Maximum Likelihood Estimates of Poisson Multilinear Models

スポンサーリンク

概要

• 論文の詳細を見る

• In the Poisson Regressions, which belong to a class of the Generalized Linear Model, all independent variables, except a scale (offset) variable, often consist of category variables. The typical cases are two-way or multi-way contingency tables, in which each count value of a cell follows the Poisson law with a mean determined by its category position in the table. This paper treats, specifically, the algorithms for obtaining the maximum likelihood estimates (MLE) of such all category Poisson Regression cases, which can be expressed generally as multilinearforms with respect to unknown parameters. For finding the MLEs of the Poisson multilinear structures, a simple and globally stable Gauss-Seidel algorithm exists. The algorithm, based on partial MLE equations, is sufficiently fast in smallsample cases. When it is preferable to accelerate the steady but not fast Gauss-Seidel algorithm, a highly efficient algorithm, which utilizes Hessian information of likelihood functions without matrix inversions, is derived and examined on some small examples. This acceleration algorithm can be interpreted as an advance version of the Successive Over-Relaxation (SOR) method. Hence, the acceleration principle developed in this paper has applicability to a wider class of problems such as optimizations or solving systems of equations, in which the Gauss-Seidel, the SOR, or the other methods are currently used. The simplicity and efficiency of the algorithms enable to introduce the Poisson multilinear models into automatic and problem-oriented applications, typically implemented in relatively slow-speed environments such as on-Web or spreadsheet platforms.

• 上智大学の論文
• 2006-03-01

著者

• 大西 博

  上智大学経済学部経済学科

関連論文

• Maximum Likelihood Estimates of Poisson Multilinear Models
• 連続型構成水準データの次元縮小観察における全体規模と構成バランスの分離バイプロット手法
• 一般均衡型地域間I/Oモデルと制約付き均衡問題の解法
• Lotus1-2-3とコンピュ-タ利用教育 (経済学の新しいツ-ル)
• 日本におけるマクロ合理的期待仮説の統計的検証
• 最小二乗推定量の良さについて:モデル推定における標本の特殊性と非確率誤差 (伊藤長正教授還暦記念号)
• 同時方程式システムの係数推定におけるバイアス消去の方法について
• 非線型回帰推定量の小標本特性

スポンサーリンク