A robust imputation method for missing responses and covariates in sample selection models.

Area of Science:

• Statistics
• Biostatistics
• Econometrics

Background:

• Sample selection bias occurs when study outcomes are partially observed.
• Existing methods for missing data are complex or rely on restrictive assumptions.
• Handling missing covariate data in selection models, especially without exclusion restrictions or normality, remains challenging.

Purpose of the Study:

• To propose a robust imputation technique for sample selection models with missing covariate data.
• To evaluate the performance of the proposed method against alternatives.
• To address limitations of existing imputation methods in the presence of missing covariates and absence of exclusion restrictions.

Main Methods:

• Development of a robust imputation technique based on the selection-t sample selection model.
• Comparison with alternative imputation methods via a simulation study.
• Application to real-world data (NHANES) with missing income and blood pressure data.

Main Results:

• The proposed robust imputation method is not sensitive to the absence of exclusion restrictions.
• The method demonstrates superior performance compared to normal-based models, even with normally distributed data.
• The technique effectively handles both missing outcome and covariate data, confirming its robustness.

Conclusions:

• The robust imputation technique offers a reliable solution for missing covariate data in sample selection models.
• The method provides a valuable alternative to existing approaches, particularly when assumptions like exclusion restrictions or normality are violated.
• Implementation in R's MICE environment facilitates practical application of this robust statistical approach.