Estimating smooth distribution function in the presence of heteroscedastic measurement errors

Area of Science:

• Biostatistics
• Biomedical Engineering
• Statistical Modeling

Background:

• Measurement error is prevalent in biomedical research, posing significant challenges when errors are heteroscedastic.
• Heteroscedasticity, where error variance differs across observations, complicates the estimation of distribution functions with single data points.

Purpose of the Study:

• To develop and evaluate methods for estimating smooth distribution functions from data contaminated with heteroscedastic errors.
• To address the challenges of analyzing biomedical data with complex error structures.

Main Methods:

• Investigated a Fourier-type deconvolution method for distribution function estimation.
• Applied and analyzed a simulation extrapolation (SIMEX) method, including obtaining asymptotic pointwise confidence bands.
• Conducted simulation studies to compare the finite sample performance of both methods.

Main Results:

• Asymptotic properties of both deconvolution and SIMEX estimators were explored.
• Asymptotic pointwise confidence bands were derived for the SIMEX estimator.
• Simulation results provided insights into the finite sample performance of the proposed methods.

Conclusions:

• The study presents viable statistical methods for handling heteroscedastic measurement errors in biomedical data.
• The findings are applicable to real-world scenarios, demonstrated by an analysis of neuro-muscular electrical stimulation experiment data in medical rehabilitation.