An Example of an Improvable Rao–Blackwell Improvement, Inefficient Maximum Likelihood Estimator, and Unbiased Generalized Bayes Estimator

Abstract

The Rao–Blackwell theorem offers a procedure for converting a crude unbiased estimator of a parameter θ into a “better” one, in fact unique and optimal if the improvement is based on a minimal sufficient statistic that is complete. In contrast, behind every minimal sufficient statistic that is not complete, there is an improvable Rao–Blackwell improvement. This is illustrated via a simple example based on the uniform distribution, in which a rather natural Rao–Blackwell improvement is uniformly improvable. Furthermore, in this example the maximum likelihood estimator is inefficient, and an unbiased generalized Bayes estimator performs exceptionally well. Counterexamples of this sort can be useful didactic tools for explaining the true nature of a methodology and possible consequences when some of the assumptions are violated.[Received December 2014. Revised September 2015.]

Deep Neural Networks Guided Ensemble Learning for Point Estimation in Finite Samples

“…We use U nif to denote the Uniform distribution, and consider the scale-uniform distribution U nif [1−k]θ, [1+k]θ with the parameter of interest θ and a known design parameter k ∈ (0, 1) (Galili and Meilijson, 2016). This type of distribution has wide applications, for example the product inventory management in economics (Wanke, 2008) and the inverse transform sampling (Vogel, 2002).…”

“…Since the support Ω x is not the same for all θ ∈ Θ with Θ as an open interval in R, this distribution family does not satisfy the usual differentiability assumptions leading to the Cramér-Rao bound and efficiency of maximum likelihood estimators (MLEs; Lehmann and Casella (2006), Galili and Meilijson (2016)).…”

“…)which is the unbiased corrected version of the MLE θ M LE = x (n) /(k + 1)(Galili and Meilijson, 2016), Utilizing our proposed method, we ensemble T 1 = θ RB and T 2 = θ M in (2) to construct a better estimator U θ RB , θ M with a smaller variance. Suppose we are interested in θ ∈ Θ = (0.2, 10) as an open interval in R with finite data size n = 2 or 10.…”

Fiducial Inference and Decision Theory

“…where θ > 0 and the generator u = (u 1 , u 2 ) is given by the largest and smallest observation from a random sample of size n from the uniform distribution on the interval (1−k, 1+k). The corresponding statistical model was recently discussed by Mandel (2020) and before that by Galili and Meilijson (2016). The existence of optimal estimators was left open by their analysis.…”

Rao–Blackwellisation in the Markov Chain Monte Carlo Era

“…The improved estimator is indeed an estimator due to the fact that T is sufficient, and hence, the conditional expectation does not depend on the unknown parameter θ . The existence of a non‐trivial sufficient statistic obviously restricts the applicability of this theorem in mathematical statistics to exponential families (Galili & Meilijson, 2016; Lehmann & Casella, 1998) and varying support extensions.…”