The rapid expansion of digital data sources, online panels, and administrative records has profoundly transformed the landscape of survey research. Traditional probability sampling, long regarded as the gold standard for population inference, is increasingly complemented or even supplanted by nonprobability samples due to cost, timeliness, and operational constraints. However, nonprobability samples pose substantial challenges for valid statistical inference, primarily because of unknown selection mechanisms and systematic selection biases. This article develops an extensive theoretical and methodological examination of data integration strategies that combine probability and nonprobability samples to support robust population-level inference. Drawing strictly on foundational and contemporary literature in survey statistics, the study synthesizes classical sampling theory with modern approaches such as mass imputation, propensity score weighting, doubly robust estimation, and statistical learning–based adjustments. The article elaborates on the conceptual underpinnings of these methods, the assumptions required for their validity, and the practical consequences of assumption violations, particularly focusing on common support, ignorability, and nonresponse mechanisms. Using the National Health and Nutrition Examination Survey as a conceptual reference framework, the paper explores how probability samples can serve as calibration anchors for integrating rich but biased nonprobability data. Rather than presenting numerical results, the analysis emphasizes interpretive insights, methodological trade-offs, and inferential implications. The discussion critically evaluates the limits of existing methods, highlighting the persistent risks of overconfidence in hybrid estimators and the need for transparency in uncertainty assessment. The article concludes by outlining future research directions, including the integration of machine learning with survey theory and the development of principled diagnostics for assessing inferential validity. Overall, this work provides a comprehensive, publication-ready contribution to the evolving field of survey data integration.
