SIMEX and standard error estimation in semiparametric measurement error models.

Area of Science:

• Statistics
• Biostatistics
• Measurement Error Theory

Background:

• Simulation Extrapolation (SIMEX) is established for parametric and non-parametric problems.
• Existing literature lacks theoretical and applied frameworks for SIMEX in semiparametric settings.
• Measurement error is a common issue in various scientific fields, including radiation dosimetry.

Purpose of the Study:

• To develop the foundational theory and application of SIMEX for semiparametric regression models.
• To address situations with mixed parametric and non-parametric modeling of mismeasured variables.
• To provide a robust method for estimating standard errors in semiparametric problems with measurement error.

Main Methods:

• Kernel-based estimation methods are employed to extend SIMEX to semiparametric models.
• Asymptotic expansions are derived to develop standard error formulae.
• Bias properties of non-parametric estimators are analyzed.

Main Results:

• A novel method for estimating the variability of non-parametric estimators in semiparametric models is presented.
• The proposed standard error method shows dramatic improvements over first-order methods in simulations and a radiation dosimetry example.
• Standard bandwidth choices are sufficient for estimating the parametric component, eliminating the need for undersmoothing.

Conclusions:

• The developed SIMEX theory provides a powerful tool for measurement error correction in semiparametric regression.
• The new standard error estimation method enhances the reliability of statistical inference in complex models.
• The findings offer broader insights into the behavior of kernel-based methods in misspecified semiparametric models.