A 'flipped classroom' creates a dynamic and effective learning environment where students learn basic knowledge at home and engage in active learning activities during the class. It has been widely used across different STEM (Science, Technology, Environment, Mathematics) disciplines. However, there is a lack of systematic and quantitative assessments of the effectiveness of the flipped classroom model on student performance. In this study, a meta-analysis was conducted with 53 research articles on the flipped classroom in STEM higher education. It showed an overall significant effect in favor of flipped learning over traditional learning models for STEM education (Hedges' g = 0.263, 95% [0.190, 0.337], Z = 7.03, p < 0.0001), with no evidence of publication bias. Moderator analysis showed that student performance was associated with the implementation strategies of flipped learning and the research control methods. Testing of the moderating factors "link activity", "pre-class activity", "in-class activity", and "post-class activity" suggests that link activities to gauge home learning at the beginning of class, a mixture of group activities and individual activities during class and setting a combination of quizzes and exercises after class, all serve to improve student achievement in STEM education.