DESIGN OF EXPERIMENTS

ANKITHA C S
❑ EXPERIMENTAL DESIGN

▪ Experimental design is the branch of statistics that deals with the

design and analysis of experiment.

▪ Design of experiment is a powerful data collection and analysis tool

that can be used in a variety of experimental situation

▪ Statistical design of experiments refers to the process of planning the

experiment so that appropriate data will be collected and analyzed by
statistical methods, resulting in valid and objective conclusions.
❑ TERMINOLOGY
❖EXPERIMENT :-
An experiment is a planned enquiry to obtain new facts or to confirm or to
deny the facts established earlier. In other words An experiment is a device or
means of getting an answer to the problem under consideration.
❖TREATMENT :-
The object of comparison is termed as treatment. For example, treatments
may be different stocking rates, feeds, fish species etc.
❖EXPERIMENTAL UNIT :-
An experimental unit is the unit of material to which a treatment is applied. In
other words The smallest division of the experimental material to which we
apply the treatments and on which we make observations on the variable
under study is termed as experimental unit
For example, the experimental unit may be pond , a cement cistern, a fish , an
animal, a piece of land etc.
❖EXPERIMENTAL ERROR :-
Experimental error is a measure of variation that exists among the
experimental units treated alike. There are two main sources of
experimental error
(i) Inherent variability in the experimental material (unit) to which
treatments are applied
(ii) variability due to lack of uniformity in the physical conduct od
experiment or in other words , failure to standardise the
experimental technique.
Experimental error provides a basis for the confidence to be placed in the
results obtained from the experiment. Therefore it is necessary to control
the experimental error. Under the principles of experimental designs are
helpful in reducing the experimental error.
❑ BASIC PRINCIPLES OF EXPERIMENTATION

According to Prof. Ronald .A. Fisher the basic principles of the design of
experiments are
1) RANDOMISATION
2) REPLICATION
FISHER’S DIAGRAM
3) LOCAL CONTROL
REPLICATION

RANDOMISATION LOCAL CONTROL

VALIDITY OF DIMINUTION OF
ESTIMATION OF ERROR
ERROR
✓AIM OF BASIC PRINCIPLES :-
i. To estimate the experimental error
ii. To reduce the experimental error

➢RANDOMISATION :-
• Randomisation is the random process of allocation of treatments to
various experimental units
• Randomisation ensures that all the experimental units have an equal
chance of receiving a particular treatment .
• It’s function is to provide unbiased estimates of treatment means and
experimental error.
• Randomisation is not an error elimination process, it is only a error
reduction process.
➢REPLICATION :-
• Repetition of the treatment under investigation is known as replication
• It also provide an estimate of experimental error and to improve the precision
of the treatment effects and hence of the experiment
• In randomisation we are getting an unbiased estimate but by replicates we
are reaching to consistent estimate
• As the precision of the experiment increases with the increase in the
number of replications
➢LOCAL CONTROL :-
• Local control is a devise of grouping experimental units into groups of
relatively homogenous units
• Local control helps in reducing the experimental error and increases
efficiency of the design
❑ EXPERIMENTAL DESIGNS

1. Completely Randomised Design (CRD)

2. Randomised complete Block Design (RBD)
3. Latin Square Design (LSD)
➢ Completely Randomised Design (CRD)
• This is the simplest of all designs which uses two principles of experimental
designs, namely replication and randomisation.
• Each treatment is applied randomly to a few experimental units
• It is not necessary that the number of replications for each treatment be
the same.
• This design is suitable only when the experimental units receiving
treatments are homogenous
• Hence the design is mostly used in laboratory experiments
oLAYOUT OF THE DESIGN :-
• The term layout refers to the placement of treatments on the experimental
units
• For example :-
It is planned to compare 4 treatments 𝐴,𝐵,𝐶, and 𝐷 with 5 replicates of
each treatment. Then 20 experimental units are required. These experimental
units are numbered from 1 to 20.Draw 5 numbers, they may turn out to be ,
say, 8,19,7,12, and 3. The experimental units bearing these numbers will
receive say treatment 𝐴.
The second set of 5 random numbers to which say treatment 𝐵 will be
applied is drawn, which may turn out to be say 1,4,13,18 and 20. The third set
of 5 random numbers is drawn which may turn out to be 9,15,17,6 and 14 and
the experimental units bearing these numbers receive say treatment 𝐶.The
treatment 𝐷 is applied to the remaining 5 experimental units.
One of the possible layouts of the design is shown below

𝟏𝑩 𝟐𝑫 𝟑𝑨 𝟒𝑩
𝟓𝑫 𝟔𝑪 𝟕𝑨 𝟖𝑨
𝟗𝑪 𝟏𝟎𝑫 𝟏𝟏𝑫 𝟏𝟐𝑨
𝟏𝟑𝑩 𝟏𝟒𝑪 𝟏𝟓𝑪 𝟏𝟔𝑫
𝟏𝟕𝑪 𝟏𝟖𝑫 𝟏𝟗𝑨 𝟐𝟎𝑩
oANALYSIS
• Let 𝒙𝒊𝒋 denote the observation on the 𝑖𝑡ℎ treatment from 𝑗𝑡ℎ replication
(𝑖 = 1,2,……𝑡 , 𝑗 = 1,2,……….𝑟)
• For analysis of this design, the following additive model is used:
𝒙𝒊𝒋 = 𝒎 + 𝒂𝒊 + 𝒆𝒊𝒋 where 𝑚 = 𝑔𝑒𝑛𝑒𝑟𝑎𝑙 𝑚𝑒𝑎𝑛
𝑎𝑖 = 𝑒𝑓𝑓𝑒𝑐𝑡 𝑑𝑢𝑒 𝑡𝑜 𝑖𝑡ℎ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡
𝑒𝑖𝑗 = 𝑒𝑥𝑝𝑒𝑟𝑖𝑚𝑒𝑛𝑡𝑎𝑙 𝑒𝑟𝑟𝑜𝑟
• Experimental error (𝑒𝑖𝑗 ) is independently normally distributed with mean 0 and
variance 𝜎 2
• The Null hypothesis to be tested is,
𝑯𝟎 : 𝑻𝒉𝒆𝒓𝒆 𝒊𝒔 𝒏𝒐 𝒔𝒊𝒈𝒏𝒊𝒇𝒊𝒄𝒂𝒏𝒕 𝒅𝒊𝒇𝒇𝒆𝒓𝒆𝒏𝒄𝒆 𝒂𝒎𝒐𝒏𝒈 𝒕𝒓𝒆𝒂𝒕𝒎𝒆𝒏𝒕 𝒆𝒇𝒇𝒆𝒄𝒕𝒔
𝑯𝟏 : 𝑻𝒉𝒆𝒓𝒆 𝒊𝒔 𝒔𝒊𝒈𝒏𝒊𝒇𝒊𝒄𝒂𝒏𝒕 𝒅𝒊𝒇𝒇𝒆𝒓𝒆𝒏𝒄𝒆 𝒂𝒎𝒐𝒏𝒈 𝒕𝒓𝒆𝒂𝒕𝒎𝒆𝒏𝒕 𝒆𝒇𝒇𝒆𝒄𝒕𝒔
• Analysis of variance technique is used to test the hypothesis
• The data collected from an experiment tabulated as shown below
Treatments Observed response from experimental units Total
1 𝒙𝟏𝟏 𝒙𝟏𝟐 𝒙𝟏𝟑 𝒙𝟏𝒋 𝒙𝟏𝒓 𝑻𝟏
2 𝒙𝟐𝟏 𝒙𝟐𝟐 𝒙𝟐𝟑 𝒙𝟐𝒋 𝒙𝟐𝒓 𝑻𝟐
.
𝒊 𝒙𝒊𝟏 𝒙𝒊𝟐 𝒙𝒊𝟑 𝒙𝒊𝒋 𝒙𝒊𝒓 𝑻𝒊
.
𝒕 𝒙𝒕𝟏 𝒙𝒕𝟐 𝒙𝒕𝟑 𝒙𝒕𝒓 𝒙𝒕𝒓 𝑻𝒕
The following computations are required:
𝑮𝟐
i) Correction Factor (CF) = Where, 𝐺 = 𝐺𝑟𝑎𝑛𝑑 𝑡𝑜𝑡𝑎𝑙 and
𝒏
𝑛 = 𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛𝑠
i.e., 𝑛 = 𝑟𝑡
ii) Total sum of square (TSS) = Sum of squares of all observations − CF
= (𝒙𝟏𝟏 𝟐 + 𝒙𝟏𝟐 𝟐 +………….+ 𝒙𝒕𝒓 𝟐 ) − 𝑪𝑭
= σ σ 𝑥𝑖𝑗 2 − 𝐶𝐹
iii) Treatment Sum of squares (TRSS)
When number of replications is the same for each treatment
𝑻𝟏 𝟐 +𝑻𝟐 𝟐 +𝑻𝟑 𝟐 +⋯……..+𝑻𝒕 𝟐
𝑻𝑹𝑺𝑺 = − 𝑪𝑭
𝒓

When number of replications is different for each treatment

𝑻𝟏 𝟐 𝑻𝟐 𝟐 𝑻𝒕 𝟐
𝑻𝑹𝑺𝑺 = + +…………+ − 𝑪𝑭
𝒓𝟏 𝒓𝟐 𝒓𝒕
iv) Error sum of squares (𝑬𝑺𝑺) = 𝑻𝑺𝑺 − 𝑻𝑹𝑺𝑺
These computations can be summarised in the form of analysis of variance
table ( ANOVA table)
The ANOVA table for CRD with 𝒕 treatments and r replications is given below
Source of Degrees of Sum of Mean square F ratio
variation freedom (df) squares (SS) 𝑺𝑺
𝑴𝑺 =
𝒅𝒇
Treatments t-1 TRSS 𝑻𝑹𝑺𝑺
= 𝑴𝟏 𝑴𝟏
𝒕−𝟏
𝑭=
Error t(r-1) ESS 𝑬𝑺𝑺 𝑬
=𝑬
𝒕(𝒓 − 𝟏)
Total rt-1 TSS

To test the hypothesis 𝐻0 , F has to be compared with the table value of 𝐹 with
(t-1) and t(r-1) df at a desired level of significance
If 𝑭 > 𝒕𝒂𝒃𝒍𝒆 𝒗𝒂𝒍𝒖𝒆 𝒐𝒇 𝑭 , We reject 𝑯𝟎
▪ When 𝑯𝟎 is rejected, the treatment means that differ significantly may be
found out. This is done by computing the critical difference (CD) or least
significant difference(LS).
▪ The formula for computing CD when the number of replications is the same
for each treatment is given by,
𝟐𝑬
𝑪𝑫 = 𝒕
𝒓
The value of 𝑡 is obtained from the t-tables at 5% level of significance with
the error degrees of freedom.
▪ If the number of replications is not the same for each treatment then CD for
comparing two treatments which have been replicated 𝑟𝑖 and 𝑟𝑗 times is
given by
𝟏 𝟏
𝑪𝑫 = + 𝑬 𝒕
𝒓𝒊 𝒓𝒋
✓ EXAMPLE :-
Five test diets were tested to ascertain the growth performance of a certain
fish in plastic pools for a period of 1 month. The daily feed provided was 50% of
the total weight of 40 fry kept in each plastic pool. The experimental design
used was completely randomised design and each treatment was replicated 4
times. Test whether there is significant difference among treatments.
Growth performance is given below
Treatments Net gain weight (g)/fish
(test diets) Rep.1 Rep.2 Rep.3 Rep.4
A 0.95 0.85 0.85 0.90
B 0.43 0.45 0.40 0.42
C 0.70 0.90 0.75 0.70
D 1.00 0.95 0.90 0.90
E 0.90 1.00 0.95 0.95
➢ RANDOMISED COMPLETE BLOCK DESIGN (RBD)
▪ Experimental design id heterogenous.
▪ Here the local control is adopted and the experimental material is grouped
into homogenous subgroups .
▪ RBD is the commonly used experimental design in agriculture.
▪ RBD is an experimental design for comparing t treatment in r blocks.
▪ The blocks consist of homogenous experimental unit.
▪ Treatments are randomly assigned to experimental units within a block, with
each treatment appearing exactly once in every block.
▪ So, complete mean that each block contain the all the treatments.
▪ Completely randomised block design mean that each block have all
treatment and the treatments are randomize with the all block.
o LAYOUT OF THE DESIGN
In agriculture experimentation the layout of RBD can be illustrated as follows.
Consider 5 treatments 𝐴, 𝐵, 𝐶, 𝐷 𝑎𝑛𝑑 𝐸 each replicated 4 times. We decide the
whole experimental area into 4 relatively homogenous blocks and subdivided
each block into 5 plots treatments are then allocated at random to the plots of
a block . Particular layout may be follows
Plots/ treatments
Blocks 1 2 3 4 5
I C E A B D
II A C D E B
III E A B D C
IV B D E C A
o ANALYSIS
• Let 𝑥𝑖𝑗 denote the observation on the 𝑖 𝑡ℎ treatment (𝑖 = 1,2, … . . 𝑡) in the 𝑗𝑡ℎ
replication (𝑗 = 1,2 … … . 𝑟). The following additive model is assumed:
𝒙𝒊𝒋 = 𝒎 + 𝒂𝒊 + 𝒃𝒋 + 𝒆𝒊𝒋 where 𝑚 = 𝑔𝑒𝑛𝑒𝑟𝑎𝑙 𝑚𝑒𝑎𝑛
𝑎𝑖 = 𝑒𝑓𝑓𝑒𝑐𝑡 𝑑𝑢𝑒 𝑡𝑜 𝑖 𝑡ℎ 𝑡𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡
𝑏𝑗 = 𝑒𝑓𝑓𝑒𝑐𝑡 𝑑𝑢𝑒 𝑡𝑜 𝑗𝑡ℎ 𝑏𝑙𝑜𝑐𝑘
𝑒𝑖𝑗 = 𝑒𝑥𝑝𝑒𝑟𝑖𝑚𝑒𝑛𝑡𝑎𝑙 𝑒𝑟𝑟𝑜𝑟
• Experimental error (𝑒𝑖𝑗 ) is independently normally distributed with mean 0
and variance 𝜎 2
• The Null hypothesis to be tested is,
𝑯𝟎 : 𝑻𝒉𝒆𝒓𝒆 𝒊𝒔 𝒏𝒐 𝒔𝒊𝒈𝒏𝒊𝒇𝒊𝒄𝒂𝒏𝒕 𝒅𝒊𝒇𝒇𝒆𝒓𝒆𝒏𝒄𝒆 𝒂𝒎𝒐𝒏𝒈 𝒕𝒓𝒆𝒂𝒕𝒎𝒆𝒏𝒕 𝒆𝒇𝒇𝒆𝒄𝒕𝒔
𝑯𝟏 : 𝑻𝒉𝒆𝒓𝒆 𝒊𝒔 𝒔𝒊𝒈𝒏𝒊𝒇𝒊𝒄𝒂𝒏𝒕 𝒅𝒊𝒇𝒇𝒆𝒓𝒆𝒏𝒄𝒆 𝒂𝒎𝒐𝒏𝒈 𝒕𝒓𝒆𝒂𝒕𝒎𝒆𝒏𝒕 𝒆𝒇𝒇𝒆𝒄𝒕𝒔
• Analysis of variance technique is used to test the hypothesis
• The data collected from an experiment tabulated as shown below
Treatments
Blocks 1 2 3 𝒊 𝒕 Total
1 𝒙𝟏𝟏 𝒙𝟏𝟐 𝒙𝟏𝟑 𝒙𝟏𝒋 𝒙𝟏𝒕 𝑻𝟏
2 𝒙𝟐𝟏 𝒙𝟐𝟐 𝒙𝟐𝟑 𝒙𝟐𝒋 𝒙𝟐𝒕 𝑻𝟐
3 𝒙𝟑𝟏 𝒙𝟑𝟐 𝒙𝟑𝟑 𝒙𝟑𝒋 𝒙𝟑𝒕 𝑻𝟑
.
.
𝒋 𝒙𝒋𝟏 𝒙𝒋𝟐 𝒙𝒋𝟑 𝒙𝒋𝒊 𝒙𝒋𝒕 𝑻𝒋
.
.
𝒓 𝒙𝒓𝟏 𝒙𝒓𝟐 𝒙𝒓𝟑 𝒙𝒓𝒋 𝒙𝒓𝒕 𝑻𝒓
Total 𝑹𝟏 𝑹𝟐 𝑹𝟑 𝑹𝒊 𝑹𝒕 𝑮
The following computations are required
𝑮𝟐
i) Correction Factor (CF) = Where, 𝐺 = 𝐺𝑟𝑎𝑛𝑑 𝑡𝑜𝑡𝑎𝑙 and
𝒏
𝑛 = 𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑜𝑏𝑠𝑒𝑟𝑣𝑎𝑡𝑖𝑜𝑛𝑠
i.e., 𝑛 = 𝑟𝑡
ii) Total sum of square (TSS) = Sum of squares of all observations − CF
= (𝒙𝟏𝟏 𝟐 + 𝒙𝟏𝟐 𝟐 +………….+ 𝒙𝒓𝒕 𝟐 ) − 𝑪𝑭
= σ σ 𝑥𝑖𝑗 2 − 𝐶𝐹
iii) Treatment Sum of squares (TRSS)
𝑻𝟏 𝟐 +𝑻𝟐 𝟐 +𝑻𝟑 𝟐 +⋯……..+𝑻𝒓 𝟐
𝑻𝑹𝑺𝑺 = − 𝑪𝑭
𝒓
σ 𝑻𝒊 𝟐
= − 𝑪𝑭
𝒓
iV) Replication Sum of squares (RSS)
𝑹𝟏 𝟐 +𝑹𝟐 𝟐 +𝑹𝟑 𝟐 +⋯……..+𝑹𝒕 𝟐
𝑹𝑺𝑺 = − 𝑪𝑭
𝒕
σ 𝑹𝒊 𝟐
= − 𝑪𝑭
𝒕
iv) Error sum of squares (𝑬𝑺𝑺) = 𝑻𝑺𝑺 − (𝑹𝑺𝑺 + 𝑻𝑹𝑺𝑺)
These computations can be summarised in the form of analysis of variance
table ( ANOVA table)
 The ANOVA table for RBD

Source of Degrees of Sum of Mean square F ratio

variation freedom squares 𝑺𝑺
𝑴𝑺 =
𝒅𝒇
(df) (SS)
Replication r-1 RSS 𝑹𝑺𝑺
= 𝑴𝟏
s 𝒓−𝟏

Treatments t-1 TRSS 𝑻𝑹𝑺𝑺

= 𝑴𝟐 𝑴𝟐
𝒕−𝟏
𝑭=
Error (r-1)(t-1) ESS 𝑬𝑺𝑺 𝑬
=𝑬
(𝒓 − 𝟏)(𝒕 − 𝟏)
Total rt-1 TSS

To test the hypothesis 𝐻0 , F has to be compared with the table value of 𝐹 with (t-1)
and (r-1)(t-1) df at a desired level of significance
If 𝑭 > 𝒕𝒂𝒃𝒍𝒆 𝒗𝒂𝒍𝒖𝒆 𝒐𝒇 𝑭 , We reject 𝑯𝟎
When 𝐻0 is rejected, it may be of interest to know which of the treatment
effects differ significantly. This is done by calculating CD using the formula.
𝟐𝑬
𝑪𝑫 = 𝒕
𝒓

The value of 𝑡 is obtained from the t-tables at 5% level of significance with

the error degrees of freedom.
✓EXAMPLE :-
To study the effect of stocking density on a certain fish species, an experiment
was conducted with 6 different stocking densities (treatments) of fingerlings in
ponds of size 0.02 ha using randomised block design with 4 replications. All
cultural practices except the stocking densities were kept the same. Harvesting
was done after 6 months of stocking. The yield (t/ha) for different stocking
densities is given below. Find out whether there is significant difference among
yields obtained at different stocking densities.
Stocking Replications
densities RI R II R III R IV
20,000 3.6 2.8 3.0 4.0
30,000 4.8 4.2 4.0 5.6
40,000 6.0 5.7 5.2 6.2
50,000 6.6 6.4 5.4 6.5
60,000 7.0 6.5 5.9 7.0
70,000 7.1 6.8 6.0 7.2