MEBoost: Variable selection in the presence of measurement error

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Covariate measurement error is a common challenge in regression analysis.
• Standard variable selection methods can be unreliable when covariates are measured with error.
• Accurate variable selection is crucial for reliable model interpretation and prediction.

Purpose of the Study:

• To introduce a novel iterative algorithm, Measurement Error Boosting (MEBoost), for variable selection in regression models with covariate measurement error.
• To evaluate the performance of MEBoost compared to existing methods like Convex Conditioned Lasso and naive Lasso.
• To demonstrate the practical application of MEBoost using data from the Box Lunch Study.

Main Methods:

• Developed an iterative algorithm, MEBoost, utilizing estimating equations that correct for covariate measurement error.
• Applied MEBoost to the Box Lunch Study dataset to identify variables related to binge eating frequency.
• Conducted a simulation study to compare MEBoost with Convex Conditioned Lasso and naive Lasso under varying degrees of measurement error.

Main Results:

• MEBoost demonstrated improved accuracy in covariate selection compared to naive Lasso, especially under higher degrees of measurement error.
• Increasing measurement error led to higher prediction error and reduced accurate covariate selection, but MEBoost mitigated this loss.
• Simulation results supported the consistency of the model selected by MEBoost.

Conclusions:

• MEBoost offers a robust approach to variable selection in the presence of covariate measurement error.
• The method provides a valuable tool for analyzing complex datasets where measurement error is a concern.
• MEBoost enhances the reliability of statistical modeling and prediction in fields like nutrition and clinical research.