Conventional quantitative methods in social science have primarily emphasized statistically testing predefined hypotheses rather than generating new theoretical insights. This chapter proposes the computing grounded theory (CGT), an algorithm-based approach that identifies potential patterns of relationships among key variables in large-scale survey data. By combining the predictive capacity of machine learning with the interpretability of model-agnostic techniques, CGT uses feature attribution to assess the relative importance of predictors, thereby identifying potential explanatory variables that have not been recognized in the existing literature. This approach advances the understanding of how data, algorithms, causality, and prediction interact in quantitative social science research.