Oracle estimation of parametric models under boundary constraints

Area of Science:

• Statistics
• Statistical Inference
• Computational Statistics

Background:

• Classical estimation problems often involve parameter spaces with boundaries.
• Standard asymptotic properties of estimators may fail when true parameters lie on the boundary.
• Confidence intervals assuming interior parameters can be overly conservative if parameters are on the boundary.

Purpose of the Study:

• To propose a penalized estimation method to address boundary issues in parameter estimation.
• To develop an adaptive lasso procedure for boundary adaptation in statistical inference.
• To achieve oracle inference that adjusts to whether true parameters are on or within the boundary.

Main Methods:

• A penalized estimation method is proposed.
• An adaptive lasso procedure is employed to shrink parameters towards the boundary.
• The method is demonstrated in frailty survival models and linear regression with order-restricted parameters.

Main Results:

• The adaptive lasso procedure yields inference that adapts to boundary conditions.
• When true parameters are on the boundary, inference matches that with a priori knowledge.
• When true parameters are in the interior, inference matches standard interior results.
• Simulation studies and real data analyses confirm good performance with realistic sample sizes.

Conclusions:

• The proposed penalized estimation method effectively handles boundary issues in parameter estimation.
• The adaptive lasso procedure offers advantages over standard methods, providing more accurate inference.
• The method demonstrates robust performance in practical scenarios like survival analysis and restricted regression.