ANOVA

One way & Two way classified

data
Dr. Mukta Datta Mazumder

Associate Professor
Department of Statistics
 ANOVA
The total variation present in a set of
observable quantities may, under
certain circumstances, be partitioned
into a number of components associated
with the nature of classification of the
data
The systematic procedure of achieving
this is called analysis of variance
(ANOVA)
 ANOVA

 ANOVA was developed by statistician and

evolutionary biologist Ronald Fisher.

The purpose of ANOVA is to test for

significant difference between means.

If we comparing two means ANOVA will

produce the same results as t- test for
independent (dependent) samples
 ANOVA

The name is derived from the fact that in

order to test for statistical significance between
means , we are actually comparing
(analyzing) variances.

Basic ANOVA Concepts:

A response variable related to one or more
explanatory variables , usually categorical .
One way & Two way Classified data

 One way or two way refers to the no. of

independent variables
 One way- one independent variable(2 level)
Ex. Brand of cereal

 Two way- two independent variables(can
have multiple levels)
Ex. Brand of cereal and calories
 One way- ANOVA
- Which science departments gives out
lowest average grade?-
Explanatory variable- Department
Response variable- student’s GPA for
individual course
-Which kind of promotional campaign
leads to greatest store income at Christmas
time
Explanatory variable-Promotion type
Response variable- daily store income
 TWO way- ANOVA
 How do the type of career and martial
 status of a person relate to the total cost of
 annual claims she/he likely to make on
 her/his health insurance

 Explanatory variables- Career and

-

 martial status

 - Response variable- health insurance

payouts
 Examples
 Students from different colleges take the
same examination. – one can see if one
college outperform other.

 A group of psychiatric patients are trying

three different therapies- counseling,
medication and biofeedback.- one can check if
one therapy is better than the others.
 ANOVA
 Summary
 Analysis of variance is a statistical method used
to test two or more means
 Provide statistical test whether population means
of several group are equal
 Generalizes the t-test to more than one group
 Inferences about means are drawn by analyzing
the variance
 Assumptions
 The experimental error are normally distributed

 Equal variances between

treatments(Homoscedasticity)

 Independence of Samples
 Assumptions

 -Independent Observations
 -Normality

 -Homogeneity :variance within all subpopulation

must be equal

 -Independence of errors –errors are

independently distributed
 Assumptions
 -
 Absence of outliers- Outlying score have been
removed from data

 If the assumptions hold then under null

hypothesis F follows F distribution with DF
between & DF within SS
 Logic of ANOVA

 ANOVA focuses on variability, it involve calculation

of several measures of variability

-Partitioning total variation into two components

-components due to difference between means &
component due to within SS (true random error)
 ANOVA Test
 To find out survey or experiment results are
significant(reject null hypothesis or accept
alternative hypothesis)

Testing groups to check if there is a difference

between them
 Hypotheses

 Null hypothesis
 H0: µ 1 = µ2 = µ3 = µ

 Alternative hypothesis

 H1: at least one population mean is different

from one another


Partitioning the total variation
 Total variation

 Between Gr variation Within Gr variation

 - variances are compared in F ratio to
determine mean differences (MS
between) are significantly bigger than
chance (MS within)
.

 F= (MS bet gr)/(MS within gr)

 MS bet gr= SS bet gr/DF bet gr

 MS within gr= SS within gr/DF within gr
 Total SS= bet SS + within SS
 Total DF= bet DF+ within DF
 F ratio is always positive as F ratio computed
from two variance
 Numerical Example

 Suppose the National Transportation Safety

Board wants to examine the safety of cars Type
A, cars Type B and cars Type C . It collects a
sample of three for each of the treatments (cars
types).
 Using the hypothetical data provided below, test
whether mean pressure applied to the driver’s
head during crash test is equal for each types of
car.
Table
Cars Type A Cars Type B Cars Type C
643 469 484
655 427 456
702 525 402
 STEPS
 1. State Null & Alternative hypotheses

 In ANOVA null hypothesis – population means are

equal
 H0: µ 1 = µ 2 = µ 3 (Mean head pressure is
statistically equal across the 3 types of cars)
 Since in null hypothesis assume all means are
equal ,we could reject the null hypothesis if one
mean is different thus
.

 -Alternative hypothesis –
 H1: at least one mean pressure is not
statistically equal

 To test – we calculate appropriate test
statistics
 under H0
 F= (MS bet gr)/(MS within gr) follows F
distribution
 Total SS – Total variation in data.
 - It is the sum of between and within variation

 SST= Σ Σ(Yij- ͞Y )² ͞Y = 529.22

 = 96303.55
.
 Between SS (or Treatment SS) –
Variation in the data between
different samples (or treatments)
 SSTr = 86049.55
 Within variation (or error SS)-
Variation in the data from each
individual treatment)
 Error SS (SSE)= 10254
 Mean squares
 -

 Next step in ANOVA - to compute Mean

squares :
 Total mean square MST= SST/ N-1 ,
N= Total no of observations
 MST= 96303.55/(9-1) = 2037.94
.
 Mean square Treatment (MSTr) = SSTr/(k -1) ,
 k= No of treatments ( in Ex no of columns )

 MSTr= 86049.55/(3 -1) = 43024.78


 Mean square error (MSE)= SSE/ (N-k)
 MSE= 10254/ (9 -3) = 1709

 NOTE:SST= SSTr + SSE but

 MST ≠ MSTr + MSE

 . TEST Statistic
 Next step – Calculate TEST Statistic

 F0= MSTr/MSE =25.17

 Obtain the critical value: To find critical value

from F distribution it is required to know DF of
numerator(DF1) & DF denominator (DF2)
along with significance level
 -F (critical )has DF1=( k-1) &
 DF2= (N-k) ,
 α = 5% or 1%
 In our example , DF1=3-1=2,
 DF2= (9-3) =6
 α = 5%
 . F (critical)
 Hence we need to find F (critical) with
 2 and 6 DF at 5% level of significance

 Using F table , F (critical) with 2 and 6

DF at 5% level of significance = 5.14
 Decision Rule
 - We reject H0 at α level of significance
-
 F (observed) > F(critical)

 In our example 25.17 > 5.14

 We may reject the null hypothesis α
level of significance
 Interpretation
 We are 95% confident that the mean
head pressure is statistically not equal
for cars Type A, Cars Type B and cars
Type C.
 However only one mean must be
different to reject the null, we do not
know which mean(s) is/are different.
.
 In ANOVA test provide us at least one
mean is different ,additional test must
be conducted to determine which
mean(s) is/are different
 Most common test – Least
significant difference(LSD) test

 ANOVA Table
SV SS DF MS VR (F) F (Cr )
Between 86049.55 2 43024.7 25.17 5.14
Gr 8
Within 10254 6 1709
Gr
Total 96303.55 8