本研究以代理人基模型探討媒體在總統選舉過程中所公布的民調有效程度以及選民對媒體的依賴程度對民意的影響力。利用種族分離模型 (Segregation Model) 與近五次總統選舉的資料來建構社會網絡的結構。以代理人參考媒體公布民調的程度j = 0.2為對照組，按四種不同的民調品質，分析j = 0.5 , 0.8的影響。 本文得到的結論是 (1)不論是有效民調或是無效民調在j = 0.2時，民調對代理人已經有影響力，j的數值愈高，影響力愈大。 (2)條件j=0.2和j=0.5之下的無效民調對優勢黨尚無影響力，對劣勢黨則有影響；但是在j=0.8之下出現相反的現象，這表示過度相信無效民調會誤導選情。 (3)民調愈接近真實得票率，模擬結果的政黨支持率也愈可能代表真實得票率。 (4)模擬結果顯示有效民調與無效民調的政黨支持率分布較其他兩者集中，說明即使是無效民調也可以產生民意的共識。 In this study, an agent-based model is applied to explore the role of poll from media and degree of reliance on the media for public opinion formation during Taiwan’s Presidential Elections. We use Schelling’s segregation model and the data of last five Presidential Elections to construct the structure of social networks. Through degree of reliance on the media published poll, j = 0.2, as the control group, according to four different polls quality to analyze the effect of j = 0.5 and j = 0.8. In this study, the conclusion is that (1) Whether valid or invalid polls in j = 0.2, the agents have been influenced by polls, the higher value of j, the greater the influence. (2) Under j = 0.2 and 0.5, invalid polls have no influence on the advantage party (candidate), but have significant influence on the disadvantage party; however, under j = 0.8, the opposite phenomenon occurs, indicating that the intensive invalid polls mislead the election results. (3) For the more accurate polls, the simulation results of each party’s support rate is also more likely to represent the real voting results. (4) The simulation results show that the political support distributions of the valid polls and invalid polls are more concentrated than the other two, indicating that even invalid polls also can produce a consensus of public opinion.