References & Citations
Statistics > Methodology
Title: Median bias reduction of maximum likelihood estimates
(Submitted on 16 Apr 2016 (this version), latest version 2 Mar 2017 (v4))
Abstract: For regular parametric problems, we show how median centering of the maximum likelihood estimate can be achieved by a simple modification of the score equation. For a scalar parameter of interest, the estimator is second-order median unbiased and equivariant under interest respecting parameterizations. With a vector parameter of interest, componentwise equivariance and second-order median centering are obtained. The new method does not depend on the maximum likelihood estimate and is effective in preventing infinite estimates, like Firth's (1993, Biometrika) implicit method for bias reduction, although with a reduced shrinkage effect. Simulation results for continuous and discrete models, including binary regression, confirm that the method succeeds in solving the infinite estimate problem and in achieving median centering, while keeping comparable dispersion and the same approximate distribution as its main competitors.
Submission history
From: Nicola Sartori [view email][v1] Sat, 16 Apr 2016 16:02:16 GMT (21kb)
[v2] Wed, 4 May 2016 23:10:20 GMT (28kb)
[v3] Wed, 1 Mar 2017 12:08:46 GMT (62kb)
[v4] Thu, 2 Mar 2017 14:18:20 GMT (62kb)
Link back to: arXiv, form interface, contact.