We study two-player pie-splitting games in which one player knows the pie and the other knows only its probability distribution. We compare treatments in which incentive-efficient strikes (disagreements) are possible with alternatives in which efficiency forbids strikes. We find that incentive-efficiency is very helpful in explaining when strikes occur. There is also evidence of substantial heterogeneity in the subjects' altruism and in their risk preferences. This means that the common-knowledge assumptions of game theory cannot be controlled in experiments; but in our experiments the main theoretical conclusions seem robust to violations of these assumptions.

Much effort has recently been devoted to the development of strategic models of bargaining with private information. At the same time, empirical analyses of data on collective bargaining in the labor market have generated an interesting list of facts to be explained. In this paper we assess the prospective interactions between these theoretical and empirical developments. The style here is expository: detailed arguments have been ruthlessly suppressed. A more complete treatment can be found in our 1989 article. The theoretical models deal with conflicts between rational bargainers in situations where at least one of the bargainers has private information, as in the case where an employer knows more about the value of labor's product than the workers do. We start by explaining why private information is important in building models of strikes. In a naive model of collective bargaining, the union starts with a tough demand, and

The authors study two-player, pie-splitting games in which one player knows the pie and the other knows only its probability distribution. The authors compare treatments in which incentive-efficient strikes (disagreements) are possible with alternatives in which efficiency forbids strikes. They find that incentive-efficiency is very helpful in explaining when strikes occur. There is also evidence of substantial heterogeneity in the subjects' altruism and in their risk preferences. This means that the common-knowledge assumptions of game theory cannot be controlled in experiments; but in the authors' experiments the main theoretical conclusions seem robust to violations of these assumptions.