Describe how Rao-Blackwell theorem enables one to find the UMVUE.
5 marks · directive: Describe
The core of a good answer
Rao-Blackwell says that conditioning any unbiased estimator T on a sufficient statistic S gives φ(S) = E(T | S), which is still unbiased and has variance no larger than T; if S is also complete, Lehmann-Scheffé makes φ(S) the unique UMVUE, so the recipe is: find a complete sufficient statistic, take any unbiased estimator, and condition on it.
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 descriptive answer for 5 marks: state the Rao-Blackwell theorem precisely, give the one-line reason for the variance reduction, and then explain why completeness (Lehmann-Scheffé) is needed to reach the UMVUE. A short worked example strengthens the answer.
Answer outline
Introduction
Let X₁, …, Xₙ have a distribution indexed by θ, and let g(θ) be the parametric function to be estimated. A UMVUE is an unbiased estimator whose variance is least among all unbiased estimators for every θ.
Statement of Rao-Blackwell
- If T is unbiased for g(θ) with finite variance and S is sufficient for θ, then φ(S) = E(T | S) is a statistic (free of θ, because S is sufficient).
- E[φ(S)] = E[E(T | S)] = E(T) = g(θ), so φ is unbiased.
- V(T) = E[V(T | S)] + V[E(T | S)] ≥ V[φ(S)], with equality only if T is already a function of S.
From improvement to UMVUE
- Rao-Blackwell alone only improves an estimator; it does not say the result is best.
- Lehmann-Scheffé: if S is complete and sufficient, any unbiased function of S is unique, so φ(S) is the UMVUE whatever T we started from.
Recipe
- 1. Find a complete sufficient statistic S (factorisation theorem; exponential family).
- 2. Find any simple unbiased estimator T of g(θ).
- 3. Compute φ(S) = E(T | S), or directly find a function of S that is unbiased.
Worked illustration
- Poisson(θ), g(θ) = P(X = 0) = e^{−θ}; T = 1 if X₁ = 0, else 0; S = ∑Xᵢ is complete sufficient.
- Given S = s, X₁ is Binomial(s, 1/n), so E(T | S = s) = (1 − 1/n)^s.
- UMVUE of e^{−θ} is (1 − 1/n)^{∑Xᵢ}.
Conclusion
Rao-Blackwellisation turns a crude unbiased estimator into a function of the sufficient statistic with smaller variance; completeness guarantees that this function is the unique UMVUE.
Examples and data to cite
- The Poisson example above, where conditioning the indicator of X₁ = 0 on ∑Xᵢ yields (1 − 1/n)^{∑Xᵢ}.
- For a normal sample, X̄ is the UMVUE of μ because it is an unbiased function of the complete sufficient statistic (∑Xᵢ, ∑Xᵢ²).
Facts to remember
- Rao-Blackwell theorem (C. R. Rao 1945, D. Blackwell 1947): V[E(T | S)] ≤ V(T).
- Law of total variance: V(T) = E[V(T | S)] + V[E(T | S)].
- Lehmann-Scheffé theorem: an unbiased estimator that is a function of a complete sufficient statistic is the unique UMVUE.
- Sufficiency is what makes E(T | S) free of θ.