様子見は付き合いに何をもたらすか：行動エラー下での離脱・復帰可能な繰り返し囚人のジレンマ

While an iterated Prisoners’ Dilemma (= IPD) has been regarded as a one-on-one relationship model, its real relationship in many situations has not only “Cooperate” and “Defect” as action options but also temporally “Escape” from the interaction as the third action option. We clarify what strategies are effective in IPD with “Escape” Option (= IPDEO) using ESS-type evolutionary simulation with action error, where the possible strategies are 19683 Markov strategies. The results of the analysis have the three points. (1) If the benefit of cooperation is not small, “nice lose-shift” strategies are effective which have the property (nice) never being the first to defect and the property (lose-shift) to change action from being sucked or mutual defection. (2) Concretely speaking, such strategies are “win-stay lose-shift strategy” (= Pavlov) and “win-Escape lose-shift” strategy (= Pot & Run); the former “Pavlov” sucks the opponent if possible and the latter “Pot & Run” escapes after sucking the opponent. However, if the benefit of cooperation is not small, Pavlov is the most effective strategy in IPDEO. (3) Though TFT strategy is not effective because of the mutual defection rock in, the most effective strategy, a kind of Pavlov, has the conditional retaliatory property similar to TFT. Therefore it is concluded that the strongest strategy is the “Retaliatory Pavlov”.