How are empirical successes based on idealized models such as those in economics and other special sciences possible? And what does the manner in which those successes are achieved tell us about the models' role and status? Existing answers to these questions, I argue, are unable to make sense of a prominent and typical case of application of game theory - the design of an institution. This dissertation seeks to bring philosophical accounts of models and their application in line with this practice. In chapter 1 I examine accounts currently on offer. The first account views models as making ceteris paribus claims and their application as satisfaction of the relevant assumptions of a model by the target system. I review two ways of identifying the relevant assumptions - by prediction and by robustness analysis - and find them both wanting. The second is the credibility account which I claim does not tell us how models become bases for explanation and prediction. On the third account models make tendency claims and they apply to a phenomenon if the model plus additions to it identify the relevant tendencies and disturbing factors underlying the phenomenon. However, this account assumes knowledge of stable causal relations, which is often unjustified in special sciences. Chapter 2 presents a study of a design of a social institution on the basis of game theory - the FCC spectrum auctions. I describe the respective roles of theory and experiment in this case, and argue that none of the accounts reviewed in chapter 1 can make sense of this example. Chapter 3 develops an account of theory application that fits the auction design case and is flexible enough also to capture the use of theory in special sciences in general. On this view, models should be treated not as ceteris paribus or capacity claims but as open formulae for causal hypotheses, which we apply by specifying an empirical realization of the causal relation in the hypothesis. This account is in line with the practice of experimental economics, and provides normative constraints for the use of rational choice models for historical explanation