Variable selection problem and evaluation of measuring uncertainty in small area estimation

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K13667
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 07030:Economic statistics-related
• Research Institution: Chiba University
• Principal Investigator: Kawakubo Yuki 千葉大学, 大学院社会科学研究院, 准教授 (80771881)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: 小地域推定 / 変数選択 / 変量効果モデル

Outline of Final Research Achievements

In this research project, we addressed the problem of variable selection in mixed effects models used in small area estimation. Small area estimation is a statistical method that attempts to improve the accuracy of estimation at the small area level, where the sample size is small in a sample survey, by using a statistical model called mixed effects models. We worked on the development of variable selection criteria for selecting combinations of auxiliary variables to be included in the mixed effects models.
Within the general framework of variable selection in small area estimation, we addressed several problems in each issue, which were published in international peer-reviewed journals and reported at academic conferences.

Free Research Field

統計学

Academic Significance and Societal Importance of the Research Achievements

いくつかの本研究成果の共通した着眼点は，変数選択法と，予測量の不確実性の評価との関連である。小地域推定においては，各小地域の推定対象の値を言い当てること（点予測）だけでなく，その不確実性を見積もることを重視している。不確実性の評価方法として，平均二乗予測誤差（MSPE）と呼ばれる指標が一般的であるが，既存手法のほとんどは，候補モデルが真であるという仮定のもとでMSPEを評価していた。しかし本研究においては，この仮定をおかずに，変数選択の不確実性を明示的に考慮したMSPEの評価を行った。従来手法はMSPEを過小評価している可能性が高いことから，本研究成果は学術的にも社会的にも意義が大きい。