In Actuarial Science and insurance, models are often used to assess risk using factors such as age, level of education, gender, environment, etc. In practice, there might be unobserved factors.  When data are gathered, the accuracy of these models should be investigated. If the model is not well fitted by the data, subsequent influences could be badly wrong. In this proposal, we shall examine tests to check the model validity. In addition, we examine the effect on inference of a failure in the model. The unobserved factors pose special problems which can sometimes be addressed using Bayesian procedures. I have done some of this work in my thesis and will develope it further.