The demand for group recommendations in the field of recommendation systems is steadily increasing. In group recommendation, how to accurately aggregate the preferences of group members to infer group decisions has become the core issue. Currently, various deep learning methods are applied to group recommendation problems. Among them, attention based methods dynamically aggregate group member preferences by distinguishing the importance of different members, which greatly improves group recommendation performance. However, attention mechanism methods cannot avoid the negative impact of potential confounding factors. That is to say, the correlation between group members and the learned candidate projects cannot accurately reflect the impact of group members on the group recommendation results, leading to false correlation. This affects the accuracy of group representation learning. To tackle this challenge, the paper introduces a model named Causal Attentive Group Recommendation(CAGR). This model incorporates causal inference within an attention network to tailor the group representation, effectively addressing the problem of capturing erroneous correlations. Building upon the potential outcome framework, CAGR utilizes the concept of individual treatment effects (ITE) to quantify the causal relationship between each group member and the outcome. Our objective is to capture the authentic influence of group members on the desired outcome. To integrate causal insights into the group representation learning process, we introduce regularization that aligns the distance between the ITE of group members and the conventional attention weights, correct the importance of group members, and obtain a more accurate causal correlation between group and group members. Comprehensive experiments conducted on two authentic datasets validate the superiority of our proposed model in the realm of group recommendation.