INCOMPLETE BLOCK DESIGNS
Balanced Incomplete Block Design (BIBD)
1. Introduction to Incomplete Block Designs
In many agricultural, industrial and biological experiments, it is not possible to accommodate
all treatments in each block due to:
• Limited block size
• Heterogeneity within blocks
• Practical constraints
When each block does NOT contain all treatments, the design is called an Incomplete Block
Design.
Among them, the most important and widely studied design is:
Balanced Incomplete Block Design (BIBD)
2. Definition of BIBD
A Balanced Incomplete Block Design is an incomplete block design satisfying:
1. v treatments
2. b blocks
3. Each block contains k treatments (k < v)
4. Each treatment appears in r blocks
5. Every pair of treatments occurs together in exactly λ blocks
It is denoted by:
BIBD(v, b, r, k, λ)
3. Parameters of BIBD
The five fundamental parameters are:
v = number of treatments
b = number of blocks
r = number of replications of each treatment
k = block size
λ = number of blocks in which every pair appears together
4. Relationships Among Parameters
The parameters of BIBD are not independent. They satisfy two fundamental relationships.
4.1 First Fundamental Relation
Total treatment occurrences counted in two ways:
Total number of treatment appearances:
vr
Total number of treatment positions in blocks:
bk
Hence,
vr = bk (1)
4.2 Second Fundamental Relation
Consider a particular treatment.
It occurs in r blocks.
In each block, it appears with (k − 1) other treatments.
Total number of pairs involving that treatment:
r(k − 1)
But each of the remaining (v − 1) treatments appears with it λ times.
Thus:
r(k − 1) = λ(v − 1) (2)
5. Necessary Conditions for Existence of
BIBD
From (1):
b = vr/k
From (2):
r = λ(v − 1)/(k − 1)
Thus r and b must be integers.
These are necessary (but not always sufficient) conditions.
6. Incidence Matrix of BIBD
Let N be a v × b matrix defined as:
n_ij = 1 if treatment i appears in block j
n_ij = 0 otherwise
This matrix is called the Incidence Matrix.
7. Properties of Incidence Matrix
Property 1: Row Sum
Each treatment appears r times.
Therefore, sum of elements in each row = r
Property 2: Column Sum
Each block contains k treatments.
Thus sum of elements in each column = k
8. Matrix Relations
Let N be the incidence matrix.
8.1 NN' (v × v matrix)
The (i, j)th element of NN' is:
• r if i = j
• λ if i ≠ j
Thus,
NN' = (r − λ)I_v + λJ_v
Where:
I_v = Identity matrix of order v
J_v = Matrix of all ones of order v
8.2 N'N (b × b matrix)
Diagonal elements = k
Off-diagonal elements = number of treatments common to two blocks
9. Proof of NN' Structure
Consider two distinct treatments i and j.
Number of blocks in which both appear = λ.
Thus (i, j)th element of NN' = λ.
For i = j:
Number of blocks in which treatment i appears = r.
Hence diagonal element = r.
Therefore:
NN' = (r − λ)I_v + λJ_v
10. Eigenvalues of NN'
Since:
NN' = (r − λ)I_v + λJ_v
Eigenvalues of J_v:
• v (multiplicity 1)
• 0 (multiplicity v − 1)
Thus eigenvalues of NN':
1. r − λ + λv = r + λ(v − 1)
2. r − λ (multiplicity v − 1)
11. Statistical Model for BIBD
Model:
Y_ij = μ + τ_i + β_j + ε_ij
Where:
μ = overall mean
τ_i = treatment effect
β_j = block effect
ε_ij ~ N(0, σ²)
Constraints:
Σ τ_i = 0
Σ β_j = 0
12. ANOVA for BIBD
Total SS = Treatment SS + Block SS + Error SS
Degrees of Freedom:
Total = vr − 1
Treatment = v − 1
Block = b − 1
Error = vr − v − b + 1
13. Estimation of Treatment Effects
Normal equations in matrix form:
Cτ = Q
Where:
C = rI_v − λJ_v
Q = Treatment totals adjusted for blocks
The matrix C is called the Information Matrix.
14. Variance of Treatment Difference
Variance of difference between two treatment means:
Var(τ_i − τ_j) = 2σ²(r − λ)/rλ
This shows precision depends on r and λ.
15. Special Case: Symmetric BIBD
If v = b and r = k,
Then design is called Symmetric BIBD.
In this case:
r(k − 1) = λ(v − 1)
16. Example of BIBD
Suppose:
v = 7, k = 3, λ = 1
Then from:
r(k − 1) = λ(v − 1)
r(2) = 6
r=3
From vr = bk:
7×3=b×3
b=7
Thus design is BIBD(7, 7, 3, 3, 1)
17. Advantages of BIBD
• Controls block variation
• Balanced comparison of treatments
• Efficient estimation of treatment contrasts