<p>Deep multi-view clustering aims to uncover the latent structure of data collected from multiple perspectives. However, most existing approaches overlook the intrinsic statistical relationships across views, leading to limited semantic alignment and representational coherence. This paper proposes a novel deep multi-view clustering framework that explicitly incorporates statistical dependency as the foundation for consistent representation learning across views. Firstly, statistical dependency is defined and investigated to model deep-level associations that reflect shared semantics and alignment across views. A joint entropy minimization strategy is designed to directly model statistical dependency across views, thereby enhancing the consistency and compactness of the learned representations. Secondly, a comprehensive loss function is formulated by integrating contrastive loss, probabilistic loss, and joint entropy loss. Furthermore, the theoretical lower bound of the loss function is analyzed, and the significance of soft clustering assignments is highlighted. A comprehensive deep multi-view clustering framework is proposed by considering both shared and complementary features, as well as the statistical dependency between different views. Finally, extensive experiments on seven public datasets demonstrate that the proposed method achieves state-of-the-art clustering performance, validating the effectiveness of incorporating statistical dependency into deep multi-view clustering.</p>