It is assumed that players bundle nodes in which other players must move into analogy classes, and players only have expectations about the average behavior in every class. A solution concept is proposed for multi-stage games with perfect information: at every node players choose best-responses to their analogy-based expectations, and expectations are correct on average over those various nodes pooled together into the same analogy classes. The approach is applied to a variety of games. It is shown that a player may beneÞt from having a coarse analogy partitioning. And for simple analogy partitioning,(1) initial cooperation followed by an end opportunistic behavior may emerge in the Þnitely repeated prisoner's dilemma (or in the centipede game), (2) an agreement need not be reached immediately in bargaining games with complete information.