|
Activity Number:
|
233
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Tuesday, August 5, 2008 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #302666 |
|
Title:
|
An EM Algorithm for Zero-Inflated Negative Binomial Regression Based on a Poisson Mixture Representation
|
|
Author(s):
|
Chih-Nan Chen*+ and Xiao-Li Meng
|
|
Companies:
|
Cambridge Health Alliance and Harvard University
|
|
Address:
|
Center for Multicultural Mental Health, Somerville, MA, 02143,
|
|
Keywords:
|
Zero-Inflated negative binomial regression model ; Overdispersion ; Count data with excess zeros ; EM algorithm
|
|
Abstract:
|
Zero-inflated negative binomial regression model is quite useful for count data with extra zeros and over-dispersion. Fitting such a model, however, is not a trivial task, particularly when the sample sizes are small. We found that an EM algorithm based on a mixture Poisson representation of the negative binomial density can help to reduce some instability of existing algorithms, because the new implementation takes advantage of the well-behaved Poisson regression fitting in its M step. Application is given to the analysis of visit numbers in the service use of mental health care for the National Latino and Asian American Study (NLAAS), a problem that originally motivated us to develop this algorithm because of the instability problems we encountered with several existing algorithms.
|
