This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Lehmann–Scheffé theorem | |
|---|---|
| Name | Lehmann–Scheffé theorem |
| Field | Statistics |
| Introduced | 1950s |
| Authors | Erich Leo Lehmann; Henry Scheffé |
Lehmann–Scheffé theorem The Lehmann–Scheffé theorem is a fundamental result in mathematical statistics that characterizes uniformly minimum-variance unbiased estimators within families possessing complete sufficient statistics. It provides a route from the Rao–Blackwell theorem to optimal unbiased estimation and underpins many classical inferential procedures in the works of Fisher, Neyman, and Pearson.
Let X denote a sample from a parametric family indexed by θ. Suppose T(X) is a complete and sufficient statistic for θ, and let δ(X) be any unbiased estimator of a parametric function g(θ). The Lehmann–Scheffé theorem asserts that the conditional expectation φ(T) = E[δ(X) | T(X)] is the unique uniformly minimum-variance unbiased estimator (UMVUE) of g(θ). This conclusion combines ideas related to sufficiency from Ronald Fisher, completeness concepts used by Jerzy Neyman, and variance reduction from the C. R. Rao inequality; it refines results of Henry Scheffé and Erich Leo Lehmann.
Understanding the theorem requires familiarity with several foundational contributions and persons: the sufficiency principle of Ronald Fisher, the concept of unbiasedness promoted by Jerzy Neyman and Egon Pearson, and the Rao–Blackwell improvement associated with C. R. Rao and David Blackwell. Technical prerequisites include measure-theoretic probability influenced by Andrey Kolmogorov, the theory of exponential families studied by Harold Hotelling and S. N. Roy, and completeness properties investigated in the context of linear models by R. A. Fisher and George Box. Key examples arise from the exponential family framework tied to work by Wilks, Samuel S. and Ronald A. Fisher and from classical hypothesis testing contexts developed by Jerzy Neyman and Egon Pearson.
The proof proceeds by combining Rao–Blackwellization with a completeness argument. Starting with any unbiased estimator δ(X), apply the Rao–Blackwell procedure attributed to David Blackwell and C. R. Rao to form φ(T)=E[δ|T]; φ(T) is unbiased and has variance no greater than that of δ(X). Completeness of T, a notion refined in studies by Henry Scheffé and Erich Leo Lehmann, implies uniqueness: if ψ(T) is another unbiased estimator, then ψ(T)−φ(T) has zero expectation for all θ and therefore vanishes almost surely, yielding ψ(T)=φ(T). The structure of the argument mirrors techniques used by Andrey Kolmogorov in ergodic decompositions and adopts conditioning strategies seen in the work of Paul Lévy and Doob, Joseph L..
Classic applications appear in parametric families where complete sufficient statistics are explicit. For the normal model with unknown mean and known variance studied by Carl Friedrich Gauss and Pierre-Simon Laplace, the sample mean is the UMVUE of the population mean; the theorem confirms results used by Gauss in least squares. For exponential families articulated by Harold Hotelling and Samuel S. Wilks, natural sufficient statistics yield UMVUEs for canonical parameters and functions, echoing results in the literature of Jerzy Neyman and Egon Pearson. In the Poisson model associated with Siméon Denis Poisson, the sample total is complete and sufficient, so functions such as the probability of zero counts obtain UMVUEs via Lehmann–Scheffé techniques. The theorem also supports estimation in linear models developed by Francis Galton and Karl Pearson, and it underlies unbiased risk estimation ideas related to Charles Stein.
Several extensions and related theorems interact with Lehmann–Scheffé. The Basu theorem connecting ancillarity and independence draws on completeness themes explored by Debabrata Basu. Extensions to semiparametric and nonparametric settings relate to work by Peter J. Bickel and Lehmann collaborators. The complete class theorems in decision theory by Bertrand de Finetti and Wald, Abraham connect minimaxity, admissibility, and UMVUE properties, and modern developments in empirical Bayes methods by Bradley Efron and Herbert Robbins explore trade-offs when completeness fails. The Rao–Blackwell theorem of David Blackwell and C. R. Rao is the immediate precursor; uniqueness aspects tie to completeness notions advanced by Henry Scheffé and Erich Leo Lehmann.
The theorem bears the names of Erich Leo Lehmann and Henry Scheffé for their roles in formalizing completeness and uniqueness arguments in mid-20th-century statistical theory. Its conceptual roots trace to Ronald Fisher’s sufficiency principle and the Rao–Blackwell improvement from David Blackwell and C. R. Rao. The broader development of unbiased estimation and decision-theoretic perspectives involved contributions by Jerzy Neyman, Egon Pearson, Abraham Wald, and contemporaries whose work at institutions such as Princeton University and University of California, Berkeley shaped modern mathematical statistics. Lehmann’s expositions and Scheffé’s methodological papers popularized the result in textbooks and courses across academic centers including Harvard University and Columbia University.
Category:Statistical theorems