Completely Randomised

UNIT 9 COMPLETELY RANDOMISED Design

DESIGN
Structure
9.1 Introduction
Objectives
Planning of an Experiment
Classification of Experimental Designs
9.2 Basic Definitions of Experimental Design
9.3 Principles of Design of Experiments
Randomisation
Replication
Local Control
9.4 Size and Shape of the Plots
9.5 Completely Randomised Design
Layout of Completely Randomised Design
Statistical Analysis of Completely Randomised Design
Least Square Estimates of Effects
Variance of the Estimates
Expectation of Sum of Squares
9.6 Suitability of CRD
9.7 Summary
9.8 Solutions/Answers

9.1 INTRODUCTION
The modern concepts of experimental designs were primarily given by Ronald
A. Fisher in the 1920s and 1930s at “Rothamasted Experimental Station”, an
agricultural research station of London. In Fisher’s first book on design of
experiments, he showed how valid conclusions could be drawn efficiently
from experiments with natural fluctuation such as temperature, soil conditions
and rainfall, that is, in the presence of nuisance variables. The known nuisance
variables usually cause systematic biases in groups of results (e.g. batch-to-
batch variables). The unknown nuisance variables usually cause random
variability in the results and are called inherent variability or noise.
The experimental design was first used in an agricultural context, the method
has been applied successfully in the military and in industry since the 1940s.
Besse Day, working at U. S. Naval Experimentation Laboratory, used
experimental designs to solve problems such as finding the cause of bad welds
at the naval shipyards during World War II. George Box, employed by
Imperial Chemical Industries before coming to the United States, is a leading
developer of experimental design produced for optimizing chemical process.
W. Edwards Deming taught statistical methods, including experimental
designs, to Japanese scientist and engineers in the early 1950’s at a time when
“Made in Japan” meant poor quality. Genichi Taguchi, the most well known
of this group of Japanese scientists is famous for his quality improvement
methods. One of the companies where Taguchi first applied his methods was 5
Toyota. Since the late 1970’s, U.S. industry has become interested again in
Design of Experiments
quality improvement initiatives, now known as “Total Quality” & “Six-
sigma” programs. Design of experiments is considered an advanced method in
the six sigma programs, which were pioneered at Motorola & GE.
According to Bernad Ostle, “The design of experiment is, the complete
sequence of steps taken ahead of time to ensure that the appropriate data will
be obtained in a way which permits an objective analysis to valid inferences
with respect to stated problem”.
In any field of study either in life sciences or some other, it is essential to plan
an experiment, i.e. what is the object and which type of data is required. In
order to make use of time and energy spent on experiment, it should be
planned with a careful designing. Once a design of experiment is decided, the
observations are obtained from it and with the technique of analysis of
variance, the data is analysed.
Basic definitions of experimental design are described in Section 9.2. The
principles of design of experiments i.e. randomisation, replication and local
control are explained in Section 9.3 whereas size and shape of plots are
described in Section 9.4. In Section 9.5, layout and statistical analysis of
completely randomised design are explained. The least square estimates of
effects, variance of the estimates and expectation of sum of squares are also
given in Section 9.5. Suitability of CRD is descrived in Section 9.6.

Objectives
After studying this unit, you would be able to

 describe the experimental design;

 explain the planning and classification of experimental designs;
 describe the principles of design of experiments;
 explain the completely randomised design;
 describe the layout of CRD;
 explain the statistical analysis of CRD; and
 explain the advantages and disadvantages as well as the suitability of
CRD.
9.1.1 Planning of an Experiment
There are some basic points regarding the planning of an experiment, which
should be under consideration. These are as follows:
1. The Experiment should be Free from Bias
An experiment must be planned so that it gives an unbiased estimate of the
values we wish to measure. It is a matter of the design being such that no
bias on the part of the experimenter can possibly enter into the results.
This is achieved mainly by randomisation.
2. There must be a Measure of Error
The true experiment is one that is strictly objective. It should furnish a
measure of error and this error alone should be the measuring stick of
significance.

6
3. There must be a Clearly Defined Objective Completely Randomised
Design
For an experiment it is essential to specify the objects perfectly. In other
words the objective of the experiment should be clearly defined.
4. The Experiment should have Sufficient Accuracy
The accuracy of an experiment can be brought by the elimination of
technical errors and by increasing replications. The number of replications
should be decided to produce a given degree of accuracy.

9.1.2 Classification of Experimental Designs

Statisticians by themselves do not design experiments, but they have
developed a number of structured schedules called “experimental designs”,
which they recommend for the taking of measurements. These designs have
certain rational relationships to the purposes, needs and physical limitations of
experiments. Designs also offer certain advantages in economy of
experimentation and provide straightforward estimates of experimental effects
and valid estimates of variance. There are a number of ways in which
experiment designs might be classified, for example, the following:
1. By the number of experimental factors to be investigated (e.g., single-
factor versus multifactor designs)
2. By the structure of the experimental design (e.g., blocked, factorial,
nested, or response-surface design)
3. By the kind of information which the experiment is primarily intended to
provide (e.g. estimates of effects, estimates of variance, or empirical
mappings).

9.2 BASIC DEFINITIONS OF EXPERIMENTAL

DESIGN

Several fundamental terms are widely used throughout this section. They may
be defined as follows:
1. Treatment
In an experiment, there are some variants under study, the effects of which
are measured and tested (compared). These variants will be referred to as
treatments. For example, to test the effects of three fertilizers, i.e.,
Nitrogen, Phosphorus and Potash on the yield of a certain crop. Then
Nitrogen, Phosphorus and Potash are called treatments.
2. Yield
The response of the treatment is measured by some indicator such as crop
production, milk production, body temperature, mileage of engine set, etc.
Such an indicator is called yield. The treatments are applied to some units
such as field plots, sample of cows, sample of patients, sample of engine,
sets, etc. and the effect on the yield is observed.
3. Experimental Units
A unit to which one treatment applied is called experimental unit. It is the
smallest division of an experimental material to which the treatment
applied and on which the variable under study is measured. In carrying out 7
an experiment, we should clear as to what constitute the experimental unit.
Design of Experiments
It can be understood that in a field of agriculture it is called plot, in the
field of animal husbandry it may be a cow (cattle), in the field of medicine
it may be a patient and in the field of automobile industry it may be engine
set and so on.
4. Experimental Material
We have already explained the concept of experimental unit. The
experimental material is nothing but a set of experimental units. For
example, a piece of land, a group of cows, a number of patients and a
group of engine sets, etc. Actually, an experimental material is that
material on which some set of treatments are applied and tested.
5. Blocks
The experimental material is divided into a number of groups or strata
which are so formed that they are within homogeneous and between
heterogeneous. These groups or strata are called blocks.
6. Experimental Error
There is always a variation between the yields of the different plots even
when they get the same treatment. This variation exists due to non-
assignable causes, which cannot be detected and explained. These are
taken to be of random type. This unexplained random part of variation is
termed as experimental error. This include all types of extraneous
variation due to, (i) inherent variability in the experimental units, (ii) error
associated with the measurement made and (iii) lack of representativeness
of the sample of the population understudy.
7. Precision
The precision of an experiment is measured by the reciprocal of the
variance of a mean, i.e.
1 1 n
 2  2
v( x )  x 
As n, the replication number increases, precision also increases.
8. Uniformity Trial
We know that to increase the efficiency of a design, the plots should be
arranged into homogeneous blocks. It can be done only if we have a
correct idea about the fertility variation of the field. This is achieved
through uniformity trial. It is known that fertility of soil does not increase
or decrease uniformly in any direction but it is distributed over the entire
field in an erratic manner. By a uniformity trial, we mean a trial in which
the field (experimental material) is divided into small units (plots) and the
same treatment is applied on each of the units and their yields are
recorded. From these yields we can draw a fertility control map which
gives us a graphic picture of the variation of the soil fertility and enables
us to form a good idea about the nature of the soil fertility variation. This
fertility control map is obtained by joining the points of equal fertility
through lines.
A uniformity trial gives us an idea about the
1. Fertility gradient of the field,
2. Determination of the shape of the plots to be used,
3. Optimum size of plots,
4. Estimation of number of replications required for achieving certain
8
degree of accuracy.
 Completely Randomised
9.3 PRINCIPLES OF DESIGN OF EXPERIMENTS Design

Good experimentation is an art and depends heavily upon the prior knowledge
and abilities of the experimenter. Designing an experiment means deciding
how the observations or measurements should be taken to answer a particular
question in a valid, efficient and economical way. If a design is properly
designed, then there will exists an appropriate way of analsing the data. From
an ill-designed experiment no conclusion can be drawn.
The fundamental principles in design of experiments are the solutions to the
problems in experimentation posed by the two types of nuisance factors and
serve to improve the efficiency of experiments. For the validity of the design
Prof. R.A. Fisher gave three principles of design of experiments, those
fundamental principles are:

 Randomisation
 Replication
 Local Control

9.3.1 Randomisation
The principle of randomisation is essential for a valid estimate of the
experimental error and to minimize the bias in the results. In the words of
Cochran and Cox, “Randomisation is analogous to insurance in that it is a
precaution against disturbances that may or may not occur and they may or
may not be serious if they do occur”. Thus, randomisation is so done that each
treatment should get an equal chance. We mean that the treatments should be
allocated randomly, i.e., by the help of random numbers. The following are
the advantages of randomisations:
1. It provides a basis for the test of significance because randomisation
ensures the independence of the observations which is one of the
assumptions for the analysis of variance.
2. It is also a device for eliminating bias. Bias creeps in experiment, when
the treatments are not assigned randomly to the units. This bias may be
personal or subjective. The randomisation ensures the validity of the
results.

9.3.2 Replication
“Replication” is the repetition, the rerunning of an experiment or
measurement in order to increase precision or to provide the means for
measuring precision. A single replicate consists of a single observation or
experimental run. Replication provides an opportunity for the effects of
uncontrolled factors or factors unknown to the experimenter to balance out
and thus, through randomisation, acts as a bias-decreasing tool. Suppose a
pain relieving drug A is applied to 4 patients, we say that drug A is replicated
four times. By repeating a treatment it is possible to obtain a more reliable
estimate because it reduces the experimental error. Further by repeating a
treatment number of times we can judge the average performance of a
treatment and the situation becomes clearer. Basically there are following uses
of replication:
9
Design of Experiments
1. It enables us to obtain a more precise estimate of the treatments effects.
2. The next important purpose of replication is to provide an estimate of the
experimental error without which we cannot test the significance of the
difference between any two treatments. The estimate of experimental error
is obtained by considering the difference in the plots receiving the same
treatment in different replications and there is no other alternative of
obtaining this estimate.
3. For a desired amount of precision, the minimum number of replications
can be obtained.
9.3.3 Local Control
This method is used to attain the accuracy or to reduce the experimental error
without increasing unduly the number of replications. Local control is a
technique that handles the experimental material in such a way that the effects
of variability are reduced. In local control, experimental units are divided into
a number of homogeneous groups called blocks. These blocks are so formed
that they are homogeneous within and heterogeneous between. This blocking
of experiment may be row-wise, column-wise or both according to the
number of factors responsible for heterogeneity. Different types of blocking
constitute different types of experimental designs. The following are the
advantages of local control:
1. By means of local control, the experimental error is reduced considerably
and the efficiency of the design is increased.
2. By means of local control the test procedure becomes more sensitive or
powerful.
Besides the above three principles, there are some other general principles in
designing an experiment. Familiarity with the treatments and experimental
material is an asset. Selection of experimental site is an asset. Selection of
experimental site should be carefully done. Within block variability should be
reduced.

9.4 SIZE AND SHAPE OF THE PLOTS

In field experiments, the size and shape of plots as well as of blocks influence
the experimental error. The total available experimental area remaining fixed,
an increase in size of plots will automatically decrease the number of plots
and indirectly increases the block size. In order to reduce the flow of
experimental material from one plot to another, it is customary to leave out
strips of land between consecutive plots and also between blocks. These non-
experimental areas are known as guard area. The size and shape of the plot
should be such that we make a compromise between statistical and practical
requirements i.e. if plot size is x and the variance of the plot is V(x), then V(x)
is minimum (statistical consideration) and there should be no disturbance for
agricultural operations (practical requirements).
The size and shape of block will ordinarily be determined by the size and
shape of plots and the number of plots in a block. It is desirable from the point
of view of error control to have small variations among the plots within a
block and large variation among the blocks i.e. in general the division of
10
experimental material into blocks is made in such a way that plots within
blocks are as homogeneous as possible.
Different Experimental Designs Completely Randomised
Design
Based on these fundamental principles, we have certain designs. The analysis
of those designs is based on the theory of least squares which gives the best
estimates of the treatments effects and was initiated by Fisher (1926) followed
by Yates (1936), Bose & Nair (1939) and Rao (1976). The following three
designs are frequently used:
1. Completely Randomised Design
2. Randomised Block Design
3. Latin Square Design

9.5 COMPLETELY RANDOMISED DESIGN

The simplest of all the design is completely randomised design (CRD) which
is applied in the case when the experimental materials are homogeneous. CRD
is based on two principles i.e. randomisation and replication. The third
principle, i.e. local control is not used because it is assumed that experimental
materials are homogeneous. In this, the treatments are allocated randomly to
the experimental units and each treatment is assigned to different
experimental units completely at random (can be repeated any number of
times) that is why it is called completely randomised design.
Suppose we have k treatments under comparison and the ith treatment is to be
replicated ni times for i =1, 2, …, k, then the total number of units required for
k
the design are n   n i . We allocate the k treatments completely at random
i1
to n units such that ith treatment appears n i times in the experiment.

9.5.1 Layout of Completely Randomised Design

The term layout refers to the allocation of different treatment to the
experimental units. We have already said that treatments are allocated
completely at random to the different experimental units. Every experimental
unit has the same chance of receiving a particular treatment.
Suppose we want to test the effect of three pain relieving drugs A, B and C on
twelve patients. Then we first number all the patients (units) from 1 to 12.
Then from a random number table of one digit we pick up 12 numbers which
are less than 4. Suppose the numbers are 1, 3, 2, 1, 3, 2, 1, 3, 2, 2, 3, 1. Thus
the drug A is allotted to patient 1, drug C is allotted to patient number 2 and
so on. It can be shown below:
(1) (2) (3) (4) (5) (6)
A C B A C B
(7) (8) (9) (10) (11) (12)
A C B B C A

It is clear from the above layout that the replications of A, B and C are equal.
If the number of replications for each treatment is 5, 4 and 3 respectively, we
number the experimental units in a convenient way from 1 to 12. We then get
a random permutation of the experimental units. To the first 5 of the units in
the random permutation we assign treatment A, to the next 4 units treatment B 11
is assigned and the treatment C is assigned to the remaining 3 units.
Design of Experiments
9.5.2 Statistical Analysis of Completely Randomised Design
Statistical analysis of a CRD is analogous to the ANOVA for one-way
classified data for fixed effect model, the linear model (assuming various
effect to be additive) becomes
yij = µ + i + eij , i = 1, 2, 3, …, k; j = 1, 2, 3, ..., ni …(1)
where yij is the yield or response from the jth unit receiving the ith treatment, µ
is the general mean effect, i is the effect due to the ith treatment, where µ and
k
i are constants so that  n i α i = 0 and eij is identically and independently
i 1
k
distributed (i.i.d.) N(0, e2). Then, n = n i is the total number of
i 1
experimental units.

The analysis of model given in equation (1) is as same as that of fixed effect
model of one-way classified data, discussed in Unit 6 of MST-005. If we write

 y
i j
ij  y ..  G = Grand total of the n observations, and
ni

y ij  y i.  Ti. = Total response in the units receiving the ith treatment,

j1

Then, as in ANOVA (one-way classified data),

k ni 2 k ni 2 k

 y
i 1 j1
ij  y..     y ij  y i .  
i1 j1
 n y
i 1
i i.  y .. 
2

i.e. TSS = SSE + SST

where, TSS, SST and SSE are the Total Sum of Squares, Sum of Squares due
to Treatments (between treatments SS) and Sum of Square due to Error
(within treatment SS) given respectively by
k ni
TSS =  y
i 1 j1
ij  y.. 
2

k
SST =  n y
i 1
i i.  y..   S2T
2

k ni
and SSE =  y
i 1 j1
ij  y i.   S2E
2

ANOVA Table for CRD

Source of DF SS MSS Variance
Variation Ratio (F)
Treatments k-1 SST = ST2 S2T
MSST= MSST
k  1 FT =
Error n-k 2 MSSE
SSE = S E S2E
MSSE =
n  k 
Total n-1 S2T + S2E

12
Under the null hypothesis, H0 : 1 = 2 = …. = k against the alternative that Completely Randomised
Design
all ’s are not equal, the test statistic
MSST
FT = ~ F(k-1, n-k)
MSSE
i.e., FT follows F distribution with (k-1, n- k) df.
If FT  F(k−1, n− k) () then H0 is rejected at  level of significance and we
conclude that treatments differ significantly. If FT  F(k−1, n−k) () then 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.
If the treatments show significant effect then we would be interested to find
out which pair of treatments differs significantly. For this instead of
calculating Student's t-test for different pairs of treatment means we calculate
the least significant difference at the given level of significance. This least
difference is called as critical different (CD) and CD at α level of significance
is given by

CD = Standard error of difference between two treatment means 

tα/2 for error degrees of freedom.
We have
σ2 σ2 1 1 
Var ( yi.  y. j ) = e  e = σ e2  
n 
ni nj  i nj 
1/ 2
1 1 
Standard Error ( y i.  y. j ) = σ e  
n 
 i nj 
Hence, the critical difference (CD) for ( yi.  y. j )
1/ 2
 1 1 
= tα/2 (for error df)  MSSE  
n 
  i n j 
Since MSSE provides an unbiased estimate of σe2.
If each treatment is replicated n times, that is ni = n for i=1, 2, ..., k then
1/ 2
 2 
CD for difference of mean = (tα/2 for error df)  MSSE   
 n 
9.5.3 Least Square Estimates of Effects
The completely randomised model in equation (1) in Sub-section 9.5.2 is a
fixed effect model. Proceeding exactly as in Section 6.4 of Unit 6, we shall get
y..
̂ = = y.. and α̂ i = αi  y i.  y.. … (2)
n

9.5.4 Variance of the Estimates

Proceeding exactly as in Section 6.7 of Unit 6, we shall get
 e2 k
Var ( ̂ ) =
n
; where n = n
i 1
i … (3)

13
Design of Experiments
 
 
2 1 1 
and Var ( αi ) = Var( αi ) = σ e  … (4)
 ni k 



i 1
ni 

If we assume that each treatment is replicated an equal number of times i.e., if
k
ni = n, (say), i = 1, 2, …, k; then n = n
i 1
i = nk

Hence, from equations (3) and (4), we get

σ 2e  k 1 
Var ( ̂ ) = and Var ( α̂i ) = Var ( α i ) = σ e2   … (5)
nk  nk 

9.5.5 Expectation of Sum of Squares

Proceeding exactly as in Section 6.7 of Unit 6 [fixed effect model for one-way
classified data], we get
k k
E (SST) = E [  n i  y i.  y..  ] = (k-1)  2e +
2 2

i 1
n α
i 1
i i

 S2  1 k
E (MSST) = E  T  =  2e +  n i α i2 … (13)
 k 1  k  1 i1
2
E (SSE) = (n- k)  e
 S2 
E (MSSE) = E  E  =  2e … (14)
 n  k  
The method of analysis of completely randomised design would be similar to
one-way ANOVA, which has been illustrated below with the following
example:

Example 1: A person wanted to purchase a lot of electric drills. He got

quotations from five manufacturers. For the selection, he wanted to conduct
an experiment to estimate the time taken by each making a hole in a metallic
sheet. As the sheet might not be uniform all over in respect of thickness and
hardness, he marked 20 places on the sheet and applied four drills from each
concern in 4 randomly selected places to make holes. The time for making
each hole was recorded and these formed the observations. The observations
in seconds are shown below in brackets along with marks of the drills denoted
by D1, D2, D3, D4 and D5.

D1 19 D3 22 D 4 20 D1 20 

D5 29  D 2 24 D5 30  D3 24

D 2 26 D 4 25 D1 16 D 2 22 
D5 28 D3 25 D5 31 D 4 28

D 4 27  D1 16 D 2 27  D3 20

14 Conduct the experiment by adopting a completely randomised design.

Solution: The analysis of the given design is done by one-way analysis of Completely Randomised
variance method. The data is analysed and computation results are given as Design
below:
The totals of time records for 4 holes by each of the different makes are
denoted by T1, T2, T3, T4 and T5 are shown below.
T1 = 71, T2 = 99, T3 = 91, T4 = 100, T5 = 118
Grand Total (G) = 479

G 2 479
2
Correction Factor (CF)    11472.05
N 20
Total Sum of Squares (TSS) = 212+ 182 + 222 +… +312 + 202 − 11472.05
= 11847 – 11472.05 = 374.95
Sum of Squares due to Makes (SSM)

=
712  992  912  1002  1182  11472.05
4
= 11761.75 – 11472.05 = 289.70
Sum of Squares due to Error (SSE) = TSS – SSM
= 374.95 – 289.70 = 85.25
Analysis of Variance Table
Sources of Variation DF SS MSS F
Makes 4 289.7 72.425 12.75
Error 15 85.25 5.68
Total 19 374.95

The tabulated value of F at 1 per cent level of significance for 4 and 15 df is

4.89. Thus, the calculated value of F viz. 12.75 shows that Make to Make
variation is highly significant thereby indicating that the hypothesis that the
time periods taken by the different Makes in boring a hole are, on an average,
the same, is rejected. So multiple camparison test will be applied for different
Makes.
Mean for Different Makes
Makes

D1 D2 D3 D4 D5

17.74 24.75 22.75 25.00 29.50

2MSSE 2  5.68
SE =   1.69
n 4
Critical difference at 1 % level of significance
15
CD = t/2 (for error df)  SE = 3.055  1.69 = 5.16
Design of Experiments
The initial difference indicates that the Make D5 is significantly better than all
the other Makes.
Pair of Difference CD Inference
Treatments
D1, D2 D1  D 2 = 7.01 5.16 Significant

D1, D3 D1  D 3 = 5.01 5.16 Insignificant

D1, D4 D1  D 4 = 7.26 5.16 Significant

D1, D5 D1  D 5 =11.26 5.16 Significant

D2, D3 D 2  D3 = 2.00 5.16 Insignificant

D2, D4 D 2  D 4 = 0.25 5.16 Insignificant

D2, D5 D 2  D5 = 4.75 5.16 Insignificant

D3, D4 D3  D 4 = 2.25 5.16 Insignificant

D3,D5 D3  D5 = 6.75 5.16 Significant

D4, D5 D 4  D5 = 4.5 5.16 Insignificant

E1) Carryout the ANOVA for the given following data of yields of 5
varieties, 7 observations on each variety:
Observations
Variety
1 2 3 4 5 6 7
1 13 15 14 14 17 15 16
2 11 11 10 10 15 9 12
3 10 13 12 15 14 13 13
4 16 18 13 17 19 14 15
5 12 12 11 10 12 10 10

9.6 SUITABILITY OF CRD

The following are some situations, in which one can apply the complete
randomised design:
1. The CRD is used in the situations where experimental materials are
homogeneous. That is why, CRD is mostly used in chemical, biological
and banking experiments, where the experimental material is thoroughly
mixed powder, liquid or chemical.
2. The CRD is used in the situations where the observations on some units
are missing or destroyed. This feature of missing observation does not
disturb the analysis of the design.
16
3. In agricultural experiments, this design is not used because experimental Completely Randomised
material is not homogeneous. Design

9.6.1 Advantages and Disadvantages of CRD

Advantages of CRD
1. In this design any number of treatments and replications can be used.
There may be different number of replications for different treatments.
2. Analysis is simple and easy even if the number of replication is unequal
for each treatment. In such case experimental error will differ from
treatment to treatment.
3. If some of the observations are missing or destroyed or not available due
to some reasons, the analysis can be done without any problem.
4. It provides large degree of freedom for error sum of squares. This
increases the sensitivity of the experiment.
5. In CRD there is no condition on the number of replication of the
treatments, they can be increased or decreased according to the need of the
experimenter. Thus, the design is flexible.
Disadvantages of CRD
1. The main disadvantage of CRD is that the principle of local control has
not been used in this design. Due to this fact, the experimental error is
inflated. This is the main reason for the criticism of CRD.
2. In agricultural experiments, the design is seldom used because the
experimental material is not homogenous.

9.7 SUMMARY
In this unit, we have discussed:
1. The experimental design;
2. The planning and classification of experimental designs;
3. Principles of design of experiments;
4. Completely randomised design;
5. Layout of CRD;
6. The statistical analysis of CRD; and
7. Advantages and disadvantages as well as suitability of CRD.

9.8 SOLUTIONS / ANSWERS

E1) The analysis of the given design is done by one-way analysis of variance
method. The data is analysed and computation results are given as
below:
Correction Factor (CF) = 6072.03
Raw Sum of Squares (RSS) = 6293
Total Sum of Squares (TSS) = 220.97 17
Design of Experiments
Sum of Squares due to Variety (SSV) = 138.40
Sum of Squares due to Error (SSE) = TSS – SSV
= 220.97-138.40 = 82.57
ANOVA Table
Source of Variance Ratio
DF SS MSS
Variation Calculated Tabulated
Variety 4 138.40 34.60
Error 30 82.57 2.75 12.58 2.66
Total 34 220.97

Null Hypothesis H0: 1 =  2 = … = 5

Since, calculated value of F is greater than the tabulated value of F, we
reject the null hypothesis and conclude that variety effects are
significantly different.
Mean for Different Varieties
Varieties
D1 D2 D3 D4 D5
14.86 11.14 12.86 16.00 11.00
2MSSE 2  13.34
SE =   1.95
n 7
Critical difference at 1 % level of significance
= t/2 (for error df)  SE = 3.055  1.95 = 5.96
The initial difference indicates that the Variety D4 is significantly better
than all the other Varieties.
Pair of Difference CD Inference
Treatments
D1, D2 D1  D 2 = 3.72 5.96 Insignificant

D1, D3 D1  D 3 = 2.00 5.96 Insignificant

D1, D4 D1  D 4 = 1.14 5.96 Insignificant

D1, D5 D1  D 5 =3.86 5.96 Insignificant

D2, D3 D 2  D3 = 1.72 5.96 Insignificant

D2, D4 D 2  D 4 = 4.86 5.96 Insignificant

D2, D5 D 2  D5 = 0.14 5.96 Insignificant

D3, D4 D3  D 4 = 3.14 5.96 Insignificant

D3,D5 D3  D5 = 1.86 5.96 Insignificant

D4, D5 D 4  D5 = 5.00 5.96 Insignificant

18
Completely Randomised
Design

19