Design of Experiment DOE

Suggested Books
• Statistical Procedures for Agricultural Research, Second Edition, by
Kwanchai A. GOMEZ & Arturo A. GOMEZ

• Statistical Methods & Data Analysis, Edition 2014 by Faqir

Muhammad
• Chapter 8 – Analysis of Variance
• Chapter 9 – Basic Design
• Chapter 10 – Factorial Experiments and Nested Classifications
Design of Experiment
• Science is concerned with understanding variability in nature
• Statistics concerned with making decisions about nature in the
presence of variability and uncertainty
• Experimental design is concerned with reducing and controlling
variability in ways, which make statistical theory applicable to
decisions made about nature.
Some useful definitions
• An experiment is a planned inquiry to discover new facts, or to
confirm or deny the results of previous investigations.
• A treatment is a procedure whose effect on the experimental
material is to be measured.
• A particular class of related treatments is often called factor.
• The states of a factor, i.e., the treatments within the class, are called
the levels of the factor.
• An Experimental Unit (EU) is the piece of experimental material to
which one trial of a single experiment is applied.
• A group of homogenous experimental units is called a block.
Some useful definitions……..
• Actual placement of the treatment to the experimental units is called
Layout.
• An experimental design is a set of rules by which the treatments to be
used in an experiment are assigned to the experimental units.
• If treatments are assigned to a set of units in such a way that every unit is
equally likely to receive any treatment, the assignment is said to be
random.
• When a treatment appears more than once in an experiment, the
treatment is said to be replicated.
• Experimental error is the variation among experimental units, which have
been treated alike.
Source of Variation
Variation comes from two main sources.
• The inherent variability that exists in the experimental material to
which treatments are applied i.e., genetic or environmental
differences among crops or animal units – which are beyond the
control of the experimenter
• There is the variation, which results from any lack in uniformity in the
physical conduct of the experiment.
Major Goal – reduce experimental error
One of the major goals of a sound research procedure is to reduce
experimental error and thus increase precision. Experimental error
can be reduced in the following ways:
• Blocking
• Concomitant variable – inclusion of variable correlated with the study
variable i.e., covariance analysis ANCOVA
• Increase the number of replication
• Proper use of efficient statistical design
• Size and shape of experimental units
Basic principles of Experimental design
• Replication of treatments
• Randomization of treatments
• Local control of variation among experimental units
Replication
• Repetition of a treatment to the experimental units, i.e., two units
treated alike constitute two replications
• Estimate experimental error
• Precise estimates of treatment effects as any effects due to uncontrolled factors
go into experimental error and only treatment effects remain in the estimates of the
treatment effects.
Randomization
• The allocation of treatments to the experimental units in such a way
that each unit has an equal chance of receiving any treatment is
called randomization.
• Eliminates any personal bias in the allocation of treatments
• Every treatment gets the same chance to go to any experimental
unit/plot
• Why randomize: validate statistical analysis (statistical reasoning:
Independence of errors)
Local Control
• Blocking, proper choice of size and shape of experimental units
• Refinement of experimental technique may help to control the
variability in the data
Basic Experimental designs
There are three basic experimental designs

• Completely Randomized Design (CRD)

• Randomized complete block design (RCBD)
• Latin square design
Completely Randomized Design (CRD)
• Simplest and least restrictive
• Every plot/experimental unit is equally likely to be assigned to any
treatment
• Experimental units are essentially homogenous
• Used in laboratory experiments since the quantity of material is
thoroughly mixed
Advantages of CRD
 Flexibility
– Any number of treatments and any number of replications
– Don’t have to have the same number of replications per treatment (but
more efficient if you do)
 Simple statistical analysis
– Even if you have unequal replication
 Missing plots do not complicate the analysis
 Maximum error degrees of freedom
Disadvantage
• Low precision if the plots are not uniform
Random allocation of treatment
i. Lottery method
Suppose we have an experiment to test 3 varieties in order to find out
which grows best in our area i.e., t=3, r=4
• Make 12 slips and write A on 4, B on 4 and C on 4
• Place them in a bowl
• Draw them one by one with out replacement
• Assign the treatment on the first slip to the 1st Experimental unit and so on
A1 2 3 4
A
A A
5 6 7 8
A
9 10 11 12
Random allocation of treatment…
ii. Random number table method
• Take a start in the Random number table and take 12 random
numbers
• Assign serial numbers 1 – 12 to the random digits in ascending order
• Assign Treatment A to first Four experimental units and B to second
four and so on
Hypothesis & ANOVA table
Ho : 1=2=3=4=…
(i.e., Mean yield from all treatments are same)
H1: At least two ’s are different
(i.e., Mean yield for at-least 2 treatments are not same)

Source df SS MS F
Total n–1

Treatments t–1

Within treatments n–t

(Error)

Decision Rule:- Reject Ho if Fcal  F(t-1,n-t)

Example - CRD
Larval population (%) for three different wheat varieties (V1= LU-26S, V2 = Pak-
81, V3 = Punjab-85, V4 = Faisalabad-83) with three plots each was observed and
is given below:- Varieties
• Test the hypothesis that all varieties have the same larval V1 V2 V3 V4
population? 43 59 49 48
31 45 60 44
• Calculate Co-efficient of variation (CV) and interpret it 33 60 60 38
• If the result is significant than apply the pairwise
comparison test to find out the significantly different pairs

Muhammad Usman
Calculations Varieties
V1 V2 V3 V4
𝐺. 𝑇 2 5702
𝐶𝐹 = = = 𝟐𝟕𝟎𝟕𝟓 43 59 49 48
𝑛 12
31 45 60 44
𝑻𝑺𝑺 = 𝟒𝟑𝟐 + 𝟑𝟏𝟐 + ⋯ + 𝟑𝟖𝟐 − 𝑪𝑭 33 60 60 38
Totals 107 164 169 130 570
= 𝟏𝟐𝟏𝟓
Means 35.7 54.7 56.3 43.3
𝟏𝟎𝟕𝟐 𝟏𝟔𝟒𝟐 𝟏𝟔𝟗𝟐 𝟏𝟑𝟎𝟐
𝑺𝑺𝒕𝒓𝒆𝒂𝒕 = + + + − 𝑪𝑭
𝟑 𝟑 𝟑 𝟑
= 𝟖𝟔𝟎. 𝟑

𝑺𝑺𝒓𝒆𝒔𝒊𝒅𝒖𝒂𝒍 = 𝑻𝑺𝑺 − 𝑺𝑺𝒕𝒓𝒆𝒂𝒕 = 𝟑𝟓𝟒. 𝟕

Hypothesis & ANOVA table
Ho : 1=2=3=4
(i.e., Mean Larval population for all varieties are same)
H1: At least two ’s are different
(i.e., Mean Larval population of at-least 2 varieties are not same)

Source df SS MS F
Total 11 1215

Treatments 3 860.3 286.7 6.47

Error 8 354.7 44.3

Decision Rule:- Reject Ho if Fcal  F  ( 3 , 8 ) F(3 ,8 ) = 4.066

Coefficient of Variation
MSE 44.3
C.V  100  100  14.01
GrandMean 47.5
• The C.V indicates the degree of precision with which the treatments are
compared and is a good index of the reliability of the experiment. The CV
expresses the amount of experimental error relative to the size of the data;
small value of CV indicates the reliability of the experiment.
• The C.V varies greatly with the nature of the experiment and the character
measured, so there is no hard and fast rule for the acceptability of a particular
CV value. The experiment is considered as reliable for any value of CV range
from 3% to 20 % (depending upon the nature of the experiment). For the
laboratory experiments, less than 10% is an acceptable value of CV.
Mean comparison test
• Tuckey’s Test (equal replications) • Tuckey’s Test (un-equal replications)

𝑀𝑆𝑒𝑟𝑟𝑜𝑟 𝑀𝑆𝑒𝑟𝑟𝑜𝑟 1 1
𝑤= 𝑞 𝑡, 𝑑𝑓 ;𝛼 𝑤= 𝑞 +
𝑟 𝑡, 𝑑𝑓 ;𝛼
2 𝑟𝑖 𝑟𝑗
44.3
𝑤= 𝑞 4, 8 ;0.05
3
44.3
𝑤 = 4.53
3
𝑤 = 17.41
Muhammad Usman
Comparisons
V1 V4 V2 V3
Means 35.7 43.3 54.7 56.3
b ab a a
V3 – V1 = 20.6*
V3 – V4 = 13
V3 – V2 = 1.6
V2 – V1 = 19*
V2 – V4 = 7.4
V4 – V1 = 7.6 Muhammad Usman
Exercise - CRD
Larval population (%) for three different wheat varieties (V1= LU-26S, V2 = Pak-
81, V3 = Punjab-85, V4 = Faisalabad-83, V5 = Faisalabad-85, V6 = Wadanak-85)
with three plots each was observed and is given below:-

• Test the hypothesis that all varieties have

Varieties
the same larval population? V1 V2 V3 V4 V5 V6
• Calculate Co-efficient of variation (CV) and 13 30 44 19 20 19
interpret it 7 16 45 31 8 15
• If the result is significant than apply the 5 30 32 30 9 10
pairwise comparison test to find out the
significantly different pairs

Muhammad Usman
Example
The effect of 4 different Feeds (A, B, C, D) were to be tested on 12 animals. Each
feed were given to 3 patients and the weight gain is recorded in kgs. Feed A &C
are old while B&D are developed recently. Feeds
• Test the hypothesis that the mean weight gain is same for A B C D
all four feeds 23.6 26.6 23.7 27.8
• If the result is significant than test which Feeds are 25.1 25.9 22.6 26
24 27.7 23.8 27.3
significantly different from each other
• Calculate Co-efficient of variation (CV) and interpret it
• Set up orthogonal contrasts for treatments and test its significance
• Compare new brands with old?
• Which one of the new brands of feed is more effective?
• Which one of the old brand is better?Muhammad Usman
Calculations Feeds
A B C D
23.6 26.6 23.7 27.8
25.1 25.9 22.6 26
24 27.7 23.8 27.3
Totals 72.7 80.2 70.1 81.1 304.1
Means 24.2 26.7 23.4 27
• Correction Factor=(G.T)2/Total observation= (304.1)2/12 = 7706.40
• Total SS = (23.6)2+(25.1)2 …(27.3)2 – CF= 7741.65 – 7706.40 = 35.25

•
• ErrorSS = TotalSS – TreatSS = 5.47
Hypothesis & ANOVA table
Ho : 1=2=3=4
(i.e., Mean relief time from all tablets are same)
H1: At least two ’s are different
(i.e., Mean relief time of at-least 2 tablets are not same)

Source df SS MS F
Total 11 35.25

Treatments 3 29.78 9.93 14.53

Within treatments 8 5.47 0.68

(Error)

Decision Rule:- Reject Ho if Fcal  F(2,8)

Coefficient of Variation
MSE 0.68
C.V  100  100  3.25
GrandMean 25.34
• The C.V indicates the degree of precision with which the treatments are
compared and is a good index of the reliability of the experiment. The CV
expresses the amount of experimental error relative to the size of the data;
small value of CV indicates the reliability of the experiment.
• The CV varies greatly with the nature of the experiment and the character
measured, so there is no hard and fast rule for the acceptability of a particular
CV value. The experiment is considered as reliable for any value of CV range
from 3% to 20 % (depending upon the nature of the experiment). For the
laboratory experiments, less than 10% is an acceptable value of CV. In this
example, the value of CV is not high so the experiment is reliable.
Important Points
• Error degrees of freedom should not be less than 6 in any case. This
rule may also guide you about the required no. of replications for
each treatment
• There should not be less than 2 replications but try to have 3 or 4
replications
• Calculations for un-equal replications e.g., death of an animal during
experiment or no germination in a specific plot
• Mean Comparison test for un-equal replications
• LSD is objectionable and not recommended. The DMR test and
Tucky’s test are useful for inference beyond ANOVA
Muhammad Usman