Paper Title
ASSESSING THE PERFORMANCE OF MAXIMUM LIKELIHOOD METHODS FOR MISSING DATA IN REAL-WORLD RESEARCH SCENARIOS

Abstract
Approximate methods such as Maximum likelihood (ML) have been adopted widely in empirical research as key tools for dealing with missing data situations and yielding reasonable and reliable statistical approximations from even incomplete data sets. This study discusses results of the use of the ML methods in different research contexts while laying emphasis on how these methods perform in studies with missing data patterns such as MCAR, MAR, and MNAR. I then discuss how the optimization of parameter estimation in the presence of missing data contributes to the reduction of sampling bias by comparing ML approach to other widely used techniques such as listwise deletion and mean imputation common in large population studies that have one or multiple missing data points, or even a missing data set. The accuracy of several ML methods is compared in small and large sample using simulation studies in terms of bias, efficiency, and coverage probability of confidence intervals. Moreover, explorative studies in the area of healthcare, social sciences, and environmental research provide insight into the performance of an ML in actual use. The results present evidence that, despite the advantages of ML methods in preserving the statistical inference, there are specific assumptions related to the missing data mechanism. Misapplied, ML risks yielding out wrong inference, mostly when data is not MAR. The findings of this research indicate the importance of researchers to understand the factors that affect the robustness of ML techniques and offers guidelines for managing missing value in practical research environments. Keywords - Maximum Likelihood, Missing Data, Bias, Parameter Estimation, Real-World Research.