Two Way/Two Factor Classification

Two Way Fixed Effects Study

Solution: ANOVA TABLE
Factor A: _____- fixed Source of Variation SS df MS
Treatments: Factor A SSA a-1 MSA
Factor B: _____ - fixed Factor B SSB b-1 MSB
Treatments: A×B Interactions SSAB (a-1)(b-1) MSAB
Treatment Combination: ab Error SSE ab(n-1) MSE
Experimental/Sampling Units: abn Total SST abn-1

Data Type: Step 1: Check for normality

Respond Variable: If distribution is normal -> Step 2
Scale: If distribution is not normal -> Nonpara Statistic
Parameter:
Estimator/Sample Statistic: Step 2: Test for equal variances
If variances are equal -> Step 3
Summary Statistics: If variances are not equal-> Nonpara Statistic
SST=
SSA= Test of Homogeneity of Variance: Bartlett Test
SSB= A hypotheses test for equal variance:
SSE= Hypotheses:
H0 : σ21 = σ22 = … = σ2a = σ2
Study Objective: (population variances are equal)
vs
To study the effects of Factor A and Factor B simultaneously on
H1 : at least one σ i differ from another : σ2i ’
2
(Respond Variable).
Test Statistic:
Two-way Fixed Efects Model:

Model: Y i j k = µ+ α i +β j +(αβ) i j +ε i j k Rejection region:

i=1,2,…,a, j=1,2,…,b, k=1,2,…,n
H0 if 𝜒2 ≱ 𝜒2α , a-1
Assumptions:
Conclusion:
1. ∑𝑎𝑖=1 𝛼𝑖 , fixed Factor A main effects
H0 is not rejected at α=___. The a population have equal
2. , fixed Factor B main effects
variance.
3. , fixed interaction A×B effects A parametric test is appropriate.
4. εijk are N=abn independent random variables and εijk ̴
N(0, σ2e)
Hence Y i j k ̴ N(µ+ α i +β i j +(αβ) i j , σ 2 e )
Step 3: Test for Interaction Effects
Hypotheses:
H 0 : (αβ) 11 = (αβ) 12 = … = (αβ) ab = 0
vs H 1 : at least one (αβ) ij ≠ 0
(no A×B Interactions effects)

Test Statistic:
𝑀𝑆𝐴𝐵
F ab = 𝑀𝑆𝐸

Rejection Region:
If F ab ≱ F α , (a-1)(b-1), ab(n-1) -> without interaction

Conclusion:
H0 is not rejected at α=___. There is no interaction effect between Factor A and Factor B.
The difference in (respond variable) does not depend on Factor B.
(No Interaction, Follow Up Analysis on Factor Main effects) Do Step 4

If F ab ≥ F α , (a-1)(b-1), ab(n-1) -> with interaction

Conclusion:
H0 is rejected at α=___. There is interaction effect between Factor A and Factor B. The difference in (respond variable) depend on
Factor B.

(Interaction, Follow up Analysis on Factor Specific Effects) Do Step 5

Step 4: Without Interaction-Test of Factor Main Effects

Step 4a: Test of Factor A Main Effects Step 4b: Test of Factor B Main Effects
Profile Plot:(Graphical Display) Profile Plot:(Graphical Display)
The plot shows that the line … The plot shows that the line …

Hypothesis: Hypothesis:
H0 : α1 = … = αa = 0 vs H1 : at least one αi ≠ 0 H0 : β1 = … = βa = 0 vs H1 : at least one βi ≠ 0
Test Statistic: Test Statistic:
𝑀𝑆𝐴 𝑀𝑆𝐵
Fa = 𝑀𝑆𝐸 Fb = 𝑀𝑆𝐸
Rejection Region: Rejection Region:
If F a ≱ F α , (a-1), ab(n-1) -> H0 : α1 = … = αa = 0 If F b ≱ F a, (b-1), ab(n-1) -> H0 : β1 = … = βa = 0
(no follow up analysis) (no follow up analysis)
Conclusion: Conclusion:
H0 is not rejected at α=___. There is no difference in mean H0 is not rejected at α=___. There is no difference in mean
(Respond Variable) among Factor A. (Respond Variable) among Factor B.

Rejection Region: Rejection Region:

If F a ≥ F α , (a-1), ab(n-1) ->Reject H0 : α1 = … = αa = 0 If F b ≥ F a, (b-1), ab(n-1) -> Reject H0 : β1 = … = βa = 0
(follow up analysis) (follow up analysis)
Conclusion: Conclusion:
H0 is rejected at α=___. There is difference in mean (Respond H0 is rejected at α=___. There is difference in mean (Respond
Variable) among Factor A. Variable) among Factor B.

Follow Up Analysis: Follow Up Analysis:

Pairwise comparison: Pairwise comparison:
H 0 : µ i . = µ i ’ . vs H 1 : µ i . ≠ µ i ’ . H 0 : µ . j = µ . j ’ vs H 1 : µ . j ≠ µ . j ’
For all a*= ( 𝑎2) pairs of Factor A main effects. For all b*= ( 𝑏2) pairs of Factor B main effects.

Comparison Procedure Comparison Procedure

1. Fisher LSD Procedure 1. Fisher LSD Procedure

2. Duncan Multiple Range Test 2. Duncan Multiple Range Test

Step 5: with interaction-Test of Factor Specific Effects

Step 5a: Factor A Specific Effects Step 5a: Factor B Specific Effects
Profile Plot Profile Plot
… …
Pairwise comparison: Pairwise comparison:
Hypothesis: Hypothesis:
H 0 : µ i j = µ i ’ j vs H 1 : µ i j ≠ µ i ’ j H 0 : µ i j = µ i j ’ vs H 1 : µ i j ≠ µ i j ’
For all ba*= b( 𝑎2) pairs of Factor A main effects. For all ab*=a( 𝑎2) pairs of Factor B main effects.

Comparison Procedure: Comparison Procedure:

1. Bonferroni-t Procedure 1. Bonferroni-t Procedure

2. Duncan Multiple Range Test 2. Duncan Multiple Range Test

Bonferroni-t Table
Factor A/B Factor B/A (Ȳ ij.- Ȳ i’j. )/ (Ȳ ij.- Ȳ ij’. ) Bonferroni 100%(1-α) C.I. Conclusion
(-,+),(-,-),(+,-),(+,+) S/NS

Duncan Table
p
(r α , p , f )
𝑀𝑆𝐸
√
𝑛
Rp

A Graphical Representation of Factor A/B Specific Effects

Two-Way Random Effects Study Two Way Mixed Effects Study

Solution: Solution:
Factor A: _____- random Factor A: _____- fixed
Treatments: Treatments:
Factor B: _____ - random Factor B: _____ - random
Treatments: Treatments:
Treatment Combination: ab Treatment Combination: ab
Experimental Units: abn Experimental Units: abn

Data Type: Data Type:

Respond Variable: Respond Variable:
Scale: Scale:
Parameter: Parameter:
Estimator/Sample Statistic: Estimator/Sample Statistic:

Summary Statistics: Summary Statistics:

SST= SST=
SSA= SSA=
SSB= SSB=
SSE= SSE=

Study Objective: Study Objective:

To study the effects of Factor A and Factor B simultaneously on To study the effects of Factor A and Factor B simultaneously on
(Respond Variable). (Respond Variable).

Two-way Random Efects Model: Two-way Mixed Efects Model:

Model: Yijk = µ+ αi+βj+(αβ)ij+εijk Model: Yijk = µ+ αi+βj+(αβ)ij+εijk

i=1,2,…,a, j=1,2,…,b, k=1,2,…,n i=1,2,…,a, j=1,2,…,b, k=1,2,…,n

Assumptions: Assumptions:
1. α1, α2, …, αa is a (RS)a from a N(0, σ2a) 1. , fixed Factor A effects
2. β 1, β2, …, β b is a (RS)b from a N(0, σ2b) 2. β 1, β2, …, β b is a (RS)b from a N(0, σ2b)
3. (αβ) 1 1 , , (αβ) 1 2 , …, (αβ) a b is a (RS)ab
3. (αβ) 1 1 , , (αβ) 1 2 , …, (αβ) a b is a (RS)ab
from a a N(0, σ2ab)
from a a N(0, σ2ab)
4. ε i j k are N=abn independent random variables and ε i j k ̴
4. ε i j k are N=abn independent random variables and ε i j k ̴
N(0, σ2e)
5. all a+b+ab+abn random variables are independent N(0, σ2e)
Hence Y i j k ̴ N(µ, σ2a, σ2b, σ2ab, σ2e) 5. all a+b+ab+abn random variables are independent

Hence Y i j k ̴ N(µ, σ2a, σ2b, σ2ab, σ2e)

ANOVA TABLE ANOVA TABLE

Source of Variation SS df MS Source of Variation SS df MS
Factor A SSA a-1 MSA Factor A SSA a-1 MSA
Factor B SSB b-1 MSB Factor B SSB b-1 MSB
A×B Interactions SSAB (a-1)(b-1) MSAB A×B Interactions SSAB (a-1)(b-1) MSAB
Error SSE ab(n-1) MSE Error SSE ab(n-1) MSE
Total SST abn-1 Total SST abn-1
.

Step 1: Check for normality Step 1: Check for normality

If distribution is normal -> Step 2 If distribution is normal -> Step 2
If distribution is not normal -> Nonpara Statistic If distribution is not normal -> Nonpara Statistic

Step 2: Test for equal variances Step 2: Test for equal variances
If variances are equal -> Step 3 If variances are equal -> Step 3
If variances are not equal-> Nonpara Statistic If variances are not equal-> Nonpara Statistic
Step 3: Test of Factor A random effects Step 3: Test of Interaction Effects
Hypotheses: Hypotheses:
H 0 : σ 2 a = 0 vs H 1 : σ 2 a > 0 H 0 : σ 2 a b = 0 vs H 1 : σ 2 a b > 0
(Factor A do not vary) (No A×B interaction effect)
Test Statistic: Test Statistic:
𝑀𝑆𝐴 𝑀𝑆𝐴𝐵
Fa = 𝑀𝑆𝐴𝐵 Fab = 𝑀𝑆𝐸
Rejection Region: Rejection Region:
If F a ≱ F α , (a-1), (a-1)(b-1) -> H0 : σ a = 0
2
If F ab ≱ F α , (a-1)(b-1), ab(n-1) -> H0 : σ2ab = 0
Conclusion: Conclusion:
H0 is not rejected at α=___. There is no variation among Factor H0 is not rejected at α=___. There is no interaction between Factor
A. A and factor B.
(without interaction)
Rejection Region: Rejection Region:
If F a ≥ F α , (a-1), (a-1)(b-1) -> Reject H0 : σ a = 0
2
If F ab ≥ F α , (a-1)(b-1), ab(n-1) -> Reject H0 : σ2ab = 0
(Factor A vary)
Conclusion: Conclusion:
H0 is rejected at α=___. There are variation among Factor A. H0 is rejected at α=___. There is interaction between Factor A
and Factor B.

Step 4: Test of Factor B random effects Step 4: Test of Factor B random effects
Hypotheses: Hypotheses:
H0 : σ2b = 0 vs H1 : σ2b> 0 H0 : σ2b = 0 vs H1 : σ2b> 0
(Factor B do not vary) (Factor B do not vary)
Test Statistic: Test Statistic:
𝑀𝑆𝐵 𝑀𝑆𝐵
Fb = 𝑀𝑆𝐴𝐵 Fb = 𝑀𝑆𝐸
Rejection Region: Rejection Region:
If F b ≱ F α , (b-1), (a-1)(b-1) -> H0 : σ2b = 0 If F b ≱ F α , (b-1), ab(n-1) -> H0 : σ2b = 0
Conclusion: Conclusion:
H0 is not rejected at α=___. There is no variation among Factor H0 is not rejected at α=___. There is no variation among Factor
B. B.
Rejection Region: Rejection Region:
If F b ≥ F α , (b-1), (a-1)(b-1) -> Reject H0 : σ2b = 0 If F b ≥ F α , (b-1), ab(n-1) -> Reject H0 : σ2b = 0
(Factor B vary) (Factor B vary)
Conclusion: Conclusion:
H0 is rejected at α=___. There is variation among Factor B. H0 is rejected at α=___. There is variation among Factor B.
Step 5: Test of factor A×B interaction effects Step 5: Test of Factor A Fixed Effects
Hypotheses: Hypotheses:
H0 : σ2ab = 0 vs H1 : σ2ab> 0 H0 : α1 = … = αa = 0 vs H1 : at least one αi ≠ 0
(No A×B interaction effect) (Factor A effects do not differ)
Test Statistic: Test Statistic:
𝑀𝑆𝐴𝐵 𝑀𝑆𝐴
Fab = Fa = 𝑀𝑆𝐴𝐵
𝑀𝑆𝐸
Rejection Region: Rejection Region:
If F ab ≱ F α , (a-1)(b-1), ab(n-1) -> H0 : σ2ab = 0 If F a ≱ F α , (a-1), (a-1)(b-1) -> H0 : α1 = … = αa = 0
Conclusion: Conclusion:
H0 is not rejected at α=___. There is no Factor A and Factor B H0 is not rejected at α=___. There is no difference in mean
interaction effect. Factor A does not depend on Factor B and Factor (Respond Variable) among Factor A. No Follow-up Analysis.
B variations does not depend on Factor B. (without interaction)
Rejection Region: Rejection Region:
If F ab ≥ F α , (a-1)(b-1), ab(n-1) -> Reject H0 : σ2ab = 0 If F a ≥ F α , (a-1), (a-1)(b-1) -> Reject H0 : α1 = … = αa = 0
Conclusion: Conclusion:
H0 is rejected at α=___. There is Factor A and Factor B H0 is rejected at α=___. There is a difference in mean (Respond
interaction effect. Factor A depend on Factor B and Factor B Variable) among Factor A. (Follow-up Analysis)
variations depend on Factor B. (with interaction)

Step 6: Follow-up Analysis

Variance Components:
Factor A (MSA-MSAB)/bn
Factor B (MSB-MSAB)/an
A×B Interaction (MSAB-MSE)/n
Error MSE
Contribution Of Variance Component to Total Variance
Total Variance
σ y = σ2a + σ2b + σ2ab + σ2e
2

Proportion of Var(Yijk) due to

Factor A σ2a / σ2y
Factor B σ2b / σ2y
A×B Interaction σ2ab / σ2y
Error σ2e / σ2y
Conclusion: DIY (Do It Yourself)