Abstract
an incompletely specified model is assumed to be a true model of data in our hand, in which it is uncertain whether the model is a zero-hypothesis or an alternative-hypothesis, and the question is to select the true model.It has been impossible to accept a zero-hypothesis by such traditional methods based on a completely specified model as testing of significance or selecting a statistical model of estimate, which has any latent loss based on ignorance of zero-hypothesis.The purpose of the present paper is to give a criterion for selecting the true model by testing of the both statistical hypotheses proposing a generalized loss of testing hypotheses, which is associated with a loss of testing in a form of probability and loss of estimate in quadratic form.
