Experiment:

An experiment is a process to have a series of trials or observations taken under some conditions
specified by the experimenter to confirm something doubtful and also to discover some unknown
principles or effects, or to test some suggested known truth. Examining the truth of a statistical
hypothesis, relating to some research problem, is known as an experiment. For example, an
experiment to examine the usefulness of certain newly developed drug.
There are two types of experiment, such as
1. An absolute experiment
2. A comparative experiment
An Absolute Experiment:
An absolute experiment is one in which the absolute value of some characteristic is determined.
Example: If we want to determine the impact of a fertilizer on the yield of a crop, it is a case of
absolute experiment. (Design of Sample Survey).

A Comparative Experiment:
A comparative experiment is one where two or more varieties or treatments are compared to
assess the significance of difference among the varieties. Example: if we want to determine the
impact of one fertilizer as compared to the impact of some other fertilizer, our experiment then
will be termed as a comparative experiment. (Design of Experiment)

Treatment: Various objects of comparison in a comparative experiment are called treatments. For
example, in an agricultural experiment, different fertilizers or different varieties of crop are
treatments. In medical experiment different doses of a medicine or diets are the treatments.

Experimental unit: The smallest subdivision of the experimental material to which the treatments
are applied and on which the variable under study is measured is called an experimental unit. Thus
in an agricultural experiment the plot of land on which the treatment is applied is an experimental
unit. In human experiments in which the treatment affects the individual, the individual will be the
experimental unit.
Yield:
The quantity which is measured from an experimental unit after the application of a treatment there
in is called a yield or response. Examples:
 The production of paddy from different agriculture plots of same size.
 The gain in weight of an experimental animal in a biological experiment.
 The IQ of a student in a psychological experiment.
Blocks:
A block is a group of experimental unit which are internally homogeneous in respect of the
characteristic affecting the yields and externally heterogeneous. Examples:
 In agricultural experiment, a block refers to a group of adjacent plots having uniform
fertility.
 In animal feeding experiments, a group of animals same age, weight, sex, breed litter.
 In a medical experiment, the patients of same symptoms having same age group, same sex
etc.

What Are The Main Reasons of Blocking?

There are certain reasons of blocking which are as follows:
 The technique of blocking increases the precision of an experiment by reducing error
variation.
 The treatments are compared in stable condition as treatment comparisons are made within
blocks of uniform units.
 Blocking sometime provides increased information from an experiment.
Experimental Error:
The yields of an experiment are usually influenced by some extraneous variations may or may not
be controlled by the experimenter. The uncontrolled variations are often called the experimental
errors.
Examples:
Experimental Error is the random variation present in all experimental results. For a homogeneous
experimental unit divided into different plots of equal sizes and different treatments are applied to
these plots; the yields of these plots will not be same. The difference of the yields may be due to
difference of treatments or due to difference of inherent soil structure or fertility condition of the
soil. In field experiment, experience tells us that even same treatments are used on all the plots,
the yield would still vary due to these sources of variations.

What Are The Main Sources Of Experimental Error?

Experimental error arises from the following factors:
 The Inherent variability in the experimental material to which the treatments are applied.
 The lack of uniformity in the methodology of conducting experiment.
 Lack of representativeness of the sample to the population under study.

Describe the Basic Principles of Design of Experiment.

According to Prof. R.A. Fisher, the basic principles of design of experiments are:
 Replication
 Randomization
 Local control
Replication:
The repetition of the treatments under investigation to more than one experimental plots is known
as replication. Examples: A treatment is allotted to r plots of an experimental unit then it can be
said that the treatment is replicated r times. The experiment should be repeated more than once
to increase the statistical accuracy of the experiments. In all experiments, some variation exists
because the experimental units, such as, individuals or plots of land, cannot be physically identical.
This variation is removable by using a number of experimental units. Therefore, the basic
experiment is performed repeatedly. Replication helps in: obtaining an accurate estimate of the
experimental error; decreasing the experimental error, thereby increasing precision; and obtaining
a more precise estimate of the mean treatment effect

Randomization:

Professor R.A. Fisher introduced the principle of randomization in modern experimental design.
Randomization is the process of distributing the treatments to the experimental units purely by
chance mechanism in such a way that any experimental unit is equally likely to receive any
treatment. This process randomly assigns treatments to the experimental units. It implies that every
allotment of treatments ends up with the same probability. Randomizations purpose is to remove
bias and other sources of extraneous variation, which are uncontrollable. It is the basis of any valid
statistical test. Therefore, the treatments must be assigned randomly to the experimental units.

Local Control: Randomization and Replication do not remove all extraneous sources of variation.
A more refined experimental technique is required for that. A design should be chosen such that
all the extraneous sources of variation come under control. For this purpose, local control, which
refers to the amount of balancing, blocking and grouping of the experimental units, is used.
Balancing implies that the treatments should be assigned to the experimental units such that the
result is a balanced arrangement of treatments. Blocking means that, similar experimental units
should be collected together to form a relatively homogeneous group. The main purpose of local
control is to increase the efficiency of an experimental design by minimizing the experimental
error. In this case, local control should not be confused with the word control. Control in
experimental design is used for a treatment. It does not receive any treatment, but the effectiveness
of other treatments should be found through comparison.

Design of Experiment:
Design of experiment is the plan used in experimentation. More specifically, design of experiment
is the formulation of a set of rules and principles according to which an experiment is to be
conducted to collect appropriate data whose analysis will lead to valid inferences for the problem
under investigation.

What are the Purposes of Experimental Design?/ Design of experiment

The purposes of experimental design are given below:
 The main purpose of design of experiment is to collect maximum amount of necessary
information for the problem under consideration, at a minimum cost in terms of time and
resources.
 Deign of experiment is needed to ensure that the requisite assumptions for analysis and
interpretation of data are fulfilled.
 Design of experiment is essential to increase the accuracy of the results of an experiment.

What are the Important Steps in Design of Experiment?

Following are the important steps to be considered by an experimenter to have a good design of
experiment:
 Choosing a set of treatments for comparison.
 Selection of experimental units to which chosen treatments will be applied.
 Specification of the number of experimental units for inclusion in the experiment.
 Specification of the method of allocating the treatments to the experimental unit.
 Specification of the measurement to be obtained from each experimental unit.
 Specification of the groping of experimental units to control extraneous sources of
variation.
Precision:
Precision refers to the closeness with which various effects are estimated. In experimental design,
precision is measured by the reciprocal of the variance of the treatment mean.
In an experiment, if a treatment is replicated r times, then the precision is given by
1 1 r
 2  2 ; where  2 is the error variance.
V x   
r
Efficiency of a Design:
Efficiency of a design in comparison with another may be defined as the ratio of the precision of
the design to that of the other design with which the comparison is made. For example let D1 and

D2 be two designs with error variances  12 and  22 and replications r1 and r2 respectively.

Then the efficiency of design D1 with respect to D2 as defined as

r1
 12
E
r2
 22
Thus efficiency of D1 with respect to D2 is the ratio of precision of D1 and D2
If E  1 , then D1 and D2 are equally efficient
If E  1 , then D1 is more efficient than D2
If E  1 , then D1 is less efficient than D2
Linear Statistical Model:
A linear Statistical model represents a linear relation of the effects of a member of factors with
different levels in an experiment and also involves one or more error terms. The model
y ij     i   j  eij ; i  1 1 p , j  1 1q

Where,
y ij is the yield corresponding to j th plot where i th treatment is used.

 is the general mean effect.

 i is the effect due to i th level of the factor A

 j is the effect due to j th level of factor B and

eij is the random error component.

 This model represents a Linear Statistical model.
The manner in which the levels of the factors are chosen leads to classify the linear model as
i. Fixed Effects Model
ii. Random Effects Model
iii. Mixed Effects Model
What do you mean by Fixed, Random, and Mixed Effects Models in Design of Experiment?
Fixed Effect Model:
A model in which all the assignable factors have fixed effects and only the error effect is random
is called a fixed effect model.
Example:
In an agricultural experiment, suppose there are five levels of irrigation and five levels of
fertilizer. Then if we use all the level of irrigation and all the level of fertilizer, the model is
said to be fixed effect model.
The model
y ij     i   j  eij ; i  1 1 p , j  1 1q

Where,
y ij is the yield corresponding to j th plot where i th treatment is used.

 is the general mean effect.

 i is the fixed effect due to i th level of the factor A

 j is the fixed effect due to j th level of factor B and

eij is the random error component.

Assumptions

i)  , i ,  j are unknown parameter

ii)  i   j 0
i j

iii) eij  NID (0, 2 )

Random Effect Model:
The model in which all the factors have random effect is known as a random effect model.

Example:
If we select the level of irrigation and the level of fertilizer at random then model is known
as a random effect model.
y ij     i   j  eij ; i  1 1 p , j  1 1q

Where,
y ij is the yield corresponding to j th plot where i th treatment is used.

 is the general mean effect.

 i is the random effect due to i th level of the factor A

 j is the random effect due to j th level of factor B and

eij is the random error component.

Assumptions

i)  unknown parameter

 i  NID (0, 2 )
ii)  j  NID (0, 2 )
eij  NID (0, 2 )

iii)  i ,  j eij are mutually independent

,
Mixed Effect Model:
A model in which some factors have fixed effects and some factors have random effects, is called
mixed effect model.
Example:
If we use all the level of irrigation but the levels of fertilizer are chosen randomly then the
model is known as mixed effect model.
 y ij     i   j  eij ; i  1 1 p , j  1 1q

Where,
y ij is the yield corresponding to j th plot where i th treatment is used.

 is the general mean effect.

 i is the fixed effect due to i th level of the factor A

 j is the random effect due to j th level of factor B and

eij is the random error component.

Assumptions

i)  , i are unknown parameter

ii) 
i
i 0

iii)  j  NID (0, 2 )andeij  NID (0, 2 )

iv)  j eij are independent

,
What Are The Differences Among The Models Of Different Type?
The differences among the models of different type are given below:
Fixed effect model Random effect model Mixed effect model
The observations have The observations have same The observations have
different expectations. expectation and constant different expectations.
variance.
The observations are The observations are The observations are
independently distributed. dependent. dependent.
The model parameters are Variance components are Both unknown parameters
estimated by least square estimated by analysis of and variance components are
method. variance method. estimated.
We test hypothesis about We test hypothesis about Both types are tested.
unknown parameters. components of variance.
One-Way Classified Data:
When data obtained from an experiment can be classified and arranged on the basis of one factor
only, the data are referred to as one-way classified data.
Linear model of one-way classification is given by:
y ij     i  eij ; i  1 1 p , j  1 1q

Where,
th
y ij is the yield corresponding to j th plot where i treatment is used.
 is the general mean effect.
 i is the effect due to i th level of the factor A
eij is the random error component.

Two-Way Classified Data:

When data obtained from an experiment can classified and arranged on the basis of two factors,
the data are known as two-way classified data.
Linear model for two-way classification is given by
y ij     i   j  eij ; i  1 1 p , j  1 1q

Where,
y ij is the j th observation due to i th class.

 is the general mean effect.

 i is the effect due to i th level of the factor A

 j is the effect due to j th level of factor B and

eij is the random error component.

M-way Classified Data:

If data obtained can be arranged on the basis of m factors simultaneously in general, then data are
termed as m-way classified data.
Requirements of a Good Experiment:
A good experiment should satisfy the following conditions:
 Absence of bias or free from systematic errors: It is essential to plan an experiment so
that unbaised estimates of treatment differences and treatment effects can be obtained from
the data of the experiment. The absence of bias is achieved by the principle of
randomization.
 Measure of experimental error: science the treatments under comparison apparently
produce different results; test of significance is needed to assess the nature of treatment
differences. Hence an experiment should furnish a measure of experimental error.
Replication provides an estimate of experimental error and thus makes a test of significance
poissible.
 Precision: Precision refers to the closeness with which different effects are estimated.
Estimates of treatment effects and of treatment differences should be precise and precision
is measured by the reciprocal of variance which in turn depends on experimental error.
Important methods of increasing precision are increased replication, refinement of
experimental technique, blocking in various efficient designs, use of confounding in
factorial experiments.
 Clearly defined objective: Every experiment should have clearly defined objective on
which the design and analysis of data considerably depend.
 Simplicity: experimental design should be very simple and consistent with the objectives
of experiment and lead to simple analysis of data.
 Scope or range of validity: The conclusions drawn from experimental data should have a
wide range of validity.
Analysis of Variance:
The Analysis of variance is the systematic procedure of partitioning the total variation present in a set of
observation into number of components associated with the nature of classification of data.
For one way classified data, the total variation can be partitioned into two components, namely variation
due to single factor and the other is due to error variation. For two way classified data, involving factor A
and B , the total variation can be partitioned into three components, namely variation due to factor A,
variation due to factor B and the other is due to error variation. Similarly it can be done for m way classified
data. It is a powerful statistical tool for testing the significance of various experimental treatment effects.
The Objectives of Analysis of Variance:
 It identifies the causes of variation and sort out corresponding components of variation with
associated degrees of freedom.
 It provides far test of significance based on F distribution.

Assumptions Made in Analysis of Variance:

 Observations and errors are independently distributed.
 Observations and errors conform to normal distributions with equal variances.
 Treatments and environmental effects are additive.

One-Way Classification:
For analysis of data by ANOVA technique, the arrangement of observation in various classes on
the basis of a single factor or criterion is called a one-way classification.

Examples: Some Examples of One-Way Classified Data are as Follows:

1. One may like to compare the results of some examination of students belonging to different
colleges.
2. One may like to compare the average yield per acre of several varieties of paddy in the
environments of Bangladesh.
3. One may be interested in comparing the mean incomes of the inhabitants of several cities
in a country.
4. One may want to examine if there is really a difference in the average mileage obtained
with different kinds of gasoline.
5. One may like to compare average life in hours of different types of electric bulbs.

Two-Way Classification (With One Observation per Cell):

Let us consider an experiment where two factors A and B at p and q levels respectively affect the
outcome of the experiment. For analysis of data, taking from this experiment, by ANOVA technique, the
arrangement of observation in various classes on the basis of two factors or criteria A and B is called a
two-way classification.
If for the effect of 1 level of factor A and 1 level of factor B , there is a single observation then it is called
two-way classification with single observation per cell. Examples:
1. The variety of paddy and the type of fertilizer used both affect the yield of a plot.
2. The yield of milk may be affected by breed and stock of the cows.
What are the basic designs used in design of experiments?

Basic designs: There are three basic designs used in design of experiments.

1. Completely Randomized Design (C.R.D)

2. Randomized Block Design (R.B.D)

3. Latin square Design (L.S.D)

Define Completely Randomized Design (C.R.D). Describe the lay-out of C.R.D

Completely randomized design

Completely Randomized Design (CRD) is the basic single factor design. In this design the
treatments are assigned completely at random so that each experimental unit has the same chance
of receiving any one treatment. But CRD is appropriate only when the experimental material is
homogeneous. As there is generally large variation among experimental plots due to many
factors CRD is not preferred in field experiments. In laboratory experiments and greenhouse
studies it is easy to achieve homogeneity of experimental materials and therefore CRD is most
useful in such experiments.

Layout

The step-by-step procedure for randomization and layout of a CRD are given here for a pot
culture experiment with four treatments A, B, C and D, each replicated five times.

Step 1. Determine the total number of experimental plots (n) as the product of the number of
treatments (t) and the number of replications (r); that is, n = rt. For our example, n = 5 x 4 = 20.
Here, one pot with a single plant in it may be called a plot. In case the number of replications is
not the same for all the treatments, the total number of experimental pots is to be obtained as the
t
sum of the replications for each treatment. i.e., n   ri where ri is the number of times the ith
i 1

treatment replicated.

Step 2. Assign a plot number to each experimental plot in any convenient manner; for example,
consecutively from 1 to n.

Step 3. Assign the treatments to the experimental plots randomly using a table of random
numbers as follows. Locate a starting point in a table of random numbers by closing your eyes
and pointing a finger to any position in a page. For our example, the starting point is taken at the
intersection of the sixth row and the twelfth (single) column of two-digit numbers. Using the
starting point obtained, read downward vertically to obtain n = 20 distinct two-digit random
numbers. For our example, starting at the intersection of the sixth row and the twelfth column,
the 20 distinct two-digit random numbers are as shown here together with their corresponding
sequence of appearance.

Random number : 37, 80, 76, 02, 65, 27, 54, 77, 48, 73,
Sequence : 1, 2, 3, 4, 5, 6, 7, 8, 9, 10,
Random number : 86, 30, 67, 05, 50, 31, 04, 18, 41, 89
Sequence : 11, 12, 13, 14, 15, 16, 17, 18, 19, 20

Step 4.Rank the n random numbers obtained in ascending or descending order. For our example,
the 20 random numbers are ranked from the smallest to the largest, as shown in the following:

Random Sequence Rank Random Sequence Rank

Number Number

37 1 8 86 11 19

80 2 18 30 12 6

76 3 16 67 13 14

02 4 1 05 14 3

65 5 13 50 15 11

27 6 5 31 16 7

54 7 12 04 17 2

77 8 17 18 18 4

48 9 10 41 19 9

73 10 15 89 20 20
Step 5.Divide the n ranks derived into t groups, each consisting of r numbers, according to the
sequence in which the random numbers appeared. For our example, the 20 ranks are divided into
four groups, each consisting of five numbers, as follows:
 Group Ranks in the Group
Number

1 8 13 10 14 2

2 18 5 15 3 4

3 16 12 19 11 9

4 1 17 6 7 20

Step 6. Assign the t treatments to the n experimental plots, by using the group number as the
treatment number and the corresponding ranks in each group as the plot number in which the
corresponding treatment is to be assigned. For our example, the first group is assigned to
treatment A and plots numbered 8, 13, 10, 14 and 2 are assigned to receive this treatment; the
second group is assigned to treatment B with plots numbered 18, 5, 15, 3 and 4; the third group is
assigned to treatment C with plots numbered 16, 12, 19, 11 and 9; and the fourth group to
treatment D with plots numbered 1, 17, 6, 7 and 20. The final layout of the experiment is shown
below.

Plot no 1 2 3 4
Treatment D A B B

5 6 7 8
B D D A

9 10 11 12
C A C C

13 14 15 16
A A B C

17 18 19 20
D B C D
Figure 1.1. A sample layout of a completely randomized design with four treatments (A, B, C
and D) each replicated five times.
Advantage of CRD:

i) CRD is the basic and the simplest design.

ii) Number of treatment can be repeated i.e. complete flexibility regarding the
replication.

iii) Error variance is minimum.

iv) In case of CRD experimental units are homogeneous.

v) Analysis of data is quite simple and straight forward even if different treatments have
unequal number of replication.

vi) The analysis is easier if there is some missing observations. In fact properly of
orthogonality is not lost by missing values in a CRD.

Disadvantage of CRD:

i) CRD is relatively inefficient design as local control method is not adopted to reduce
error variation in this design .

ii) It is seldom used in field experiments because homogeneous unis over the whole
experimental area is rarely available in practice.

iii) Due to inflated error variations there is greater change of wrongly accepting null
hypothesis.

iv) Precision of CRD is less than other design.

Application or uses of CRD:

i) CRD is most useful in laboratory technique and methodological studies e.g. in

physics, chemistry, in chemical and biological experiments, in some green house
studies etc.

ii) CRD is conveniently used in situation having homogeneous experimental units.

iii) It is suitable in situations where a large fraction of experimental units may not
respond or may be lost in course of experiment.

iv) It is advantages for small experiments because it furnishes maximum number of error
degrees of freedom.
Stating the underlying assumptions, discuss the method of analysis of data of completely
Randomized Design.

Model The additive model of C.R.D with unequal observation is

yij     i  eij ; i  1 1 k , j  1 1 ni

Where, y ij is the i th treatment in the j th replication

 is the general mean effect

 i is the effect due to i th treatment

eij is the random error component

Assumption of completely Randomized Design.

The main assumption of completely Randomized Design is as follows

1. The observations are selected at random

2. All observations are independent.

3. The treatment effects are additive in nature 

i
i 0

4.

eij ~ NID 0,  2 .
5. The model considered is fixed effect model.
Layout of One-Way Classified Data:

Data of completely Randomized Design with K -treatments can be presented in tabular form as
shown below:

Treatment
A1 A2 … Ai … Ak
1 y11 y 21 … yi1 … y k1
2 y12 y 22 … yi 2 … yk 2
      
j y1 j y2 j y ij y kj
replication
    
ni y1n1 y 2n2 … y ini … y knk
Total y1 . y2 . … yi . … yk .
Mean y1 . y2 . … yi . … yk .
Where n1  n2    nk  N = total number of observations

k ni
G = Grand total =  y
i 1 j 1
ij

ni

y ij
yi . j 1
= The mean value of the ith treatment =
ni

G
y. .   Grand mean
N

Partition of the Total Variation:

The total variation can be partitioned into two components as follows

2

Total variation    y ij  y..     y ij  y i .  y i .  y.. 

k ni k ni
2

i 1 j 1 i 1 j 1

 ni 
   y i .  y..     y ij  y i .     y ij  y i .   0
k ni k ni
2 2

i 1 j 1 i 1 j 1  j 1 

  ni  y i .  y..     y ij  y i . 
k k ni
2 2

i 1 i 1 j 1

Total SS  Treatment SS  error or Within SS

Define Randomized Block Design (RBD), Write down the merits demerits and uses of
R.B.D.

Randomized block design

A randomized block design (RBD) is a design in which the whole set of experimental units
arranged in several blocks which are internally homogeneous and extremely heterogeneous and
then the selected treatments are randomly allocated to the experimental units within each block
such that each treatment occurs one or same number of times in each block.

Advantages of RBD:

i) RBD is more efficient than CRD and thus provides more accurate and precise results
than CRD.

ii) Any number of blocks and any number of treatment can be used in RBD except the
restriction that at least two replicates are needed to a carry out the test of significance.

iii) Analysis of data is simple and straight forward in RBD.

iv) RBD provides a method of eliminating or reducing the effects of trends.

v) It is flexible readily adoptable and easy to analyze.

Disadvantage of RBD:

i) RBD is not suitable for large number of treatments as large error variation may arise
in such case and when the blocks are within heterogeneous.

ii) Property of orthogonality is lost by missing values in a RBD leading to complicated

analysis of data.

iii) RBD has less error d.f. than comparable to CRD.

iv) Since RBD controls variability due to one extraneous factor it is unsatisfactory when
several extraneous factor exists among the experimental unit.

v) The efficiency of RBD decreases as the block size increases.

Uses of RBD:

i) RBD removes one extraneous source of variation from experimental error and so
increases precision. Thus RBD is used to increases precision.

ii) Its use is found to be satisfactory in many experimental situations and thus it avoids
the necessity of using more complex designs.

Page no 1
 iii) RBD provides unbiased estimates of block means in addition to that of treatment
means and thus furnishes additional information from the experiment. It is not
necessary that all blocks be conducted at the same location or at same time.

Reasons of blocking:

i) One major reason for use of blocks is to make inferences over a large number of
environmental conditions.

ii) Another major reason is to reduce error variation by removing an unwanted source of
variation from error variation.

Discuss the method of analysis of data of R.B.D setting the necessary assumptions./ Set up a
linear model to analyze the data obtained from R.B.D with one observation per cell.

The linear model in RBD for one observation per experimental unit is given by

yij     i   j   ij ; i  1 1 p
j  1 1q

Where,

y ij is the observation corresponding to the ith Treatment in the jth Block

 is the general mean effect

 i is the fixed effect due to ith treatment

 j is the fixed effect due to jth Block

eij is the random error component.

Assumption:

The main assumptions of R.B.D is as follows

1. The observations are selected at random

2. There is no interaction between blocks and treatment so that they are independent.
p q
3. treatment effects and block effects are additive in nature, i.e   i  0 and
i 1

j 1
j 0

4.  ,  i and  j are unknown parameters.

Page no 2
 
5. eij ~ NID 0,  2 . 
Layout of Data:

Block
Marginal
B1 B2 … Bj … Bq Mean
Total
Treatment

A1 y11 y12 … y1 j … y1q y1 . y1 .

A2 y 21 y 22 … y2 j … y2q y2 . y2 .

      

Ai yi1 yi 2 … yij … yiq yi . yi .

      

Ap y p1 y p2 … y pj … y pq yp. yp.

Marginal
y. 1 y. 2 … y. j … y. q G = y. .
Total

Mean y. 1 y. 2 … y. j … y. q y. .

here yi . = total value of the observations corresponding to ith treatment

y. j = total value of the observations corresponding to jth block

y
j 1
ij
th
yi . =Mean value corresponding to i treatment =
q

th
y
i 1
ij
y. j = Mean value corresponding to j block =
p

p q p q
G=  yij   yi.   y. j = Grand total
i 1 j 1 i 1 j 1

Page no 3
Latin square design (LSD):
Latin square design is a design in which experimental units are arranged in complete blocks in
two different ways, called rows and columns and then the selected treatments are randomly
allocated to the experimental units within each row and column such that each treatment appears
exactly once in each row and once in each column. Since this design is a square arrangement
where the treatments are denoted by Latin letters, so this design is named Latin square design.
The term Latin square was first used in analysis of variance context by R. A. Fisher who
borrowed it from Swiss mathematician Leonard Euler (1707-1783).
In general a r  r Latin square is an arrangement of r letters in r rows and r columns such that
each letter appears once in each row and once each column.
Thus a 3  3 Latin square with treatments A, B, and C is given by

A B C
B C A
C A B
Example:
i) In agricultural field experiments, LSD is used to eliminate the variation due to soil
fertility difference in two perpendicular directions and then to compare the yields of
several varieties of paddy or wheat.
ii) In animal feeding experiments LSD may be used to remove the variation due to
breeds and ages of cows and then to compare the yields of milk from cows fed on
different nations.

Advantages of LSD:
i) LSD is more efficient than RBD and CRD. Since it control more of the variation than
CRD or RBD.
ii) Statistical analysis of data remains simple even with missing observations.
iii) LSD is an complete layout needs less number of observations tan the corresponding
complete layout. So LSD has adequate economy in the use of experimental material.
iv) LSD covers relatively complete situations where factors can be studied
simultaneously.

Disadvantages of LSD:
i) LSD is not suitable for large member of treatments.
ii) Analysis of data in a LSD depends on the assumption that there in no interaction
among rows, columns and treatments. So LSD is not appropriate when interactions
are present in data.
iii) Error d.f is relatively small in a LSD. In fact, there is no error d . f for 2  2 latin
square.
iv) Property of orthogonality is lost by missing values in LSD.
 Uses of LSD: Latin square design is used in experimentation in different way:
i) Glass house experiments, where there may exists variation across the house due to
light differences and along the house due to treatment differences.
ii) Cow feeding experiment.
iii) Used to eliminate two extraneous source of variability.
iv) Field experiment.

Why this design names "Latin square design"?

Since this design is a square arrangement where the treatments are denoted by Latin letter, so this
is named by Latin square design.

Why L.S.D is called incomplete?

A complete experimental design is a design which contains at least one observation for every
possible combination of the levels of the factors. In a L.S.D there are involving three factors such
as row, column, and treatment each with same number of levels r provides only r2 observation
out of r 3 possible level of combinations. That is why L.S.D is called incomplete or incomplete
three ways lay out.

How does LSD or incomplete three way classification differ from complete three way
classification?
A complete three way classification involves r 3 possible level combinations. While a LSD or
incomplete three way classification is a design involving r 2 observations out r 3 possible level
combinations.

Difference between LSD and RBD:

LSD differs from RBD in the following points:
i) Where in RBD, treatments are arranged in complete blocks in one direction to
eliminate one extraneous source of variability. In LSD treatments are arranged in
complete blocks in two directions to remove two extraneous sources of variation.
ii) The number of blocks and treatments need not be equal in RBD. While member of
rows, columns and treatments must be equal in LSD.
iii) RBD is a complete layout while LSD is incomplete layout viewed from the types of
blocks used.
Types of Latin squares: According to Fisher and Yates various types of Latin squares are as
defined below:

Standard Square: A square is said to be Standard Square if the first row and the first column are
ordered alphabetically or numerically.
For example,
1 2 3  A B C
   
 2 3 1  or ,  B C A 
 3 1 2 C A B 
   

Conjugates square: Two standard squares are said to be conjugate if the row of one square are
the columns of the other. For example,
 A B  A B
  and   are conjugate square.
 B A  B A

Self Conjugate Square: A square is called self Conjugate Square if its arrangement of rows and
columns are the same. For example,
 A B C
 
 B C A  is self conjugate square.
C A B 
 

Orthogonal Latin squares: Two Latin squares of same size are said to be orthogonal Latin
squares if each letter of one square appears exactly once with each letter of the other square
when the two Latin squares are superimposed.
For example,
1st latin square 2nd latin square
 A B C   
  
  
are orthogonal latin square.
 B C A 
C A B     
  
 Difference between CRD and LSD
CRD LSD
The additive model of C.R.D with unequal 1. A Linear additive model for LSD is given by
observation is yijl     i   j   l  eijl ; i, j , l  1, 2,..., r.
yij     i  eij ; i  1 1 k , j  1 1 ni
where,
Where, y ij is the i th treatment in the j th yijl  observation in the i th row, j th column and l th treatment.
replication
  general mean effect
 is the general mean effect
 i is the effect due to i th treatment  i  fixed effect of i th row
eij is the random error component  j  fixed effect of j th column
 l  fixed effect of l th treatment.
eijl  random error component.

2. In CRD the number of replication varies 2. In LSD the number of replication must be the same as the number
from treatment to treatment. of treatment.

3. For missing observation CRD does not loss 3. LSD losses its orthogonality for missing observation.
its orthogonality.
4. CRD does not control any external sources 4. LSD controls the external sources of variation in two
of variation and it is less efficient than LSD perpendicular direction for which and it is more efficient than CRD
Stating the necessary assumptions, discuss the method of analysis of data of L.S.D.
Statistical analysis of LSD:
A fixed effect model for LSD is given by
yijl     i   j   l  eijl ; i, j , l  1, 2,..., r.
where,
yijl  observation in the i th row, j th column and l th treatment.
  general mean effect
 i  fixed effect of i th row
 j  fixed effect of j th column
 l  fixed effect of l th treatment.
eijl  random error component.
Assumption:
i)  , i ,  j and  l are unknown parameters
ii) There is no interaction effect between rows, columns and treatment.
r r r
iii) i    j    l  0
i 1 j 1 l 1

eijl ~ NID  0,   so that E  e   0,

2
ijl
iv)
V  eijl    2 and Cov  e , e   0 ; i  i , j  j , l  l .
ijl ij l 
 Data of L.S.D with r rows, r columns and r treatments can be treated as a case of incomplete
three way classification. In the L.S.D r treatments are arranged in r rows and r columns
Column Marginal
1 2 … j … r Mean
Row Total

1 y11. y12. … y1 j . … y1r . y1. . y1. .

2 y 21. y 22. … y 2 j. … y 2r . y 2 .. y 2 ..

      

i y i1. y i 2. … y ij . … y ir . y i... y i...

      

r y r1. y r 2. … y rj. … y rr. y r.. y r ..

Marginal
y.1. y. 2. … y. j . … y. r . G = y...
Total

Mean y.1. y. 2. … y. j . … y.r . y...

Where r2= Total no. of observations

yi...   yijl = total of ith row

j (l )

yi..
yi...   Mean of ith row
r
y. j .   y ijl
i (l )
= total of jth column

y. j.
y. j .   Mean of jth column
r
y..l   y ij (l )  total of l th treatment
ij

y..l
y..l   mean of l th treatment
r
G  y...   y ijl  Grand total of all observation
i jl

y...
y...   Grand mean
r2