Testing and Estimating Shape-Constrained Nonparametric Density and Regression in the Presence of Measurement Error.

Area of Science:

• Statistics
• Nonparametric Statistics
• Econometrics

Background:

• Many applications require density functions or regression curves to meet specific shape constraints, such as monotonicity or unimodality.
• Existing nonparametric methods for shape-constrained estimation and testing assume precise covariate observation.
• Measurement errors in covariates invalidate these traditional methods, necessitating new approaches.

Purpose of the Study:

• To develop a novel statistical methodology for hypothesis testing and function estimation under shape constraints in the presence of measurement errors.
• To provide a robust framework for analyzing data where covariates are observed with inaccuracies.
• To extend the applicability of shape-constrained nonparametric methods to real-world scenarios with imperfect data.

Main Methods:

• A novel approach based on "tilting" an estimator (of density or regression mean) until it satisfies the desired shape constraint.
• Utilizing the distance of tilting as a test statistic for hypothesis testing.
• Employing bootstrap methods for calibrating the hypothesis test.
• Developing constrained curve estimators also based on the tilting principle.

Main Results:

• The proposed method is valid in the context of measurement errors, addressing a significant limitation of existing techniques.
• The tilting approach effectively handles shape constraints (e.g., monotonicity, log-concavity) even with noisy data.
• Bootstrap calibration provides a reliable way to assess the significance of the test statistic.

Conclusions:

• The developed methodology offers a powerful new tool for statistical inference under shape constraints when data is subject to measurement error.
• This work bridges the gap between theoretical nonparametric methods and practical data analysis scenarios with imperfect measurements.
• The tilting-based approach provides a unified framework for both testing and estimation under shape constraints in the presence of measurement errors.