Applications of Multilevel Zero-Inflated Poisson Model -Analyzing occurrence and frequency of juvenile risky behavior-

초록

In much Education research, count data of interest such as runaway frequency have more zeros than can be inferred from Poisson distribution. One of the main reasons of this kind of zero-inflated data is because data are generated from two latent groups - runaway risk group and risk-free group, for example. Using Zero-Inflated Poisson (ZIP) model, it is possible to simultaneously estimate parameters from the two latent group (the probability of risk group membership and the event frequency under risk group) using all the cases from a sample. Moreover in Education research, ZIP model needs to be extended to multilevel to incorporate nested data structure where independence assumption can be violated otherwise. By extending ZIP model to multilevel, it is also possible to incorporate group level correlates of risk group membership and event frequency. In this study, Bayesian approach to ZIP regression (Ghosh et al., 2006) is extended to multilevel to provide methodological tool for studying individual- and group-level correlates of event occurrence and frequency simultaneously. Exemplifying ‘runaway’ analysis used data from Korean Education and Employment Panel (KEEP) provided by Korean Research Institute for Vocational Education and Training.

키워드