MA6452 STATISTICS AND NUMERICAL METHODS

UNIT II

By
Dr V Valliammal
Assistant Professor
Department of Applied Mathematics
SVCE
Agricultural research is the one which gave rise to the
development of theory and practice of
experimental design.

Important examples of experimental designs are

1.Completely randomised design (C.R.D)
2.Randomised block design (R.B.D)
3.Latin square design (L.S.D)
4.22 Factorial design
 DESIGN OF EXPERIMENT
Suppose there is an agricultural experiment for a
certain area of land.
Let there be four different chemical treatments
of soil. These different chemical treatments may
produce different yields.
In order to obtain as much information as
possible the details of an experiment must be
carefully planned in advance.
This planning is referred to as the design of
experiment.
Treatments: Various objects of comparison in a comparative
experiment are called treatments.
(i.e) in an agricultural experiment, different fertilizers used, or
different types of crops or different methods of cultivation are
the treatments.

Experimental unit: The smallest division of the experimental

material to which we apply the treatments and on which we
make observations on the experimental variables under the
investigation is called experimental unit.
I n an agricultural experiment, the plot of land is the
experimental unit.
 Blocks: In an agricultural experiment ,we divide the whole
experimental units or fields, into relatively homogeneous
subgroups.
These subgroups which are more homogeneous amongst
themselves than the whole field are called Blocks.
Experimental error:
In our agricultural experiment, even if the same
treatement,say, fertilizier,is used on all the plots of the lands,
the yield would vary from plot to plot, due to the difference in
soil fertility. such variation in the yield from plot to plot is due
to randomcauses.this is termed as experimental error
Here the error does not mean a mistake but it is due to
extraneous variables
 Basic principles in the design of
experiment
1.Randomisation
2.Replication
3.Local control
Randomization: A set of objects is said to be
randomized when they are arranged in order.
Though our desire is to eliminate completely the effect
of extraneous variables in any experiment, it is not
possible. so ,we try to control them by randomization.
Randomization is like insurance .It is always a good
idea and even some times better than we accept
 Replication means the repetition of the treatments under investigation.

Replication is necessary to increase the accuracy of estimates of the

treatment effects.
LOCAL CONTROL: To control the effect of of extraneous variables, we use a
principle called local cotrol in the experimental design.

NOTE:1.To estimate the magnitude of an effect in an experiment the principle of

randomisation and replication are applied.

2. Randomisation by itself is not necessarily sufficient to yield a valid experiment.

3. Randomisation must be invariably accompanied by sufficient replication so as to

ensure validity in an experiment.
To plan an experiment the following three are essential.

1.A statement of the objective. Statement should clearly

mention the hypothesis to be tested.

2.A description of the experiment. Description should

include the type of experimental material, size of the
experiment and the number of replications.

3.The outline of the method of analysis. The outline of

method consists of analysis of variance.
 Analysis of Variance (ANOVA) is a technique that
will enable us to test for the significance of the
difference among more than two sample means.
Analysis of variance is useful, for example, for
determining (i) which of various training methods
produces the fastest learning record,
(ii) whether the effects of some fertilisers on the
yields are significantly different,
(iii) whether the mean qualities of outputs of
various machines differ significantly etc.
1.Each of the samples is drawn from a normal
population.
2.The variances for the population from which
samples have been drawn are equal.
3.The variation of each value around its own
grand mean should be independent for each
value.
 Determine
1.One estimate of the population variance from
the variance among the sample means.
2.Determine a second estimate of the
population variance from the variance within
the sample.
3.Compare these two estimates if they are
approximately equal in value, accept the null
hypothesis.
The technique of analysis of variance has
been discussed for

(i)one-way classification and

(ii)two way classification separately

(iii)Randomized block design

(iv) Latin Square

 In one way classification the data are classified according
to only one criterion (or) factor .
Working procedure:
Step1:Let H0 :There is no significant difference between
the treatments.
Step2: Let H1: There is significant difference between the
treatments.
Step3:find N he number of observations.
Step4:Find T,the total value of all observations.
 Step:5 calculate the total sum of squares.
TSS=

Step 6:calculate the column sum of squares

Here N is the number of elements in each column.

SSE=TSS-SSC

Step7:prepare the ANOVA table to calculate F- ratio

 ANVOA TABLE

Sum of squares Degrees of freedom Mean sum Variance ratio

squares
Between SSC c-1 MSC=SSC/c - 1 Fc= MSC/MSE
samples
Within samples SSE nc MSE=SSE/n - c
Total TSS n -1

Step 8: Find the table value;

Step 9: Conclusion :
If Fc < F Tab Difference not significant
If Fc > F Tab Difference is significant.
 A completely randomized design experiment with 10 plots and 3
treatments gave the following result.

Plot No 1 2 3 4 5 6 7 8 9 10

Treatment A B C A C C A B A B

Yield 5 4 3 7 5 1 3 4 1 7

Analyse the result for treatment effects.

Null hypothesis H0 : There is no significant difference among the average yields in

the 3 treatment.

Treatments Yields from plots

A 5 7 3 1
B 4 4 7 -
C 3 5 1 -
 Treatment Treatment Treatment
A B C
X1 X12 X2 X22 X3 X32
5 25 4 16 3 9
7 49 4 16 5 25
3 9 7 49 1 1
1 1 - - - 0
16 84 15 81 9 35
X1 X12 X2 X22 X3 X32
 Tabulated F for (7,2) d.f at 5% level of signficance is
19.35. since call F < tab F, we accept the null hypothesis

i.e., there is not significant difference among the average

yields in the 3 treatment.

i.e., the treatments are not significantly different.

This is a term which stems from agricultural research in which
several variables or treatments are applied to different blocks of
land for repetition or replication of the experimental design are as
follows.
Merits of RBD
(a)Easy to lay out.
(b)Allows flexibility.
(c)Simple statistical analysis
(d)The lots of information due to missing data is smaller than with
any other design.

But this design is usually suited (i) only for small number of
treatments and (ii) for homogeneous experimental material.
 Two way classification
In two way classification of analysis of variance, we consider one
classification along column wise and the other along row wise
Procedure for testing means
H0 : There is no significant difference between column means as well as
between row means
H1: There is significant difference between column means and between
row means
Step I : Find N
Step II : Find T
Step III : Find T2/N
Step IV : Find TSS
Step V : Find SSC
Step VI : Find SSR (Row Sum of Squares )
Then TSS = SSC + SSR + SSE
 ANVOA TABLE

Source of Sum of squares Degrees of Mean sum Variance ratio

variation freedom squares

Between SSC c1 MSC=SSC/c 1 Fc =MSE/MSC

columns

Between rows SSR r1 MSR=SSR/r 1 FR= MSE/MSR

Residual(error) SSE N-c-r+1 MSE=SSE/N-c-

r+1
 Three varieties A,B,C of a crop are tested in a randomized block
design with four replications. The plot yields in pounds are as follow.

A6 C5 A8 B9
C8 A4 B6 C9
B7 B6 C 10 A6
Analysis the experimental yield and state your consolation.

Null hypothesis H0 : There is no significant difference between varieties (rows) and

between (blocks)
Varieties Yields
1 2 3 4 Total
A 5 7 3 1 24 (V1)
B 4 4 7 - 28 (V2)
C 3 5 1 - 32 (V3)
Total 21 (B1) 15 (B2) 24 (B3) 24 (B4) 84 (T)
 Source
Of
variation

Between blocks
(Coloumns)

Between
varieties
(Rows)

Residual

(i) The tabulated F for (3,6) d.f at 5% level of signficance is 4.76. since call F tab =
4.76. Since FC < F Tab, we accept the null hypothesis H0. That is there is no
significant difference between yields.

(ii) The tabulated F for (2,6) d.f at 5% level of signficance is 5.14. since call F tab =
5.14. Since FR < F Tab, we accept the null hypothesis H0. That is there is no
significant difference between varieties.
It may be necessary to control two sources of error or variability at
the same time as the difference in rows and the difference in
columns.
For example, errors in different rows and columns could be due to
change in soil fertility in different plots of the land.
In such a case, therefore, it is desirable that each treatment should
occur once in each row and once in each column as in the following
figure.
This arrangement is called a Latin Square.

Z X Y W

X Z W Y

I Y W Z X

W Y X Z
In this design there have to be as many replications as there are
treatments.
The experimental area is divided into plots arranged in a square in
such a manner that there are as many plots in each row as there
are in each column, this number being equal to the number of
treatments.

Then the plots are assigned to various treatments such that every
treatment occurs only once in each row and once in each column.

There is a large number of ways of doing this and the way it is to

be done in any particular layout must be determined randomly.
Latin square designs are used in industrial, laboratory field, green
house, educational, medical, marketing and sociological
experimentation in addition to agricultural problems.

Advantages of the Latin Square design over other designs are

(a)Latin square controls more of the variation than the completely

randomised block design with a two way stratification.

(b)The analysis is simple.

(c)Even with missing data the analysis remains relatively simple.

Source of
variation

Rows

Columns

Treatment

Residual or
error
 Analyses the following results of a Latin square experiments.
1 2 3 4
1 A (12) D (20) C (16) B (10)
2 D (18) A (14) B (11) C (14)
3 B (12) C (15) D (19) A (13)
4 C (16) B (11) A (15) D (20)

The letters A, B, C, D denote the treatments and the figures in brackets denote the
observations.

Null hypothesis H0 : There is no significant difference between rows, between

coloumns and treatments. Code the data by subtracing 15 from every value, we have
Source of Variaons

Rows

Columns

Treatments

Residual
 Conclusion:

Tabulated value of F for (3,6) d.f at 5% level of signficance

is 4.76. since call F tab = 4.76. Since FR < F Tab, we accept the
null hypothesis H0. That is there is no significant difference
between rows.

Tabulated value of F for (6,3) d.f at 5% level of signficance

is 8.94. since call F tab = 4.76. Since FC < F Tab, we accept the
hypothesis. That is there is no significant difference
between columns.

Since FT > F Tab, we accept the null hypothesis H0. That is

there is no significant difference between treatments.
 In earlier sections, we considered the testing of a number of treatments (not
necessarily related to each other) in Randomised Blocks or latin squares etc.

when dealing with industrial applications, we often come across several factors which
may affect the characteristics of our interest.

Now, we would like to estimate the effects of each of the factors and how the effect of
one factor varies over the level of the others.

To see this in detail, let us consider the yield of a chemical process.

This yield may be affected by several cofactors such as the levels of pressure,
temperature, concentration, radiations etc.

A natural temptation to analyze this sort of situations is to test each of the factors
separately by keeping the other factors as constrains in a given experiment.

However such an experiment might not give the necessary required information.
The procedure of varying all factors simultaneously is termed as Factorial Experiment.
1.Find out the main effects and interactions in the following 22 - factorial
experiment and write down the analysis of Variance table :-

(i) 00 (ii) 10 (iii) 01 (iv) 11

Block I 64 25 30 6

II 75 14 50 33

III 76 12 41 17

IV 75 33 25 10
Solution:- Taking deviation from y = 37..

Blocks
Treatment
combination I II III IV
total
X1 X2 X3 X4 X12 X 22 X 32 X 24

(y1) (1) 27 38 39 38 142 729 1444 1521 1444

(y2) a -12 -23 -25 -4 -64 144 529 625 16

(y3) b -7 13 4 -12 -2 49 169 16 144

(y4) ab -31 -4 -20 -27 -82 961 16 400 729

Total -23 24 -2 -5 -6 1883 2158 2562 2333

Step 1: N = 16
Step 2 : T= -6
Step 3 : C.F = T2 / N
= 36/16
= 2.25
Step 4 : TSS = X 2
1 + X 2
2 + X 2
3 + X 2
4 T 2
/N

= 1883+2158+2562+2333-2.25
= 8933.75
Step 5 : SSC = X12 + X 22 + X 32 + X 24 2 T / N
N1

N1=no.elements in each coloumn..

= (X ) + (X ) + (X
1
2
2
2
3 )2 + ( X 4 )2

T2
N1 N

= { [ (-23)2 + (24)2 +(-2)2 +(-5)2 ] / 4 } -2.25

 = { 1134 / 4 } -2.25
=281.25

(Y ) 2
+ (Y2 ) 2 + (Y3 ) 2 + (Y4 ) 2 T2
Step 6 : SSR = 1

N2 N

N2 = no of elements in each row..

= { [ (142)2 + (-64)2 + (-2)2 + (-82)2 ] / 4 } - 2.25
= 7744.75

Step 7 : SSE = TSS - SSC - SSR.

= 8933.75 - 281.25 - 7744.75
= 907.75
For 22 expriment..

A = [a+ab-b-(1)]
= -64-82+2-142
= -286

B = [ b + ab - a - (1) ]
= -2-82+64-142
= -162

AB = ab + (1) - a - b
= -82 +142 + 64 + 2
=126

Main Effect A = 1/2 [a + ab - b - (1) ]

= - 143
 1. B = 1/2 [ b + ab - a - (1) ]
= - 81
2. AB = 1/2 [ab + (1) - a -b ]
= 63
3.SSA = [a+ab-b-(1)]2 / 16
= (-286)2 / 16
= 5112.25
4.SSB = [ b + ab - a - (1) ]2 / 16
= (-162 )2 / 16
= 1640.25
5.SSAB = [ ab + (1) - a - b ]2 / 16
= (126)2 / 16
= 992.25
 Variance Table value of F
S.v d.f S.S MSS
ratio 5% 1%

100.86/93.83
Blocks 3 281.5 93.83 F(9,3)=8.81 27.35
=1.075

2581.88/100.86
Treatments 3 7744.75 2581.58 F(3,9)=3.86 6.99
=25.60

5112.25/100.86
A 1 5112.25 5112.25 FA(1,9)=5.12 6.99
=50.69

1640.25/100.86
B 1 1640.25 1640.25 FB(1,9)=5.12 6.99
=16.26

992.25/100.86
AB 1 992.25 992.25 FAB(1,9)=5.12 6.99
=9.84

Error 9 907.75 100.86

 Error ( d.f ) = N - C - r + 1...
= 16 - 4 - 4 +1
=9

Cal FA > Tab FA

Cal FB >Tab FB
Cal FC >Tab FC ..