FAR EASTERN UNIVERSITY – DEPARTMENT OF MEDICAL TECHNOLOGY

Analysis of Variance (ANOVA)

I. Introduction
A. Statistical Test Measurement of Serum
1. One Mean - Z Test and T Test Testosterone Levels (mg/dL)
2. Two Means - Z Tests and T Test (Independent and Paired)
3. Two or More Means - ANOVA (One-Way, Two-Way, Repeated measures)
B. Rationale
• A study was conducted to determine whether cigarette smoking is
associated with reduced serum-testosterone levels (mg/dL) in men aged
35 to 45. Is there enough evidence to say that serum testosterone differs
across smoking habits?
a. Nonsmokers who had never smoked
b. Former smokers who had quit for ≤6 months prior to the study
c. Light smokers, defined as those who smoked ≤10 cigarettes per day
d. Heavy smokers, defined as those who smoked >10 cigarettes per day
• Multiple Independent T Test can be done, but it increases the probability
that some comparison will result in a type I error.
C. Errors in Hypothesis Test Multiple Independent T-Test
1. TYPE II error or β
a. Fail to reject (Accepted) the false H0
b. Rejected the True Hɑ
2. TYPE I error or α
a. Rejected the true H0
b. Accepted the false Ha
Condition of Null Hypothesis
True False
Fail to Type II error
Possible Correct Action
reject H0 or β (↓ ɑ)
Action
Type I error
Reject H0 Correct Action
or α (↑ ɑ)
• What is needed is to simultaneously compare these
means in one step by a method is called ANOVA

II. DEFINITION AND TYPES OF ANOVA

A. ANOVA (Analysis of Variance)
Two-sample t-test (with pooled variance)
•
Used when the question involves the comparisons of means from ≥2 independent groups.
•
The total variation between groups and within groups is determined
•
B. Types of ANOVA

One-Way ANOVA Two-Way ANOVA

Definition A test that allows one to make A test that allows to make compare between the
comparisons between the means of means of three or more groups of data, where
three or more groups of data. two independent variables are considered.
No. of Independent Variables One. Two.
What is Being Compared? The means of three or more groups of The effect of multiple groups of two
an independent variable on a independent variables on a dependent variable
dependent variable. and on each other.
Number of Groups of Samples Three or more. Each variable should have multiple samples.

AARON JAN PALMARES, RMT, MSMST_11.04.19 Page 1

 FAR EASTERN UNIVERSITY – DEPARTMENT OF MEDICAL TECHNOLOGY

III. ANOVA Hypothesis Testing

A. ANOVA Hypothesis Testing
1. Hypotheses: H0 & Ha 4. Critical Region 7. Conclusion
2. Level of Significance 5. Compute test statistic
3. Test Statistic 6. Statistical decision
IIIa. One-Way ANOVA
• One source of variation is investigated. Extension of t-test (≥3 samples)
A. Assumptions:
1. Independent variable should consist of ≥2 categorical (nominal or
ordinal) groups (ethnicity, profession, physical activity)
2. Dependent variable should be measured at the scale
(interval or ratio) level (IQ score, weight)
3. Dependent variable should be normally distributed (randomly
assigned) for each category of the independent variable
4. Independence of observations, no significant outliers & homogeneity
of variances (Non-Parametric Counterpart: Kruskal-Wallis)
B. Procedure:
• Determine if the mean hours of relief from pain differ among the five
different brands of headache tablets (A to E).
• n = observations / group; N = total no. of observations; k = groups
F-distribution with 2 sets of df
Hours Relief from headache per Tablet
N A B C D E
1 5 9 3 2 7
2 4 7 5 3 6
3 8 8 2 4 9
4 6 6 3 1 4
5 3 9 7 4 7
Total 26 39 20 14 33 132
Mean 5.2 7.8 4 2.8 6.6 5.28
1. State the Hypotheses
a. H0: µ1 = µ2 = µ3 = µ4 ….µk or 𝛕𝐣 = 0 (There is no significant difference among the mean hours of relief from pain
provided by the various brands of headache tablets)
b. Ha: not all 𝛕𝐣 = 0 (There is a significant difference in at least one mean hour of relief from pain provided by a
certain brand of headache tablet)
2. Level of Significance (ɑ): 0.05, 0.01, 0.1 4. Critical Region
3. Test Statistic: 𝑘−1 5−1=4
• 𝐹𝑁−𝑘 = 𝐹25−5=20 = 2.87
MSA
• One-Way ANOVA, F (Variance of Ratio, V.R) = /MSW
5. Compute Test Statistic

Sources of variation Sum of squares Degrees of freedom Mean square Computed f

𝒌
SSA (Among / 𝟐 𝑺𝑺𝑨
∑ 𝒏(𝒙𝒋 − 𝒙
̅⋯ ) 𝒌−𝟏 MSA = 𝒌 −𝟏
Between groups) 𝒋=𝟏 𝑴𝑺𝑨
𝒌 𝒏𝒋 F = 𝑴𝑺𝑾
SSW (SST-SSA) 𝟐 𝑺𝑺𝑾
∑ ∑(𝒙𝒊𝒋 − 𝒙
̅𝒋 ) 𝑵−𝒌 MSW = 𝑵 −𝒌
Within groups
𝒋=𝟏 𝒊=𝟏
𝒌 𝒏𝒋
SST 𝟐
∑ ∑(𝒙𝒊𝒋 − 𝒙
̿) 𝑵−𝟏
Total sum of squares
𝒋=𝟏 𝒊=𝟏

AARON JAN PALMARES, RMT, MSMST_11.04.19 Page 2

 FAR EASTERN UNIVERSITY – DEPARTMENT OF MEDICAL TECHNOLOGY

5. Compute Test Statistic

Source of variation Sum of squares Df Mean square Computed f
𝑺𝑺𝑨 𝟕𝟗.𝟒𝟒
SSA 79.44 5–1=4 MSA = 𝒌 −𝟏 = 𝟓−𝟏
= 19.86 𝑴𝑺𝑨 𝟏𝟗.𝟖𝟔
F = V.R = 𝑴𝑺𝑾 = 𝟐.𝟖𝟖
= 6.90
𝑺𝑺𝑾 𝟓𝟕.𝟔𝟎
SSW 57.60 25 – 5 = 20 MSW = 𝑵 −𝒌 = = 2.88
𝟐𝟎
SST 137.04 25 -1 = 24
6. Statistical Decision
• Reject H0 if the computed F (V.R) is ≥ the critical F
• Since V.R (6.9) is ≥ FCRIT (2.87), reject H0
7. Conclusion
There is a significant difference in at least one mean
hour of relief from pain provided by a certain brand of
headache tablet.

IIIb. Two-way ANOVA

• Examines the influence of two (categorical) independent
variables on one (continuous quantitative) dependent variable,
as well as interaction between the two independent variables
• The technique to be used for the analysis of data below is by
randomized complete block design
• Units to which the treatments are applied are subdivided into
groups called blocks
A. Assumptions
1. No significant outliers
2. Dependent variable should be normally distributed for each
group of the 2 independent variables
3. Homogeneity of variances for each combination of the
groups of the two independent variables
B. Procedure
• A physical therapist wished to compare 3 methods for teaching
patients to use a certain prosthetic device. Use 95% CI
• 3 treatments, and 5 blocks (age groups).
Time (in Days) Required to Learn the Use of a Prosthetic Device
Teaching Method Total Mean
Age Group
A B C
< 20 7 9 10 26 8.67
20 – 29 8 9 10 27 9.00
30 – 39 9 9 12 30 10.00
40 – 49 10 9 12 31 10.33
≥ 50 11 12 14 37 12.33
Total 45 48 58 151
Mean 9.0 9.6 11.6 10.07
1. State the Hypotheses
a. H0: 𝝉𝒋 = 0 (There is no significant difference among the mean days of learning in the use of prosthetic device)
b. Ha: not all 𝛕𝐣 = 0 (There is a significant difference in at least one mean day of learning in the use of prosthetic
device)
2. Level of Significance: ɑ = 0.1, 0.05, 0.01
3. Test Statistic: 4. Critical Region
𝑀𝑆𝑇𝑟 𝑘−1
• Two Way ANOVA, Fc = V.R. = 𝑀𝑆𝐸 • Fc ≥ 𝐹(𝑛−1)(𝑘−1) 𝐹 3−1=2
(5−1)∗(3−1) =8
• Treatments (k – 1) and Residual error (n – 1)(k – 1)

AARON JAN PALMARES, RMT, MSMST_11.04.19 Page 3

 FAR EASTERN UNIVERSITY – DEPARTMENT OF MEDICAL TECHNOLOGY

Compute Test Statistic

5.
Sources of Computed
SS DF MS
variation F
𝑆𝑆𝑇𝑟
Treatments SSTr k – 1 → (3 – 1) → 2 MSTr = 𝑘 −1 𝑀𝑆𝑇𝑟
𝑆𝑆𝐵𝐼
V.R = 𝑀𝑆𝐸
Blocks SSBI n – 1 → (5 – 1) → 4 MSBI = 𝑛 −1
𝑆𝑆𝐸
Residual SSE (n – 1)(k – 1) → (4)(2) → 8 MSE = (𝑛 −1)(𝑘 −1)
Total SST kn – 1 → (3*5) – 1 → 14 SST= SSBl + SSTr + SSE
6. Statistical Decision
• Reject H0 if the computed V.R. ≥ critical F (4.46). Since the V.R (20.91) ≥ FCRIT (4.46), reject H0
7. Conclusion
• The mean days of learning is not the same for all teaching methods
• There is a significant difference in at least one mean day of learning in the use of prosthetic device

IIIc. Repeated Measures ANOVA (with Single Factor)

• Frequently used experimental designs in health sciences
• Repeated measurements of the same variable are made on ≥3 occasions.
1. change over time (e.g. longitudinal study)
2. change under different conditions
• Subject serves as its own control for extraneous variation among subjects
• Major disadvantage
1. Carry-over effect - occurs when ≥2 treatments are being evaluated.
2. Position effect - a subject’s response to a treatment, that was
experienced last in a sequence might be different if it has been
experienced first in the sequence.
A. Assumptions
1. The subjects are assigned by simple random sampling (randomness)
2. Each observation is independent (size 1 from each of kn)
3. The populations (kn) must have the same variance (normality)
4. The treatments (k) are fixed (the only interest in the study)
5. There is no interaction between treatments and subjects. Treatment effects are additive.
6. Dependent variable is measured in scale
7. Homogeneity of covariance (sphericity)
• Covariance (correlations) exist among the repeated measures since it was taken on the same individual
B. Procedure
• In a study, 18 subjects with chronic low back pain were examined.
They completed a survey questionnaire assessing physical
functioning at baseline, and after 1, 3, and 6 months. Higher
values indicate better physical functioning. Determine if there is
an improvement of the values over time.
1. State the hypotheses
a. H0: µB = µM1 = µM3 = µM6 or 𝛕𝐣 = 0
(There are no significant differences among the mean
values/scores of physical functions at all time points)
b. Ha: not all 𝛕𝐣 = 0
(There is a significant difference in at least one mean
value/score of physical function at one time point)

AARON JAN PALMARES, RMT, MSMST_11.04.19 Page 1

 FAR EASTERN UNIVERSITY – DEPARTMENT OF MEDICAL TECHNOLOGY

2. Level of Significance: 0.05, 0.01, 0.1

𝑀𝑆𝑇𝑟
3. Test Statistic: V.R. = 𝑀𝑆𝐸
4. Critical Region:
𝑘−1 4−1=3
• Fc ≥ 𝐹(𝑘−1) (𝑛−1) 𝐹17∗3 =51
• Treatments: k – 1 → (4 – 1) → 3
• Residual error: (n – 1)(k – 1) → (18 – 1) (4 – 1) → 51
5. Compute Test Statistic:
Sources SS DF MS Computed F
𝑆𝑆𝑇𝑟
Treatment SSTr k–1 MSTr = 𝑘 −1 𝑀𝑆𝑇𝑟
𝑆𝑆𝐵𝐼 V.R = 𝑀𝑆𝐸
Blocks SSBI n–1 MSBI =
𝑛 −1
𝑆𝑆𝐸
Residual (error) SSE (n – 1) (k – 1) MSE = (𝑛 −1)(𝑘 −1)
Total (N) SST kn – 1 SST = SSBl + SSTr + SSE
Sources SS DF MS Computed F
Treatment 2396 4–1=3 799 5.50
Blocks 20238 18 – 1 = 17 1190 8.20
Residual (error) 7404 17 * 3 = 51 145
Total 30038 (4 * 18) – 1 = 71
6. Statistical Decision
• Reject H0 if the computed F (V.R) is ≥ the FCRIT
• Since the V.R. (5.50) is ≥ FCRIT (2.79); reject the H0
7. Conclusion
• There is a significant difference in at least one mean value/score of physical function at one time point.
• The six-month treatment had a statistically significant effect on the physical function.

IIId. Multiple Comparison Procedures

• If the means has a significant difference, we proceed with the Post-Hoc Analysis to find out which pair of means
are significantly different. E.g. Tukey’s, Bonferroni, Dunnet, Games-Howell
1. Bonferroni
• Sig. level divided by the no. of individual pairs (/k)
• More power when the number of comparisons is small
• Assumptions are not required to be met
𝑀𝑆𝐸
2. Tukey HSD (honestly significant difference) = 𝑞 √
𝑛
• More power when testing large numbers of means
• Used when homogeneity of variances is met.
If not, use Games-Howell
• All possible differences between pairs of means
are computed

• Conclusion: There is a significant difference between the mean score of reading of

Low and High SES (P = ≤0.0001) and; Middle and High SES (P = 0.009 or 0.014).

AARON JAN PALMARES, RMT, MSMST_11.04.19 Page 2