Answers are grounded in this paper and cite their sources.
Let f(x: θ) = a(x) exp\θ₁u₁(x) + θ₂u₂(x) + c(θ)\, θ = (θ₁, θ₂) ∈ ⊖ ⊂ R², be a density with respect to the Lebesgue measure on the real line which characterizes a two-parameter exponential family of distributions. Let (θ₁, η₂) be the mixed parameters, where η₂ = E₂(X)\. Assume that θ₂ can be represented as θ₂ = -θ₁ φ'(η₂) where φ'(η₂) = dφ(η₂)/dη₂ for some function φ(η₂). Let (X₁, ⋯, Xₙ) be independent random variables having a common density f(x: θ) and set Tᵢ = ∑ⁿⱼ₌₁ uᵢ(Xⱼ), i = 1, 2. It is shown that if u₂(x) is a 1-1 function then the random variables T₂ and Zₙ = T₁ - nφ(T₂/n) are independent and that the statistic Zₙ is ancillary for θ₂ in the presence of θ₁ (i.e. the density of Zₙ depends on θ₁ only). Furthermore, the density of Zₙ belongs to the one-parameter exponential family with natural parameter θ₁. These results enable us to construct uniformly most powerful unbiased (UMPU) tests for various hypotheses concerning the parameter θ₁ which are based on the statistic Zₙ.
No takes yet. Share an insight, caveat, or question.
Bar‐Lev et al. (1982) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: