Poisson regression analysis of malaria incidence in Jayapura, Indonesia

This study analyzes malaria risk factors using Poisson and Classical Regression Analysis. The distribution of the discrete dependent variable (malaria incidence) was checked to ascertain the necessity for Poisson regression. Goodness-of-fit test indicate that both the normal as well as Poisson distribution do not
describe the complete data set well. Cluster Analysis revealed that the data set (132 cases) contains three distinct .groups/clusters. Models were constructed per cluster using Poisson and Classical regression. Poisson regression and classical regression are comparable based on the mean absolute percentage error (MAPE). Poisson regression though is advantageous over classical regression in terms of parsimony. In count data, Poisson regression is not troubled by the stochastic assumptions that the data should satisfy. Classical regression however, could encounter problems due to skewness, nonnegative value of the response, and nonconstant variance inherent to discrete random variables. The paper also illustrates a practical consideration in analyzing data from a heterogeneous group.