1. Approach to ANOVA Questions
One way ANOVA
1. State the one-way ANOVA model (estimates only one effect)
π¦ππ = π + πΌπ + πππ
π€βπππ
2. State the hypotheses
By the treatment effects Treatment means
π»0: πΌ1 = πΌ2 = β― = πΌπΎ = 0 π1 = π2 = β― = ππ = π
π»π΄: some πΌπ β 0 some ππ β π
3. Rejection region /Criteria
Reject π»0 , the null hypothesis if πΉπ β₯ πΉπ = πΉβ, πβ1, πβπ
Where;
β ππ π‘βπ πππ£ππ ππ π πππππππππππ
π β 1 = πππππππ ππ πππππππ πππ ππππ‘ππ ππ πππ‘ππππ π‘
π β π = πππππππ ππ πππππππ πππ π‘βπ πππππ π‘πππ
4. Computations
We compute
(1) The total of the values of individual items in the samples, π = β β π¦ππ
π
π=1
π
π=1
where i = 1 , 2, . . . n and j = 1, 2, . . . k
(2) Correction factor =
π2
π
, where π = β ππ
(3) πππ = β β π¦ππ
2
π
π β
π2
π
(4) πππ΅ =
1
π
β ππ
2
β
π2
π
(5) πππΈ = πππ β πππ΅
(6) Draw ANOVA Table to compute the F-statistic
2. 5. Compare πΉπ with πΉπ to take Decision and Conclude.
Exercise 2
The data in the table below gives the number of hours of pain relief provided by 4 different
types of headache tablets administered to 24 people. The 24 experimental units were randomly
divided into 4 groups and each group was treated with a different brand/type. Do the different
drug types give significantly different hours of pain relief?
Typical Solution.
1. State the one-way ANOVA model (estimates only one effect)
π¦ππ = π + πΌπ + πππ
π€βπππ
π¦ππ - jth observation under the ith drug type
Β΅ - general mean
πΌπ - ith treatment effect (ith drug type effect)
βππ - error term
2. Hypothesis
Ho: Ξ±1 = Ξ±2 = Ξ±3 = Ξ±4 = 0 OR Β΅1 = Β΅2 = Β΅3 = Β΅4 = Β΅(say)
HA: Some Ξ±i β 0 Some Β΅io β Β΅
Source of variation Sum of
square
Degrees of
freedom
Mean Squares Computed F β ratio
Between samples or
categories
SSB k - 1
MSB =
1
ο
k
SSB
= SB
2
Fk-1, n-k = 2
2
W
B
S
S
Within samples or
categories
SSW N - k
MSW =
k
n
SSW
ο
=SW
2
Compare with FΞ±, (k-1), (n-k)
from F tables
Total SST N- 1
Brands
1 2 3 4
12.2 4.9 8.0 4.6
9.5 10.6 12.1 6.1
11.6 7.0 5.7 5.0
13.0 8.3 8.6 3.8
10.1 5.5 7.2 8.2
9.6 11.7 12.4 7.7
3. 3. Rejection region
Reject Ho if Fc β₯ FT (F0.05, 3, 20) = 3.10
Accept Ho if Fc < F0.05, 3, 20 = 3.10
4. Computations
We compute
(1). The total of the values of individual items in the samples, π = β β π¦ππ
π
π=1
π
π=1 =
where i = 1 , 2, . . . n and j = 1, 2, . . . k
(2). Correction factor =
π2
π
, where π = β ππ
(3). πππ = β β π¦ππ
2
π
π β
π2
π
= 1905.02 β
203 .42
24
= 1905.02 β 1723.815
= 181.205
(4) πππ΅ =
1
π
β ππ
2
β
π2
π
πππ΅ =
1
6
(662
+ 482
+ 542
+ 35.42) β
203.42
24
πππ΅ = (1804.86)β 1723.815
πππ΅ = 81.045
(5). πππΈ = πππ β πππ΅
πππΈ = 181.205 β 81.045
πππΈ = 100.160
(6) ANOVA Table
Source of variation Df SS MS F-ratio
Between treatments 3 81.045 MSB = 27.015 Fc =
27.015
5.008
= 5.1638
Within Samples Errors 20 100.160 MSE = 5.008
Total 23 181.205
5. Decision and Conclusion
Fc = 5.1638 > FT = 3.10 ; Reject Ho implying;
Drug types are significantly different in terms of the hours of pain relief they give.
Note: The sums of squares are NEVER negative.
4. Source of variation Sum of
square
Degrees of
freedom
Mean Squares Computed F β ratio
Between samples or
categories
SSB k - 1
MSB =
1
ο
k
SSB
= SB
2
Fk-1, n-k = 2
2
W
B
S
S
Within samples or
categories
SSW n - k
MSW =
k
n
SSW
ο
=SW
2
Compare with FΞ±, (k-1), (n-k)
from F tables
Total SST n - 1