To predict bureaucratic performance, and ultimately to prescribe managerial methods that aid change, a model of the internal policy processes connecting bureau inputs and outputs is necessary. Unless we create such a model, we put ourselves in the position of making prescription of an internally unknown engine: we can explore what it consumes to run, what it produces, and how efficient its outputs are relative to its inputs. But we cannot confidently predict how the engine behaves in different situations, estimate what the engine is fully capable of doing, or prescribe how to engineer improvements in efficiency. How do we explore the workings of a bureaucratic "engine" as it adopts policy? One approach is in fact deductive: specify the inputs—the preferences of staff, the supply of raw materials (e.g., money, cases, data, analysis), the preferences of the external actors who control inputs, and the external referents of the decision criteria—the bureau's outputs, and a model of how internal processes connect the two. Then test how well the model makes empirical predictions. The model connecting inputs to outputs might start with the assumption that bureau members together operate approximately as a unitary, rational, and fully informed decision-maker. Such theoretically compact models can be empirically well founded when: (1) organizational processes support fully informed decisions by an administrative "dictator" who has well-defined preferences (see Plott 1976); (2) the organizational structure "by design" (i.e., divisions of labor, specialization, redundancies in analysis, and so forth) provides a rationality-approximating decision process (see Cohen 1981); or (3) environmental feedback and ecological "selection" are sufficiently frequent, accurate, consistent, and clearly perceived so that agencies can move toward "rational" outcomes over time.
Such rational choice models are elegant, but the conditions on which they rest frequently do not promise to hold strongly in practice: a bureau's top decision-makers are often neither fully informed nor in complete, dictatorial charge of the bureau ( Lindblom 1959; Kaufman 1981); bureaucratic structures may not closely approximate rational decision processes or even be coordinated