Unit 10 Randomised Block Design: Structure
Unit 10 Randomised Block Design: Structure
10.1 INTRODUCTION
The completely randomised design was simple due to the reason that principle
of local control was not used and it was assumed that the experimental
material is homogeneous, but it is observed that the experimental material is
not fully homogeneous. In agricultural field experiments sometimes a fertility
gradient is present in one direction. In such situation the simple method of
controlling variability of the experimental material consist in stratifying or
grouping the whole experimental area into relatively homogeneous strata or
sub-groups (called blocks), perpendicular to the direction of fertility gradient.
These blocks are so formed that plots within a block are homogeneous and
between blocks are heterogeneous. In other words, there may be less
variation within a block and major difference or variation between blocks. It is
to be kept in mind that familiarity with the nature of experimental units is
necessary for an effective blocking of the material. The procedure of division
of experimental material into a number of blocks give rise to a design known
as Randomised block design (RBD) which can be defined as an arrangement
of t treatments in r blocks such that each treatment occurs precisely once in
each block.
In other words, when the experimental units are heterogeneous, a part of the
variability can be accounted for by grouping the experimental units in such a
way that those experimental units within each group are as homogeneous as
possible. The treatments are then allotted randomly to the experimental units
within each group (or block). This results in an increase in precision of
estimates of the treatment contrasts, due to the fact that error variance that is a
function of comparisons within blocks is smaller because of homogeneous
blocks.
Layout and statistical analysis of randomised block design are explained in
Sections 10.2 and 10.3. The least square estimates of effects, variance of the
estimates and expectation of sum of squares are also given in Section 10.3.
19
Design of Experiments
Missing plots techniques in RBD for one and two missing plots are described
in Section 10.4 whereas the suitability of RBD is explored in Section 10.5.
Objectives
After studying this unit, you would be able to
explain the randomised block design;
describe the layout of RBD;
explain the statistical analysis of RBD;
find out the missing plots in RBD; and
explain the advantages and disadvantages as well as the suitability of
RBD.
Block I A B D C
Block II C A D B
Block III D B C A
20
If we write that Randomised Block Design
y
i j
ij y.. G = Grand total of all the p q observations.
= q ( y i. y.. ) 2 p ( y. j y.. ) 2 ( y ij y i. y. j y .. ) 2
i j i j
The product terms vanish since the algebraic sum of deviations from mean is
zero. Thus
TSS = SSE + SSB + SST
where TSS, SST, SSB and SSE are the total sum of squares, sum of squares
due to treatments (between treatments SS), sum of squares due to blocks and
sum of squares due to error (i.e., within treatment SS) given respectively by
TSS = ( y ij y.. )2
i j
MSSE = MSSB
Error (p−1)(q−1) SE2 FB =
S2E /( p 1)(q 1) MSSE
Total pq−1
21
Design of Experiments
Under the null hypothesis, H0 : 1 = 2 = …. = p against the alternative that
all ’s are not equal, the test statistic
MSST
FT = follows F [(p −1), (p −1) (q −1)]
MSSE
i.e., FT follows F-distribution with [(p −1), (p −1) (q −1)] df.
If FT F with [(p −1), (p −1) (q −1)] df at level of significance, (Usually
5% ) then H0 is rejected and we conclude that treatments differ significantly.
If FT F with [(p −1), (p −1) (q −1)] df at level of significance then H0 may
be accepted, i.e. the data do not provide any evidence against the null
hypothesis which may be accepted.
Similarly, under the null hypothesis, H0 : 1 = 2 = …. = q against the
alternative that all ’s are not equal, the test statistic
MSSB
FT = follows F[(q −1), (p −1) (q −1)]
MSSE
and we can discuss its significance as explained above.
σ e2
Var ( ̂ ) =
pq
( p 1) 2
Var ( α̂ i ) = σe
pq
(q 1) 2
and Var( j ) = σe .
pq
10.3.3 Expectation of Sum of Squares
Proceeding exactly as in CRD, we get
E[SST] = (p 1)σ 2e q αi2
i
(SST) q
E E(MSST) σ e2 i αi2
( p 1) ( p 1)
E(SSB) = (q 1)σ e2 p β 2j
j
(SSB) p
E E(MSSB) σ 2e j β 2j
( q 1) ( q 1)
22
E(SSE) = (q 1)(p 1)σ e2 Randomised Block Design
(SSE ) 2
E
( q 1)( p 1) E(MSSE ) σ e
Hence under the null hypothesis
H0 : 1 = 2 = …. = p = 0;
H0 : 1 = 2 = …. = q = 0
E (MSST) e2 and E(MSSB) e2
i.e. each of the mean sum of squares due to treatments and blocks gives an
unbiased estimate of the error variance e2 under the null hypothesis H0 and
H0 respectively.
Example 1: There were 4 different makes of cars. A problem was posed to
estimate the petrol consumption rates of the different makes of cars for
suitable average speed and compare them. The following experiment could be
conducted for an inference about the problem:
Five different cars of each four makes were chosen at random. The five cars
of each make were put on road on 5 different days. The cars of A make run
with different speeds on different days. The speeds were 25, 35, 50, 60 and 70
mph. Which car was to put on the road on which day and what speed it should
have was determined through a chance mechanism subject to the above
conditions of the experiment. The procedure was adopted for each of the
makes of cars. For each car, the number of miles covered per gallon of petrol
was observed. The observations are presented below:
23
Design of Experiments
Here,
(337.1) 2
Correction Factor (CF) = 5681.82
20
Raw Sum of Squares = (20.6)2 (19.5)2 .... (13.7)2 (12.7)2 = 5781.41
Total Sum of Squares (TSS) = 5681.41 -5681.82 = 99.59
Sum of Squares due to Speed (SSS)
Variance Ratio
Source of
DF SS MSS
Variation
Calculated Tabulated
Treatments
(Makes) 3 28.78 9.59 23.97 3.49
Total 19 99.59
In both the cases either for speeds or for makes, calculated value of F is
greater than tabulated value of F at 5% level of significance and thus null
hypothesis is rejected.
In the above experiment, we are interested only on makes so multiple
camparison test will be applied for different makes.
2MSSE 2 0.40
SE = 0.40
5 5
The initial difference indicates that the Make A is significantly better than all
the other Makes.
Pair of Difference CD Inference
Treatments
A, B A B = 1.44 1.22 Significant
Blocks
Varieties
I II III IV
A 7 16 10 11
B 14 15 15 14
C 8 16 7 11
Solution: Let us find the block and variety totals by the following table:
Blocks
Varieties Total
I II III IV
A 7 16 10 11 44
B 14 15 15 14 58
C 8 16 7 11 102
Total 29 47 32 36 144
(144) 2
Correction Factor (CF) = 1728
12
Raw Sum of Squares (RSS) = (7)2 (14) 2 ... (14) 2 (11) 2 = 1858
Total Sum of Squares (TSS) = 1858 − 1728 = 130
25
Design of Experiments
(29) 2 (47) 2 (32)2 (36) 2
Block Sum of Squares (SSB) = CF
3
841 2209 1024 1296
= 1728
3
1790 1728 = 62
(44) 2 (58)2 (42) 2
Variety Sum of Squares (SSV) = CF
4
1936 3364 1764
= 1728
4
1766 1728 = 38
Sum of Squares due to Error (SSE) = TSS− SSV − SSB
= 130 − 62 −38 = 30
ANOVA Table
Source of Variance Ratio
DF SS MSS
Variation Calculated Tabulated
Variety 2 38 19 3.8 5.14
Blocks 3 62 20.67 4.13 4.76
Error 6 30 5
Total 11 130
In both these cases either for varieties or for blocks, calculated value of F is
less than tabulated value of f at 5% level of significance and thus null
hypothesis is accepted and inferred that variety effect and block effect are
insignificant.
where,
B1' = total of all available (p −1) observations in 1st block
2 (G ' Y )2
Total Sum of Squares (TSS) = y ij Y2 where (i,j) (1,1)
i j pq
Sum of Squares due to Error (SSE) = TSS – SSB – SST
(G ' Y )2 (B1' Y) 2 (T1' Y)2
SSE = Y 2 terms not involving Y
pq p q
For obtaining the value of Y, we minimize the sum of squares due to error
with respect to Y. This is obtained by solving the equation
27
Design of Experiments
(SSE ) 2(G ' Y) 2(B1' Y) 2(T1' Y )
2Y 0
Y qp p q
Solution: We convert the two missing plots problems into one missing plot
problem, for which we take the average of the values of I block in which x is
missing. This average is (10+12)/2 = 11. Thus, the estimate of x is taken to be
x1=11 and it is inserted in place of x and form the following table of totals:
Blocks
Treatments Total
I II III
A 12 14 12 TA = 38
B 10 y 8 TB = 18 + y
C 11 15 10 Tc = 36
Total B1 = 33 B2 = 29 + y B3 = 30 G = 92 + y
28
Thus, from the above table we get Randomised Block Design
54 87 92 49
=12.25 12
4 4
Blocks
Treatments Total
I II III
A 12 14 12 TA = 38
B 10 12 8 TB = 30
C x 15 10 Tc = 25 + x
Total B1 = 22+x B2 = 41 B3 = 30 G = 93 + x
Thus from the above table we get p = 3, q = 3, B1' =22, Tc' = 25, G ' = 93
Again applying the missing estimation formula
3 25 3 22 93
x
4
75 66 93 48
12
4 4
Thus, x2 = 12
Again using x2 =12, we estimate the second estimate of y i.e. y2 for which
B'2 = 29, TB' = 18, G ' = 92
3 18 3 29 93
ŷ
4
54 87 93 47
11.75 12
4 4
We see that the second estimate of y i.e. y2 is not materially different from y1.
Thus, we take the estimated values of x 12 and y = 12. Inserting both the
estimated values of x and y we get the following observations:
29
Design of Experiments
Blocks
Treatments Total
I II III
A 12 14 12 TA = 38
B 10 12 8 TB = 30
C 12 15 10 Tc = 37
Total B1 = 34 B2 = 41 B3 = 30 G = 105
(105) 2 11025
Correction Factor (CF) = 1225
9 9
Raw Sum of Square (RSS) = (12) 2 (10) 2 .... (8) 2 (10) 2 = 1261
Total Sum of Squares (TSS) = 1261-1225 = 36
(38) 2 (30) 2 (37) 2
Treatment Sum of Squares (SST) = CF
3
1444 900 1369
= 1225
3
3713
1225 1237.67 1225
3
= 12.67
(34)2 (41) 2 (30) 2
Block Sum of Squares (SSB) = CF
3
1156 1681 900
= 1225
3
1245.67 1225 = 20.67
Error Sum of Squares (SSE) = TSS − SST − SSB
= 36 − 12.67− 20.67 = 2.66
ANOVA Table
Variance Ratio
Source of
DF SS MSS
Variation
Calculated Tabulated
Total 8-2 =6
Treatments
Blocks
A B C D
2. The RBD is suitable only when the number of treatments is small because
as the number of treatments increases, the block size also increases and it
disturbs the homogeneity of the block.
3. RBD is suitable only when experimental material is heterogeneous with
respect to one factor only. If there is two-way heterogeneity, LSD is used.
31
Design of Experiments
Disadvantages of RBD
1. The main disadvantage of RBD is that if the blocks are not internally
homogeneous, then a large error term will result. In field experiments, it is
usually observed that as the number of treatments increases, the block size
increases and so one has lesser control over error.
2. The number of replications for each treatment is same. If replication is not
same, the only remedy is to adopt CRD.
3. It cannot control two sided variation of experimental material
simultaneously. That is why, it is not recommended when experimental
material contains considerable variability.
10.6 SUMMARY
In this unit, we have discussed:
1. The randomised block design;
2. The layout of RBD;
3. The statistical analysis of RBD;
4. The missing plot techniques in RBD; and
5. The advantages and disadvantages as well as the suitability of RBD.
ANOVA Table
2MSSE 2 0.97
SE = 0.80
q 3
E2) We have p = 3, q = 4, B'3 = 213, T2' = 337, G ' = 1207 and the value of
y = 109
Therefore,
Correction Factor = 144321.33
Raw Sum of Squares = 144442.00
Total Sum of Squares = 120.67
Treatment Sum of Squares = 76.67
Block Sum of Squares = 20.67
Error Sum of Squares = 23.33
ANOVA Table
Block means are not but treatment means are significantly different at
5% level of significance.
33
Design of Experiments
In the above experiment, we are interested only treatments, so multiple
camparison test will be applied for different treatments.
For pairwise testing, we have to find the standard error of difference of
two treatment means:
2MSSE 2 4.66
SE 1.76
p 3
CD SE t 2 at error df
= 1.76 2.447 4.31
Treatment means are
Difference of
Pair of
Treatment CD Inference
Treatments
Means
A, C A C 0 .3 4.31 Insignificant
A, D A D 2 .3 4.31 Insignificant
B, D B D 3 .7 4.31 Insignificant
34