Parametric fitting in modeling demography for epidemiology forecasting

Abstract

Central to any epidemiological forecasts are the assumptions made pertaining to the population dynamics, and consequently, to each of the demographic components. However, most of the demographic models developed so far, is less useful in epidemiological forecasting given that often, it involves time series procedures and mostly are too detailed, requiring too much parameter inputs. In this study therefore, parametric models for mortality, fertility, and migration were fitted to the Belgian demographic data from 1954 to 2006. Age-structured models requiring a minimal set of parameter inputs were developed for mortality and fertility, using Quasi-Poisson Generalized Linear and Nonlinear techniques, and for migration, using Nonlinear regression technique. Comparison of the mean structures using BIC resulted to a sum of two exponential and a Fractional Polynomial model for mortality, sum of 3 Gaussian model for fertility, and sum of an exponential and 2 Gaussians for migration.