The purpose of this project is to establish a general model to reconcile apparently contentious theoretical interpretations of the effects of similarity in resource dependence profiles on the propensity of organizations to collaborate. According to the first interpretation the more two organizations depend on similar resources - i.e., the more their niches overlap - the more intense is their rivalry. Because competition erodes social relations, the lower will be the propensity of rival organizations to cooperate. According to the second interpretation organizations made similar by overlapping niches will find it less costly to communicate across their boundaries and manage their joint resource dependencies by establishing - rather than severing - exchange relations with other organizations. We argue that the rapprochement of conflicting theoretical visions is possible in the context of a model that admits the presence of a non-monotonic relation between similarity in patterns of resource dependence and inter-organizational collaboration. In the model that we propose the propensity of organizations to collaborate is controlled by two opposing forces - opportunity for cooperation and rivalry - that depend on niche overlap. Opportunities for cooperation increase with niche overlap, but at a decreasing rate. Rivalry increases with niche overlap at an increasing rate. The main implication of these assumptions is a non-monotonic relation between similarity resource dependence profiles and the propensity of interdependent organizations to collaborate and exchange resources. We provide an empirical test of the model using data on all the organizations providing health care services in one of the largest Italian regions. We chose this particular setting because hospitals operate in environments that pose both competitive and institutional constraints. As a consequence, managing cooperation via resource exchange, and dealing with competition triggered by dependence on similar resources are equally important processes for these organizations. The target data set will contain information on dyadic relations defined in terms of exchange of patients among all the 111 hospital organizations operating in the given region. The observation time frame spans 4 years. Under assumptions of a balanced panel design the final sample will consist of 48,840 dyadic observations. The analytical part of the project involves the estimation of stochastic models for discrete counts. Because the analysis of dyadic data poses complicated inferential problems induced by the lack of independence of the observations, statistical models for social networks will also be estimated.