ANOVA

Analysis of Variance
ANOVA was developed by Sir Ronald A. Fisher. It is also called F test, this name was
given by Snedecor in Fisher’s honour. ANOVA deals with variance rather than with
standard deviations and standard errors. It useful in testing differences between two or
more means at the same time.
It is a powerful tool which aids to design studies efficiently and understand interaction
effect.

ANOVA v/s T test -

● T test is used in the case experiments involving only 2 groups
● ANOVA can be used to compare the means of more than two populations
● In T test the sample population should be of minimum 30
● ANOVA can accommodate a huge population count
Variance
Variance is a measure of the dispersion or spread of a set of scores. It describes the extent
to which the scores differ from each other. There are three types of variance -

i) Systematic Variance - It is a variability in the dependent measure due to manipulation

of the experimental variable by the experimenter.

ii) Error Variable - the results of experiments are affected by extraneous variable which
tends to mask the effect of experimental variable. The term experimental error or error
variance is used to refer to all such uncontrolled sources of variation in experiment. It
results from random fluctuations in the experiment.

Variance = SD ²

Variance is a terminology of ANOVA is more frequently called Mean Square (MS)

 Mean of Square = Variance (SS)/ df
Mean of Square (MS) is average variation per degree of freedom
In ANOVA, there are two types of variance - i) between group, ii) within group
Total variance = Between + Within
Between Group Variance- it is defined as the group means which may vary considerably
from each other. It examines how independent groups may differ from each other on a
variable. This is useful in examining the efficacy of interventions or treatments.
Within Group Variance- it is defined as members of each group may vary among
themselves average variance of the members of each group around their respective group
means. In with group difference, the researcher is not looking at the difference between
two group but the difference between the same group taken at two times points.
Factor- a factor is variable that experimenter defines and controls so that its effect can be
evaluated in the experiment
Levels- each specific variation in a factor is called the level of that factor. For Factor A,
levels would be a1, a2, a3…….
Main Effect- the difference in performance from one level to another for a particular
factor, averaged over other factors is called main effect.
Interaction Effect- interaction between two variables is said to occur when change in the
value of one variable alters the effects on the other variable. It can be evaluated in all
experiments having two or more IV.
Error- experimental error is the difference between a measured value and its value. It is
the inaccuracy that stops us from seeing an absolutely correct measurement. There are two
types of error - Type 1 and Type 2
One Way ANOVA
(Simple ANOVA, Single Factor Experimental Design, Experimental Randomised Group
Design)
One-Way Analysis of Variance (ANOVA) is a statistical method used to analyze and compare the means
of three or more groups or treatments to determine if there are significant differences among them. It is
particularly useful when you want to assess whether a single categorical independent variable (often
called a factor) has a statistically significant effect on a continuous dependent variable.

Factor

In One-Way ANOVA, you have one categorical independent variable, which is the factor under
investigation. This factor has three or more levels or categories. For example, it could be different drug
treatments, different types of fertilizer, or different teaching methods.

Dependent Variable

The dependent variable is the continuous variable that you want to compare between the different levels
of the factor. This variable could be anything measurable, such as test scores, blood pressure, or crop
yield.
Assumptions of ANOVA
1. Normality of the distribution of criterion measures -

This states that the distribution of scores within each treatment population is normal.
This assumption is satisfied when the scores within the treatment groups are from
normally distributed population.

1. Homogeneity of Variance -

Variance of scores in each of the treatment groups are homogenous. This is the
variances of the individual groups are equal. It means level of variance for a
particular variable is constant across the sample.

1. Independence of Sample -

Each sample is randomly selected and independent of each other

Randomised Group Design
Numerical Example
An investigator is interested in exploring the most effective method of instruction in the
classroom. He decides to try 3 methods - 1) Lecture 2) Seminar 3) Discussion
He randomly selects 5 students/subjects for each of the three groups from 10th grade
students. After 3 month instructions,, an achievement test was administered between three
groups.
One way ANOVA deals with one independent variable (A) - method of teaching and can
have many sublevels (a1 Lecture, a2 Seminar, a3 Discussion).
Method of Teaching - IV
Achievement scores - DV
Levels - lectures, seminar, discussion
 Methods

Subject Lecture Seminar Discussion

1 8 11 5

2 10 13 5

3 11 13 8

4 11 15 9

5 12 16 10

Total 52 68 37

Grand Total = 157

n = 5 (number of students in each group)

k = 3 (number of methods/ subgroups)
kn = 5 x 3 =15 (Total no of observations)
Partitioning of Total Variance and df -
Total sum of squares is partitioned into two component parts – one due to variation between the groups and
other due to variation within the groups.
The total variation and df will have same following partitioning -
Total
kn - 1
3 x 5 - 1 = 15 - 1 = 14

Between Groups Within Groups (error

variance)
k-1 k (n-1)
3-1 = 2 3 (5-1) = 3 x 4 = 12
Here, n = sumner of subjects in each group
k = number of subgroups
kn = number of observations (N)
Anova Summary Table

Source of SS Df MS (SS/df) F (Between

Variation group MS/
Within group
MS

Between 96.13 2 48.065 12.64**

Within (error) 45.6 12 3.8

Total 141.73 14

Consulting table B for critical value of F ratio with df 2 and 12

At 0.05 level of significance = 6.93
At 0.01 level of significance = 3.88

Interpretation- the observed values (12.64**) is significant at both the levels as it is greater than the
value (critical value) at both levels.

On the basis of the results of the experiment, we can conclude that three methods of instructions
(lecture, seminar, discussion) produced significant differences in three groups.
Randomised Block Design - One Way ANOVA
In randomised block design, it is possible to reduce error variance by forming blocks such
that the experimental units within the blocks are relatively more homogeneous. The
primary objective of creating the blocks is to eliminate the experimental error - the
variation due to difference between blocks.
Achieving homogeneity within the blocks and consequently heterogeneity between the
blocks reduces the overall error variance as the variation between the blocks is eliminated
from the error variance. Thus, it is expected that blocking will result in smaller error
variance than in a randomised group design.
The number of subjects in each block (r) is necessarily kept equal to the proposed number
of treatments (k).
Blocks are arranged on the basis of a pretest on the assumption that the experimental units
forming the blocks will be more homogenous.
Numerical Example

An investigator is interested in studying the Effects of Stress (IV) on the Work Output
(DV). The work output is the number of vowels cancelled in a cancellation task in a fixed
interval of time. The subjects are to be tested under four levels of stress (treatment, k).
The investigator decides to pretest the subjects under uniform conditions and to have 5 (r)
replications of treatment, thus a total of kr (4 x 5) 20 subjects required for the experiment.
Within each block the subjects are randomly assigned to 4 treatment conditions.
 Blocks (r) Treatments (k)

t1 t2 t3 t4 Σ

Block 1 13 9 10 7 39

Block 2 12 8 8 6 34

Block 3 10 9 9 6 34

Block 4 10 7 5 4 26

Block 5 8 6 4 4 22

Total 53 39 36 27 155 (Grand

Total)

Grand Total = 155

Blocks (r) = 5
Treatments (k) = 4
Number of subjects in each block = 4
kr = 5 x 4 = 20
 Partitioning of Total Variation and df

Total
(kr - 1)
(20 - 1 = 19

Treatment Blocks Blocks x Treatment (error)

(k-1) (r-1) (k-1)(r-1)

4-1=3 5-1=4 (4-1)(5-1)

3 x 4 = 12
k = number of treatments = 4
r = number of blocks = 5
kr (N) = total number of subjects/ observations = 20
 Summary of ANOVA for Randomised Complete Block Design -

Source of SS Df MS (SS/df) F (MS

Variation treatment) / MS
TxB (error)

Treatment 69.75 3 23.25 31.0**

Block 47.0 4 11.75

Treatment x 9.0 12 0.75

Block (error)

Total 125.75 19

Df for treatment is 3 and 12, and

Critical value from Table B, at 0.05 level the value is 5.95
at 0.01 level the value is 3.49

NOTE: We do not find the F value in respect of block differences, as the main objective in creating blocks is only
to reduce the experimental error. The blocks have been so created that the subject within the blocks are more
homogenous and consequently increase the between block difference.
Test of Significance/ Interpretation

To evaluate the observed F value for the treatment effect, we refer to the Table F
(Table B from Broota) for df 3 and 12. Our observed value (31.0) far exceeds the
tabled or critical value. Therefore we conclude that treatment means differ
significantly.
 Two Way ANOVA
Two-Way Analysis of Variance (ANOVA) is a powerful statistical technique employed to investigate the influence of
two categorical independent variables, or factors, on a continuous dependent variable. It is particularly valuable in
experimental research settings where there are two distinct sources of variation to be considered.

● Factors

In a Two-Way ANOVA, there are two categorical independent variables (factors), typically referred to as Factor A and
Factor B.

● Dependent Variable

This is the variable you are interested in studying or predicting. It is continuous and assumed to be normally
distributed.

● Interaction Effect

The interaction between Factor A and Factor B refers to whether the effect of one factor depends on the level of the
other. In other words, it assesses if the relationship between the dependent variable and one factor is affected by the
levels of the other factor. For example, method of instruction will have different effect depending upon the level of
intelligence of subject.

● Main Effects

These are the individual effects of each factor on the dependent variable, ignoring the presence of the other factor.
There is a main effect for Factor A and a main effect for Factor B.
 Numerical Example
The investigator designs a study in which the effect of two variables are studied simultaneously i.e.,
effect of intelligence and methods of instruction on the achievement scores of the students.

a1 Superior Intelligence

Factor A (level of intelligence) a2 Inferior Intelligence

Factor B (Methods of Instruction) b1 Lecture

b2 Seminar

b3 Discussion

The levels (a1, a2, b1, b2, b3) are fixed.

Factorial design for Two Way Anova is 2 x 3

It do not represent a random sampling, it means levels of factors are arbitrary chosen by the
experimenter.
A total of 3 subjects were selected, 15 of superior intelligence and 15 of inferior
intelligence. Five subjects from each of two groups were randomly assigned to
each of the 3 methods of instructions. That is, 5 subjects were randomly assigned
to each of the six treatments. After 3 months of instructions, an achievement score
was administered.
Intelligence (Factor A) Lecture (b1) Seminar (b2) Discussion (c3)

7 11 12

9 13 13
a1 (Superior
intelligence) 11 15 15

12 15 16

12 17 18

Total 51 71 74

11 8 9

12 8 10

a2 13 10 11

15 11 11

16 12 12

Total 67 49 53
Independent variables (IV) - intelligence (A)
- methods of instructions (B)

Dependent Variable (DV) - achievement scores

AB Interaction Table -

b1 b2 b3 Total

a1 51 71 74 196

a2 67 49 53 169

Total 118 120 127 365 (Grand

Total)
 Partitioning of Total Variance and df

Total
(kn - 1) = (6 x 5) - 1 = 29

Between Within (error variance)

(k-1) k(n-1)
6-1 = 5 6 (5-1) = 24

A B AxB
(r-1) (c-1) (r-1)(c-1)
2-1 = 1 3-1 = 2 (2-1)(3-1) = 2

n = number of subjects in each treatment = 5

k = number of treatments = 6
r = number of rows, representing levels of factor A = 2
k = number of columns, representing levels of factor B = 3
kn = total observations = 30 (N)
 Summary of Two way ANOVA

Source of SS Df MS (SS/df) F (MS / Within

Variation (error)

A (intelligence) 24.30 1 24.30 5.98*

B (method) 4.47 2 2.235 0.55

AB (interaction) 93.8 2 46.9 11.55**

Within (error) 97.6 24 4.06

Total 220.17 29

Df for Factor A is 1 and 24, and

Critical value from Table B at 0.05 level is 4.26
at 0.01 level is 7.82
Df for AB is 2 and 24, and
Critical Value from Table B for AB at 0.05 level is 3.40
at 0.01 level is 5.61
NOTE: F ratio for Factor B (method) is 0.55 and it is not considered because it is less than 1, hence considered
insignificant.
 ● Test of significance/ Interpretation of results for Main Effect A -

The observed value of F is 5.98 and is significant at 0.05 level (as this value is greater than tabled value). Therefore, the
mean groups selected on the basis of levels of intelligence are not the same.

● Test of significance/ Interpretation of results for Main Effect B -

The observed value of F is 0.55 and is less than 1. Therefore, we conclude that the three methods of instructions did not
affect the achievement scores differentially.

● Test of significance/ Interpretation of results for Interaction Effect AB -

The F ratio in respect to AB is 11.55 and this value far exceeds the tabled value and is significant at both the levels.
Therefore, we concluded that the effectiveness of particular method of instruction depends upon the level of intelligence.
While the students with superior intelligence do better with seminar (14.2) and discussion (14.8) methods of instruction.
While students with low intelligence gain more from lecture method (13.4).

Two way table of means

b1 b2 b3 Total

a1 51/5 = 10.2 71/5 = 14.2 74/5 = 14.8 196/15 = 13.06

a2 67/5 = 13.4 49/5 = 9.8 53/5 = 10.6 169/15 = 11.26

Total 118/10 = 11.8 120/10 = 12.0 127/10 = 12.7

Three way Anova
A three-way ANOVA (Analysis of Variance) is a statistical technique used to analyze the differences
among group means in a study with three independent categorical variables. It extends the
principles of one-way and two-way ANOVA to cases where there are three independent factors.

Independent Variables
In a three-way ANOVA, there are three independent categorical variables, often referred to as
factors. Each factor has two or more levels or categories. For example, in a study on the effects of a
new drug, the three factors could be:
● Drug Type (e.g., Drug A, Drug B, Placebo)
● Dosage (e.g., Low, Medium, High)
● Gender (e.g., Male, Female)
Dependent Variable
The dependent variable is the outcome or measurement of interest. It must be a continuous
variable. For example, in a drug study, this could be something like blood pressure, heart rate, or
some other physiological measurement.
Main Effects
A main effect is the effect of one independent variable on the dependent variable, ignoring the
effects of the other independent variables. In a three-way ANOVA, there are three main effects,
one for each factor.
Interaction Effects
Interaction effects occur when the effect of one independent variable on the dependent variable
depends on the level of another independent variable. In a three-way ANOVA, there are several
possible interaction effects, including two-way interactions and three-way interactions.

Remember, conducting a three-way ANOVA requires meeting certain assumptions, such as

normality of data, homogeneity of variances, and independence of observations.
An investigator wanted to study the effect of intelligence, socioeconomic status and gender on
academic achievement. To study the effect of three independent variables simultaneously on on a
single dependent variable, at that time we use a certain statistical analysis called three way anova.
Independent variables - A (Intelligence)
- B (Socioeconomic Status)
- C (Gender)
Dependent Variable - Academic Achievement

Levels of Factor A (Intelligence) - a1 High, a2 Average, a3 Low

Levels of Factor B (Socioeconomic Status) - b1 High, b2 Average, a3 Low
Levels of Factor C (Gender) - c1 Male, c2 Female

Factorial design of this particular case is 3 x 3 x 2

 A1 (High A2 (Average A3 (Low Intelligence)
Intelligence) Intelligence)
Total

C1 (Male) C2 C1 (Male) C2 C1 (Male) C2 (Female)

(Female) (Female)

B1 (High
socioeconomic
status )

B2 (Average
socioeconomic
status)

B3 Low
socioeconomic
status )

Total Grand
Total
Steps:
1. Correction term
2. Total sum of squares (SS)
3. Sum of Square (Factor A)
4. Sum of Square (Factor B)
5. Sum of Square (Factor C)
6. Sum of Square (A x B)
7. Sum of Square (A x C)
8. Sum of Square (B x C)
9. Sum of Square ( A x B x C)
10. Sum of Square within
11. Summary Table
12. Interpretation/ Test of Significance
Summary Table