Statistics colloquium talk at the University of Iceland will be given on Thursday December 12th 2013, at 12:00.
Speaker: Christopher Desjardins, Post-doctoral Researcher, Department of Applied Mathematics, Science Institute, University of Iceland
Abstract: The purpose of this study was to develop a statistical model that best explains variability in the number of school days suspended. Number of school days suspended is a count variable that may be zero-inflated and overdispersed relative to a Poisson model in spite of the covariates considered. In order to properly model school suspensions, four categorical models were examined: Poisson, negative binomial, Poisson hurdle, and negative binomial hurdle models. Additionally, the probability of a student being suspended at least one day was modelled using a binomial logistic regression model. Of the count models considered, the negative binomial hurdle model had superior fit based on Akaike’s Information Criterion, Vuong’s statistic, and its predicted distribution of school days suspended. However, similar fit was found between the models based on cross-validation measures considered. Modelling the probability of a student being suspended at least once using a binomial logistic regression model with an interaction fit both the training and test data and had adequate fit based on a test of deviance. Furthermore, in situations where the data are similar to the school suspension data, i.e a zero-inflated and overdispersed count variable, using a negative binomial hurdle to model the count or a binomial logistic regression to model the probability of an occurrence may be considered.