Fibre analysis and tangent models

Available methods for deriving tests and confidence regions for multiparameter models seem seriously limited in their feasibility. This paper gives preliminary survey to some developing techniques that use conditioning as a primary device to derive tests and confidence regions. The techniques usefibre analysis which involves the conditional determination of OLS (observed level of significance); the techniques are available initally for models with pivotal quantities and for continuous exponential models. The techniques can then be extended to more general models using the notion of tangent model.