CMStatistics 2015: Start Registration
View Submission - CMStatistics
B1514
Topic: Contributed on Generalized linear models and beyond Title: Deviance analysis of age-period-cohort models Authors:  Bent Nielsen - University of Oxford (United Kingdom) [presenting]
Abstract: The age-period-cohort model is often used to model cancer mortality data aggregated in Lexis diagrams. The model has an inbuilt identification problem that has caused much controversy. Four contributions are made. First, the identification problem is addressed by reparametrising the model in terms of freely varying parameters: a linear plane and double differences of the original time effects. Second, a deviance analysis is presented that unifies a wide range of sub-models from the literature. Third, a new plot of the deviation of the time effects from the unidentified linear trend is suggested and interpreted. Fourth, parametrisations of these deviations are suggested. The methods are applied to a classic Belgian lung cancer data.