Incomplete block design: In RBD and LSD, each block-row and column contains all the treatments

under comparison. But for large number of treatments in an experiment, it may be difficult or even
impossible to obtain large size homogeneous blocks to accomodate all the treatments in each block. In
such cases, the number of experimental units per block is less than the number of treatments under
consideration and thus the block is said to be incomplete block and a design constituted of such blocks is
called an incomplete block design.

Balanced Incomplete Block Design (BIBD): When all treatment comparisons are equally important, the
treatment combinations used in each block should be selected in a balanced manner. Thus, an incomplete
block design in which every pair of treatments appear together in the same number of blocks is called a
balanced design as it ensures that all pairs of treatments are compared with equal precision. In fact, a
balanced incomplete block design is an arrangement of v treatments in b blocks each consisting of k
experimental units (k < v) such that: (i) Every block contains k experimental units (k < v). (ii) Every
treatment occurs at most once in each block. (iii) Every treatment occurs exactly in r blocks (r < b). (iv)
Every pair of treatments occurs together in exactly λ blocks.

The quantities b, v ,r, k and λ which are integers are called parameters of BIBD. Thus BIBD has 5
parameters which are not independent as they satisfy the following conditions or relations: (i) bk = vr (ii)
r(k −1) = λ(v −1) (iii) r > λ (iv) b ≥ v or r≥k (Fisher’s inequality) (v) b ≥v+r−k (vi) v > k

Example: A hospital wants to compare four pain-relief treatments:

A,B,C,D

Patients can safely receive only two treatments because of time or ethical constraints. Suppose number of
patients are 6.

BIBD is

From the design:

Number of treatments, v=4

Number of blocks, b=6

Each treatment appears in 3 blocks, r=3

Every pair of treatments occurs together exactly once, λ=1

Proof of the BIBD properties:

(i) bk=vr

Proof: Since BIBD consists of b blocks of size k each, then total number of observations is bk. On the
otherhand, the design contains v treatments in which each replicated r times, so the total number of
observations is vr. Hence, bk=vr.

(ii) r(k−1)=λ(v−1)

Proof: v treatments has = Pairs where each pair occurs λ times. So, the total number of times
all the pairs occur together in the blocks of the design is . On the otherhand, each block containing
k plots has = Pairs and total number of treatment pairs in all b blocks is .
Hence, = or, bk(k−1)=λv(v−1) or, vr(k−1)=λv(v−1) ∴ r(k−1)=λ(v−1).

(iii) r>λ

Proof: We have, r(k−1)=λ(v−1)

Sincek<v, then, (k−1)<(v−1)

It follows that r>λ

Note: Incidence matrix N=(nij); i=1,2,...,b; j=1,2,...,v is a b × v matrix defined by

By using previous example,

The incidence matrix is

(iv) b ≥ v,⇒ r ≥ k
Notice that all entries above the main diagonal are zero. Therefore, this is a lower triangular matrix. The
determinant of a triangular matrix equals the product of its diagonal elements.

Note: 1. The equality rank(NTN)= rank(NNT) = rank(N) holds for every real matrix N, whether N is
square or rectangular.

2. If A is a square matrix then

(v) b ≥ v+r−k

Proof: We know that

v > k and r ≥ k so that (v−k) > 0 and (r−k) ≥ 0

∴ (v−k)(r−k) ≥ 0
⇒vr−rk−vk+k2 ≥ 0

⇒bk−rk−vk+k2≥0

⇒k(b−r−v+k)≥0

∴b≥v+r−k

Note: The determinant of an incidence matrix is an integer because all entries of the matrix are
integers.
N(N´N) = (r - 𝝀)N + 𝝀 N

[ ]

Since, for symmetrical BIBD r = k

So, N(N´N) =

[ ]
As, N(N´N)
We can write,
Theorem: For a symmetrical BIBD, N´N = NN´

Proof:

=N´N

For symmetrical BIBD as b = v so N is a square matrix and

Analysis of data of BIBD for fixed effect model:

The fixed effect model of BIBD is given by,

……(1)

…… (2)
 ……(3)

We obtain from (1)

In order to eliminate the block effects, we have to multiply equation (3) by k and equation (2) by ∑ to get
Note: What happens without the incidence matrix?

Suppose

 Treatments: A,B,C,D
 Block 1 contains only A and B.

There are no observations y13,y14.

CR is given by F ≥ with (v-1) and (N-b-v+1) d.f.

Comment:
Note:

1. What is a replication?

A replication is a complete copy of all the treatments.

2. Why is Block SS called unadjusted?

The formula is

where Bi is the total of the i-th block.

Notice that this formula uses only the block totals.

The variation among the blocks is calculated without considering which treatments are
present in the blocks.

Hence it is called the unadjusted Block SS.

3. Why is Treatment SS called adjusted?

The Treatment SS is

Notice that it is not calculated from the treatment totals Tj.

Instead, it uses Qj where the block effects have already been removed.

Therefore, Treatment SS is adjusted for block effects.

Example

Suppose a BIBD has v=4 treatments: A, B, C, D; b=6 blocks; k=2, r=3, λ=1

The observations are

Block Treatments Yield