Construction safety remains a critical issue, with the total number of safety accidents continuing to be alarmingly high and frequent. In order to improve construction safety management and effectively prevent accidents, a novel accident causation mechanism analysis framework is proposed based on STPA and an improved complex network model. This framework aims to reveal the mutual feedback relationships between multiple construction stakeholders and their causation mechanisms. The framework is designed to reveal the interactions and coupling effects among multiple construction stakeholders in typical levels and types of accidents. First, STPA is applied to analyze 1 467 construction accident cases collected from Chinese government websites between 2013 and 2023. The accident level, type, UCA, and causes leading to the UCA are extracted to construct an accident-behavior-causation network. Second, an improved complex network incorporating OR is proposed to more accurately assess the strength of associations between causal factors, reflecting the feedback relationships among stakeholders. Finally, the topological parameters of the complex network are analyzed to explore the causal mechanisms across typical accident scenarios. The analysis reveals that among the interactions and coupling effects of construction stakeholders, the lack of responsibility for safety(F331) is a key causal factor leading to accidents. It has a strong coupling effects with insufficient site management(F111), which results in safety accidents by coupling the failure to implement necessary safety precautions(B10). This study systematically analyzes the causal mechanism of construction accidents from the novel perspective of mutual feedback relationships among multiple stakeholders. It emphasizes the importance of reinforcing construction safety and responsibility management. The findings provide new insights and methodologies for advancing research in construction safety management.