GMM nonparametric correction methods for logistic regression with error contaminated covariates and partially...

Area of Science:

• Statistics
• Biostatistics
• Epidemiology

Background:

• Covariate measurement error is a common issue in logistic regression.
• Existing methods often require replicate data, which is frequently unavailable.
• This limitation hinders accurate statistical modeling and inference.

Purpose of the Study:

• To develop novel non-parametric correction methods for logistic regression with covariate measurement error.
• To address the challenge of unavailable replicate data.
• To provide robust statistical tools for analyzing data with imperfect covariate measurements.

Main Methods:

• Utilizing generalized methods of moments (GMM) with instrumental variables from a calibration subsample.
• Employing non-parametric correction techniques.
• Implementing inverse selection probability weighting and GMM on the entire sample.

Main Results:

• The proposed GMM approaches effectively correct for covariate measurement error.
• The methods demonstrate good performance in simulation studies.
• Successful application to a real-world dataset validates the approach.

Conclusions:

• The developed GMM non-parametric correction methods offer a viable solution for logistic regression with covariate measurement error.
• These methods are particularly useful when replicate data is absent.
• The study provides a valuable advancement for statistical analysis in various scientific fields.