One-Way Analysis of

Variance
Engineering Data Analysis
 Objectives
Design experiments involving several factors.
• At the end of the lesson, the students are
expected to
• Design and conduct engineering experiments
involving a single factor with an arbitrary
number of levels; and
• Understand how the analysis of variance is
used to analyze the data from these
experiments.
 Completely Randomized Single-Factor Experiment
13-2.1 EXAMPLE: TENSILE STRENGTH Treatments
A manufacturer of paper used for making grocery • levels of the factor (Hardwood Concentration)
bags is interested in improving the product’s tensile
• a very general term that can be traced to the early
strength. Product engineering believes that tensile
applications of experimental design methodology in
strength is a function of hardwood concentration in
the agricultural sciences.
the pulp and that the range of hardwood
concentrations of practical interest is between 5 and Replicates
20%. A team of engineers responsible for the study
decides to investigate four levels of hardwood • observations (6 per treatment in the tensile strength
concentration: 5%, 10%, 15%, and 20%. They example)
decide to make up six test specimens at each The role of randomization in this experiment is
concentration level by using a pilot plant. extremely important. The response for each of the a
TABLE ● 13-1 Tensile Strength of Paper (psi) treatments is a random variable.
Hardwood Concentration Observations
(%)
1 2 3 4 5 6 Totals Averages
5 7 8 15 11 9 10 60 10.00
10 12 17 13 18 19 15 94 15.67
15 14 18 19 17 16 18 102 17.00
20 19 25 22 23 18 20 127 21.17
383 15.96
 Completely Randomized Single-Factor Experiment
ANALYSIS OF VARIANCE We may describe the observations in Table 13-2 by
the linear statistical model
TABLE ● 13-2 Typical Data for a Single-Factor
𝑖 = 1, 2, . . . , 𝑎
Experiment 𝑌𝑖𝑗 = 𝜇 + 𝜏𝑖 + 𝜖𝑖𝑗 ቊ
𝑗 = 1, 2, . . . , 𝑛

Treatment Observations Totals Averages (13-1)→underlying model for single-factor experiment

1 y11 y12 … y1n y1· 𝑦ത1⋅
2 y12 y22 … y2n y2· 𝑦ത2⋅ yij – a random variable denoting the (ij)th observation
⋮ ⋮ ⋮ ⋮⋮⋮ ⋮ ⋮ ⋮ μ − a parameter common to all treatments called the
a ya1 ya2 … yan ya· 𝑦ത𝑎⋅ overall mean
y·· 𝑦ത1⋅⋅ τi − a parameter associated with the ith treatment
called the ith treatment effect
ϵij − random error component
 Completely Randomized Single-Factor Experiment
Notice that the model could have been written as Completely Randomized Design (CRD)
𝑖 = 1, 2, . . . , 𝑎
𝑌𝑖𝑗 = 𝜇 + 𝜖𝑖𝑗 ቊ • An experimental design where observations are
𝑗 = 1, 2, . . . , 𝑛 taken in random order and that the environment
μi = μ + τi – mean of the ith treatment (often called the experimental units) in which the
treatments are used is as uniform as possible.

We assume that the errors ϵij are normally and

independently distributed with mean zero and Fixed-Effects Model
variance σ2. See Fig. 13-1 (b), display of the model • Estimating the treatment effects
in Equation 13-1 for the completely randomized
single-factor experiment.
 Completely Randomized Single-Factor Experiment
Random-Effects or Components of Variance • yi· = total of the observations under the ith treatment
Model
• 𝑦ത𝑖⋅ = average of the observations under the ith
Testing hypotheses about the variability of the τi and treatment
estimating this variability
• y·· = grand total of all observations
Analysis of Variance (ANOVA) for the Fixed-
• 𝑦ത⋅⋅ = grand mean of all observations
Effects Model
• N = an = total number of observations
In the fixed-effects model, the treatment effects τi
are usually defined as deviations from the overall • The “dot” subscript notation implies summation over
mean μ, so that 𝑎
the subscript that it replaces.
෍ 𝜏𝑖 = 0
𝑖=1
(13-2)
𝑛

𝑦𝑖⋅ = ෍ 𝑦𝑖𝑗 𝑦ത𝑖⋅ = 𝑦𝑖⋅ Τ𝑛 𝑖 = 1, 2, . . . , 𝑎

𝑗=1 𝑎 𝑛

𝑦⋅⋅ = ෍ ෍ 𝑦𝑖𝑗 𝑦ത⋅⋅ = 𝑦⋅⋅ Τ𝑁

𝑖=1 𝑗=1
(13-3)
 One-Way ANOVA
We are interested in testing the equality of the a Equivalently,
treatments μ1, μ2,…, μa. Using Equation 13-2, we
find that this is equivalent to testing the hypotheses H0: μ1 = μ2 = … = μa
H0: τ1 = τ2 = … = τa = 0 H1: At least one mean is different from the others
H1: τi ≠ 0 for at least one i
(13-4) ANOVA Sum of Squares Identity: Single Factor
Experiment
If H0 is true,
• Each observation consists of the overall mean μ 𝑎 𝑛 𝑎 𝑎 𝑛
2 2 2
plus a realization of the random error component ෍ ෍ 𝑦𝑖𝑗 − 𝑦ത⋅⋅ = 𝑛 ෍ 𝑦ത𝑖⋅ − 𝑦ത⋅⋅ + ෍ ෍ 𝑦𝑖𝑗 − 𝑦ത𝑖⋅
ϵij 𝑖=1 𝑗=1 𝑖=1 𝑖=1 𝑗=1
• All N observations are taken from a normal (13-5)
distribution with mean μ and variance σ2
or symbolically
• Changing the levels of the factor has no effect on
the mean response SST = SSTreatments + SSE
(13-6)
 One-Way ANOVA
ANOVA F-Test • The computations for this test procedure are
𝑆𝑆TreatmentsΤ 𝑎 − 1 𝑀𝑆Treatments usually summarized in tabular form as shown
𝐹0 = =
𝑆𝑆𝐸 Τ 𝑎 𝑛 − 1 𝑀𝑆𝐸 in Table 13-3. This is called an analysis of
(13-7) variance (or ANOVA) table.

Computing Formulas for ANOVA: Single Factor TABLE ● 13-3 Analysis of Variance for a Single-
with Equal Sample Sizes
𝑎 𝑛 Factor Experiment, Fixed-Effects Model
2
2 𝑦⋅⋅
𝑆𝑆𝑇 = ෍ ෍ 𝑦𝑖𝑗 − Source of Sum of Degrees of Mean F0
𝑁 Variation Squares Freedom Square
𝑖=1 𝑗=1
(13-8) Treatments SSTreatments a−1 MSTreatments MSTreatments/MSE
𝑎
𝑦𝑖⋅2 𝑦⋅⋅2 Error SSE a(n − 1) MSE
𝑆𝑆Treatments =෍ −
𝑛 𝑁
𝑖=1 Total SST an − 1
(13-9)
The error sum of squares is obtained by subtraction as
SSE = SST − SSTreatments
(13-10)
 One-Way ANOVA
13-1/545 Tensile Strength ANOVA We can use α = 0.01. From Equations 13-8, 13-
We can use the analysis of variance to test the 9, and 13-10, 4 6
hypothesis that different hardwood concentrations 2 𝑦⋅⋅2
do not affect the mean tensile strength of paper. 𝑆𝑆𝑇 = ෍ ෍ 𝑦𝑖𝑗 −
𝑁
𝑖=1 𝑗=1
2
Hardwood Observations 2 2 2
383
Concentration = 7 + 8 + ⋯ + 20 −
(%) 1 2 3 4 5 6 Total Average 24
𝑆𝑆𝑇 = 512.96
5 7 8 15 11 9 10 60 10.00 4
𝑦𝑖⋅2 𝑦⋅⋅2
10 12 17 13 18 19 15 94 15.67 𝑆𝑆Treatments =෍ −
𝑛 𝑁
15 14 18 19 17 16 18 102 17.00 2 2 2𝑖=1 2 2
60 + 94 + 102 + 127 383
= −
20 19 25 22 23 18 20 127 21.17 6 24
𝑆𝑆𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠 = 382.79
383 15.96
𝑆𝑆𝐸 = 𝑆𝑆𝑇 − 𝑆𝑆𝑇𝑟𝑒𝑎𝑡𝑚𝑒𝑛𝑡𝑠
𝑆𝑆𝐸 = 512.96 − 382.79
The hypotheses are
𝑆𝑆𝐸 = 130.17
H0: τ1 = τ2 = … = τa = 0
H1: τi ≠ 0 for at least one I
 One-Way ANOVA
The ANOVA is summarized in Table 13-4. Because P = P(F3,20 > 19.60) ≈ 3.59×10−6 (From computer
f0.01,3,20 = 4.94, we reject H0 and conclude that software)
hardwood concentration in the pup significantly
affects the mean strength of the paper. We can also Practical Interpretation: There is strong evidence
find a P-value for this test statistic as follows. to conclude that hardwood concentration has an
• Table ● 13-4 ANOVA for the Tensile Strength Data effect on tensile strength. However, the ANOVA
does not tell as which levels of hardwood
Source of Sum of Degrees of Mean f0 P-value
Variation Squares Freedom Square
concentration result in different tensile strength
means. We see how to answer this question in
Hardwood 382.79 3 127.60 19.60 3.59 E-6
concentration Section 13-2.3.
Error 130.17 20 6.51

Total 512.96 23
 One-Way ANOVA
Computing Formulas for ANOVA: Single Factor
with Unequal Sample 𝑎Sizes
𝑛𝑖
2 𝑦⋅⋅2
𝑆𝑆𝑇 = ෍ ෍ 𝑦𝑖𝑗 −
𝑁
𝑖=1 𝑗=1
(13-13)𝑎
𝑦𝑖⋅2 𝑦⋅⋅2
𝑆𝑆Treatments =෍ −
𝑛 𝑁
𝑖=1
(13-14)
SSE = SST − SSTreatments
(13-15)
 One-Way ANOVA
13-1/554 Consider the following computer output.
Source DF SS MS F P-value
Factor ? 117.4 39.1 ? ?
Error 16 396.8 ?
Total 19 514.2

(a) How many levels of the factor were used in this

experiment?
(b) How many replicates did the experimenter use?
(c) Fill in the missing information in the ANOVA
table. Use bounds for the P-value.
(d) What conclusions can you draw about
inferences in the factor-level means?
 One-Way ANOVA
13-5/555 In Design and Analysis of Experiment, 8th (a) Does cotton percentage affect breaking
edition (John Wiley & Sons, 2012), D. C. strength? Draw comparative box plots and
Montgomery described an experiment in which the perform an analysis of variance. Use α = 0.05.
tensile strength of a synthetic fiber was of interest to
the manufacturer. It is suspected that strength is (b) Plot average tensile strength against cotton
related to the percentage of cotton in the fiber. Five percentage and interpret the results.
levels of cotton percentage were used, and five
replicates were run in random order, resulting in the (c) Analyze the residuals and comment on
following data. model adequacy.
Cotton Observations
Percentage
1 2 3 4 5
15 7 7 15 11 9
20 12 17 12 18 18
25 14 18 18 19 19
30 19 25 22 19 23
35 7 10 11 15 11
 One-Way ANOVA
13-17/557 An experiment was run to determine
whether four specific firing temperatures affect the
density of a certain type of brick. The experiment
led to the following data.
Temperature Density
(°F)
100 21.8 21.9 21.7 21.6 21.7 21.5 21.8
125 21.7 21.4 21.5 21.5 ― ― ―
150 21.9 21.8 21.8 21.6 21.5 ― ―
175 21.9 21.7 21.8 21.7 21.6 21.8 ―

(a) Does the firing temperature affect the density of

the bricks? Use α = 0.05.
(b) Find the P-value for the F-statistic computed in
part (a).
(c) Analyze the residuals from the experiment.