graeco-Latin Squares offer an efficient way to control three sources of nuisance
variation simultaneously. They maintain treatment balance through orthogonality and allow
researchers to obtain more precise estimates of treatment effects using fewer experimental runs.

SEE IF YOU CAN DO THIS!

1. Construct a 4×4 Graeco-Latin Square using letters A-D and Greek symbols 𝛼 − 𝛿.
2. Identify the number of degrees of freedom for error in a 5×5 Graeco-Latin Square.
3. Explain why the design does not exist for p = 6.
REFERENCES

• Montgomery, D. C. (2012). Design and analysis of experiments (8th ed.). John Wiley
& Sons.
• Cox, D. R. (1958). Planning of experiments. Wiley.
• Cochran, W. G., & Cox, G. M. (1957). Experimental designs (2nd ed.). Wiley.
• Kuehl, R. O. (2000). Design of experiments: Statistical principles of research design
and analysis (2nd ed.). Duxbury Press.
• Mead, R., Curnow, R. N., & Hasted, A. M. (2002). Statistical methods in agriculture
and experimental biology (3rd ed.). Chapman & Hall/CRC.

Prepared By: FAITH M. MAHILUM BS STAT 4

4.4 BALANCED INCOMPLETE BLOCK DESIGNS

In many experiments, including all treatments within each block, is often impractical due
to limitations in space, materials, time, cost, or equipment capacity. A Balanced Incomplete
Block Design (BIBD) addresses this by using smaller “incomplete” blocks while maintaining
fairness and balance in treatment comparisons.

A design is considered “balanced” if:

• Each treatment appears in exactly r blocks.

• Each block contains exactly k treatments.

• Every pair of treatments appears together in exactly λ blocks.

Although not all treatments appear in every block, BIBDs ensure that each pair of
treatments is compared the same number of times. This reduces experimental error, provides
equal comparison opportunities, and produces unbiased estimates of treatment effects, making
it a practical solution when full blocking is impossible.

OBJECTIVES

Students should be able to:

1. Define a Balanced Incomplete Block Design.

2. Identify parameters (v, b, r, k, λ) of a BIBD.
3. Explain why incomplete blocks are useful and necessary.
4. Determine whether a set of parameters satisfies BIBD conditions.
5. Recognize situations where BIBD is superior to a complete block design.

TRY THIS!

“Balanced Incomplete Block Designs – Applying them to Rice Varieties”

You’re working on an experiment where planting every rice variety in each field is not
possible due to irrigation limits. Instead of a complete block design, you decide to use a
Balanced Incomplete Block Design (BIBD). In this type of design:
 • Not all treatments (rice varieties) appear in each block (field).

• Each treatment appears in the same number of blocks.

• Every pair of treatments occurs together in the same number of blocks.

Your job is to understand why the BIBD structure is useful and what its key components
are before diving into the subsections of this chapter.

1. What are the variables in the experiment?

2. Why can’t a randomized complete block design be used in this situation?

3. How could a BIBD help in testing these rice varieties?

4. What might be the values of r, k, and λ in this experiment?

5. What would be an appropriate research question for a BIBD study in this context?

THINK AHEAD!

“Digging Deeper – Understanding the Statistical Analysis of BIBD”

You performed the ANOVA for a Balanced Incomplete Block Design, partitioning
variation into Treatment, Block, and Error sources based on the BIBD structure.
1. Choosing the Test:

• Why did you use the BIBD ANOVA instead of a standard randomized block
ANOVA?

• What evidence (design parameters, balance conditions) confirmed this choice?

2. Facing Uncertainty:

• Were you unsure about how the sums of squares or degrees of freedom were
computed?

3. Reflecting on design choice:

• Were there other designs you considered, and why was BIBD preferred?

• How does understanding λ, r, and k help you interpret results accurately?

4. Applying your knowledge:

• In your own words, explain why BIBD might be chosen over a complete block
design in a field experiment.

• How might this understanding influence planning or analyzing future

experiments?
READ AND PONDER
A BIBD is defined by the following parameters:
• v = number of treatments
• b = number of blocks
• k = number of treatments per block
• r = number of blocks each treatment appears in
• λ = number of times each pair of treatments appears together

The balance conditions are:

1. vr = bk
2. λ(v - 1) = r(k - 1)

Even though blocks are incomplete, these relationships ensure equal representation of
treatments. Applications include agriculture (plots with limited size), medicine (limited
subjects per trial), and manufacturing (testing machinery with limited capacity).

CONCLUSION
BIBDs provide a structured and efficient solution when complete blocks are not feasible.
Through mathematical balance, they ensure fair treatment comparisons despite incomplete data
patterns.

SEE IF YOU CAN DO THIS!

1. A design has v = 5, b = 10, k = 2, r = 4. Does it satisfy BIBD conditions?

2. How many times does each treatment pair occur together if λ = 1?
3. Give two real-world examples where BIBD is appropriate.
REFERENCES

• Montgomery, D. C. (2012). Design and analysis of experiments (8th ed.). John Wiley
& Sons.
• Cochran, W. G., & Cox, G. M. (1957). Experimental designs (2nd ed.). Wiley.
• Gomez, K. A., & Gomez, A. A. (1984). Statistical procedures for agricultural research
(2nd ed.). John Wiley & Sons.
• Hinkelmann, K., & Kempthorne, O. (2005). Design and analysis of experiments:
Volume 1 – Introduction to experimental design (2nd ed.). Wiley-Interscience.
• Kuehl, R. O. (2000). Design of experiments: Statistical principles of research design
and analysis (2nd ed.). Duxbury Press.

Prepared By: FAITH M. MAHILUM BS STAT 4

4.4.1 STATISTICAL ANALYSIS OF BIBD

INTRODUCTION

After constructing a Balanced Incomplete Block Design, the next step is analyzing the
data. Because the blocks do not contain all treatments, standard ANOVA must be adjusted.
This module focuses on computing sums of squares, determining treatment effects, and
performing hypothesis tests in BIBDs.

OBJECTIVES

Students will be able to:

1. Write the statistical model for a BIBD.

2. Distinguish between block and treatment effects.
3. Compute sums of squares using adjusted treatment totals.
4. Construct and interpret the ANOVA table for a BIBD.
5. Perform hypothesis testing for treatment differences.

TRY THIS!

Suppose 4 treatments are arranged in a BIBD where each block contains 3 treatments.

1. How many treatments are missing in each block?

2. Why do we need “adjusted” treatment totals?

3. What happens if we use unadjusted totals?

 THINK AHEAD

1. Why is blocking still essential even if blocks are incomplete?

2. What would happen to treatment comparisons if block effects were ignored?

3. Why is the error degrees of freedom smaller in BIBDs compared to complete

blocks?

READ AND PONDER

Statistical Model:

𝑦𝑖𝑗 = 𝜇 + 𝜏𝑖 + 𝛽𝑗 + 𝜀𝑖𝑗

where 𝜏𝑖 denotes the treatment effect and βⱼ is the block effect.

Total Sum of Squares:

(𝑦. . )2
𝑆𝑆𝑇 = ∑ 𝑦 2 −
𝑁

Block Sum of Squares:

1 2
(𝑦. . )2
𝑆𝑆𝐵 = ∑(𝑏𝑙𝑜𝑐𝑘 𝑡𝑜𝑡𝑎𝑙𝑠) −
𝑘 𝑁

Adjusted Treatment Totals:

1
𝑄𝑖 = 𝑇𝑖 − ∑(𝑛𝑖𝑗 𝐵𝑗 )
𝑘
Treatment SS (Adjusted):

𝑘
𝑆𝑆𝑇𝑟𝑡(𝑎𝑑𝑗) = ∑ 𝑄𝑖2
λ

Error SS:

𝑆𝑆𝐸 = 𝑆𝑆𝑇 − 𝑆𝑆𝐵 − 𝑆𝑆𝑇𝑟𝑡(𝑎𝑗𝑑)

ANOVA Table includes treatments, blocks, error, and total.

CONCLUSION
BIBD analysis relies on adjusting treatment totals to account for incomplete
representation within blocks. The resulting ANOVA allows fair comparison of treatments even
when blocks do not contain all of them.

SEE IF YOU CAN DO THIS!

1. Explain why Qᵢ is necessary for treatment comparisons.

2. Identify df for treatment, block, and error in a BIBD with v = 6 and b = 10.
3. Compute SST if ∑ 𝑦 2 = 980 and 𝑦. . = 100 with N = 20.

REFERENCES

• Montgomery, D. C. (2012). Design and analysis of experiments (8th ed.). John Wiley
& Sons.
• Cochran, W. G., & Cox, G. M. (1957). Experimental designs (2nd ed.). Wiley.
• Hinkelmann, K., & Kempthorne, O. (2005). Design and analysis of experiments:
Volume 1 – Introduction to experimental design (2nd ed.). Wiley-Interscience.
• Gomez, K. A., & Gomez, A. A. (1984). Statistical procedures for agricultural research
(2nd ed.). John Wiley & Sons.
• Mead, R., Curnow, R. N., & Hasted, A. M. (2002). Statistical methods in agriculture
and experimental biology (3rd ed.). Chapman & Hall/CRC.

Prepared By: FAITH M. MAHILUM BS STAT 4

4.4.2 LEAST SQUARES ESTIMATION OF PARAMETERS

INTRODUCTION

In Balanced Incomplete Block Designs, the incomplete nature of blocks complicates

estimation of treatment effects. Least Squares (LS) provides a systematic method of estimating
these effects using adjusted totals and exploiting the balance conditions of the design.

OBJECTIVES

Students will be able to:

1. Write the LS normal equations for treatments and blocks.

2. Explain how balance simplifies estimation.
3. Compute treatment effect estimates using Qᵢ values.
4. Interpret LS estimates in terms of treatment differences.

TRY THIS!

You are analyzing 5 fertilizers applied using a BIBD, where each block contains only
3 fertilizers.

1. Why might simple averages mislead you?

2. How does LS estimation help?

3. What parameters must be estimated?

THINK AHEAD!

“Digging Deeper - Least Squares Estimation in BIBD”

You estimated treatment and block effects in a Balanced Incomplete Block Design
(BIBD) using least squares, following the constraints required to make the model identifiable.
Use this sheet to reflect on the estimation process and its implications.

1. Choosing the Method:

• Why was least squares estimation used to estimate treatment and block
effects?

• What model assumptions supported the LS approach?

2. Evaluating the Estimation Process:

• Did you face difficulty understanding the constraint (e.g., ∑ 𝜏𝑖 = 0)?

• What would happen if each treatment did not appear in exactly r blocks?

3. Interpreting LS Estimates:

• How confident were you in interpreting treatment effect estimates?

• Did the estimated effects make sense given the raw data?

4. Beyond Estimation:

• Why are BIBD parameter conditions (k, r, λ) essential for unique solutions?
 5. Reflecting on Practice:

• How might understanding these constraints influence planning future

experiments with limited resources?

• In your own words, explain why LS estimation combined with BIBD design
provides reliable treatment effect estimates.

READ AND PONDER

Least Squares Normal Equations for treatments (with constraint ∑ 𝜏𝑖 = 0):

𝑟𝜏𝑖 − ∑(𝑛𝑖𝑗 𝛽𝑗 ) = 𝑄𝑖

Estimated Treatment Effect:

𝑘𝑄𝑖
𝜏̂𝑖 =
λ𝑣

Interpretation:
The treatment effect estimator depends on:
• The number of times a treatment appears (k)
• How often each pair appears together (λ)
• The overall balance v

LS estimation ensures fairness, even though treatments occur in incomplete blocks.

CONCLUSION
Least Squares provides a stable and unbiased way to estimate treatment effects in BIBDs.
The use of adjusted totals 𝑄𝑖 ensures that treatment comparisons remain valid despite
incomplete blocks.
SEE IF YOU CAN DO THIS!

1. Compute 𝜏̂𝑖 if k = 3, λ = 2, v = 5, and 𝑄𝑖 = 4.

2. Explain why 𝜏̂𝑖 becomes smaller if λ increases.
3. Write the constraint used in estimating treatment effects.

REFERENCES

• Montgomery, D. C. (2012). Design and analysis of experiments (8th ed.). John Wiley
& Sons.
• Hinkelmann, K., & Kempthorne, O. (2005). Design and analysis of experiments:
Volume 1 – Introduction to experimental design (2nd ed.). Wiley-Interscience.
• Kuehl, R. O. (2000). Design of experiments: Statistical principles of research design
and analysis (2nd ed.). Duxbury Press.
• Mead, R., Curnow, R. N., & Hasted, A. M. (2002). Statistical methods in agriculture
and experimental biology (3rd ed.). Chapman & Hall/CRC.
• Cochran, W. G., & Cox, G. M. (1957). Experimental designs (2nd ed.). Wiley.

Prepared By: FAITH M. MAHILUM BS STAT 4

4.4.3 RECOVERY OF INTERBLOCK INFORMATION

INTRODUCTION

Incomplete blocks cause loss of information. However, if block effects are random,
additional information about treatment differences can be “recovered” using interblock
comparisons. This approach, introduced by Yates, leads to more precise estimates of treatment
effects.

OBJECTIVES

After completing this module, students will be able to:

1. Explain the idea of interblock information.

2. Distinguish fixed vs. random block effects.
3. Describe Yates’ method of recovering information.
4. Identify conditions under which recovery is beneficial.
5. Interpret recovered treatment estimates.

TRY THIS!

“Understanding Random and Fixed Effects in BIBD

Imagine each block represents a day in a factory, and day-to-day variation is random.
This exercise helps you think about how BIBD handles block effects in different scenarios.

Your job is to understand the role of fixed and random effects and how they influence treatment
comparisons.

1. Should day be treated as a fixed or random effect?

2. Why might treating day as a random effect allow recovery of information?

 3. Would the same reasoning apply if blocks were fixed instead of random?

4. What implications does this have for planning experiments with day-to-day
variability?

THINK AHEAD!

“Digging Deeper - Recovery of Interblock Information”

Sometimes, incomplete blocks naturally lead to some loss of information, but

additional precision can be gained by recovering comparisons between blocks,
especially when block differences are small.

1. Understanding Information Loss:

• Why do incomplete blocks naturally lead to loss of information?

• How does ignoring interblock comparisons limit the precision of treatment

estimates?

2. Conceptual Understanding of Recovery:

• What is gained by using interblock comparisons in BIBD analysis?

• Were there aspects of adjusted totals or covariances that clarified how recovery
improves precision?
 3. Were Recovery may not be Advisable:

• Under what circumstances would recovering interblock information not be

recommended?

• How might block heterogeneity or extreme variability affect the usefulness of

recovery?

4. Practical Implication:

• How does recovering interblock information influence real-world decision-

making and interpretation of treatment effects?

READ AND PONDER

If block effects are random, then:

𝑉𝑎𝑟(𝛽𝑗 ) = 𝜎𝛽2 and 𝑉𝑎𝑟(𝜀𝑖𝑗 ) = 𝜎 2

Using this, interblock information can be incorporated by estimating 𝜎 2 and 𝜎𝛽2 .

Recovered estimators combine: intrablock information (within blocks), interblock information
(across blocks). Yates’ method creates weighted treatment estimates using both sources. This
reduces the variance of treatment contrasts.

CONCLUSION
Recovery of interblock information improves precision in treatment comparisons
when block effects are random. It expands the usefulness of BIBDs, especially when blocks
are too small to contain all treatments.
SEE IF YOU CAN DO THIS!

1. Give an example where block effects are random.

2. Explain why recovery does not work when block effects are fixed.
3. Describe one benefit of using recovered treatment estimates.

REFERENCES

• Montgomery, D. C. (2012). Design and analysis of experiments (8th ed.). John Wiley
& Sons.
• Yates, F. (1936). A new method of arranging variety trials involving a large number of
varieties. Journal of Agricultural Science, 26(3), 424–455.
• Hinkelmann, K., & Kempthorne, O. (2005). Design and analysis of experiments:
Volume 1 –Introduction to experimental design (2nd ed.). Wiley-Interscience.
• Kuehl, R. O. (2000). Design of experiments: Statistical principles of research design
and analysis (2nd ed.). Duxbury Press.
• Mead, R., Curnow, R. N., & Hasted, A. M. (2002). Statistical methods in agriculture
and experimental biology (3rd ed.). Chapman & Hall/CRC.

Prepared By: FAITH M. MAHILUM BS STAT 4