No Zero Left Behind: Comparing the Fit for Zero-Inflation Models as a Function of Skew and Proportion of Zeros
by Jeffrey M. Miller, M. David Miller.
Several models have been proposed for analyzing data characterized by a preponderance of zeros.
Substantively, the choice between these models should be based solely on the data generating process.
However, datasets can vary as a function of both the proportions of zeros and the distribution for the nonzeros.
A Monte Carlo design was used to sample 1,000 cases from each of six distributions and five proportions of zeros.
Between-model superiority was tested using deviance statistics and AICs over 2,000 simulations per condition.
The results suggest that the best-fitting zero-inflated model sometimes depends on the proportion of zeros
and the distribution for the nonzeros. In fact, there are situations where the zero-inflated models are not necessary.
Zero-Inflation, ZIP, Hurdle, Zero-Inflated Poisson, Poisson
Jeffrey M. Miller, email@example.com
M. David Miller, firstname.lastname@example.org
Joseph W. McKean, email@example.com
READING THE ARTICLE: You can read the article in
portable document (.pdf) format (427133 bytes.)
NOTE: The content of this article is the intellectual property of the authors, who retains all rights to future publication.
This page has been accessed 2732 times since OCTOBER 26, 2008.
Return to the Home Page.