The following data come from a two-factor random effect model :
Factor A \ Factor B | B₁ | B₂ | B₃
A₁ | 6, 8 | 10, 12 | 14, 16
A₂ | 12, 14 | 16, 18 | 20, 22
(i) Given that the experimentor suspects interaction, write the linear model stating all the assumptions for the above data.
(ii) Clearly stating the hypothesis, carry out the tests for the main effects and interactions giving the complete ANOVA table for the above data.
20 marks · directive: Write
The core of a good answer
Model y_ijk = μ + aᵢ + bⱼ + (ab)ᵢⱼ + e_ijk with independent normal random effects; the ANOVA gives SS_A = 108, SS_B = 128, SS_AB = 0, SS_E = 12 (total 248), so F_AB = 0 (no interaction), and since MS_AB = 0 the A and B tests are made against the pooled error MS 1.5: F_A = 72 > F₀.₀₅(1, 8) = 5.32 and F_B = 42.67 > F₀.₀₅(2, 8) = 4.46, so both variance components σ_a² and σ_b² are significant.
An outline written by upsc.help, not an official UPSC answer; every fact checked by a separate review. How we work
What the question asks
Part (i): the random-effects model with interaction, its distributional assumptions and the expected mean squares. Part (ii): hypotheses on variance components, the computation of the two-way ANOVA with 2 replicates per cell, and the correct denominators (in a random model the main effects are tested against MS_AB). Here MS_AB = 0 exactly, which must be handled explicitly. Marks 5 + 15.
Answer outline
Introduction
a = 2 levels of A, b = 3 levels of B, n = 2 replicates per cell, N = 12. Grand total G = 168, correction factor CF = 168²/12 = 2352.
Part (i): Model and assumptions
- y_ijk = μ + aᵢ + bⱼ + (ab)ᵢⱼ + e_ijk; i = 1, 2; j = 1, 2, 3; k = 1, 2.
- aᵢ ~ N(0, σ_a²), bⱼ ~ N(0, σ_b²), (ab)ᵢⱼ ~ N(0, σ_ab²), e_ijk ~ N(0, σ²), all mutually independent.
- Levels of A and B are random samples from larger populations of levels.
- E(MS_A) = σ² + nσ_ab² + bnσ_a²; E(MS_B) = σ² + nσ_ab² + anσ_b²; E(MS_AB) = σ² + nσ_ab²; E(MS_E) = σ².
Hypotheses
- H₀A: σ_a² = 0; H₀B: σ_b² = 0; H₀AB: σ_ab² = 0 (each against > 0).
Step 1: Totals
- Row totals A₁ = 66, A₂ = 102; column totals B₁ = 40, B₂ = 56, B₃ = 72.
- Cell totals: 14, 22, 30 / 26, 34, 42.
- ∑y² = 2600; TSS = 2600 − 2352 = 248.
Step 2: Sums of squares
- SS_A = (66² + 102²)/6 − 2352 = 108.
- SS_B = (40² + 56² + 72²)/4 − 2352 = 128.
- SS_cells = (14² + 22² + 30² + 26² + 34² + 42²)/2 − 2352 = 236; SS_AB = 236 − 108 − 128 = 0.
- SS_E = 248 − 236 = 12.
Step 3: ANOVA table
- A: SS 108, d.f. 1, MS 108.
- B: SS 128, d.f. 2, MS 64.
- AB: SS 0, d.f. 2, MS 0.
- Error: SS 12, d.f. 6, MS 2.
- Total: SS 248, d.f. 11.
Step 4: Interaction test
- F_AB = MS_AB/MS_E = 0/2 = 0 < F₀.₀₅(2, 6) = 5.14: H₀AB accepted; the cell means are exactly additive (each B step adds 4, A₂ exceeds A₁ by 6).
Step 5: Main effects
- The textbook ratios MS_A/MS_AB and MS_B/MS_AB are undefined here (division by 0).
- Since interaction is absent, pool AB with error: SS = 12, d.f. 8, MS = 1.5.
- F_A = 108/1.5 = 72 > F₀.₀₅(1, 8) = 5.32: reject H₀A.
- F_B = 64/1.5 = 42.67 > F₀.₀₅(2, 8) = 4.46: reject H₀B.
- Against unpooled MS_E the ratios are 54 and 32, also significant.
- Variance component estimates: σ̂² = 1.5 (pooled); σ̂_a² = (108 − 1.5)/6 = 17.75; σ̂_b² = (64 − 1.5)/4 = 15.63.
Conclusion
Interaction is not significant (its sum of squares is exactly zero), while both factors contribute significant variance components. All sums of squares and F ratios were computed in Python; critical values are from the F table in the paper.
Examples and data to cite
- The cell means 7, 11, 15 (A₁) and 13, 17, 21 (A₂) differ by exactly 6 in every column, which is why SS_AB = 0.
Facts to remember
- Two-way ANOVA with n replicates: SS_A = ∑Aᵢ²/(bn) − CF, SS_B = ∑Bⱼ²/(an) − CF, SS_AB = ∑T_ij²/n − CF − SS_A − SS_B.
- Random model: test A and B against MS_AB, test AB against MS_E.
- Pooling AB with error assumes σ_ab² = 0; here SS_AB = 0 exactly.
- F₀.₀₅(2, 6) = 5.14, F₀.₀₅(1, 8) = 5.32, F₀.₀₅(2, 8) = 4.46.