Advances and Applications in Statistics
Volume 34, Issue 2, Pages 137 - 165
(June 2013)
|
|
A BAYESIAN ANALYSIS OF US MORTALITY CURVES FOR RACE-SEX DOMAINS BY SMALL AREA
Rong Wei, Balgobin Nandram and Dilli Bhatta
|
Abstract: Fitting of mortality curves across all ages in race-, gender- and state-specific populations is one of the key procedures to generate and publish life tables at the National Center for Health Statistics (NCHS). We use the eight-parameter Heligman-Pollard (HP) model to fit these curves to US mortality data for 1999-2001 summed. Because the data are studied in these small domains, there is the possibility for unrepresentative zero deaths at some ages. This creates difficulties for fitting the HP model. To minimize the estimation and computational difficulties, we do not fit the death counts directly to the HP model. Rather we first estimate the probabilities of deaths for each race, sex and state over age range (0-80 years) using a Bayesian beta-binomial model with a correlation across this age range. This procedure provides smoothed estimates of the probabilities. Then, in a second stage, we fit the HP model to these smoothed estimates using nonlinear least squares, thereby avoiding the difficulties in fitting the HP model directly to the death counts. We present estimates of the mortality curves for a few selected small areas by race (black, white) and gender (male, female). |
Keywords and phrases: state mortality, mortality curve, Heligman-Pollard model, small area, griddy Gibbs sampler, Markov model. |
|
Number of Downloads: 346 | Number of Views: 883 |
|