Models as measuring instruments: measurement of duration dependence of unemployment

Nancy Cartwright views models as blueprints for nomological machines - machines that, if properly shielded, generate law-like behaviour or regularities. Marcel Boumans has argued that we can look for devices inside models, which enable us to measure aspects of these regularities. Therefore, if models do produce regular behaviour (Cartwright), they might perhaps generate numbers about phenomena in the world, provided we can locate a good measuring device in the model (Boumans). How do they do this? Models are often understood to consist of internal principles and bridge principles. Both of these play a role in the measuring process. This paper suggests that we can understand the internal principles to be responsible for generating the regularities displayed by the model - and for many users of the model, this is sufficient use. The bridge principles (following Cartwright - 'the real bridge principles') are a matter of picking the right mathematical representation to make the model work with respect to some real world case at hand. These 'real' bridge principles may enable the model to generate numbers and so serve to make the model into a measuring device. This paper explores this construction of a measuring device of duration dependence of unemployment. Since search theory cannot be made operational for this purpose duration models, which model of the outflow process of unemployment with a Weibull function, are taken as a representation of reservation wage setting in search models. As I show, this Weibull function serves as a bridge principle to make a measure of the elasticity of response between unemployment benefit and duration of unemployment. It enables measurement to take place by modelling behaviour according to some assumptions, which operate as constraints. While this Weibull function serves this purpose, its assumptions are arbitrary rather than connected to theory, and measurements generated with this bridge principle turn out to be highly sensitive to its specification. The Weibull function does enable the model to function as a measuring device, but the lack of ability to make newly invoked unobservable variables operational makes it of dubious worth.