本研究利用實驗經濟的方法來檢驗雙佔Bertrand市場中，廠商的決策行為是否會因為匿名 與否以及是否事先知道賽局回合數及何時結束而有顯著的差異。本文有別於過去相關賽局文獻 以合作為實驗分析重點，我們改用\"背叛''一即不合作為分析重點，並再將之細分為弱式和強式 背叛進行檢定。實驗的結果顯示匿名與否會造成受試者的決策行為有顯著的差異；亦即在不匿 名的情況下，即使是與對手只有一面之緣，受試者探取背叛策略的次數顯著較少，故從實驗結 果確定見面三分情的效果的確顯著。此外，本文亦針對在重覆賽局中，若受試者是否因知道賽 局何時結束，而在賽局結束前幾回合即提前背叛，即所謂「終局效果」來進行檢定。經無母數 的檢定結果證實在匿名時，終局效果是顯著的，受試者在遭對手背叛之前會選擇先下手為強； 但在不匿名的情況下，可能受到見面三分情的影響，終局效果並不顯著。

This experimental study aims to examine the end-game effects of prisoner’s dilemma games presented in a form of Bertrand competition with the face-to-face versus anonymity treatments in a repeated game. Our experimental findings are summarized as follows. First, it is found that firm's decisions to defect in the face-to-face condition are statistically different from that in the anonymous condition. Specifically, the frequencies of firm's defections in the face-to-face condition are significantly less than the ones in the anonymous condition. Second, the results also confirm the end-effects as predicted by the reputation model in which cooperation occurs in early rounds of the game and then is followed by defection in rest of rounds in a finitely repeated game under the anonymity condition. However, the end-game effects are attenuated under the face-to-face condition.