Due to the absence of probabilistic mathematical foundations, determining sample size in nonprobability sampling remains a methodological challenge. This article introduces a novel framework, emph{Sample Size Determination Based on Modeling Objectives and Predictive Power}, shifting the paradigm from population-based to model-based inference. Within this framework, sample size is redefined based on optimizing model performance—specifically, the minimum size needed to detect significant relationships and ensure generalizable predictive power. Combining simulation and power analysis, this approach provides a systematic, R-powered solution for non-probability designs, enhancing research validity and advancing standardization in this domain. Simulations confirm its efficacy; however, its application is limited by the prerequisite need for valid auxiliary data and prior knowledge of key model parameters.