Transcript ch_13

Chapter 13
Comparing Groups: Analysis of
Variance Methods
 Learn
….
How to use Statistical inference
To Compare Several Population
Means
Agresti/Franklin Statistics, 1 of 82
 Section 13.1
How Can We Compare Several
Means?
One-Way ANOVA
Agresti/Franklin Statistics, 2 of 82
Analysis of Variance

The analysis of variance method
compares means of several groups
• Let g denote the number of groups
• Each group has a corresponding
population of subjects
• The means of the response variable for the
g populations are denoted by µ1, µ2, … µg
Agresti/Franklin Statistics, 3 of 82
Hypotheses and Assumptions
for the ANOVA Test

The analysis of variance is a significance
test of the null hypothesis of equal
population means:
• H0: µ1 = µ2 = …= µg

The alternative hypothesis is:
• Ha: At least two of the population means
are unequal
Agresti/Franklin Statistics, 4 of 82
Hypotheses and Assumptions
for the ANOVA Test
The assumptions for the ANOVA test
comparing population means are as
follows:
1. The population distributions of the
response variable for the g groups are
normal with the same standard
deviation for each group
Agresti/Franklin Statistics, 5 of 82
Hypotheses and Assumptions
for the ANOVA Test
2. Randomization:
•
•
In a survey sample, independent
random samples are selected from
the g populations
In an experiment, subjects are
randomly assigned separately to the
g groups
Agresti/Franklin Statistics, 6 of 82
Example: How Long Will You
Tolerate Being Put on Hold?

An airline has a toll-free telephone
number for reservations

The airline hopes a caller remains on
hold until the call is answered, so as
not to lose a potential customer
Agresti/Franklin Statistics, 7 of 82
Example: How Long Will You
Tolerate Being Put on Hold?

The airline recently conducted a
randomized experiment to analyze
whether callers would remain on hold
longer, on the average, if they heard:
• An advertisement about the airline and its
•
•
current promotion
Muzak (“elevator music”)
Classical music
Agresti/Franklin Statistics, 8 of 82
Example: How Long Will You
Tolerate Being Put on Hold?

The company randomly selected one out
of every 1000 calls in a week

For each call, they randomly selected one
of the three recorded messages

They measured the number of minutes
that the caller stayed on hold before
hanging up (these calls were purposely
not answered)
Agresti/Franklin Statistics, 9 of 82
Example: How Long Will You
Tolerate Being Put on Hold?
Agresti/Franklin Statistics, 10 of 82
Example: How Long Will You
Tolerate Being Put on Hold?

Denote the holding time means for
the populations that these three
random samples represent by:
• µ1 = mean for the advertisement
• µ2 = mean for the Muzak
• µ3 = mean for the classical music
Agresti/Franklin Statistics, 11 of 82
Example: How Long Will You
Tolerate Being Put on Hold?

The hypotheses for the ANOVA test
are:
• H0: µ1=µ2=µ3
• Ha: at least two of the population
means are different
Agresti/Franklin Statistics, 12 of 82
Example: How Long Will You
Tolerate Being Put on Hold?

Here is a display of the sample means:
Agresti/Franklin Statistics, 13 of 82
Example: How Long Will You
Tolerate Being Put on Hold?

As you can see from the graph on
the previous page, the sample
means are quite different

This alone is not sufficient evidence
to enable us to reject H0
Agresti/Franklin Statistics, 14 of 82
Variability between Groups and
within Groups

The ANOVA method is used to
compare population means

It is called analysis of variance
because it uses evidence about two
types of variability
Agresti/Franklin Statistics, 15 of 82
Variability Between Groups and
Within Groups

Two examples of data sets with equal
means but unequal variability:
Agresti/Franklin Statistics, 16 of 82
Variability Between Groups and
Within Groups

Which case do you think gives
stronger evidence against
H0:µ1=µ2=µ3?

What is the difference between the
data in these two cases?
Agresti/Franklin Statistics, 17 of 82
Variability Between Groups and
Within Groups

In both cases the variability between pairs
of means is the same

In ‘Case b’ the variability within each
sample is much smaller than in ‘Case a.’

The fact that ‘Case b’ has less variability
within each sample gives stronger
evidence against H0
Agresti/Franklin Statistics, 18 of 82
ANOVA F-Test Statistic

The analysis of variance (ANOVA) F-test
statistic is:
Between groups variability
F
Within groups variabili ty

The larger the variability between groups
relative to the variability within groups, the
larger the F test statistic tends to be
Agresti/Franklin Statistics, 19 of 82
ANOVA F-Test Statistic

The test statistic for comparing
means has the F-distribution

The larger the F-test statistic value,
the stronger the evidence against H0
Agresti/Franklin Statistics, 20 of 82
ANOVA F-test for Comparing
Population Means of Several Groups
1. Assumptions:
 Independent random samples
 Normal population distributions with
equal standard deviations
Agresti/Franklin Statistics, 21 of 82
ANOVA F-test for Comparing
Population Means of Several Groups
2. Hypotheses:
• H0:µ1=µ2= … =µg
• Ha:at least two of the population
means are different
Agresti/Franklin Statistics, 22 of 82
ANOVA F-test for Comparing
Population Means of Several Groups
3. Test statistic:
Between groups variabili ty
F
Within groups variabili ty
F- sampling distribution has df1= g -1,
df2 = N – g = (total sample size – no. of
groups)
Agresti/Franklin Statistics, 23 of 82
ANOVA F-test for Comparing
Population Means of Several Groups
4. P-value: Right-tail probability above
the observed F- value
5. Conclusion: If decision is needed,
reject if P-value ≤ significance level
(such as 0.05)
Agresti/Franklin Statistics, 24 of 82
The Variance Estimates and the
ANOVA Table


Let σ denote the standard deviation
for each of the g population
distributions
One assumption for the ANOVA F-test
is that each population has the same
standard deviation, σ
Agresti/Franklin Statistics, 25 of 82
The Variance Estimates and the
ANOVA Table



The F-test statistic is the ratio of two
estimates of σ2, the population variance for
each group
The estimate of σ2 in the denominator of the
F-test statistic uses the variability within
each group
The estimate of σ2 in the numerator of the
F-test statistic uses the variability between
each sample mean and the overall mean for
all the data
Agresti/Franklin Statistics, 26 of 82
The Variance Estimates and the
ANOVA Table

Computer software displays the two
estimates of σ2 in the ANOVA table

The MS column contains the two
estimates, which are called mean squares

The ratio of the two mean squares is the
F- test statistic

This F- statistic has a P-value
Agresti/Franklin Statistics, 27 of 82
Example: ANOVA for Customers’
Telephone Holding Times

This example is a continuation of a
previous example in which an airline
conducted a randomized experiment to
analyze whether callers would remain on
hold longer, on the average, if they heard:
• An advertisement about the airline and its
•
•
current promotion
Muzak (“elevator music”)
Classical music
Agresti/Franklin Statistics, 28 of 82
Example: ANOVA for Customers’
Telephone Holding Times

Denote the holding time means for
the populations that these three
random samples represent by:
• µ1 = mean for the advertisement
• µ2 = mean for the Muzak
• µ3 = mean for the classical music
Agresti/Franklin Statistics, 29 of 82
Example: ANOVA for Customers’
Telephone Holding Times

The hypotheses for the ANOVA test
are:
• H0:µ1=µ2=µ3
• Ha:at least two of the population
means are different
Agresti/Franklin Statistics, 30 of 82
Example: ANOVA for Customers’
Telephone Holding Times
Agresti/Franklin Statistics, 31 of 82
Example: ANOVA for Customers’
Telephone Holding Times

Since P-value < 0.05, there is
sufficient evidence to reject
H0:µ1=µ2=µ3

We conclude that a difference exists
among the three types of messages
in the population mean amount of
time that customers are willing to
remain on hold
Agresti/Franklin Statistics, 32 of 82
Assumptions for the ANOVA F-Test
and the Effects of Violating Them
1. Population distributions are normal
•
Moderate violations of the normality
assumption are not serious
2. These distributions have the same
standard deviation, σ
•
Moderate violations are also not serious
Agresti/Franklin Statistics, 33 of 82
Assumptions for the ANOVA F-Test
and the Effects of Violating Them


You can construct box plots or dot
plots for the sample data distributions
to check for extreme violations of
normality
Misleading results may occur with the
F-test if the distributions are highly
skewed and the sample size N is
small
Agresti/Franklin Statistics, 34 of 82
Assumptions for the ANOVA F-Test
and the Effects of Violating Them

Misleading results may also occur
with the F-test if there are relatively
large differences among the standard
deviations (the largest sample
standard deviation being more than
double the smallest one)
Agresti/Franklin Statistics, 35 of 82
The 1998 General Social Survey asked subjects how many
friend they have. Is this associated with the respondent’s
astrological sign (the 13 symbols of the Zodiac)? The ANOVA
table for the data reports F=0.61
State the null hypothesis
a. The population means for all 12 Zodiac signs
are the same.
b. At least two population means are different.
Agresti/Franklin Statistics, 36 of 82
The 1998 General Social Survey asked subjects how many
friend they have. Is this associated with the respondent’s
astrological sign (the 13 symbols of the Zodiac)? The ANOVA
table for the data reports F=0.61
State the alternative hypothesis
a. The population means for all 12 Zodiac signs
are the same.
b. At least two population means are different.
Agresti/Franklin Statistics, 37 of 82
The 1998 General Social Survey asked subjects how many
friend they have. Is this associated with the respondent’s
astrological sign (the 13 symbols of the Zodiac)? The ANOVA
table for the data reports F=0.61
Based on what you know about the F
distribution would you guess that the test
value of 0.61 provides strong evidence against
the null hypothesis?
a. No
b. Yes
Agresti/Franklin Statistics, 38 of 82
The 1998 General Social Survey asked subjects how many
friend they have. Is this associated with the respondent’s
astrological sign (the 13 symbols of the Zodiac)? The ANOVA
table for the data reports F=0.61
The P-value associated with the F-statistic is
0.82. At a significance level of 0.05, what is the
correct decision?
a. Reject Ho
b. Fail to Reject Ho
c. Reject Ha
d. Fail to Reject Ha
Agresti/Franklin Statistics, 39 of 82
Section 13.2
How Should We Follow Up an
ANOVA F-Test?
Agresti/Franklin Statistics, 40 of 82
Follow Up to an ANOVA F-Test

When an analysis of variance F-test
has a small P-value, the test does not
specify which means are different or
how different they are

We can estimate differences between
population means with confidence
intervals
Agresti/Franklin Statistics, 41 of 82
Confidence Intervals Comparing
Pairs of Means

For two groups i and j, with sample means
yi and yj having sample sizes ni and nj, the
95% confidence interval for µi - µj is:
1 1
y  y  t s 
n n
i
j
.025
i

The t-score has df = N – g
Agresti/Franklin Statistics, 42 of 82
j
Confidence Intervals Comparing
Pairs of Means



Some software refers to this
confidence method as the Fisher
method
When the confidence interval does
not contain 0, we can infer that the
population means are different
The interval shows just how different
the means may be
Agresti/Franklin Statistics, 43 of 82
Example: Number of Good
Friends and Happiness


A recent GSS study asked: “About
how many good friends do you
have?”
The study also asked each
respondent to indicate whether they
were ‘very happy,’ ‘pretty happy,’ or
‘not too happy’
Agresti/Franklin Statistics, 44 of 82
Example: Number of Good
Friends and Happiness

Let the response variable y = number
of good friends

Let the categorical explanatory
variable x = happiness level
Agresti/Franklin Statistics, 45 of 82
Example: Number of Good
Friends and Happiness
Agresti/Franklin Statistics, 46 of 82
Example: Number of Good
Friends and Happiness


Construct a 95% CI to compare the
population mean number of good friends
for the two categories: ‘very happy’ and
‘pretty happy.’
95% CI formula:
1 1
y  y  t s 
n n
i
j
.025
i
Agresti/Franklin Statistics, 47 of 82
j
Example: Number of Good
Friends and Happiness

First, use the output to find s:
s  MS error  234.2  15.3

Use software or a table to find the t-value of
1.963
Agresti/Franklin Statistics, 48 of 82
Example: Number of Good
Friends and Happiness

Next calculate the interval:
1
1
(10.4  7.4)  1.963(15.3)

276 468
 3.0  2.3, or (0.7, 5.3)

Since the CI contains only positive numbers,
this suggest that, on average, people who are
very happy have more good friends than
people who are pretty happy
Agresti/Franklin Statistics, 49 of 82
Controlling Overall Confidence
with Many Confidence Intervals

The confidence interval method just
discussed is mainly used when g is
small or when only a few
comparisons are of main interest

The confidence level of 0.95 applies
to any particular confidence interval
that we construct
Agresti/Franklin Statistics, 50 of 82
Controlling Overall Confidence
with Many Confidence Intervals

How can we construct the intervals so that
the 95% confidence extends to the entire
set of intervals rather than to each single
interval?

Methods that control the probability that all
confidence intervals will contain the true
differences in means are called multiple
comparison methods
Agresti/Franklin Statistics, 51 of 82
Controlling Overall Confidence
with Many Confidence Intervals



The method that we will use is called
the Tukey method
It is designed to give overall
confidence level very close to the
desired value (such as 0.95)
This method is available in most
software packages
Agresti/Franklin Statistics, 52 of 82
Example: Multiple Comparison
Intervals for Number of Good Friends
Agresti/Franklin Statistics, 53 of 82
Section 13.3
What if There Are Two Factors?
Two-Way ANOVA
Agresti/Franklin Statistics, 54 of 82
ANOVA

One-way ANOVA is a bivariate method:

Two-way ANOVA is a multivariate method:
• It has a quantitative response variable
• It has one categorical explanatory variable
• It has a quantitative response variable
• It has two categorical explanatory
variables
Agresti/Franklin Statistics, 55 of 82
Example: How Does Corn Yield Depend
on Amounts of Fertilizer and Manure?

A recent study at Iowa State University:
•
•
•
•
A field was portioned into 20 equal-size plots
Each plot was planted with the same amount of
corn seed
The goal was to study how the yield of corn later
harvested depended on the levels of use of
nitrogen-based fertilizer and manure
Each factor (fertilizer and manure) was measured
in a binary manner
Agresti/Franklin Statistics, 56 of 82
Example: How Does Corn Yield Depend
on Amounts of Fertilizer and Manure?

What are the four treatments you can
compare with this experiment?
Agresti/Franklin Statistics, 57 of 82
Inference about Effects in
Two-Way ANOVA

In two-way ANOVA, a null hypothesis
states that the population means are
the same in each category of one
factor, at each fixed level of the other
factor
Agresti/Franklin Statistics, 58 of 82
Example: How Does Corn Yield Depend
on Amounts of Fertilizer and Manure?

We could test:
H0: Mean corn yield is equal for plots
at the low and high levels of fertilizer,
for each fixed level of manure
Agresti/Franklin Statistics, 59 of 82
Example: How Does Corn Yield Depend
on Amounts of Fertilizer and Manure?

We could also test:
H0: Mean corn yield is equal for plots
at the low and high levels of manure,
for each fixed level of fertilizer
Agresti/Franklin Statistics, 60 of 82
Example: How Does Corn Yield Depend
on Amounts of Fertilizer and Manure?

The effect of individual factors tested
with the two null hypotheses (the
previous two pages) are called main
effects
Agresti/Franklin Statistics, 61 of 82
Assumptions for the Two-way
ANOVA F-test
1. The population distribution for each
group is normal
2. The population standard deviations
are identical
3. The data result from a random
sample or randomized experiment
Agresti/Franklin Statistics, 62 of 82
F-test Statistics in Two-way
ANOVA

For testing the main effect for a factor, the
test statistic is the ratio of mean squares:
MS for the factor
F
MS error
Agresti/Franklin Statistics, 63 of 82
F-test Statistics in Two-way
ANOVA

When the null hypothesis of equal
population means for the factor is true,
the F-test statistic values tend to fluctuate
around 1

When it is false, they tend to be larger

The P-value is the right-tail probability
above the observed F-value
Agresti/Franklin Statistics, 64 of 82
Example: Testing the Main
Effects for Corn Yield

Data and sample statistics for each
group:
Agresti/Franklin Statistics, 65 of 82
Example: Testing the Main
Effects for Corn Yield

Output from Two-way ANOVA:
Agresti/Franklin Statistics, 66 of 82
Example: Testing the Main
Effects for Corn Yield

First consider the hypothesis:
H0: Mean corn yield is equal for plots
at the low and high levels of fertilizer,
for each fixed level of manure
Agresti/Franklin Statistics, 67 of 82
Example: Testing the Main
Effects for Corn Yield

From the output, you can obtain the
F-test statistic of 6.33 with its
corresponding P-value of 0.022

The small P-value indicates strong
evidence that the mean corn yield
depends on fertilizer level
Agresti/Franklin Statistics, 68 of 82
Example: Testing the Main
Effects for Corn Yield

Next consider the hypothesis:
H0: Mean corn yield is equal for plots
at the low and high levels of manure,
for each fixed level of fertilizer
Agresti/Franklin Statistics, 69 of 82
Example: Testing the Main
Effects for Corn Yield

From the output, you can obtain the
F-test statistic of 6.88 with its
corresponding P-value of 0.018

The small P-value indicates strong
evidence that the mean corn yield
depends on manure level
Agresti/Franklin Statistics, 70 of 82
Exploring Interaction between
Factors in Two-Way ANOVA

No interaction between two factors
means that the effect of either factor
on the response variable is the same
at each category of the other factor
Agresti/Franklin Statistics, 71 of 82
Exploring Interaction between
Factors in Two-Way ANOVA
Agresti/Franklin Statistics, 72 of 82
Exploring Interaction between
Factors in Two-Way ANOVA

A graph showing interaction:
Agresti/Franklin Statistics, 73 of 82
Testing for Interaction

In conducting a two-way ANOVA,
before testing the main effects, it is
customary to test a third null
hypothesis stating that their is no
interaction between the factors in
their effects on the response
Agresti/Franklin Statistics, 74 of 82
Testing for Interaction

The test statistic providing the sample
evidence of interaction is:
MS for interactio n
F
MS for error

When H0 is false, the F-statistic tends to be
large
Agresti/Franklin Statistics, 75 of 82
Example: Testing for
Interaction with Corn Yield Data

ANOVA table for a model that allows
interaction:
Agresti/Franklin Statistics, 76 of 82
Example: Testing for
Interaction with Corn Yield Data

The test statistic for H0: no interaction
is
•
•
F = 1.10 with a corresponding P-value
of 0.311
• There is not much evidence of
•
interaction
We would not reject H0 at the usual
significance levels, such as 0.05
Agresti/Franklin Statistics, 77 of 82
Check Interaction before
Main Effects

In practice, in two-way ANOVA, you
should first test the hypothesis of no
interaction
Agresti/Franklin Statistics, 78 of 82
Check Interaction before
Main Effects

If the evidence of interaction is not
strong (that is, if the P-value is not
small), then test the main effects
hypotheses and/or construct
confidence intervals for those
effects
Agresti/Franklin Statistics, 79 of 82
Check Interaction before
Main Effects

If important evidence of interaction
exists, plot and compare the cell
means for a factor separately at each
category of the other factor
Agresti/Franklin Statistics, 80 of 82
An experiment randomly assigns 100 subjects
suffering from high cholesterol to one of four
groups: low-dose Lipitor, high-dose Lipitor, lowdose Pravachol and high-dose Pravachol. After
three months of treatment, the change in
cholesterol level is measured.
What is the response variable?
a. Cholesterol level
b. Drug dosage
c. Drug type
Agresti/Franklin Statistics, 81 of 82
An experiment randomly assigns 100 subjects
suffering from high cholesterol to one of four
groups: low-dose Lipitor, high-dose Lipitor, lowdose Pravachol and high-dose Pravachol. After
three months of treatment, the change in
cholesterol level is measured.
What are the factors?
a. Cholesterol level and drug type
b. Drug dosage and cholesterol level
c. Drug type and drug dosage
Agresti/Franklin Statistics, 82 of 82