In order to improve the government's ability to manage social risk from major decision-making based on the higher-order Markov chain theory, the variable weight method and set pair analysis method are used to build the prediction model for social risk from major decision-making. Firstly, the state sets of social risk indexes before and after making major decisions are composed a set pair. Then the variable weight method is used to calculate the connection degree of the set pair and state transferring probability matrix in different time to overcome the problem that the constant weight can not reflect the order importance of the value of index. Finally, the higher-order Markov chain is proposed to predict the connection degree of the set pair and to analyze social risk situation. In addition, the higher-order coefficients of the Markov chain are calculated by the similarity between state transferring probability matrixes. The model presented is verified and compared by an illustrative example of PX project decisions. The results show that the proposed model is more accurate and efficient to predict social risk compared with the traditional model. Furthermore, the risk attitude of experts has great influence on short-term social risk analysis and prediction, while the long-term social risk analysis and prediction are less affected.