BES – LAB 5

Multiple Comparisons
1. Objectives
- Determine when to use multiple comparison tests.
- Conduct multiple comparison tests with Bonferroni and Tukey’s HSD method.
- Produce and interpret the results.
2. Exercises
Firstly, remember to set your own working directory.

Exercise 1. Kudzu is a plant that was imported to the United States from Japan and now covers
over seven million acres in the South. The plant contains chemicals called isoflavones that have been
shown to have beneficial effects on bones. One study used three groups of rats to compare a control
group with rats that were fed either a low dose or a high dose of isoflavones from kudzu.csv. One
of the outcomes examined was the bone mineral density in the femur (in grams per square centimeter).

a. Use graphical and numerical methods to describe the data.

b. Examine the assumptions necessary for one-way ANOVA. Summarize your findings.
c. Run one-way ANOVA and report the results.
d. Use a multiple-comparisons method to compare the three groups.
e. Write a short report explaining the effect of kudzu isoflavones on the femur of the rat.

Let’s run the following codes:

Ø kudzu<-read.table("kudzu.csv", header=TRUE, sep = ",",

stringsAsFactors = F)
Ø str(kudzu) #to see the structure of the data frame
Ø kudzu$treatment<-factor(kudzu$treatment, levels =
c("Control","LowDose","HighDose"))
Ø kudzu$treatment
Ø table(kudzu$treatment) #check the sample sizes

Just follow the instructions from Lab 4 for part a and b. In this lab, we focus on multiple
comparison tests, and so we must conduct one-way ANOVA first to see whether significant
difference exists among 3 groups.

Ø aovKudzu <- aov(boneMineralDensity~treatment,data=kudzu)

Ø summary(aovKudzu)

You should get the following output:

Df Sum Sq Mean Sq F value Pr(>F)

treatment 2 0.003186 0.0015928 7.718 0.0014 **
Residuals 42 0.008668 0.0002064
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

1|Page
 BES – LAB 5

Based on this result, we can conclude that there exists a significant difference among 3 groups
of Control, LowDose, and HighDose. Then we can apply a multiple comparison method to identify
where this difference comes from.
For Bonferroni method, we will use the function pairwise.t.test() for comparing all pairs
of means simultaneously. The format of the code is as the following:

Ø pairwise.t.test(outcome, grouping factor, p.adjust.method =

p.adjust.methods, paired = FALSE)

Below are the meanings of the arguments:

outcome Name of your outcome variable
grouping factor Name of the factor that defines the groups
p.adjust.method Method for adjusting p-value. Use "bonferroni" if you want Bonferroni
method.
paired A logical statement indicating whether you want paired-samples t-tests

So the code to run the Bonferroni method for our exercise will be:

Ø pairwise.t.test(kudzu$boneMineralDensity, kudzu$treatment, paired

= F, p.adjust.method = "bonferroni")
Here is the ouput:
Pairwise comparisons using t tests with pooled SD
data: kudzu$boneMineralDensity and kudzu$treatment
Control LowDose
LowDose 1.0000 -
HighDose 0.0107 0.0022

P value adjustment method: bonferroni

For Tukey’s HSD method, we need the multcomp package. First, install the package. Then use
library(multcomp)to load the package. The appropriate function is glht(), the format of
which is as follows:

Ø mcpModel <- glht(aovModel, linfct = mcp(grouping factor =

"method"))
Ø summary(mcpModel)

where:
mcpModel Object containing information of the multiple comparison procedure used
aovModel Name of the model that has been created with the aov() function.
grouping factor Factor that defines the groups
linfct = Specifies the method to adjust p-value. If you want Tukey’s HSD, replace
mcp(grouping "method" by "Tukey"
factor="method")
summary(mcpModel) Print output of the multiple comparison procedure contained in mcpModel

2|Page
 BES – LAB 5

So the code to run the Tukey’s HSD method for our exercise will be:
Ø library(multcomp)
Ø tukeyKudzu <- glht(aovKudzu, linfct = mcp(treatment = "Tukey"))
Ø summary(tukeyKudzu)
You will see the following output:
Simultaneous Tests for General Linear Hypotheses

Multiple Comparisons of Means: Tukey Contrasts

Fit: aov(formula = boneMineralDensity ~ treatment, data = kudzu)

Linear Hypotheses:
Estimate Std. Error t value Pr(>|t|)
LowDose - Control == 0 -0.002933 0.005246 -0.559 0.84234
HighDose - Control == 0 0.016200 0.005246 3.088 0.00961 **
HighDose - LowDose == 0 0.019133 0.005246 3.648 0.00200 **
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Adjusted p values reported -- single-step method)

Question 1. In which case should we choose Bonferroni or Tukey’s HSD method? Which method
is more appropriate for this problem?
Question 2. Read and interpret the outputs for Bonferroni and Tukey’s HSD methods. Compare
the 2 outputs and make comments about their differences.
Exercise 2. If a supermarket product is frequently offered at a reduced price, do customers expect
the price of the product to be lower in the future? This question was examined by researchers in a
study conducted on students enrolled in an introductory management course at a large midwestern
university. For 10 weeks, 160 subjects read weekly ads for the same product. Students were
randomly assigned to read 1, 3, 5, or 7 ads featuring price promotions during the 10-week period.
They were then asked to estimate what the product’s price would be the following week.
(a) Make a Q-Q plot for the data in each of the four treatment groups. Summarize the
information in the plots and draw a conclusion regarding the normality of data.
(b) Summarize the data with a table containing the sample size, mean, and standard deviation for
each group.
(c) Is the assumption of equal standard deviations reasonable here? Explain why or why not.
(d) Carry out a one-way ANOVA. Give the hypotheses, the test statistic with its degrees of
freedom, and the p-value. State your conclusion.
(e) Use the Bonferroni or Tukey’s HSD procedure to compare the group means. Summarize the
results and support your conclusions with a graph of the means.
Follow the instructions from Lab 4 to do parts a, b, and c. After importing data from
pricepromo.csv, you’re also expected to produce output for d and e as follows:

3|Page
 BES – LAB 5

Df Sum Sq Mean Sq F value Pr(>F)

Promotions 3 8.361 2.7868 25.66 1.52e-13 ***
Residuals 156 16.946 0.1086
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Pairwise comparisons using t tests with pooled SD

data: price$Price and price$Promotions

1ads 3ads 5ads

3ads 0.2695 - -
5ads 4.5e-07 0.0023 -
7ads 1.7e-12 9.6e-08 0.1242

P value adjustment method: bonferroni

Simultaneous Tests for General Linear Hypotheses

Multiple Comparisons of Means: Tukey Contrasts

Fit: aov(formula = Price ~ Promotions, data = price)

Linear Hypotheses:
Estimate Std. Error t value Pr(>|t|)
3ads - 1ads == 0 -0.1490 0.0737 -2.022 0.18438
5ads - 1ads == 0 -0.4163 0.0737 -5.648 < 0.001 ***
7ads - 1ads == 0 -0.5885 0.0737 -7.985 < 0.001 ***
5ads - 3ads == 0 -0.2672 0.0737 -3.626 0.00204 **
7ads - 3ads == 0 -0.4395 0.0737 -5.963 < 0.001 ***
7ads - 5ads == 0 -0.1722 0.0737 -2.337 0.09417 .
---
Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
(Adjusted p values reported -- single-step method)

4|Page