Applications of Time Series Methods for Obtaining Complete Mortality Counts

• Activity Number: 589
• Type: Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Diba Khan*+ and Myron J. Katzoff
• Companies: Centers for Disease Control and Prevention/NCHS and Centers for Disease Control and Prevention/NCHS
• Keywords: mortality ; correlation ; forecasting ; nonlinear models

Abstract

The National Center for Health Statistics collects state level mortality data by cause of death through the National Vital Statistics System. The data used in our analyses are the weekly mortality counts in each of three selected states that cover a span of 8 years for a selected cause of death. Several methods exist in the literature for estimating complete mortality counts but, in this paper, only two approaches for modeling and predicting mortality time series are examined. In each approach, the non-Gaussian structure of the data is taken into account. In the first approach, models are developed using the quasi-likelihood method separately for each state with no provision for the possibility of cross-state dependence and are, therefore, of a univariate type. The second approach uses a multivariate model that exploits the cross-state correlation structure for adjoining states within a region. Results are then compared when the correlation structure is explicitly modeled in the multivariate approach with those when it is excluded from the models using the univariate approach.