|
Abstract:
|
Abortion is one of the top social issues in the US. There are nearly 1,000,000 abortions per year on average. In 2015, approximately 35% of all pregnancies in New York City ended in abortion (CDC). For African American women in NYC, more abortions happen than babies born. A timely prediction of the prevalence of Induced Terminations of Pregnancy incidents is highly desirable. However, the modeling of discrete incidents is not trivial and often exhibits nonlinear features across time and locations. Hence, demanding computations are inevitable to estimate the nonlinear Spatio-temporal model. In this research, we apply a functional data analysis that assumes the incidents as a function of time and interprets the incident nonlinearly at each time. Under the Bayesian framework, we utilize a radial basis function network with latent variables to capture the nonlinear pattern of spatial incidents and estimate the model through a dynamically weighted particle filter for efficient computation in the temporal change of pattern. The proposed methodology analyzes yearly Induced Terminations of Pregnancy incidents data collected in the Texas Department of Health Services from 2008 to 2020.
|
