Let Y₁, Y₂, ⋯, Yₙ be n independent observations from a population with mean μ and variance σ². Obtain the best linear unbiased estimator of μ and an unbiased quadratic estimator for σ².
10 marks · directive: Obtain
The core of a good answer
Among linear estimators ∑aᵢYᵢ with ∑aᵢ = 1 (unbiasedness), the variance σ²∑aᵢ² is least when every aᵢ = 1/n, so Ȳ is the BLUE of μ with variance σ²/n; the quadratic form Y′AY with A = (I − J/n)/(n − 1) satisfies E(Y′AY) = σ², giving the unbiased quadratic estimator S² = ∑(Yᵢ − Ȳ)²/(n − 1).
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
Two short derivations. For the BLUE, impose unbiasedness on a linear form and minimise variance subject to that constraint (Lagrange multiplier or Cauchy-Schwarz). For σ², use E(Y′AY) = σ² tr(A) + μ² 1′A1 and choose A to make this σ² for every μ.
Answer outline
Introduction
Write Y = (Y₁, …, Yₙ)′, E(Y) = μ1, D(Y) = σ²I, where 1 is the vector of ones and J = 11′.
Step 1: Linear unbiased estimators
- T = a′Y = ∑aᵢYᵢ; E(T) = μ∑aᵢ = μ for all μ ⇔ ∑aᵢ = 1.
- V(T) = σ²∑aᵢ².
Step 2: Minimise
- Minimise ∑aᵢ² subject to ∑aᵢ = 1: φ = ∑aᵢ² − 2λ(∑aᵢ − 1), ∂φ/∂aᵢ = 0 ⇒ aᵢ = λ, so aᵢ = 1/n.
- Or Cauchy-Schwarz: 1 = (∑aᵢ)² ≤ n∑aᵢ², so ∑aᵢ² ≥ 1/n, equality at aᵢ = 1/n.
- BLUE: Ȳ, V(Ȳ) = σ²/n.
Step 3: Expectation of a quadratic form
- E(Y′AY) = tr[A D(Y)] + E(Y)′A E(Y) = σ² tr(A) + μ² 1′A1.
Step 4: Choose A
- Need 1′A1 = 0 and tr(A) = 1.
- A = (I − J/n)/(n − 1): (I − J/n)1 = 0, and tr(I − J/n) = n − 1.
- Y′(I − J/n)Y = ∑(Yᵢ − Ȳ)², so S² = ∑(Yᵢ − Ȳ)²/(n − 1) and E(S²) = σ².
Conclusion
Ȳ is the BLUE of μ and S² = ∑(Yᵢ − Ȳ)²/(n − 1) is an unbiased quadratic estimator of σ². No normality is needed for either result.
Examples and data to cite
- This is the simplest case of the Gauss-Markov theorem: Y = 1μ + e with X = 1.
Facts to remember
- E(a′Y) = a′E(Y); V(a′Y) = a′D(Y)a.
- E(Y′AY) = tr(AΣ) + μ′Aμ for E(Y) = μ, D(Y) = Σ.
- ∑(Yᵢ − Ȳ)² = Y′(I − J/n)Y.
- Cauchy-Schwarz inequality: (∑aᵢ)² ≤ n∑aᵢ².