2020 Fiscal Year Final Research Report
High dimensional multivariate linear mixed model and application to small area estimation
Project/Area Number |
19K23242
|
Research Category |
Grant-in-Aid for Research Activity Start-up
|
Allocation Type | Multi-year Fund |
Review Section |
0107:Economics, business administration, and related fields
|
Research Institution | Tokyo Medical and Dental University (2020) The Institute of Statistical Mathematics (2019) |
Principal Investigator |
Ito Tsubasa 東京医科歯科大学, M&Dデータ科学センター, 助教 (90849001)
|
Project Period (FY) |
2019-08-30 – 2021-03-31
|
Keywords | 小地域推定 / Fay-Herriotモデル / 高次元データ / 平均二乗誤差 / 信頼区間 / 2次補正 |
Outline of Final Research Achievements |
Multivariate small are estimation problem, especially the Fay-Herriot model in which we only obtain aggregated data for each area is considered in the setting where the ratio of the square of the dimension of the observed vectors and the number of areas converges to a constant. The convergence rates of the estimators of the parameters in the model are shown, noting that the dimensions of these parameters also increase in this setting. Small area mean vectors are predicted by the empirical best linear unbiased predictors, which can be obtained by substituting the estimators of the parameters into its bayes estimator. As its prediction risks, the mean squared error matrix and the confidence interval are constructed such that the approximation errors are of second order regarding the number of areas. It is shown that the additional terms appear in both compared with the results obtained when the dimension of the observed vectors is fixed.
|
Free Research Field |
数理統計学
|
Academic Significance and Societal Importance of the Research Achievements |
標本数が小さい地域の平均所得などを推定する際、小地域推定の手法による予測量はリスクを改善するため有用であり、相関をもった多次元データに対しては、多変量モデルから得られる予測量がよりリスクを改善する。一方で高次元モデルについての研究はない。小地域推定の手法は漸近近似に基づくが、観測値が高次元の場合は近似誤差が大きくなるため、従来の方法では誤った結果を得る恐れがある。本研究では、より高次元のデータをモデルに組み込むことでさらに予測リスクを改善でき、高次元データに対しても頑健なリスク評価を行っているため誤った結果を得る恐れが少ないため、公共機関が小地域に対してより正しい政策決定をする際の助けとなる。
|