Abstract:
|
Age-period-cohort (APC) analysis of incidence and/or mortality data has received much attention in the literature. To circumvent the non-identifiability problem inherent in APC models, additional constraints need to be imposed on the parameters of the model. However, different constraints often lead to drastically different results. We advocate to set the constraint to reflect the different nature of the three temporal variables. Recognizing the age effects to be deterministic, we do not constrain the age parameters. For the stochastic period and cohort effects, we set a constraint of constant relative variation (CRV) on the period and cohort trends. The CRV constraint dictates that between the two stochastic effects, the one with a larger curvature gets a larger (absolute) slope, and that the one with zero curvature gets no slope. The latter property should prove useful in that if any of them is null and void, the CRV constraint will guarantee that its slope is zero. We conduct Monte-Carlo simulations to examine the statistical properties of the CRV method. We analyze the data of prostate cancer mortality for nonwhites in the U.S. to illustrate the methodology.
|