Model fitting of zero-inflated and overdispersed count data
Researchers often encounter data which exhibit an excess number of zeroes than would be expected in a Poisson or negative binominal model. This is referred to as zero-inflation. Additionally, data may display excess variability or overdispersion. Failure to model zero-inflation and overdispersion ma...
Saved in:
Main Authors: | , |
---|---|
Format: | text |
Published: |
Animo Repository
2010
|
Online Access: | https://animorepository.dlsu.edu.ph/etd_bachelors/11326 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | De La Salle University |
Summary: | Researchers often encounter data which exhibit an excess number of zeroes than would be expected in a Poisson or negative binominal model. This is referred to as zero-inflation. Additionally, data may display excess variability or overdispersion. Failure to model zero-inflation and overdispersion may lead to serious underestimation of standard errors and misleading regression parameter estimates. Poisson, negative binomial, zero-inflated Poison (ZIP) and zero-inflated negative binomial (ZINB) regression models are applied to CBMS Pasay City Poverty Census of 2005. Barangays are ranked according to estimated proportion of households below food poverty line. Overdispersion parameters indicate that the data is overdispersed and hence, a negative binomial model is preferred over Poisson model. However, zero-inflation parameters pose no significant evidence that the data is zero-inflated. Accordingly, goodness of fit statistics for the over-al best fit model show that the negative binomial regression model is the most preferred. |
---|