Give an estimate for the population mean under systematic sampling and derive its variance in terms of the intraclass correlation coefficient.
10 marks · directive: Give
The core of a good answer
With N = nk and a random start, the estimator is the mean ȳ_sy of the selected systematic sample, which is unbiased for Ȳ, and V(ȳ_sy) = (σ²/n)[1 + (n − 1)ρ_w] = [(N − 1)/N](S²/n)[1 + (n − 1)ρ_w], where σ² = ∑(y − Ȳ)²/N and ρ_w is the intraclass correlation between pairs of units in the same systematic sample.
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
Set up the k possible systematic samples, show unbiasedness, write the variance as the variance of the k sample means, and expand the square so that the cross-product terms define the intraclass correlation.
Answer outline
Introduction
Population of N = nk units arranged as k systematic samples (columns) of n units each; yᵢⱼ is the j-th unit of the i-th sample, i = 1, …, k, j = 1, …, n. A random start i is chosen with probability 1/k.
Step 1: Estimator
- ȳ_sy = ȳᵢ = (1/n)∑ⱼ yᵢⱼ for the chosen start i.
- E(ȳ_sy) = (1/k)∑ᵢ ȳᵢ = (1/(nk))∑∑yᵢⱼ = Ȳ, unbiased.
Step 2: Variance
- V(ȳ_sy) = (1/k)∑ᵢ (ȳᵢ − Ȳ)².
- n²(ȳᵢ − Ȳ)² = [∑ⱼ(yᵢⱼ − Ȳ)]² = ∑ⱼ(yᵢⱼ − Ȳ)² + ∑∑_{j≠j′}(yᵢⱼ − Ȳ)(yᵢⱼ′ − Ȳ).
Step 3: Intraclass correlation
- ρ_w = E[(yᵢⱼ − Ȳ)(yᵢⱼ′ − Ȳ)]/E(yᵢⱼ − Ȳ)² = ∑ᵢ∑∑_{j≠j′}(yᵢⱼ − Ȳ)(yᵢⱼ′ − Ȳ) / [nk(n − 1)σ²],
- where σ² = (1/N)∑∑(yᵢⱼ − Ȳ)².
Step 4: Combine
- Summing over i: n²∑ᵢ(ȳᵢ − Ȳ)² = Nσ² + nk(n − 1)ρ_wσ².
- V(ȳ_sy) = (1/(kn²))[nkσ² + nk(n − 1)ρ_wσ²] = (σ²/n)[1 + (n − 1)ρ_w].
- In terms of S² = Nσ²/(N − 1): V(ȳ_sy) = [(N − 1)/N](S²/n)[1 + (n − 1)ρ_w].
Conclusion
ȳ_sy is unbiased for Ȳ with variance (σ²/n)[1 + (n − 1)ρ_w]. The variance depends on how alike the units within a systematic sample are: positive ρ_w inflates it, negative ρ_w reduces it.
Examples and data to cite
- The factor 1 + (n − 1)ρ_w is the same design effect that appears in cluster sampling, since a systematic sample is a single cluster of n units.
Facts to remember
- Systematic sample mean is unbiased when N = nk.
- V(ȳ_sy) = (σ²/n)[1 + (n − 1)ρ_w] (verified by brute force on test populations in Python).
- −1/(n − 1) ≤ ρ_w ≤ 1.
- σ² = (N − 1)S²/N.