Professional Documents
Culture Documents
Construction of Partially Balanced Incomplete Block Designs: Jyoti Sharma, D. K. Ghosh and Jagdish Prasad
Construction of Partially Balanced Incomplete Block Designs: Jyoti Sharma, D. K. Ghosh and Jagdish Prasad
Construction of Partially Balanced Incomplete Block Designs: Jyoti Sharma, D. K. Ghosh and Jagdish Prasad
Abstract
In this paper, some series of BIB and PBIB designs have been constructed
using either Semi-Regular Group Divisible Designs or Regular Group
Divisible Designs along with its corresponding group.
1. Introduction
Yates (1936) introduced the concept of BIBD. BIBD is an arrangement of v
treatments into b blocks each of k (<v) treatments, satisfying the following conditions:
1. Every treatment occurs at most once in each block.
2. Every treatment occurs in exactly r blocks.
3. Every pair of treatment occurs together in exactly λ blocks.
A BIBD is said to be symmetrical if v=b and r=k. The terms v, b, r, k, λ are known as
the parameters of BIBD. In this design we can estimate all possible treatment contrast
with the same precision. The rich contribution of BIBD is mainly by Fisher (1940,
1942), Fisher and Yates (1963) or Bose (1939-1959).
A BIBD is said to be resolvable if the b blocks are grouped into r classes of n blocks
each, such that each class forms a complete replication of all the v treatments and
each class however contains b / r blocks. BIBD are not available for every parametric
combination. Also even if a BIBD exists for a given no. of treatments (v) and block
size (k), it may require too many replications. To overcome this problem, Bose and
68 Jyoti Sharma et al
Nair (1939) introduced a class of binary, equi-replicate and proper designs, which we
called PBIBD with m-associate classes. Bose and Shimamoto (1952), Nair and Rao
(1942) have also contributed to the theory of PBIBD.
A PBIBD with two associate classes is an arrangement of v treatments in b blocks
such that:
1. Each of the v treatments is replicated r times in b blocks each of size k (k < v),
and no treatments appears more than once in any block.
2. There exists a relationship of association between every pair of the v
treatments satisfying the following conditions:
a. Any two treatments are either first or second associates.
b. Each treatment has exactly ni ith associates (i=1, 2).
c. Given any two treatments which are ith associates, the number of treatments
common to the jth associate of the first and kth associate of the second is pijk
and is independent of the pair of treatments. Also pijk=pikj, i, j, k=1, 2.
3. Any pair of treatments which are ith associate occur together in exactly λi
blocks for i=1,2.
Nair and Rao (1942) modified the original definition of PBIB designs. For m=2, Bose
and Shimamoto (1952) classified the known PBIB designs into
(1) Group Divisible (GD),
(2) Simple (S.I),
(3) Triangular (T),
(4) Latin Square Type (Li) and
(5) Cyclic Designs.
A group divisible design is an arrangement of v=mn treatments into b blocks such that
each block contains k(<v) distinct treatments which are partitioned into m( 2)
groups of n( 2) treatments each, further any two distinct treatments occurring
together in blocks if they belong to the same group, and in blocks if they
belong to different groups. A group divisible design is classified into
1) Singular Group Divisible Design
2) Semi-Regular Group Divisible Design
3) Regular Group Divisible Design.
2. Method of Construction
Let us consider a group (m,n) of a GD design, where v=mn treatments arranged in m
groups of n treatments each. For an example: consider a Group (2, 2). Now this group
Construction of Partially Balanced Incomplete Block Designs 69
Here the incidence matrix and concurrence matrix of d2 is denoted by N2, and N2N2׳
which are as follows:
⎡1 0⎤ ⎡1 0 1 0⎤
⎢0 1⎥
⎢0 1 0 1 ⎥⎥
N = ⎢ ⎥ N 2 N 2' = ⎢
2
⎢1 0⎥ ⎢1 0 1 0⎥
⎢ ⎥ ⎢ ⎥
⎣0 1⎦
⎣0 1 0 1⎦
Remark: If the same group is repeated p-times then N2N2 ׳is expressed as
⎡ I m .... I m ⎤
N 2 N 2' = p ⎢⎢ ... ... ... ⎥⎥ (2.1)
⎢⎣ I m ... I m ⎥⎦
Proof: Consider a SRGD design with parameters v1, b1, r1, k1, m, n, λ1, λ2 whose
incidence matrix is denoted by N1. Next add the corresponding group to this design
and let the incidence matrix of the group is N2. Now we define the incidence matrix of
resulting design N which is given by,
N = [N 1 N 2 ]v×b
'
of SRGD and hence N 1 N 1 can be expressed as
70 Jyoti Sharma et al
⎡ r1 λ2 λ1 .... λ2
⎤
⎢λ r1 λ2 .... ⎥
λ1
⎢ 2 ⎥
' ⎢λ λ2 r1 .... ⎥
λ2
N 1N 1 = ⎢ 1 ⎥ and
⎢ .... .... .... .... .... ⎥
⎢λ2 λ1 λ2 ..... r1 ⎥
⎢ ⎥
⎣⎢ ⎦⎥
⎡ 1 0 1 .... 0 ⎤
⎢ 0 1 0 .... 1 ⎥
⎢ ⎥
N 2 N '
2 = ⎢ 1 0 1 .... 0 ⎥
⎢ ⎥
⎢ .... .... .... .... .... ⎥
⎢⎣ 0 1 0 .... 1 ⎥⎦
,
Now using (2.2), NN is expressed as
⎡ r1 + 1 λ 2 λ1 + 1 .... λ 2 ⎤
⎢ λ r1 + 1 λ2 .... λ1 + 1⎥⎥
⎢ 2
⎢λ + 1 λ 2 r1 + 1 .... λ2 ⎥
NN ' = ⎢ 1 ⎥ (2.3)
⎢ .... .... .... .... ..... ⎥
⎢ λ2 λ1 + 1 λ 2 ..... r1 + 1 ⎥
⎢ ⎥
⎢⎣ ⎥⎦
Here all diagonal elements are same, that is, (r1 + 1) and off diagonal elements are λ2
and (λ1+ 1). Hence we can say N is the incidence matrix of a PBIB design with
parameters v=v1, b=b1+m, r=r1 + 1, k=k1, λ11=λ1+ 1, λ12=λ2.
Further we verified that r-λ11 >0 and (rk-vλ12) > 0 holds true and hence design is
RGD.
Example 2.1: Consider SR2 with parameters v=4, r=4, k=2, b=8, m=2, n=2, λ1=0,
λ2=2 with corresponding group (2, 2), whose blocks are as follows:
1 3 4 2 1 3 4 2
2 4 1 3 2 4 1 3
Now by adding the group to this design we obtain another design whose blocks are,
1 3 4 2 1 3 4 2 1 2
2 4 1 3 2 4 1 3 3 4
Which is a RGD design with parameters v=4, r=5, k=2, b=10, n1=1, n2=1, m=2,
n=2, λ1=0, λ2=1. This design is listed as R-3 in the Clatworthy (1973).
Construction of Partially Balanced Incomplete Block Designs 71
Example 2.2: Consider a RGD design (R-1) with parameters v=4, r=4, k=2, b=8,
m=2, n=2, λ1=2, λ2=1 along with its group (2, 2), whose blocks are as follows:
1 2 2 1 1 3 1 2
3 4 4 3 2 4 4 3
Now with this design, add its group (2, 2). we have another design whose blocks are,
1 2 2 1 1 3 1 2 1 2
3 4 4 3 2 4 4 3 3 4
The resulting design is a RGD with parameters v=4, r=5, k=2, b=10, m=2, n=2, λ1=3,
λ2=1 which is reported as R-2 in Clatworthy(1973).
NN '
= [N 1 N 2 ]⎢ N 1
' ⎥ = [
N 1N 1
'
+ N 2 N '
2 ] (2.4)
⎣⎢ N ⎥
21 ⎦
⎡ r1 + 1 λ λ + 1 .... λ ⎤
⎢ λ r1 + 1 λ .... λ + 1⎥⎥
⎢
'
⎢λ + 1 λ r1 + 1 .... λ ⎥
Hence NN = ⎢ ⎥ (2.5)
⎢ .... .... .... .... ..... ⎥
⎢ λ λ + 1 λ ..... r1 + 1⎥
⎢ ⎥
⎣⎢ ⎥⎦
Here all diagonal elements are same, that is, (r1 + 1) and off diagonal elements are of
two types which are λ + 1=λ1 and λ=λ2. Hence we can say N is the incidence matrix of
72 Jyoti Sharma et al
a PBIB design with parameters v=v1, b=b1+ m, r=r1+ 1, k=k1, m*=m, n*=n, λ1=λ+1,
λ2=λ.
Further we verified that r-λ1>0 and (rk-vλ2) > 0 holds true and hence the resulting
design is RGD.
Example 2.3: If we substitute s=3 in the above theorem, then a resolvable BIBD with
parameters v=9, b=12, r=4, k=3, λ=1 is considered, which is as follows:
1 4 7 1 2 3 1 2 3 1 2 3
2 5 8 4 5 6 6 4 5 5 6 4
3 6 9 7 8 9 8 9 7 9 7 8
Now by adding the design of group (3, 3) to this BIBD we obtained the following
design:
1 4 7 1 2 3 1 2 3 1 2 3 1 2 3
2 5 8 4 5 6 6 4 5 5 6 4 4 5 6
3 6 9 7 8 9 8 9 7 9 7 8 7 8 9
The resulting design is RGD with parameters v=9, b=15, r=5, k=3,, λ1=2, λ2=1
reported as R-59 in Clatworthy(1973).
Corollary 2.1: If there exists a GD design with parameters v, b, r, k, m, n, λ1, λ2, then
a resolvable BIBD exists provided λ11=λ12.
Proof: Consider either a SRGD or RGD design such that n=k and let the incidence
matrix of this design is denoted by N1. Next add the corresponding group to this
design and let the incidence matrix of the group is N2. Now we add N2 to N1 and then
we have the incidence matrix of resulting design N is given by
N = [N 1 N 2 ]
Next the concurrence matrix N N ׳of this design is as follows:
⎡ r1 + 1 λ 2 λ1 + 1 .... λ 2 ⎤
⎢ λ r1 + 1 λ2 .... λ1 + 1⎥⎥
⎢ 2
⎢λ + 1 λ 2 r1 + 1 .... λ2 ⎥
NN ' = ⎢ 1 ⎥
⎢ .... .... .... .... ..... ⎥
⎢ λ2 λ1 + 1 λ 2 ..... r1 + 1 ⎥
⎢ ⎥
⎣⎢ ⎦⎥
Here all the diagonal elements are r1+ 1, that is, r=r1+1. Further off-diagonal elements
are either λ11=λ2 or λ11=λ1 + 1. Since it is given that λ11=λ12=λ, and r=r1+1 and hence
NN’ becomes the concurrence matrix of a BIBD.
Example 2.4: Consider a SRGD design(SR-1) with parameters v=4, r=2, k=2, b=4,
m=2, n=2,, λ1=0, λ2=1,whose blocks are as follows:
Construction of Partially Balanced Incomplete Block Designs 73
1 3 4 2
2 4 1 3
Now by adding the group (2,2) to the above design we have another design whose
blocks are,
1 3 4 2 1 2
2 4 1 3 3 4
This gives the blocks of resolvable BIBD. Therefore the resulting design is a
resolvable BIBD with parameters v=4, r=3, k=2, b=6, λ=1.
Table 2.1
References