The research develops and tests Bayesian approaches for handling missing data in large-scale social surveys. It compares multiple imputation strategies with model-based priors that reflect plausible demographic relationships. Using simulated datasets and a real-world survey application, the paper evaluates parameter recovery, uncertainty calibration, and robustness under non-random missingness. The results show that carefully chosen priors improve inference quality and provide more reliable uncertainty estimates. The paper ends with practical guidance for analysts and survey teams.