Handling Missing Data and Attrition Bias in Unbalanced Panel Data Sets: Multiple Imputation Techniques and Inverse Probability Weighting in Longitudinal Health Economics Research

Abstract

Longitudinal health economics research faces substantial methodological challenges when dealing with unbalanced panel data sets characterized by systematic missingness and attrition bias. Traditional analytical approaches often fail to account for the complex mechanisms underlying data loss, leading to biased parameter estimates and compromised statistical inference. This paper presents a comprehensive framework for addressing missing data patterns through the integration of multiple imputation techniques and inverse probability weighting methods specifically tailored for health economics applications. The research develops novel theoretical foundations for understanding missingness mechanisms in longitudinal health data, distinguishing between missing completely at random, missing at random, and missing not at random scenarios. We propose a unified approach that combines Bayesian multiple imputation with inverse probability weighting to simultaneously address both unit non-response and item non-response while maintaining the temporal structure inherent in panel data. The methodology incorporates auxiliary variables and leverages the predictive power of observed covariates to enhance imputation accuracy. Empirical validation using simulated data sets and real-world health economics panels demonstrates substantial improvements in parameter estimation accuracy and reduction in bias compared to conventional listwise deletion and single imputation methods. The proposed framework yields consistent estimators under mild regularity conditions and provides valid statistical inference through proper uncertainty quantification. Results indicate that the integrated approach reduces bias by up to 45\% in treatment effect estimation and improves confidence interval coverage rates to nominal levels across various missingness scenarios.