Underdispersion Regresi Poisson Aplikasi Pada Kasus Kematian Bayi Di Provinsi Jawa Tengah

Abstract

In a Poisson regression analysis, the response variable (Y) must meet the assumptions equidispersion (variance value is equal to the mean). However, in real data often occurs overdispersion / underdispersion (variance value is not equal to the mean). Overdispersion is the emergence of greater diversity in the set of data compared with the expected variance based models. The implication, for the correct model, the value of Pearson Chi-square statistic divided by the degree of freedom will to be equal to 1. Overdispersion occurs if the value exceeds 1, and underdispersion occurs if the value is less than 1. Overdispersion / underdispersion cause the resulting model be less precise. One alternative to overcome that by replacing assuming a Poisson distribution with a more flexible distribution. The purpose of this article is to show the condition underdispersion Poisson regression modeling on Infant Mortality Case By Regency / City in Central Java province. The predictor variables used are the number of health facilities health centers in each district / city, village midwives availability ratio in each district/city, the percentage of births assisted health personnel, the baby's health care coverage, and number handling obstetric complications.