http://www.unaab.edu.

ng

COURSE CODE: ABG 501

COURSE TITLE: Methods of Animal Experimentation
NUMBER OF UNITS: 3 Units
COURSE DURATION: Three hours per week

COURSE DETAILS:
COURSE DETAILS:

Course Coordinator: Dr. O. Olowofeso, (B. Agric. Tech., M. Sc., Ph.D., RAS)
Email: [email protected]
Office Location: Room 6, COLANIM
Other Lecturers: Prof. M.O. Ozoje

COURSE CONTENT:

Variables, scientific data, experimental process, basic terminologies in animal

experimentation, basic procedures in animal experimentation, measure of central tendency,
testing for goodness of fit, Analysis of Variance, design of experiment, correlation and
regression, sampling and sampling methods.

COURSE REQUIREMENTS:

This is a compulsory course for all final year students in the College of Animal Science
and Livestock Production (COLANIM). The course must be registered for and passed
before students can graduate in any of the five major departments in COLANIM. To do
well in the course, students must have the basic knowledge of mathematics and
elementary statistics. It is also expected that all students must have a very good scientific
calculator and statistical table. Attendance in class is mandatory and students must take
part in all tests and final examination. The minimum score required as pass mark at the
end of the semester for the course is 40%.

READING LIST:

1. Thomas Glover and Kelvin Mitchell: An Introduction to Biostatistics, McGraw-Hill

Companies, Inc. 2001.

2. Mead, R. and Curnow, R. N: Statistical Methods in Agriculture and Experimental

Biology, ELBS Edition, 1987.

3. Cochran, W.G. and Cox, G. M:Experimental designs, A Wiley International Edition

1
 http://www.unaab.edu.ng

LECTURE NOTES

1.0 VARIABLES AND MEASUREMENTS

Statistics is essential concerned with the application of logic and objectivity in the
understanding of events. Necessarily, the events have to be identified in terms of relevant
characteristics or occurrences that could be measured in numeric or non-numeric
expressions.

The characteristics of occurrence is technically referred to as variable because it may

assume different value or forms within a given range of values or forms known as domain
of the variable. The act of attaching values or forms to variables is known as
measurement.

A variable may be qualitative or quantitative and a qualitative variable may be continuous

or discontinuous i.e. discrete. The convention is to identify a variable by the upper case
letter e.g. X and an observation of it by a corresponding lower case e.g. x.

1.1 TYPES OF VARIABLES

1.1.1 QUALITATIVE VARIABLE

A qualitative variable may have forms x that could only be described but could not be
measured numerically. It is a non-numeric entity. Nutrient when identified only as
Nitrogen P, K or Zn is a qualitative variable or factor. Variety of maize is qualitative; sex
of a farmer is qualitative. Yield scores when recorded as high, moderate or low is
qualitative. Qualitative variable has states or forms, which can be described, categorized,
classified or qualified.

1.1.2 QUANTITATIVE VARIABLE

A quantitative variable Y is numeric, it may assume values y that can be quantified. Ages,
weight, height, volume, are all quantitative variables. Production, prices, costs, sales, sizes
are also quantitative.

A qualitative x may be recorded as if a quantitative variable, for instance in a farm survey,

a male farmer may be coded as 0 and a female as 1. In an entomological experiment,
response may be coded as 0 if the plant survives a chemical treatment and 1 if otherwise.
Resistance of plants after exposure to a plant diseases may be scored as 0,1,2,3,4 and 5 in
ascending order of infection and in a taste experiment, the taster may be asked to grade
dishes 1,2,3 and 4, respectively. And on the other hand, a quantitative variable Y may be
coded qualitatively, for example in an agronomic trial, maize plot response may be coded
A for yields exceeding 7400kg per ha, B for yields in the bracket 6001-7500kg, C for
yields between 5001 and 6000kg and D for yields 5000kg and below. Forms A, B, C and
D are of the descriptive types.

2
 http://www.unaab.edu.ng

1.1.3 CONTINUOUS AND DISCRETE VARIABLES

A quantitative variable may be continuous or discontinuous i.e. discrete. The family size
of a farming household, X in a crop survey may take between values 1,2,3,4,5,6,7 and so
on. In the interval (0,100) the number if values which X can take is countable (finite), it is
101, but the size of farms Y cultivated by the household may take an uncountable
(infinite) number of values within even the shortest of intervals when measurement is not
limited by degree of accuracy nor by the precision of measuring equipment. Family size
variable X is a typical discrete variable while the farm size Y is a typical continuous one.
Let Y1 and Y2 be 2 distinct values of a continuous Y the variable may still take an
uncountable number of values between Y1 and Y2 however close the two variables might
have been to each other. On the other hand, a discrete variable X may not take more than a
known finite number of values between two distinct values X1 and X2, however distant the
two values might have been from each other. In a continuous situation, there is always
another value between two values in the discrete environment, there may not be any other
value between two values.

Implication of variable status (and measurement) on data management: The status of a

variable has several implications on data management. It has implication for methods and
other aspects of:
i. Measurement of variable
ii. Data collection
iii. Data organization
iv. Data summarization
v. Data presentation
vi. Data analysis
vii. Data interpretation
viii. Data storage and retrieval
ix. Data use

1.2 SCIENTIFIC DATA

1.2.1 Definition of scientific data
Statistics is about the theory, method and practice of data collection and analysis. It is
about procedures and formulas, and theoretical and practical method of data collection,
organization, summarization, presentation, analysis, interpretation, evaluation storage and
retrieval. A statistical data is acknowledged as any piece of information, quantitative or
otherwise, that has been generated through application of Logic and objectivity. Methods
of animal experimentation are concerned with description, analysis, prescription, and
prediction. Description is scientific reduction of a mass of data into few numbers that
represent salient characteristics of the data. It is a form of mathematical analysis.

1.2.2 Categories of Scientific data

It is convenient and instructive to categories data as follows:
 Routine records
 Survey data
 Experimental data
 Cross-sectional data
 Repeated measures (Time series data)

3
 http://www.unaab.edu.ng

1.2.3 Routine Records

These are general data that are periodically recorded essentially for administration use of
the establishment. Some statistical analysis of the data may be obtained to describe and
analyse the activities and operation and of the establishment and to initiate implement and
monitor future plans, projects and programmes. Planning research and statistics
department in ministries, parastatals, institutes and companies exist for such purpose.
Meteorological data are recorded and kept in weather stations; yield records are obtained
and kept routinely in agricultural plantation and farms.

These categories of data are cheap. But it is often collected without any definite purpose
objectives or focus. The planning may be quiet haphazard, the design or protocol
collection may not promote objectivity while the instrument of data collection may not be
comprehensive, precise and focused enough to permit credible statistical analysis. The
research design is often inadequate or inappropriate.

1.2.4 SURVEY
Surveys are investigations in which the surveyor obtains information about characteristics,
opinions, attitudes, tendencies, activities or operation of the individual unit of the
population. A population is defined as the totality of the units being surveyed or
investigated. For example, a population of households in a rural economic survey, a
population of farms in a farm survey, a population of persons living in a country, a
population of plots in farm survey, a population of trees in a plantation reserve. There are
two types of survey – namely census and sample survey. A census is a complete
enumeration in which every unit in the population is observed, they usually consume large
amount of time, human and material resources. A sample survey involves observation of
only a selected sample of the eligible population units. Sample surveys are cheaper but the
sample must be relevant, objective and representative to have any meaning.

A sample may be:

i. Statistical or non statistical
ii. Regular or adhoc
iii. Probability or non-probability
iv. Perspective or retrospective
v. Confidential or non confidential
vi. Cross-sectional or time series

1.2.5 Statistical and non-statistical samples

A statistical sample otherwise known as a probability or objective sample is such that
sample selections must have been on a random process involving the use of instruments of
chance. There are many types of statistic sample. These are distinguished by the extent of
randomness, the degree of homogeneity and for agglomeration of the population units, the
number of samples, stages, phases or occasions, the mode of selection etc. examples of
statistics samples include:

i. simple random sample

ii. stratified random sample
iii. systematic random sample
iv. cluster sample
v. multi-stage sample
4
 http://www.unaab.edu.ng

vi. multi-phase sample

vii. multi-occasion sample
viii. interpenetrating sub-sample

Non statistical sample are subjective, sample collections are non-random. A most
immediate example is quota sample, which is a common practice in opinion surveys. A
probability sample may be subjected to inferential statistical analysis. But non-statistical
samples would not permit application of probability concepts which elements form the
basis of inferential statistics. Thus such samples can only be described: no generalization
for the target population is valid.

1.2.6 Regular and Adhoc surveys

Some establishments, particularly national statistics offices, carry out surveys on regular
basis. Population censuses should be decennid events, agricultural sample/censuses should
also be decennid. National demographic and health surveys should be periodical. Adhoc
surveys are not regular, they are done when occasion demand. Indeed most research
sample surveys including crop surveys, farm surveys and epidemiological surveys are
adhoc.

1.2.7 Prospective and Retrospective Survey Data

In a prospective survey, observation are made on specified characteristic of the sample
units on a regular basis from a particular time at which the sample is determined until a
specified future point in time. The survey is retrospective when data in respect of the
specified set of characteristics are obtained on the unit from the current point back into the
past point. These two classes of survey data are common in clinical trials involving
patients whose histories are obtained into the past in retrospective survey or into the future
in a prospective survey. Such surveys investigate the progression and dynamics of medical
conditions under specific treatment regimes. Further instances of prospective or
retrospective studies are long term soil fertility experiments investigating the dynamics of
soil fertility over a period of persistent soil use and under various nutrient regimes and
agronomic practices.

1.3 Experimental Data

There are two types of experiments namely comparative experiments and absolute
experiments respectively.
Absolute experiments- Absolute experimental data come from absolute experiment
where there are no differential treatments, no comparisons are intended and no inferential
analysis could be made.
Comparative experiment-These are planned investigations into the effects of treatments
selected and applied to experimental units or plots according to specific notes or
procedures, the plots (or subject or units) may be subjected to regimes of variations that
are imposed by inherent chance factors.
On one hand and some non-chance disturbance factors like blocks, rows and columns on
the other. Designed comparative except ensure allocation of plots to treatments in away
that facilitates isolation and assessment of genuine treatment effects possible. It allows the
experimental choice of treatment for the experiment and the choice of allocation pattern of
treatment to plots. These twin aspects of experimental design are known as treatment
design and experimental design respectively.
5
 http://www.unaab.edu.ng

1.3.1 Cross-sectional and time series of repeated measures data

This terminology identifies the time frame of the data. A survey or experimental data
could be time series or cross-sectional data. A cross sectional or latitudinal data has a
single time scope or frame. The data consists of observations made on sampling or
experimental units over a single point in time, which may be a growing season, a month, a
quarter, a year.

i. A single experiment gives a cross-sectional data.

ii. A census data is cross- sectional.
iii. A farm survey in which the figures are observations made at a point in time is
cross-sectional.
iv. A series of experiment or surveys in which observations are not repeated.
Time series or repeated measures. The time series or longitudinal data are repeated
measures on same experimental or survey units.

 Sets of observations made on each of the experimental units of an experiment

at P different times constitute a time series or repeated measures.
 A long –term experiment in which yields are taken at successive periods on
the same experimental units gives a kind or time sexes data.
 Retrospective or prospective surveys in which observations in respect of
specified medical conditions of a patient are made over time are time series
data.
 Economic data that are observations on same units, economic units, like
household political entities etc. constitutes an economic series that are time
series.

1.3.2 Data collection through Experiment.

Here we are concerned with comparative experiment –explain.
* Persistence in forage to cutting frequency
* Response to different levels of protein
* Livestock systems (cage and floor)
* Breed /strain differences.
For example, we need data to investigate response of maize to varietal and nutrient regime
and weeding methods in a field of plots. The relevant questions include
i. How many varieties? How many and what levels of each of the major nutrients- N
P and K should be invalid? What are the weeding frequencies?
ii. How do we constitute treatment combinations?
iii. If there exists a trend in fertility distribution that makes the plots heterogeneous
how do we arrange the plots and allocate the treatment combinations to isolate and
eliminate effects of plot heterogeneity and facilitates isolation and inferences on
treatment effects.
iv. How we allocate treatments to plots to entrench complete objectivity and permit
statistical analysis.
v. What statistical model do we adopt and what hypothesis do we construct to
facilitate relevant comparisons and tests?

6
 http://www.unaab.edu.ng

Answers to these questions constitute the essential elements of experimental design. All
biological, materials vary; even identical twins treated the same way will have different
weaning weights. The variation is caused by many factors and the role of statistics is to:
i. Measure the amount of random variation due to inherent variability of the
materials
ii. Measure non-random, systematic variation ascribable to treatment effects.
iii. Measure the latter in relation to the former to assess the genuineness of treatment
effects. One of the procedures for this kind of investigation is experimentation, that
is comparative experimentation, which can be achieved through a process known
as experimental design and which may involve feeding trails, cultural practices,
pest control measures, perennial crops, livestock and pasture, fertilizers, plant
population/ spacing, crop rotations, intercrossing etc.

1.3.3 The experimental Process

There are three parts to an experimental process namely.
* Design
* Experiment and
* Analysis
Experiment involves the specification of the objective and formulation of intended
hypotheses. Design has two aspects, namely experiment design and treatment design.
Experiment designs has to do with
i. Treatment structure
ii. Block structure and
iii. Integration of treatment and block structures with treatment structure and it
involves:
a. Determination of plot size, shape and orientation
b. Determination of block size, shape and orientation
c. Allocation of treatments to plots and blocks including order and method or
randomization.
d. Specification of the statistical model of analysis.

Plot and block structures are constructed to ensure minimum plot variability and
maximum precision and efficiency units so dictate. Blocking is done so that units within a
block are as similar as possible e.g. animals of similar weight, age, previous milk yield, or
units of the same “fertilizer status” etc. Blocking or stratification is used routinely in
agronomy trials but this should not be so.

Effective blocking increases precision. Treatment comparisons are made within blocks;
hence variation between blocks is eliminated. Conclusions are valid over the range of
situations represented by different blocks.

Treatment Design has to do with.

i. Selection of factors and levels of factors
ii. Construction of treatments based on appropriate combinations of factor levels.
iii. Selection of treatments for the experiment and
iv. Determination of number of replicates per treatments which is
a. a function of treatment structure whether factorial or otherwise.
b. An issue to be jointly solved by the statistician and substantive field
experts.
7
 http://www.unaab.edu.ng

1.3.4 Basic terminologies

a. Plots are the smallest units of the experimental field plots in field experiments,
pots in green house experiments, pens or individual animals in animal experiments,
farms or farmers in on farm surveys/trials, patients in medical trials, farms in
diseases survey/trails etc.
b. Yields are response of experimental units to treatments. E.g. reproductive yields,
vegetative yields, disease scores, moisture content, number of stands, survival
scores etc.
c. Treatments are applications that can stimulate response. Varieties, lines, nutrients,
procedures, methods and any stimulus that can generate reaction quality as
treatments. These are test and control treatments. The later is the reference and
which the former are assessed. The performance of a test treatment is the
difference between its average yield and that of the control.
d. Replicate are simply repetitions of treatments over plots. A whole experiment may
be replicated over time and space.
e. Blocks are divisions of the experimental materials. They are supposed to be groups
of the experimental units or plots such that plots within same blocks are as
homogenous as possible in all materials aspects, while plots in different blocks are
to be as different as possible.
f. Experimental error is a measure of plot variability. It is an expression for all
variation that can be attributed to the effects of all non-treatment factors and other
unidentified disturbance factors. Experimental error variance is conventionally
denoted as σ2 and the measure of variability is the standard deviation given as σ =
σ2

1.3.5 BASIC PROCEDURES

1. Replication means that instead of having a single plot in each treatment, we have
several smaller ones known as replicates. Replicates are desirable because, it:
i. Enlarges scope of investigation since replicates stimulates varying
environmental conditions.
ii. Enhances precision and overall efficiency by effecting reduction in error by
a factor of the square root of the number of replicates.
iii. Permits determination of experimental error and therefore enables
probability statements about estimates of effects.
iv. Minimizes experimental error because it reduces plot size to a precision
enhancing form.

2. Randomization is the act of allocating plots to treatment purely on the basis of

chance. It is an objective strategy that guarantees that:
i. Every treatment has equal chance (probability) of being allocated to any
given plot.
ii. Treatment allocations are devoid of bias, subjectivity and personal
manipulations.
iii. Statistical estimation and tests of hypothesis on effects as theoretically
valid.
iv. Blocking is partition of heterogeneous plots into homogeneous groups
known as blocks to facilitate isolation of block variation that could distort
treatment effects.
8
 http://www.unaab.edu.ng

3. Blocking is used when variation among experimental units so dictates. It is done so

that units within a block are as similar as possible. In livestock experiment, blocks
may be animals of similar weight, age, previous milk yield, lactation etc.
4. Blocking factors are disturbance factors whose effects are a nuisance that must be
neutralized through appropriate experiment design. It acts as:
i. An error control strategy that when used effectively, reduces error variance
and increases precision and reliability of estimates of effects.
ii. An indirect replication of treatments or experimental design with direct
positive implication for validity and efficiency of experiments by widening
the scope/of inferences and increases precision and overall efficiency
respectively.

1.3.6 Measure of Central Tendency or Measure of Location

It is possible track of the overall picture of data by looking at all the picture at once. One
type of measure useful for summarizing data defines the centre or middle of the sample.
This type of measure is a measure of central location or tendency.
a. The arithmetic mean denoted by x is the sum of all the observations divided by
the number of observation
n

Σ is simply a short way of writing x1 + x2 +…….xn.

i=1

b. The median - the middle values when you arrange from the lowest to the highest.
When you have 2 values as the ones in the middle, the 2 values must be averaged
to get the median.
c. The mode - the most frequently occurring values among all the observation is a
sample. Some distributions have more than one mode. A distribution with one
mode is referred to as unimodal, 2 is bimodal, 3 is trimodal etc.

1.3.7 Measure of Dispersion or Measure of Variation

A measure of dispersion or a measure of variability is an indication of the scatter of
measurement around the centre of the distribution or the opposite of how clustered the
measurements are around the centre.
1. The Range: The difference between the highest and lowest measurements in a
group of data is termed the range. If the measurements are arranged in increasing
order of magnitude as if the median were about to be determined, then:
Sample range = Xn- X1
The range is a relatively crude measure of dispersion measured as it does not take into
account any measurements except the highest and the lowest.

Hypothetical Example 1

xi (g) xi-x (g) Ixi - xI(g) (xi - x)2(g)2

1.2 -0.6 0.6 0.36
1.4 -0.4 0.4 0.16
1.6 -0.2 0.2 0.04
1.8 0.0 0.0 0.00
2.0 0.2 0.2 0.04
2.2 0.4 0.4 0.16
2.4 0.6 0.6 0.36
9
 http://www.unaab.edu.ng

Σxi=12.6g  x  x   0.0 g  xx  2.4 g Σ (xi - x )2=1.12g2

= sum of squares
x = 12.6 = 1.8g
7

Range = x7 – x1 = 2.4g – 1.2g = 1.2g

Mean deviation
 x n- x
i = 2.4
7
= 0.34g

2
S 
n
X i X
2


1.12 g 2
 0.1867 g 2
i 1 n 1 6

S= 0.1867g2 = 0.43g

Sample 2.

xi (g) xi - x (g) x1 - x (g) (xi - x )2 (g2)

1.2 -0.6 0.6 0.36
1.6 -0.2 0.2 0.04
1.7 -0.1 0.1 0.01
1.8 0.0 0.0 0.00
1.9 0.1 0.1 0.01
2.0 0.2 0.2 0.14
2.4 0.6 0.6 0.36
xi = 12.6g (xi- x )= 0.0(g)  xi - x (g) =1.8g  (xi - x )2 =0.82g2

= sum of squares
x = 12.6 = 1.8g
7

Range = x7 – x1 =2.4g – 1.2g = 1.2g

Mean deviation  x -x1 = 1.8g = 0.26g

n 7

S2  
n
X iX

2
0.82 g 2
 0.1367 g 2
i 1 n 1 6

S= 0.1367g2 = 0.37g

2. The mean deviation – The sum of all deviations from the means i.e.  (xi - x )
will always equal zero, however, such summation will be useless as a measure of

10
 http://www.unaab.edu.ng

dispersion. Summing the absolute values of the deviation from the means results in a
quantity that is an expression of dispersion about the mean. Dividing this quantity by n
yields, a measure known as the means deviation or mean absolute deviation of the sample.
It has the same unit as that of the data.

Sample mean deviation (xi - x ) ……………………………….(1)

n
3. The variance – The sum of the squares of the deviations from the mean is called
the sum of squares, abbreviated as SS and defined as.
Population SS = (xi - µ)2 ………………………………. (2)
Samples SS = (xi - x )2 ………………………………….. (3)
The mean sum of squares is called the variance and for a population is denoted as  2
(sigma squared).

σ2 = (xi - µ)2 ……………………………….(4)

N

The best estimate of the population variance is the sample variance S2

S2   i
X  X 2 ............................................(5)
n 1
If in equation 4, we replace µ by x and N by n, the result is a quantity that is a biased
estimate of σ2. The dividing of the sum of squares by n-1 (called the degree of freedom
abbreviated Df) rather than n, yields an unbiased estimate. It is equation 5 that should be
used to calculate the sample variances. The calculation of S2 can be tedious for large
samples but it can be facilitated by the use of equality.

Sample SS = x12 (x1)2

 n

Sample 1
Xi (g) X2i (g2) xi (g) X2i (g2)
1.2 1.44 1.2 1.44
1.6 2.56 1.6 2.56
1.7 2.89 1.7 2.89
1.8 3.24 1.8 3.24
1.9 3.61 1.9 3.61
2.0 4.00 2.0 4.00
2.4 5.76 2.4 5.76
xi=12.6g x i=23.50g2
2
xi=12.6g x i=23.50g2
2

n=7 n=7

x = 12.6 = 1.8g x = 12.6 = 1.8g

7 7

CSS = x12 (x1)2 CSS = 23.50g2 – (12.6g)2

 n 7

11
 http://www.unaab.edu.ng

= 23.50g2 – (12.6g) = 0.82g2 (CSS is called corrected sum of squares)

7

= 23.50g2 – 22.68g2 S2 = 0.82g2 = 0.1367g2

6
= 0.82g2 S = 0.1367g2 = 0.37g

S2 = SS V = 0.37g
n-1 1.8g

= 12.6g = 0.1867g2 = 0.21 = 21%

6
S = 0.1867g2 = 0.43g

S 0.43g
V    0.24  24%
X 1.8 g

The variance has square units. If measurements are in grams, their variances will be in
grams squared or if measurements are in cubic centimeters, their variance will be in terms
of cubic centimeters squared even though such squared units have no physical
interpretation.

4. The standard deviation is the positive square root of the variance therefore, it has
the same units as the original measurements. Thus for a population.

X i
2
– (x i )2
S= N
N

And for a sample

X i
2
– (x i )2
S= n
n-1

4. The coefficient of variation or variability is defined as

S S S
V  orV  100% sin ce is generally a small quantity, it is
X X X

frequently multiplied by 100% in order to express V as a percentage.

The standard error
S.E (X) = S or SD
n n
where n is number of observation(s)

  S  SD  Standard deviation
12
 http://www.unaab.edu.ng

2.0 TESTING FOR GOODNESS OF FIT

2.1 CHI SQUARE GOODNESS OF FIT

Example 1
A plant geneticist may raise 100 progeny from a cross that is hypothesized to result in a 3:
1 phenotypic ratio of 84 yellow: 16 green is observed, although out of this total of 100
plants, the geneticists hypothesis would predict a ratio of 75 yellow: 25 green. The
question to be asked, then is whether the observed frequencies (84 and 16) deviated
significantly from the frequencies expected if the hypothesis were true (75 and 25). The
statistical procedure for attacking the question first involves the concise statement of the
hypothesis to be tested. The hypothesis in this case is that the population which was
sampled has 3:1 ratio of yellow-flowered to green-flowered plants. Statistically, this is
referred to as null hypothesis (abbreviated Ho) because it is a statement of “no difference”
in this instance, we are hypothesizing that the population flower colour ratio is not
different from 3:1. If it is concluded that Ho is false, then an alternative hypothesis
(abbreviated HA) will be assumed to be true. In this case, H A would be that the population
sampled has a flower colour ratio which is not 3 yellow: 1 green. One states a null
hypothesis and an alternative hypothesis for every statistical test performed and all
possible outcomes are accounted for by the two hypothesis.
k

X2 =  (fif- fi)
i-1
2

Here, fi is the frequency or number of counts, observed in class i

f is the frequency expected in class I if the null hypothesis is true.
The summation is performed over all K categories of data.
Calculation of Chi square of fit data consisting of 100 flower colours to a hypothesized
colour ratio of 3:1.
Ho: The sample data came from a population having 3:1 ratio of yellow to green
flowers.
HA: The sample data came from a population not having a 3:1 ratio of yellow to green
flowers.

Category (flower colour)

Yellow Green n
fi 84 16 100
(f) (75) (25)
Degree of freedom = v = k – 1 = 2 – 1 = 1
k

X2 = 2
 (fif- fi)
i-1

= (84-75)2 + (16-25)2
75 25
92 + 92
75 25

13
 http://www.unaab.edu.ng

= 1.080 + 3.240
= 4.320
0.025 < P < 0.05

Therefore, reject Ho

Example 2

In series of matings involving black-polled cattle, a researcher found 54 black-pooled, 18

red-pooled, 25 black horned and 8 red-horned offspring. Two genes are known to be
involved and a ratio of 9:3:3:1 was to be expected. Were these results in agreement with
this ratio?
Observed (O) Expected (E) O–E
54 59.0625 -5.0625
18 19.6875 -1.6875
25 19.6875 5.3125
8 6.5625 1.4375
X2 = (-5.0625)2 + (-1.6875)2 + (5.3125)2 + (1.4375)2
59.0625 19.6875 19.6875 6.5625
= 2.327.

Since calculated X2 is less than X2 .05. (4-1) = 7.815, we do not reject the hypothesis i.e. the
observed results are in agreement with the expected 9:3:3:1 ratio.

2.2 CONTINGENCY TABLES

Experimental subjects may be classified according to several attributes and enumeration
obtained on each cell. Such data are arranged in contingency tables and the researcher is
interested in testing for independence of the classification variables.
e.g.
Consider a population of 150 animals challenges with a certain disease. The following
data represent numbers of males and females resistant and susceptible to the disease. The
hypothesis to test is that resistance or susceptibility is independent of sex.

Resistamt Susceptible Total

Male 35 28 63
Female 50 37 87
Total 85 65 150

Above is referred to as a 2 x 2 contingency table.

The expected frequency in a contingency table are obtained by the formula
Fij = (RiCj)
n

Where Ri is the total frequency in row i, Cj is the total frequency in column j and n is the
total sample size. The number of degrees of freedom in a contingency table is
given as
d.f.= (r-1) (c-1)
Where r= number rows and C = number of columns in the table,
For example, expected cell frequencies are as follows
14
 http://www.unaab.edu.ng

Resistant Susceptible Total

Male 35.7 27.3 63
Female 49.3 37.7 87
Total 85.0 65.0 150

The formula for the Chi-square test statistic for contingency tables is
x2= ΣiΣj (Observed frequency – Expected frequency)
Expected frequency

Here x2 = (35 -35.7)2 + (28 – 27.3)2 + (50 – 49.3)2 + (37 – 37.7)2

35.7 27.3 49.3 37.7
= 0.014 + 0.018 + 0.010 + 0.013
= 0.055

Since calculated x2 = 0.055 is smaller than x2(0.05,1) = 3.841

The test is not significant and we conclude that the frequencies of resistant and susceptible
individuals are the same for both sexes. In order words, resistance and susceptibility are
independent of sex.

2.3 STATISTICAL ERRORS IN HYPOTHESIS TESTING

One needs an objective criterion for rejecting or not rejecting the null hypothesis for a
particular statistical test. Theoretically, a very large x2 values might be calculated even
when the hypothesis is true although the larger the Chi-square, the smaller is the
probability that Ho is true. But how small a probability (i.e how large x2) shall we require
to reject the null hypothesis? As explained earlier, a probability of 5% or less is commonly
used as the criterion for rejection of Ho. The probability used as the criterion for rejection
is turned termed the significance level denoted by α, and the value of the test statistic
corresponding to this probability (e.g. x2 = 3.841 for the 5% significance level) is the
critical value of the statistic.

It is very important to realize that a true null hypothesis occasionally will be rejected
which of course means that we have committed an error. Moreover, this error will be
committed with a frequency of α. That is, if Ho is in fact a true statement about a statistical
population, it will be concluded (erroneously) to be false 5% of the time. The rejection of
a null hypothesis when it is in fact true is a Type 1 error (also called an error, or an α error
of the first kind). On the other hand, if Ho is in fact false, our test may occasionally not
detect this fact, and we shall have reached an erroneous conclusion by not rejecting Ho.
This error of not rejecting the null hypothesis when it is in fact false is a Type II error
(also called a β error, or an error of the second kind).
The two types of errors in hypothesis testing
If Ho is true If Ho is false
If Ho is rejected Type I error No error
If Ho is not rejected No error Type II error

15
 http://www.unaab.edu.ng

2.4 COMPARING MEANS

To test one-way paired sample hypothesis –concerning treatments or sample, means, the
students t-statistic due to William Gosset is used.
e.g.
consider the following data on weaning weight (kg) of creep fed and non creep fed calves.

Creep fed Non creep fed

84.5 83.1
86.9 50.0
75.0 61.4
88.7 74.7
85.4 62.5
88.9 62.5
78.5 46.9

x1 = 83.9857 x2 = 63.0143
S1 = 5.291 S2 = 12.729

HYPOTHESIS 1: HYPOTHESIS INVOLVING THE MEAN (µ)

Suppose the researcher wants to determine whether the average weight of his creep fed
calves compares favourably with literature reports of 87kg. This is a one sample
hypothesis .
Ho: µ = C i.e 83.9857 = 87.0
Ho:   C i.e 83.9857 = 87.0

The test statistic is

t=x– c
Sx
Which is compared with t a from a table of the t- distribution

e.g. /t/ = 83.9857 – 87.0

5.291 SD = -1.507
√7 √n
since this t value is smaller than t(0.05,6) = 2.447 we do not reject Ho and conclude that
there is no difference statistically between the two means being tested.
i.e tc<tt or -1.507<2.447

HYPOTHESIS II: Test of hypothesis involving two samples

If the researcher wants to test the mean weaning weight of creep-fed calves is significantly
different from that on non-creep fed individuals, he performs a two sample or two
treatment t test.
Ho: µ1 = µ2
Ho: µ1 ≠ µ2
x1 – x2
SD1 _ SD2
The test statistic is t- √n1 √n2

16
 http://www.unaab.edu.ng

Where SD1 _ SD2 = the standard difference between means or S x1 – x2

√n1 √n2

= S2 p + S2 p

n1-1 n2-1
Where n1 and n2 = the total number of observations used to compute the first and second
means, respectively and
S2p = Pooled sample variance

= SS1 + SS2 = (n1 – 1)S12 + (n2 – 1)S22

n1 + n2 - 2
For our example
S2p= 6(27.99468) + 6(162.02744)
7 + 7 -2
= 95.01106

Hence S x1 – x2 = 95.01106 x 2 = 5.6276

6

Thus calculated t = 83.9857 – 63.0143 = 3.7265

5.6276

Since this t is greater than ta, n1 + n2 – 2 or t(0.05,12) = 2.179. we reject Ho and conclude that
the two means are significantly different. In other words, creep fed calves had
significantly higher weaning weights than non-creep fed calves.

3.0 CORRELATIONS
When pairs of observation are available from a bivariable normal distribution, the
researcher may be interested in determining the strength of the relationship between the
two variables concerned. The quantity that describes this is the correlation coefficient. Its
sign indicates the direction of the relationship. Such relationship as between body weight
and egg products, body weight gain and feed consumption, milk yield and butterfat
percent, body weight and egg weight, body weight and height at withers. Shell thickness
and % calcium in plasma, sperm concentration and fertility and so on are of considerable
interest to the Animal Scientist.

The sample correlation coefficient is computes by:

r= Σxy – (Σ x) (Σy)/n
√Σx2 – (Σx)2 √Σy2 – (Σy)2
n n

Compute the sample correlation coefficient from the following data on body weight (BW)
and shank length (SL) of 20 six week old broilers.

17
 http://www.unaab.edu.ng

BW (kg) SL (cm) BW(kg) SL(cm)

0.63 4.8 0.90 5.2
0.80 4.0 0.72 4.7
0.78 5.1 0.66 4.7
0.83 5.0 0.66 5.0
0.74 4.7 0.66 4.7
0.77 4.7 0.60 4.6
0.74 4.8 0.9 5.1
0.85 5.2 0.83 5.0
0.66 4.5 0.77 5.0
0.89 5.2 0.84 5.2
Let BW = X and SL =Y

Compute the following

ΣX = 15.23
ΣY = 97.20
ΣX2 = 11.7651
ΣY2 = 474.12
ΣXY = 74.30
Then
r= 20x74.3 – (15.23)(97.20)
-
√[(20x11.7651) 15.232][√(20x474.12)– (97.20)2

1486  1480.36
  0.52
3.3491X 34.56

The standard error of the estimate is

Sr = 1 – r2
n–2

= 1 - (0.52)2 = 0.041
18
To test whether the correlation coefficient was significant or not, we compute the t statistic
t= r = 0.52 = 12.68
Sr 0.041
And compare with t  , n-2, here t(0.05,18) = 2.101. We conclude that the correlation was
highly significant P< 0.01.

4.0 ANALYSIS OF VARIANCE

Analysis of variance is essentially a process of partitioning the variations in observed data
into its various components. It is normally presented in the form of a summary table
known as the ANOVA table.
An analysis of variance table can be for one-way or two classification depending on the
number of criteria used for classifying the data. In one-way analysis, we consider only one
criterion or factor as affecting yield, for example. In a two-way analysis we consider two
factors as affecting yield.

18
 http://www.unaab.edu.ng

4.1. One way classification

Consider the yield equation:
Yij = µ+ ti + Σij
Where µ = population mean- the inherent effect
ti = the treatment effect
Σij = the error effect which is due to nature, randomness or any other factors
beyond human control.
In the above equation, the only factor considered to affect yield is the treatment
applied. The equation is based on the following assumptions.
1. The treatment effect, t is fixed.
2. The total effect of the treatment is equal to zero i.e. Σij =0.
3. The expected value of error is equal to zero i.e. ΣE =0
4. The error is normally and independently distributed with mean 0 and
variance σ2. i.e: LIN (0, σ2).
In other to test the significance of the treatment effect on yield, we use the F-test. The
following is a summary of the ANOVA.
ANOVA Table – One way classification
Source Degree of Sum of squares Mean square F-ratio
freedom
n
Total adjusted n-1 - -
ΣYij – CT
for the mean ij

Treatment t-1 SS Treatment MS treatment

ΣT
i r
i
2
– CT
DF treatment MS error
Residual n-t SS Total – SS SS residual -
Treatment DF residual
Where
CT = Correction term
(ΣYij)2
N
N= total number of observations
t= number of treatments
r= total number of observations per treatment.
ΣYi = ΣYij Therefore, the sum of all observations per treatment.
i

ΣYij2 = sum of the squares of all the observations per treatment

DATA TABLE
Treatment 1 2 3 4 5 Total Yi
1 Y11 Y12 Y13 Y14 Y15 Y1
2 Y21 Y2
3 Y3 Y3
4 Y45 Y4
5 Y53 Y5
Total Yj Y4 Y
Y1 = Sum of observation on the first treatment
Y2 = Sum of observation on the second treatment
19
 http://www.unaab.edu.ng

Y3 = Sum of observation on the third treatment

Y4 = Sum of observation on the fourth treatment
Y5 = Sum of observation on the fifth treatment

Example of one-way classification,

Replication (Block)
1 2 3 Total Yi
1 7 2 4 13
Treatment 2 6 4 6 16
3 8 4 5 17
4 7 4 2 13
Toyal Y j 28 14 17 59

Summary of the data

1. ΣYij2 = 72 + 22 + 22 + 42 + 62 + 42 + 62 + 82 + 42 + 52 + 72 + 42 + 22 = 331
2. C.T. = (ΣYij2)
N
2
=59
12
= 290.0833
3. ΣYij2 = 132 + 162 + 172 + 132
R 3 3 3 3
= 56.3 + 85.3 + 96.3 + 56.3
= 294.2
4. ΣYij2 = 282 + 142 + 172
t 4 4 4

ANOVA TABLE
Source of d.f s.s m.s F-ratio
variation
Total a.f.m 11 331-290=41 3.7 -
Treatment 3 4.2 1.4 14<1 = 46
0.304
Residual 8(n-t) (41-4.2)36.8 4.6

4.2 Statistical inference

The purpose of the analysis is to be able to draw inferences on the significance of
the effect of the factor we are considering. In doing this we follow the following
procedure.
i. Setting up of a hypothesis: Firstly, we set up a null hypothesis that the
treatment effect is not significant i.e.
Ho:ti = 0
We also set up an alternative hypothesis that the treatment effect is
significant thus:
HA:ti = 0

20
 http://www.unaab.edu.ng

ii. We decide on the level of significance (not given) and find the tabulated F-
ratio from the Table.
iii. We compare the tabulated F- ration with the computed F – ratio.
iv. Conclusion. We accept the null hypothesis if the F-ratio calculated is less
than the F-ratio tabulated. Otherwise, we accept the alternative hypothesis.
In the above example for the one-way classification, we are testing for the
treatment effect and our null hypothesis will be:
Ho: t1 = t2 = t3 = t4 = 0
i.e. there is no significant difference between the treatments or that the
treatment has no effect.
The alternative hypothesis is
HA:µ = µ = i
At  (level of significance) = 5%
F0.05 (3,8) = 4.07
F computed = 14
46
Therefore, FC< FT, we accept the hypothesis that there is no treatment effect.

4.3 Two –way classification

The two-way classification of variance is basically the same with the one-way
classification except for the fact that while only one factor is considered to affect yield in
the one way, two factors are considered to affect yield in the two way analysis. It is very
useful where variations in observed data exist due to blocking or grouping of the
experimental units. The two-way classification excluded such variation from those due to
the applied treatments and thereby increases the precisions of the conclusions on the
experiment.
The yield equation is
Yield = Population mean + treatment effect + Blocking effect + error term
Yij    Ti  Pj   ij

Where Pj = the blocking effect. The basic assumptions underlying the equation are the
same with the one-way classification.
The working formulas for the degree of freedom, sum of squares, mean squares, and F-
ratio are given in the table below:
Correction term C.T. = (ΣYij2)
N
Source of Degree of Sum of squares Mean square F-ratio
variation freedom
n
Total adjusted n - 1 - -
ΣYij2 – CT
for the mean ij

Treatment t-1 SS Treatment MS treatment

ΣY
i b
i
2
– CT
DF treatment MS error
Block B-1 SS block MS block
ΣY
i t
i
2
– CT
DF residual MS error
Error (b-1)(t-1) By subtraction

21
 http://www.unaab.edu.ng

CT = Correction term
N = total number of observation
B = number of blocks
T = number of treatments.

Numerical Example:
Given the data table used in the one-way classification, the ANOV table is given below.
Source of Degree of Sum of squares Mean square F-ratio
variation freedom
Total adjusted 11 41 3.7
for the mean
Treatment 3 4.2 1.4 14<1= 8.5
16
Block 2 27.25 13.65
Residual 6 9.55 4.6

Test of hypothesis.
For blocks.
H0:p1 = p2 = p3 i.e There is no block effect
HA: p1 = p1: 1 = 1 i.e There is block effect.

F 0.05 (2,6) = 5.14

F computed F tabulated (8.5 5.14). There is block effect.
We reject that there is no difference between the block means as that the block means are
the same.

4.4 Two-way Classification with Sub-samples.

When more than one observation is made per treatment an additional source of
variation exists.
This is the variation which exists between units (sub samples) which are given the same
treatment. Separately this source of errors from the residual error, enables us to arrive at a
more valid conclusion.
This analysis has the advantage of being more precise because it recognizes the fact that
observations made in any experimental unit cannot be exactly the same.
The yield equation is given below.
Yijk = µ + Ti+ (tp)ij + Σijk

Where (tp) ij = Interaction effect.

The formulas for computing the ANOVA table are given in the following table.

Source Degree of Sum of squares Mean E(MS) F-ratio

freedom square
Total n-1 n TSS - -
adjusted for m=rts Σ Yijk2 – CT
ijk
the mean
Treatment t-1 Σ Yi..2 – CT TrSS σ2 ε + S tp σ2 + RMS
rs T-1 rs σt2

22
 http://www.unaab.edu.ng

Block r-1 Σ Yi..2 – CT =BSS BSS σ2 ε + S tp σ2 + BMS

rs r-1 Stp σ2 RMS
Treatment (r-1)(r-1) Σ Yij2 – CT – BSS – TrSS TrB – SS 2
σ ε + S tp σ 2
TB.MS
Block t-1 Ij s (t-1)(r-1) RMS
Interaction r-1
Error n-(t-1) TSS – TrSS – BSS –Tr X BSS σ2 ε -
n-(r-1) BSS (d.f)R
-(r-1)
(r-1) + 1

The Treatment by Block interaction (i.e. T X B) is a sort of Residual. From the table,
when dealing with treatment, what we want to test is rso2 y T i.e whether the effect due to
treatment is significant or not (or to find out whether it is 0 or not, thus we find from the
chart or ANOVA Table, under E(MS) one equation which contains σ2ε and sσ2 TP to make
σ2 ε & σ 2T = 1 and the error or equation to use is TXB effect.

This E (MS) of the TXB interaction is used to test the E (MS) of both the treatment
and the block, effect, this is because the E (MS) of TXB interaction contains both σ2ε &
SσTp 2 i.e. the overall error and the joint error respectively which occur in both the
treatment effect which is Sσ 2Tp, we use the Residual error which is σ 2ε.
The E (MS) is a guide to enable one tell what to test what.
Note:
σ 2ε = Overall effect
SσTp 2 = joint effect
rso2τ = Treatment effect alone
St σ 2 = block effect alone

CT = Y…..2
Rts

4.5 ANOVA FOR A LATIN SQUARE EXPERIMENT.

Here we look at
(a) Row = r
(b) Columns = c
(c ) treatments = t
What differentiates the Latin square from the 2 way classification is that in the way, we
can have different numbers of blocks, but in the Latin square experiment. The numbers of
rows must be equal to the numbers of column.
i.e. r = c ---------- 2 way classification
r=c=t ------------ Latin Square
Thus if there are 4 treatments, there must be 4 rows and 4 columns.
It is recorded as columns Treatments
23
 http://www.unaab.edu.ng

1 2 3 4 Yi
1 A B C D
2 D A B C
3 C D A B
4 B C D A
Y.j Y..
From the above chart, there must be 1 treatment/row/column i.e. all treatments must be
represented in every row and column in such a way that no treatment occurs twice in any
row or column.
In the above treatments A, B,C,D are observations.
The column and row totals are computed as Yj & Yi respectively.
For the Treatment total Tr T, there will be one for
AT = A1 + A2 + A3 + A4
BT = B1 +B2 +B3+B4
CT = C1 + C2 +C3 + C4
DT = D1 + D2 +D3 +D4
i.e. pick all the As i.e. A1+A2+A3+A4 = ass up for AT. Same for B, C, D.
Once this is done the ANOVA is simple.
Source Degree of Sum of squares Mean E(MS) F-ratio
freedom square
Total n= r x c TSS - -
adjusted for n - 1 Σ Yij2 – CT n-1
the mean
Row r-1 Σ Yi..2 – CT RowSS σ2 ε + cσ2p RowMS
C r-1 RMS
Column c-1 Σ Yi..2 – CT ColSS σ2 ε + σ 2 CMS
j r c-1 RMS
Treatment t -1 Σ Yj2 – CT Tr. SS σ2 ε + σ 2 Tr.MS
j r (t-1) RMS
Error n- r – c- t By substitution Substitution σ2 ε -
+2
* Yt = YtA + Ytb + Y
Summary by spacing
A (2”) = 203 + 283 + 279 + 334 + 250 = 1,349
B (4”) = 257 + 252 + 266 + 280 + 259 = 1,314
C (6”) = 231 + 204 + 280 + 287 + 260 = 1,262
D (8”) = 245 + 271 + 227 + 246 + 202 = 1,191
E (10”) = 182 + 230 + 245 + 195 + 228 = 1,188
Calculations:
C.T = Y.02 = (6,304)2 1,589,617
rc 25
Total = Yij2 -CT = (257) 2 +……….+ (338)2 - CT = 36,571
Row = Yi2 -CT = (1,255) 2 +……….+ (1440)2 - CT = 13,601
C 5
24
 http://www.unaab.edu.ng

Column = Yj2 - CT = (1,228) 2 +……….+ (1309)2 - CT = 6,146

5
Spacing = Yt – CT = A + B + C + D2 + E2 - CT
2 2 2 2

r
Residual = 36,571 – (13,601 + 6146 + 4156) = 12,668

5.0 Design of Experiment

In carrying out scientific experiments we are very often interested in comparing the effects
of certain treatments on the elements we are investigating. Alternatively, we may concern
ourselves with the determination of some properties of our elements of interest. In
carrying out the investigation, we will discover that certain factors will affect our results.
Some of these factors are controllable while others are not. By designing the experiment in
a particular way we may be able to observe the factors that we can control as against those
we cannot control and thereby be able to make inferences on them. The right design has to
be employed in any experiment if the results are to give valid conclusions. However, what
type of design we employ will depend on the particular experiment and on the
characteristics we are investigating.

Time personnel and funds are also important.

Requirements of a good experiment

To be able to carry out a good experiment the following conditions are important.
(1) Absence of systematic error: this implies that the experimental units which
are designed to receive one treatment should not be different in any
systematic way from those receiving other treatments. If they do, a source
of variation (systematic error) is introduced and this will affect the validity
of inferences made from the result of the experiment.
(2) Precision: The treatment comparison must be as precise as possible. By a
measure of precision of the estimate, e.g. mean X however, precision
refers to a measure of how close to the set of possible sample estimates for
a particular sample design may be expected to come to.
(3) Range of validity: the experiment must allow for a wide range of validity
on the conclusion. This means that as the units for the experiment is carried
out; the wider is the range of validity of the conclusions.
(4) Simplicity: As much as possible, the experimental arrangement should be
simple. If the experiment is not simple, them the procedure might be
difficult to follow especially if the experiment is to be carried out by
unskilled people.

25
 http://www.unaab.edu.ng

(5) Also, for a good experiment, the error must be assessable. We must be able
to measure the amount of uncertainty involved in the experiment. This
means that we must have a set of experimental units responding
independently to the same treatment and such units must differ only
randomly from the units of other treatments.
(6) The experiment must not only be simple, precise and its error measurable,
it must also be cheap.

5.1 Completely Randomized Design

The completely randomized design is appropriate when the experimental units are not
known or anticipated to be different from one another. In this case any variation in the
results observed from the units will be largely due to the treatment applied.

In the completely randomized design, the experimental units are numbered and the
treatments are randomly assigned to them. This means that all the units will have the same
chance of being given or not given a particular treatment. The one way classification is
used in the analysis of variance for this design.

The completely randomized design has the following advantages:

1. It is simple to carry out and analyse.
2. The design is flexible in that the number of treatments and replicates are limited
only by the number of experimental units.
3. The loss of information due to missing data is small related to other designs.

Disadvantages
The mean disadvantage of this design is that it does not take cognizance of other source of
variation. Any variation is assumed to be due to be treatments applied. Where the
experimental units are not homogenous (apart from treatments) this can lead to erroneous
conclusions.

5.2 Randomized Complete Block Design

This design is used where there are other sources of variation apart from the treatment. By
the use of this design, we are able to determine the significance of those additional sources
of variation. (e.g. the block effect). If the block effect is significant, it means that the
precision of the experiment has been increased by the use of this design.

The first step in the design is to group the experimental units into blocks. The main
purpose of grouping is to make sure that all the units in a block are as uniform as possible
so that observed variation between blocks will be due to the treatment effect. After the
blocking, the treatments are then randomly applied to the units of each block. For
example, we might be interested in observing the effects of various feed components
(treatments) on the weight gain of cattle. If we have a herd of cattle consisting of animals
26
 http://www.unaab.edu.ng

of difference age group variations in the weight gain by the animals might not be due to
the differences in age. To be able to observe the variation due to age, it then becomes
necessary first to group the animals according to age before applying the various feeds.

The two-way classification is used in the analysis of randomized complete block design.

5.3 Factorial
In factorial analysis we deal with factors.
Let us look at a two-factor factorial.
Let Factor A = Variety (2 varieties)
Factor B = Fertilizer (N) (2 levels of N. fertilizer).
In this case we talk of a 2 X 2 or 22 factorial.

If we call the first level of factor A 0 and its 2nd is 1, we therefore have.
0 0 00
1 01
1 0 10
1 11

A N O VA
Source d.f
Total 3
A 1
B 1
Residual AB 1
Let us assume that we planted something at the green house and the same thing is repeated
on the farm, i.e. there now a 3rd factor which is location, we would then have 3-factor
factorial.
Thus A = Variety (3) 15 15 15
B = Fertilizer NPK (3) 15 20 15
C = Location (2) 15 30 15

This is a 32 by 2 (3x3x2) or 32 x 2factorial. We can make a design for this type of factorial
thus:
A factorial experimental is one in which an equal number of observations is made for all
possible combinations of the levels.

Design of the experiment

Variety 0 1 2
Fertilizer 0 1 2 0 1 2 0 1 2
Location 0000 010 020 100 110 120 200 210 220
1001 011 021 101 111 121 201 211 221
After this, randomize following the same procedure

27
 http://www.unaab.edu.ng

ANOVA
Source Df SS MS
3 3 3
Total 3 x 3 x 2 -1 = 17
ΣΣΣ y2 ijk – CT
i j k
3
Variety 3 -1 = 2
Σ yi2 – CT
i =1
6 j k
Fertilizer 3–1=2 Σ y.j2 – CT
i
6
V&F (3- 1)(3-1)
(2) (2)
=4

0 1 2
0 00 10 20 Y.1
1 01 11 21 Y.2
2 02 12 22 Y.3
Y1… Y2… Y3… Y….

Factorial experiment is one designed to examine the effect of one or more factors, each
factor being applied at two levels at least so that differential effects can be observed. It
investigates all possible treatment combinations which may be formed from the factors
under investigation. The level of a factor denotes that intensity with which it is brought to
bear. It may be measured quantitatively as when fertilizer is applied to plots in a given
wt/unit area or qualitatively as when patients are considered two levels inoculated and not
inoculated.
If we have 2 factors A and B each with 2 levels-0,1 and 0, 1 in effect we shall have 4
treatments -00, 01,10, and 11 as seen on the chart.
2 X 2
A B
0 1
0 00 01
1 10 11

If this supposed experiment is repeated 4 times so that we have 4 blocks –or replicates as
shown in the table below:

Treatments
00 01 10 11
1
Block 2
3
4

We can thus set up our ANOVA table.

28
 http://www.unaab.edu.ng

ANOVA
Source d.f SS MS F-ratio
Total 15(16-1)
Blocks 3(4-1) MSn MSB
MSE
Treatments 3(4-1) MSr MST
MSE
Factor A 1 MSA MSA
MSAB
Factor B 1 MSB MSB
MSAB
A and B 1 MSAB MSAB
MSE
Residual 9 (15-3-3) MSE

In calculating the F-ratio, we test the blocks and treatment with the Residual which is
above (9). In calculating the F-ratio for the Factors A&B, we use the MS for AB to test
for factors A,B while we use the overall residual to test for the AB.
Example
N = 2 levels
K = 2 levels
Coded yield of maize

Replicate Treatment
00 01 10 11 Total
1 2 3 3 4 12
2 5 7 4 5 21
3 2 4 5 6 17
4 4 3 2 1 10
Total 13 17 14 16 60
N2 = A
P
0 1
0 00 10 27
13 14
1 01 11 33
17 16
30 30 60
=B

ANOVA
RSS = 122 + 212 + 172 + 102 + 602
4 4 4 4 4
TSS = 132 + 142 + 172 + 162 + 602
4 4 4 4 4
ASS = 302 + 302 + 3600 = 0
4 4 16

29
 http://www.unaab.edu.ng

BSS = 272 + 332 - 3600 = 4.2

8 8 16

Source Df SS MS F
Total 15 19 - -
Replicate 3 10.5 3.5 3.5/7 Significant
Treatment 3 2.5 0.83 0.83/0.7
A 1 0 0.0 0
B 1 2.2 2.2 2.2/0.3 Significant
AB 1 0.3 0.3 2.2/0.7
Residual 9 6 0.7

3x3 Factorial i.e.32 – 2 factors each at 3 levels.

If we have 2 factors with each at 3 levels
The 2 factors are A&B
3 levels are 0, 1, 2.

The treatments will be 9 treatments thus

0 1 2
012
00, 01, 02,10,11,12,20,21,22
23-3 factors at 2 levels.
0 000
0 1 001
0
0 0 010
1 011

0 100 8 treatments
0 1 101
0
0 0 110
1 111

Here there are 23, 3 factors each at 2 levels and thus we have 8 treatments. If we assume
there are 5 replicate each with 8 treatments.
1 2 3 4 5 6 7 8 Y1
Replicates 1
2
3
4
5 40
Y5 Y

We can set up the ANOVA Table.

There are 3 models which we can use in this.

30
 http://www.unaab.edu.ng

In model 1- we use the Residual to test all eff4ects; in models II and II we use the
expected mean square as a guide for determining appropriate denominator for testing.
The ANOVA table is as follows.

ANOV TABLE
Source d. f SS MS F-ratios
Total 39
Replicate. 4
Treatment 7
A 1
B 1
AB 1
C 1
AC 1
BC 1
ABC 1
Residual 28

Model I- All factors fixed

Model II – All factors random
Model III - Some factors random some fixed.

A 0 1
B 0 1 0 1
C 1 0 1 0 1 0
ABC 001 10 011 100 101 110 111
000

Exercise:
1. Explain what we mean by Factorial Design
2. Differentiate between Factorial design and completely randomized design.

6.0 LATIN SQUARE DESIGN

The LS design has two basic restrictions in that the number of rows = number
of columns = number of treatment and no one treatment must appear twice in the same
row the same column. Such a design is presented below.

A B C D E A E D C B
E A B C D B A E D C
D E A B C C B A E D
C D E A B D C B A E
B C D E A E D C B A

We must now randomize such that the restrictions above still hold. The randomization can
be by column or by row. This can be done by changing the column to row. The analysis of
variance for the Latin square Design is the same as done in the past
The Latin square design is like randomized complete block treatment but different
in that no one treatment must appear twice in the same row or column. Thus the Latin
square is a diagonal effect and vertical effects.
31
 http://www.unaab.edu.ng

7.0 SPLIT PLOT DESIGN

Suppose there are 2 factors A & B. A has 3 and B has 2 levels- and we want to use
Latin methods. To use the Latin square we find the treatment combination so as to
randomize.

A 3 levels
6 treatments
B 2 levels
For the square design, we shall have 6 rows, 6 columns and 6 treatments.

A O 1 2
B O 1 0 1 0 1
OO O1 10 11 20 21
We assume that factor A requires a large piece of land while factor B does not. We can
divide the whole piece of land into Blocks e.g. the piece of land is divided into 3 blocks.
Each block can be divided up into whole plots and each whole plot is divided again into
sub plots. This processed will happen to each block.

Whole

A C D

Sub Plot a1
a2
a3
a4

If the factor A is the methods of land preparation and each factor A is divided into 4 plots
we have a1,a2 , a3, a4. while factor B is the varieties of crop (maize) to plant and because
there are 2 levels of B or 2 Sub plots per whole we have b1, b2. we randomize factor A in
to each whole plot. i.e. each whole plot is now a1, a2, a3, a4. after this randomization of
factor A among whole plots we now randomize factor (i.e. Varieties) into the sub plots.
Thus this is a 2- stage randomization procedure.
A 4 - a1, a2, a3, a4.
B 2 - b1, b2
This shows that each factor A/ level will contain all the levels of factors B. i.e. a1 contains
b1 & b2. in considering the yield in a b1.. The yield is thus made of
Y ij =

Where Bk = sub plot effect

U = overall mean
Ti = Block effect
Pi = Whole plot effect
Sij = error due to block and whole plot interaction effect
(PB) jk = error due to whole plot and sub plot interaction
Σijk = overall error.
The split plot design is different from the randomized complete block design R.C.B.
(1) It splits the factors
32
 http://www.unaab.edu.ng

(2) It incorporates the principles of R.C.B. design e.g. if there ae 2 factors A &
B and A has 4 and B 2 levels there will be 8 treatments. In the R.C.B. design the 8
treatments will be randomized completely c within in the blocks, but in the split plot
design, the factors will be split into A and B and one factor will be randomized within the
other factor i.e. B. within A.

In the R.C.B. design we have

Y ijk = µ+ ti + pj + (tp)ij + Σijk
Where ti, i.e. treatment effect will be the 8 treatments
i.e. block effect will be the 4 blocks

But in the split plot design we have

Yijk = µ+ t1 + pi§ij + βk + (tp)ij + Σijk

ANOVA
Sources D.f
Y Total 23
T Replicates 2
P whole plots 3
S Error a 6
B Sub plots 1
(rp) Interaction 3
Eijk Error (b) 8

8.0 REGRESSION
Regression is a technique for quantifying the relationship among variables, one of which
(the response or dependent variable) is known to be functionally dependent on one or
more other variables called explanatory or independent variable:

Regression has considerable application in Animal Science Research. For example, in

studying growth rate in animals, the researcher regresses body weight on time. The
regression coefficient obtained represents the change in body weight per unit time which
is growth rate. A physiologist may want to determine the rate of disappearance of a certain
drug from the blood. He may also wish to study the changes in blood cholesterol
concentration in animals following administration of increasing doses of a certain drug.
The nutritionist may be interested in the change in feed conversion with increasing levels
of a certain nutrient or increasing ambient temperature. All these studies involve
functional relationship between variables and regression techniques are very useful. The
type of regression which a researcher may be interested in clued simple linear regression,
multiple regression, polynomial and non-linear regression.

Simple Linear Regression (SLR)

Here, only two variables are involved, one response and one explanatory. The model is:
yi = bo + b1x 1 +ei
Where bo and b1 are the intercept and slope respectively and are estimates of the
parameters, ßo and ß1

33
 http://www.unaab.edu.ng

From sample data, numeric values for the statistics can be obtained by solving a set of
normal simultaneous equations in two unknowns. However, easy-to-use formulae are
available for hand computations.
b1 = Σxy – (Σx. Σx)/n
Σx2 – (Σx)2/n
and
bo=y – b1x
If it sis known apriori that will be zero when x is zero, then regression is done through the
origin. The model becomes:
y1 = bx1 + e1
Simple estimated b y:
b = Σxy
Σx2

e.g. fit a regression line through the pairs of measurements

X Y XY X2 Y2
1 0.06 0.06 1 0.0036
2 0.10 0.02 4 0.01
3 0.19 0.57 9 0.0361
4 0.34 1.36 16 0.1156
5 0.47 2.35 25 0.2209
6 0.63 3.78 36 0.3969
7 0.84 5.88 49 0.7056
8 1.09 8.72 64 1.1881
2 2
 X  36  Y  3.72  XY  22.92  X =204  Y =2.6768
b= 22.92 – (36)(3.72)/8
204-(36)2/8
=6.18/42 = 0.147
b0= 0.465 – (0.147)(4.5)= -0.197
Hence, our model becomes
y1 = -0.197 + 0.147x
The line corresponding to this equation is obtained by plotting the (xi,yi) points.
The regression coefficient obtained is interpreted to mean that body weight was increasing
by 0.147kg per week on the average. In other words, the
Total SS = Σy2 – (Σy)2/n
Σy2

SS = (Σxy – (Σx. Σy)/n)2

Σx2 – (Σx)2/n
or (Σxy)2 with I D.f for regression through origin
2
Σx

Residual SS = Total SS- Regresion SS with n-2 or n- 1 D.f as appropriate

Corresponding mean squares are obtained as used and a variance ratio, f computed as:
F = Regresion Mean Square
Residual Mean Square
This F is then compared with tabulated F as appropriate degrees of freedom.

34
 http://www.unaab.edu.ng

From our example

Total SS = 2.6768 – (3.72)2
8
= 0.947 with 7 D.f
Regression SS = (22.92-(36) (3.72)/8)2
204-(36)
8
= 38.1924
42
= 0.909 with 1 D.f
Residual SS = 0.97-0.909
= 0.038 with 6 D.f

Our regression ANOVA table thus becomes

SOURCE D.F. SS MS F

Total 7 0.947
Regression 1 0.909 0.909** 151.5
Residual 6 0.038 0.006

The analysis shows that the regression is highly significant (P< 0.01)
The standard error of the regression coefficient is given as:
S.E (b)= Sy.x/ √Σ(x1 – x)2
Where Sy. x = Residual MS or
One commonly used criterion for determining the adequacy of a fitted regression model is
the R2 or r2 (coefficient of multiple determination) or 1-R2 (coefficient of allenation). R2 is
the proportion of the variability in Y explained by the fitted regression model.
R2 = Regression SS
Total SS, corrected for y
And R, the correction between x and y = +√R2 -. The sign will correspond to the sign of
the value Σxy – (Σx)( Σy)/n the square of which is the numerator of regression SS. in our
example:
R2 = 0.909 = 0.96 or 96%
0.947
and R = 0.98

9.0 SAMPLING

9.1 What is a sample?

In conducting an experiment it is usually not possible to get information on the entire
population we are interested in. what is done then is to select a manageable portion of the
population which is referred to as a SAMPLE.
A sample must be a true representative of the population if it is to lead to valid inferences.
35
 http://www.unaab.edu.ng

9.2 WHY DO WE SAMPLE?

1. The main objective for sampling is to reduce the population to a
manageable proportion.
2. The population may be homogenous that the result obtained from the
sample is as equally useful as that obtained from the population.
3. The whole population might be inaccessible.
4. Collecting information from the whole population might be practically
impossible in terms of the personnel required.
5. It may be too costly money wise to work with the whole population.

9.3 SAMPLING METHODS

Many methods exist for selecting the sample from a population and which method is used
will depend on the particular situation, the type of data required and any other constraints
on the researcher. Sampling methods are:
1. Simple Random Sampling
Simple random sampling is a procedure whereby every element of the population is given
the chance of being included or excluded from the sample. In making such a selection, the
following procedure is followed:
a. A list of all the elements in the population is obtained.
b. The various elements are identified with numbers e.g. 1, 2, 3, 4, 5 etc.
c. The size of the sample is decided upon.
d. The sample is selected with the use of the table of random numbers.
The major advantage in simple random sampling is that it gives all the members of the
population a chance of being included or excluded from the sample. However, it has a
disadvantage where the population is so large since the identification of each element in
the population might be impracticable. Also, the use of simple random sampling is limited
only to finite population.
2. Stratified Sampling Technique
In this method, the population is first partitioned or stratified into sub- populations known
as strata and simple random sampling is then used to obtain samples from each stratum.
The purpose of stratifying is to ensure that certain characteristics of the population which
may be eliminated by the use of simple random sampling and which we want to included
in the sample are taken care of . Stratification could be carried out on the basis of sex,
religion, ethnic groups, income etc. the number of the sub-sample from each stratum will
depend on the proportion of that group in the population.
Before stratified sampling technique could be used, the following conditions must be
satisfied:
a. The population must be heterogeneous with respect to the variable of
interest.
b. The population must be divisible with subsets.

36
 http://www.unaab.edu.ng

c. The elements of each sub-group or stratum must be heterogeneous between

(among) themselves. In other words, the subgroups must have different
means and possibly zero variances.

3. Cluster Sampling Technique.

As in the stratified sampling technique, the population is partitioned into clusters before
sub-samples are taken from each cluster. Unlike a stratum however, the elements in a
cluster ate heterogonous while the clusters are homogeneous between or among
themselves. Where the population is large, each cluster can be further sub-divided into
sub-clusters or strata before carrying out a simple random sampling on each sub-group.

4. Purpose Sampling Technique

Here the sample taken depends mainly on the purpose for which the survey is carried out.
What elements included or excluded for the sample if left to the judgment of the
researcher (usually an expert) (e.g. political campaigns)-18 years old and above.

5. Systematic Sampling Technique.

There are times whereby it might not be possible to group or list the elements in a
population. This is commonly the case in market surveys and sociology. Systematic
sampling involves selecting the elements for our sample at regular interval but in a
consistent way. In a market survey for example, we may decide to talk to every 5th or 10th
person we come across in the market.

6. Multi-stage or Nested Sampling Technique

The procedure involves more than one sampling stage. Any of the other sampling
procedures could be used in any of these stages.

Suppose we want to relate the performance of students in the University of Agriculture,

Abeokuta to their ethnic background, the first stage will involve grouping the students
according to ethnic groups. The second stage might involve grouping elements in each
cluster according to some social factors such as religion, sex, income etc. the third stage
will then involve taking a random sample from each of the strata.
Thus in the multi-stage sampling procedure we might use.

Cluster sampling - first stage

Stratified sampling - Second stage
Simple random sampling - Third Stage.

37