Derive the variance of the sample mean under simple random sample without replacement (SRSWOR).
10 marks · directive: Derive
The core of a good answer
Under SRSWOR of n from N, V(ȳ) = [(N − n)/(Nn)] S² = (1 − f)S²/n, with f = n/N and S² = ∑ᵢ₌₁ᴺ(Yᵢ − Ȳ)²/(N − 1); the factor 1 − f is the finite population correction.
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
A standard derivation. The cleanest method uses inclusion indicators: write ȳ = (1/n)∑ aᵢYᵢ where aᵢ = 1 if unit i is in the sample, find E(aᵢ), V(aᵢ) and Cov(aᵢ, aⱼ), and sum. Show that ȳ is unbiased first.
Answer outline
Introduction
Population values Y₁, …, Y_N, mean Ȳ, S² = ∑(Yᵢ − Ȳ)²/(N − 1). All C(N, n) samples are equally likely.
Step 1: Indicators
- aᵢ = 1 if unit i is selected, 0 otherwise; ∑aᵢ = n.
- P(aᵢ = 1) = n/N, so E(aᵢ) = n/N and V(aᵢ) = (n/N)(1 − n/N).
- P(aᵢ = aⱼ = 1) = n(n − 1)/[N(N − 1)], so Cov(aᵢ, aⱼ) = −(n/N)(1 − n/N)/(N − 1).
Step 2: Unbiasedness
- ȳ = (1/n)∑aᵢYᵢ; E(ȳ) = (1/n)(n/N)∑Yᵢ = Ȳ.
Step 3: Variance
- Work with deviations: ȳ − Ȳ = (1/n)∑aᵢ(Yᵢ − Ȳ) since ∑aᵢ = n.
- V(ȳ) = (1/n²)[∑(Yᵢ − Ȳ)²V(aᵢ) + ∑∑_{i≠j}(Yᵢ − Ȳ)(Yⱼ − Ȳ)Cov(aᵢ, aⱼ)].
- ∑∑_{i≠j}(Yᵢ − Ȳ)(Yⱼ − Ȳ) = −∑(Yᵢ − Ȳ)², because ∑(Yᵢ − Ȳ) = 0.
Step 4: Simplify
- V(ȳ) = (1/n²)(n/N)(1 − n/N)∑(Yᵢ − Ȳ)²[1 + 1/(N − 1)] = (1/n²)(n/N)(1 − n/N)(N/(N − 1))(N − 1)S².
- V(ȳ) = (1/n)(1 − n/N)S² = [(N − n)/(Nn)]S².
Step 5: Estimation and comparison
- s² = ∑(yᵢ − ȳ)²/(n − 1) is unbiased for S², so v(ȳ) = (1 − f)s²/n.
- SRSWR gives V(ȳ) = σ²/n = (N − 1)S²/(Nn), larger than under SRSWOR.
Conclusion
V(ȳ) = (1 − f)S²/n. When the sampling fraction f is small the correction is negligible and the formula reduces to S²/n; when n = N the variance is 0, as it should be.
Examples and data to cite
- Example: N = 1000, n = 100, S² = 25 gives V(ȳ) = 0.9 × 25/100 = 0.225.
Facts to remember
- SRSWOR inclusion probabilities: πᵢ = n/N, πᵢⱼ = n(n − 1)/[N(N − 1)].
- V(ȳ) = (1 − f)S²/n, f = n/N (finite population correction).
- s² is unbiased for S² under SRSWOR.
- Cov(aᵢ, aⱼ) is negative: selecting one unit lowers the chance of selecting another.