5.

Randomized block design

With a randomized block design, the experimenter divides subjects into subgroups called
blocks, such that the variability within blocks is less than the variability between blocks. Then,
subjects within each block are randomly assigned to treatment conditions. Compared to
a completely randomized design, this design reduces variability within treatment conditions
and potential confounding, producing a better estimate of treatment effects.
The randomized block design evaluates differences among more than two groups that contain
matched samples or repeated measures that have been placed in blocks. Blocks are
heterogeneous sets of items or individuals that have been either matched or on whom repeated
measurements have been taken. Blocking removes as much variability as possible from the
random error so that the differences among the groups are more evident.
Although blocks are used in a randomized block design, the focus of the analysis is on the
differences among the different groups. As is the case in completely randomized designs,
groups are often different levels pertaining to a factor of interest. A randomized block design
is often more efficient statistically than a completely randomized design and therefore produces
more precise results.

Model
The model for RBD is
𝑌𝑖𝑗 = 𝜇 + 𝛼𝑖 + 𝛽𝑗 + 𝜖𝑖𝑗 for 𝑖 = 1, 2, . . . , 𝑡; 𝑗 = 1, 2, . . . , 𝑟
Here 𝑌𝑖𝑗 is the jth observation for the ith treatment, μ is the grand mean, the ith treatment effects
𝛼𝑖 are subject to the restriction ∑𝑡𝑖=1 𝛼𝑖 = 0 and the 𝜖𝑖𝑗 are values of independent random
variables having identical normal distribution with zero means and the variance 𝜎 2 .

Advantages and disadvantages

Advantages
(i) Accuracy: Blocking can increase precision by removing one source of variation from
experimental error.
(ii) Flexibility: There is no restriction on the number of treatments or number of blocks so
long as each treatment is replicated the same number of times in each replication.
(iii) Easy computation: Statistical analysis is relatively simple. Moreover any number of
treatments may be omitted from the analysis without complicating it.
(iv) It is possible to separate sum of squares for error into components corresponding to
particular treatment effect.

Disadvantages
(i) The efficiency of the design decreases as the number of treatments and, hence, block
size increases.
(ii) Missing data can cause some difficulty in the analysis.
 (iii) The design is less efficient than others in the presence of more than one source of
variation.

Hypothesis
The null hypothesis for this design is against alternative that all treatments are not equal.

Layout Plan
Following steps is to be followed for layout of the experiment with 8 treatments and 3
replications.
(i) Form three blocks of eight plots each, perpendicular to the gradient.
(ii) Number the plots from 1 to 8 within each block.
(iii) Assign the treatments randomly to the units of first block.
(iv) Repeat step (iii) for remaining three blocks.
(v) A separate randomization is used in each block.

Block I Block II Block III

1 T4 1 T7 1 T3
2 T7 2 T3 2 T7
3 T3 3 T6 3 T1
4 T2 4 T2 4 T4
5 T5 5 T8 5 T6
6 T1 6 T5 6 T2
7 T6 7 T4 7 T8
8 T8 8 T1 8 T5
Gradient
Fig.4.1 Assignment of numbers of plots and treatments with in each block

Structure of ANOVA table and analysis for RBD

Let there be t treatments replicated r times each.
Table 5.1: Structure of ANOVA for RBD
Source of Degree of Sum of Mean Sum of Square F
Variation Freedom (df) Square(SS) (= SS)
df
Replications 𝑟−1 SSR SSR 𝑠𝑅2
𝑠𝑅2 =
r−1 𝑠𝑒2
Treatment 𝑡−1 SST SST 𝑠𝑇2
𝑠𝑇2 =
t−1 𝑠𝑒2
Error (𝑡 − 1)(𝑟 − 1) SSE SSE
𝑠𝑒2 =
t(r − 1)
Total rt-1 TSS
Analysis: Each treatment repeated r times. Total observations
(𝐺𝑇)2
Grand total (GT) = ∑𝑡𝑖 ∑𝑟𝑗 𝑋𝑖𝑗 Correction factor = 𝑁
Total Sum of Squares (TSS) = ∑𝑡𝑖 ∑𝑟𝑗 𝑋𝑖𝑗
2
− 𝐶𝐹
 ∑𝑡𝑖 𝑇𝑖2
Treatment Sum of Squares (SST) = − 𝐶𝐹
𝑟
∑𝑟𝑗 𝑅𝑗2
Replication Sum of Squares (SSR) = − 𝐶𝐹
𝑡
Error Sum of Squares (SSE) = TSS- SSR-SST
Mean Sum of Squares and F test are computed as given in Table 5.1.
If Fcalculated ≥ F∝%,[(𝑡−1),(𝑡−1)(𝑟−1)] , then H0 is refused at level of significance and we
conclude that treatments differ significantly. If Fcalculated < F∝%,[(𝑡−1),(𝑡−1)(𝑟−1)] , H0 may be
accepted, i.e. the data do not provide any evidence to prefer one treatment to the other and as
such all of them can be considered alike.

Standard Error of difference

When the null hypothesis is rejected the treatment differences are compared through comparing
the means by critical difference.
Standard Error of difference between any two treatment means is given by
1
2𝑆𝑒2 2
𝑆𝐸(𝑡𝑖̅ − 𝑡𝑗̅ ) = [ ]
𝑟

Critical Difference (CD) for the significance of the difference between any two treatment
means at 𝛼% level of significance is
1
2𝑆𝑒2 2
𝐶𝐷(𝑡𝑖̅ − 𝑡𝑗̅ ) = 𝑡(𝑒𝑟𝑟𝑜𝑟 𝑑𝑓,𝛼/2%) × [ ]
𝑟

Relative Efficiency of RBD relative to CRD

𝑟(𝑡−1)𝑆𝑒2 +(𝑟−1)𝑆𝑏2
𝑅𝐸 = (𝑟𝑡−1)𝑆𝑒2
If blocking has been effective in increasing precision, then RE will be greater than one. The
quantity (𝑅𝐸 − 1) × 100 measures the percentage increase in precision due to blocking.

Estimation of Missing value in RBD

If only one observation is missing the procedure of analysis the experiment is first estimate the
value using the formula
𝑡𝑇𝑖.′ +𝑟𝑅.𝑗
′
−𝐺 ′
𝑦̂𝑖𝑗 = (𝑡−1)(𝑟−1)

Where 𝑇𝑖.′ = the sum of known observations getting ith treatment

𝑅.𝑗′ = the sum of known observations in jth block

𝐺 ′ = the sum of all the known observations

then substitute the estimated value 𝑦̂𝑖𝑗 in the missing place and carried out the ANOVA as
usual.
Because of the substitution, however, one degree of freedom is lost from both the error
and the total in the ANOVA. Note that the substitution of the value adds no information to the
experiment. Its only purpose is to make the data orthogonal for valid analysis of variance. It
should not be used in the computation of means.
 When two or three values, say 𝑎, 𝑏, 𝑐 are missing, we first guess a value for 𝑏 and 𝑐 and
use the formula for to find an approximation for 𝑎. With this value for 𝑎 and guessed value for
𝑐 we use the formula to solve for 𝑏. Then with these two values we use the equation to find an
approximation for 𝑐. This cycle is repeated until the new values do not differ appreciably from
previous ones. Then the ANOVA is computed with these values. One degree of freedom for
each substituted value must be subtracted from the total and error lines in the ANOVA.
When all of the missing values are in the same block or treatment the simplest solution
is to act as if the block or treatment had not been included in the experiment. The ANOVA is
run with the entire block or treatment omitted.

Table 5.2 Structure of ANOVA for RBD when k observations are missing and estimated
Source of Degree of Freedom Sum of Mean Sum of Square F
Variation Square
𝑆𝑆𝑅 S2R
Replications 𝑟−1 SSR 𝑆𝑅2 = (𝑟−1)
S2e
Treatment 𝑡−1 SST 𝑆𝑆𝑇 S2T
𝑆𝑇2 = (𝑡−1) S2e
Error (𝑡 − 1)(𝑟 − 1) − 𝑘 SSE 𝑆𝑆𝐸
𝑆𝑒2 = (𝑡−1)(𝑟−1)−𝑘
Total 𝑟𝑡 − 1 − 𝑘 TSS

The adjusted Treatment Sum of Squares (SST) is obtained by subtracting the adjustment factor
from the treatment SS.
2
′
(𝑅.𝑗 +𝑡𝑇𝑖.′ −𝐺 ′ )
𝐴𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡 𝑓𝑎𝑐𝑡𝑜𝑟 = 𝑡(𝑡−1)(𝑟−1)2

If the treatments show significant effect, then the S.E. of difference between two treatment
means is
1
2 𝑡 2
𝑆𝐸𝑑 = [𝑆𝑒2 (𝑟 + 𝑟(𝑟−1)(𝑡−1))]

If one of the treatments corresponds to the missing observation.

We subtract 1 d.f. for each estimated value from the Total SS and Error SS.