HANDLING OF OVERDISPERSION CASES IN MORBIDITY DATA IN SELUMA REGENCY

The problem of overdispersion as a violation of the assumption of equidispersion in Poisson regression is generally caused by sources of unobserved heterogeneity, missing observations on predictor variables, outliers in the data, errors in the specification of the bridging function, and many observed values that are zero. The purpose of this study is to find out the right model and the variables that affect data that occurs overdispersion and excess zero in the case of the number of days of disruption at work, school, or other daily activities due to health complaints. The methods used were Poisson Regression, Negative Binomial Regression, Hurdle Poisson Regression, Zero Inflated Poisson Regression, Zero Inflated Negative Binomial Regression, and Hurdle Negative Binomial Regression. The data used were morbidity taken from data on the number of days of disruption at work, school or other daily activities due to health complaints in Seluma district, Bengkulu Province. It was found that the best model is Zero Inflated Negative Poisson with the smallest Akaike Information Criterion (AIC) value of 1620.609 and the variables that have a significant effect on the log model and the logit model are marital status and work variables.