Activity Number:
|
512
- Bayesian Model Selection
|
Type:
|
Contributed
|
Date/Time:
|
Wednesday, August 2, 2017 : 10:30 AM to 12:20 PM
|
Sponsor:
|
Section on Bayesian Statistical Science
|
Abstract #324314
|
|
Title:
|
Inflated Values Selection Method on Multiple-Inflated Poisson Model
|
Author(s):
|
Qiuya Li* and Kwok Fai Geoffrey TSO and Yichen Qin and Yang Li
|
Companies:
|
and City University of Hong Kong and University of Cincinnati and Renmin University of China
|
Keywords:
|
Count data ;
Multiple inflation ;
Poisson distribution ;
Inflated values selection ;
Adaptive lasso
|
Abstract:
|
Count data with inflations arise in economics, public health and other fields. Zero-inflated Poisson model (ZIP) is used to model count data with excess zeros, while multiple-inflated Poisson model (MIP) could be applied to analyze count data with multiple inflated values. MIP assumes a finite mixture model of a Poisson distribution and a set of degenerate distributions. The existing modeling methods to determine the inflated values are mainly conducted by inspecting the histogram of the count response and fitting the model with various combinations of inflations to select the best model, which are relatively computationally complicated and easy to overlook the inflated values. Selection of the multiple inflated values still receives much attention currently. We address inflated values selection by adopting the adaptive lasso regularization scheme on the mixing proportion that performs both inflated values selection and parameters estimation simultaneously. Simulation studies are carried out in different model settings and demonstrate the excellent performance in inflated values selection and proposed method is used to investigate the health service data.
|
Authors who are presenting talks have a * after their name.