Differences in panel-data results according to different methods of missing value handling

This study compares the differences in results based on missing value handling methods using actual panel data from a study on school dropout youth in Korea. The 5th wave 2017 data from the “Dropout Youth Panel Survey” and “Study on Customized Support Policy According to Performance Channels of Out-of-School Youth” were analyzed. Multiple regression was used to examine the influence of gender, career exploration behavior, self-esteem, career resilience, and social stigma on future expectations. Different missing value handling methods (listwise deletion, pairwise deletion, mean imputation, EM, MI, K-nearest algorithm and random forest) were applied, and their results compared. The findings demonstrate that the method of handling missing values can significantly impact research results. Traditional methods failed to detect the effect of career exploration behavior, while EM, MI and machine learning method revealed a statistically significant positive influence. This study aims to warn that the results may vary depending on how missing value is handled. Future studies should consider the pattern and causes of missingness for more accurate results.