T-type Corrected-Loss Estimation for Error-in-Variable Model.

Area of Science:

• Statistics
• Econometrics
• Biostatistics

Background:

• Linear models are widely used but sensitive to errors in covariate measurements.
• Measurement errors, especially those following a Laplace distribution, can significantly bias estimation results.
• Outliers in practical data can further compromise the robustness of standard estimation techniques.

Purpose of the Study:

• To propose a novel t-type corrected-loss estimation method for linear models with Laplace-distributed measurement errors.
• To assess the asymptotic normality and robustness of the proposed estimator.
• To demonstrate the practical utility and performance of the method using simulations and a real-world application.

Main Methods:

• Development of a t-type corrected-loss function to handle Laplace-distributed measurement errors.
• Theoretical analysis to establish the asymptotic normality of the proposed estimator.
• Simulation studies to evaluate the estimator's performance against vertical outliers.
• Application to Six-Minute Walk test data for practical validation.

Main Results:

• The proposed t-type corrected-loss estimator is shown to be asymptotically normal.
• Simulation results indicate strong resistance to vertical outliers, preserving estimation accuracy.
• The method demonstrates good performance in the real-world Six-Minute Walk test application.

Conclusions:

• The proposed t-type corrected-loss estimation offers a robust and reliable approach for linear models with Laplace-distributed covariate errors.
• The estimator's resilience to outliers makes it suitable for practical data analysis where such deviations are common.
• The successful application to the Six-Minute Walk test highlights its potential in various scientific fields.