Obtain the likelihood ratio test (LRT) for testing H₀ : θ ≤ θ₀ against H₁ : θ > θ₀, using a random sample of size n taken from P(θ).
10 marks · directive: Obtain
The core of a good answer
With x̄ the sample mean, the likelihood ratio is Λ = 1 if x̄ ≤ θ₀ and Λ = (θ₀/x̄)^{nx̄} e^{n(x̄ − θ₀)} if x̄ > θ₀; Λ decreases in x̄ beyond θ₀, so the LRT rejects H₀ when T = ∑Xᵢ > k, with k chosen from the Poisson(nθ₀) distribution so that P(T > k) ≤ α (randomising at k for exact size).
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
The method is: write the Poisson likelihood, maximise it over the whole parameter space and over H₀ (a constrained maximum, because H₀ is one-sided), form the ratio, and turn 'Λ small' into a condition on the sufficient statistic ∑Xᵢ. The size condition must be stated using the distribution of ∑Xᵢ at θ = θ₀.
Answer outline
Introduction
P(θ) is the Poisson distribution: f(x; θ) = e^{−θ}θˣ/x!, x = 0, 1, 2, …; Θ = (0, ∞), Θ₀ = (0, θ₀].
Step 1: Likelihood
- L(θ) = e^{−nθ} θ^{∑xᵢ} / ∏xᵢ!; log L = −nθ + nx̄ log θ − ∑ log xᵢ!.
- Unrestricted MLE: θ̂ = x̄ (log L increases for θ < x̄ and decreases after).
Step 2: Maximum under H₀
- If x̄ ≤ θ₀, the maximum over (0, θ₀] is at x̄; if x̄ > θ₀, it is at θ₀ (log L is increasing on (0, x̄)).
Step 3: Ratio
- Λ = sup_{Θ₀} L / sup_Θ L.
- x̄ ≤ θ₀: Λ = 1 (never reject).
- x̄ > θ₀: Λ = e^{−nθ₀}θ₀^{nx̄} / (e^{−nx̄}x̄^{nx̄}) = (θ₀/x̄)^{nx̄} e^{n(x̄ − θ₀)}.
Step 4: Monotonicity
- log Λ = n[x̄ log(θ₀/x̄) + x̄ − θ₀]; d(log Λ)/dx̄ = n log(θ₀/x̄) < 0 for x̄ > θ₀.
- So Λ < c ⇔ x̄ > c′ ⇔ T = ∑Xᵢ > k.
Step 5: Critical value
- T ~ Poisson(nθ); P_θ(T > k) increases in θ, so the size is attained at θ = θ₀.
- Choose the smallest k with P(T > k | nθ₀) ≤ α; to get exact size α, reject with probability γ when T = k, where P(T > k) + γP(T = k) = α.
- Large n: reject if (T − nθ₀)/√(nθ₀) > z_α.
Conclusion
The LRT rejects H₀ for large values of ∑Xᵢ. Since the Poisson family has monotone likelihood ratio in ∑Xᵢ, this test coincides with the uniformly most powerful test of the one-sided hypothesis.
Examples and data to cite
- Example: n = 10, θ₀ = 1, α = 0.05: T ~ Poisson(10) under θ₀, and P(T ≥ 16) ≈ 0.049, so reject when ∑Xᵢ ≥ 16 (value from the Poisson(10) distribution).
Facts to remember
- Poisson MLE: θ̂ = x̄.
- Likelihood ratio: Λ = sup_{Θ₀} L(θ) / sup_{Θ} L(θ); reject H₀ for small Λ.
- ∑Xᵢ ~ Poisson(nθ) for a Poisson sample.
- Karlin-Rubin: in a family with monotone likelihood ratio in T, 'reject for large T' is UMP for one-sided hypotheses.