Detecting and Correcting Estimation Bias Caused by Content Deletion in User-generated Content Studies

User-generated content (UGC) platforms increasingly enable content deletion to manage the online environment and user-posted content, which contributes to missing data issues in empirical research involving UGC data. Such missingness may be correlated with the unobserved data, resulting in the most challenging missingness type–Missing Not at Random (MNAR), which can lead to biased estimates in empirical analyses. Despite its prevalence, limited scholarly attention has been given to the missing data challenges arising from content deletion, and researchers lack theoretical guidance for diagnosing and handling such missingness. In this study, we conduct an in-depth investigation of the reasons for content deletion across various UGC platforms and identify variables susceptible to the MNAR issue in empirical studies. We also develop a theoretical framework for diagnosing missingness mechanisms and selecting appropriate correction methods. Most importantly, we propose theoretical formulas to correct biased estimation in two MNAR scenarios for scale-level studies. Extensive simulations and sensitivity tests demonstrate that our proposed methods effectively reduce the estimation bias under scale-level MNAR compared to benchmark approaches. Collectively, our work advances the missing data literature and provides practical implications for empirical researchers.