Count Regression Models with Application to Caries Experience for Children Attending Lady Northey Dental Clinic in Nairobi

Abstract

Count regression models were developed to model data with integer outcome variables. These models can be employed to examine occurrence and frequency of occurrence. Four common types of count regression models are applied to caries data among children aged between three and six years attending Lady Northey Dental clinic between September, 2014 and November 2014. These models include Poisson, Negative Binomial (NB), Zero Inflated Poisson (ZIP) and Zero Inflated Negative Binomial (ZINB). The simplest count regression model, Poisson, was fitted first before considering other complex models. However, it did not perform better than its improved counterparts. The NB model proved to be the the simplest model that fits the data well according to Akaike Information Criterion (AIC), and was therefore employed to determine the important predictors of caries experience among the children. Model comparison was performed on the four models by use of AIC. Deviance values for various NB models were compared and the model with the least deviance value was considered to give a subset of best predictors of Early Childhood Caries (ECC). These predictors included age, gender, brushing frequency, feeding habit biscuits, feeding habit jam and highest education of the mother.