1-Way Completely Randomized ANOVA: James H. Steiger

Introduction
An Introductory Example
The Basic Idea behind ANOVA
1-Way Completely Randomized ANOVA
James H. Steiger
Department of Psychology and Human Development

Vanderbilt University
James H. Steiger 1-Way Completely Randomized ANOVA

Introduction
1-Way Completely Randomized Design

1 Introduction
2 An Introductory Example
3 The Basic Idea behind ANOVA
Calculation in R

Introduction
Introduction
In this module, we introduce the single-factor completely

randomized analysis of variance design.
We discover that there are several ways to conceptualize
the design.
For example, we can see the design as a generalization of
the 2-sample t-test on independent groups.
Or we can see it as a special case of a test for comparing
variances.

Introduction
Consider the following extremely simple, artificial data set

representing three groups, each with three subjects. The
scores are
Group1 Group2 Group3

1 1.00 4.00 7.00
2 2.00 5.00 8.00
3 3.00 6.00 9.00

Introduction
An Introductory Example Calculation in R
ANOVA:The Basic Idea
At its foundation, ANOVA attempts to assess whether a

group of means are all the same. If there are k groups, then
the null hypothesis might be written
H0 : µ1 = µ2 = . . . = µk (1)

Introduction

ANOVA does this by comparing the variability of the
sample means to what it should be if the population means
are all the same.
Suppose, for the time being, that the population means are
all the same, the sample sizes are all the same (n per
group), and the populations all have the same variance σ 2 .
Then the means represent repeated samples from the same
population.
In that case, as we recall from our earlier discussion of the
Z test, the variance of the sample means should be
approximately σM 2 = σ 2 /n.
It is crucial to realize that, under all the assumptions, that

is the smallest long run variance that these means can show
for any set of population mean values.
On the other hand, if the population means are not all the
same, then the variance of the sample means should be
larger than σ 2 /n, because the spread of the means reflects
more than sampling variability.
Introduction
The plot on the next side shows 10 sample means taken

from each of 3 groups. All populations had standard
deviations of 15 and sample sizes of 25. The lower line
shows 30 means from 3 populations with identical means of
100,100,100. Group 1 is red, Group 2 in blue, Group 3 in
green. The dispersion that you see along the line is totally
due to within group sampling variation.
The upper line shows 30 means sampled from populations
with means of 90,100,and 110. The dispersion of the means
on the top line reflects both within group sampling
variation and the spread of the means between groups.

Introduction
80 90 100 110 120

sample.mean

Introduction
As a consequence of the facts from the preceding slides, we

can conceptualize the null hypothesis in ANOVA in an
entirely different (but equivalent) way from Equation 1
To emphasize that this is a one-tailed hypothesis, I write
both the null and alternative hypotheses.
2 σ2 2 σ2
H0 : σM = H1 : σM > (2)
n n

Introduction
As a consequence of this, and a lot of statistical machinery,

we arrive at an F statistic that may be written, for k
groups,
s2 ns 2
Fk −1,k (n−1) = M = M (3)
σˆ2 /n σˆ2
We simply compute the sample variance of the sample
means, and divide by an estimate of σ 2 /n. With equal
sample sizes, σ̂ 2 is simply the mean of the group variances.

Introduction
Here are the calculations in R.

> Means <- c(mean(Group1),mean(Group2),mean(Group3))
> Variances <- c(var(Group1),var(Group2),var(Group3))
> n <- 3
> k <- 3
> MS.between <- n*var(Means)
> MS.within <- mean(Variances)
> F <- MS.between/MS.within
> df.between <- k-1
> df.within <- k * (n-1)
> c(MS.between,MS.within,F,df.between,df.within)
[1] 27 1 27 2 6

Introduction
Calculation in R
The preceding method of calculation is extraordinarily

simple. But it is only valid if each group has the same n.
The Gravetter-Walnau textbook gives computational
formulas for the analysis of variance in the general case
when n is not necessarily the same in each group. I also
provide computational instructions for the case of unequal
group n in a handout on the course website.
However, it is unlikely that, in practice, you will ever
calculate an ANOVA by hand. R makes it really easy.
First, have to enter your data in the correct, special format.
Instead of entering the data for our 3 groups in 3 columns,
we enter each score on a second line, accompanied by a
group code. The data entry process is shown on the next
slide.

Introduction
Calculation in R
It is vitally important that the factor variables are typed as
factors.
In this case, we are ready to go. The following simple
commands generate the ANOVA source table, as discussed
in your textbook.
> Group <- c(1,1,1,2,2,2,3,3,3)
> Score <- c(1,2,3,4,5,6,7,8,9)
> summary(aov(Score ~ factor(Group)))
Df Sum Sq Mean Sq F value Pr(>F)
factor(Group) 2 54 27 27 0.001 ***
Residuals 6 6 1
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The F statistic has a value of 27.00, with 2 and 6 degrees
of freedom.
Remember that this is a 1-tailed test, so the critical value
with α = 0.05 is at the 0.95 quantile. To find the critical
value, we use
> qf(0.95,2,6)
[1] 5.143253

1-Way Completely Randomized ANOVA: James H. Steiger

Uploaded by

Copyright:

Available Formats

1-Way Completely Randomized ANOVA: James H. Steiger

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1-Way Completely Randomized ANOVA: James H. Steiger

Uploaded by

Copyright:

Available Formats

Introduction

1-Way Completely Randomized ANOVA

Department of Psychology and Human Development

James H. Steiger 1-Way Completely Randomized ANOVA

1-Way Completely Randomized Design

3 The Basic Idea behind ANOVA

James H. Steiger 1-Way Completely Randomized ANOVA

In this module, we introduce the single-factor completely

James H. Steiger 1-Way Completely Randomized ANOVA

Consider the following extremely simple, artificial data set

Group1 Group2 Group3

James H. Steiger 1-Way Completely Randomized ANOVA

ANOVA:The Basic Idea

At its foundation, ANOVA attempts to assess whether a

James H. Steiger 1-Way Completely Randomized ANOVA

ANOVA:The Basic Idea

It is crucial to realize that, under all the assumptions, that

ANOVA:The Basic Idea

The plot on the next side shows 10 sample means taken

James H. Steiger 1-Way Completely Randomized ANOVA

ANOVA:The Basic Idea

80 90 100 110 120

James H. Steiger 1-Way Completely Randomized ANOVA

ANOVA:The Basic Idea

As a consequence of the facts from the preceding slides, we

James H. Steiger 1-Way Completely Randomized ANOVA

ANOVA:The Basic Idea

As a consequence of this, and a lot of statistical machinery,

James H. Steiger 1-Way Completely Randomized ANOVA

ANOVA:The Basic Idea

Here are the calculations in R.

James H. Steiger 1-Way Completely Randomized ANOVA

The preceding method of calculation is extraordinarily

James H. Steiger 1-Way Completely Randomized ANOVA

James H. Steiger 1-Way Completely Randomized ANOVA

You might also like