PROPORTIONAL EFFECT OF OUTLIERS ON OVER-DISPERSION

Abstract

The impact of outlier on analysis of time series data in causing over-dispersion was examined. The problemof overdispersion is central to all General Linear Models (GLM's) having discrete responses. If the estimated dispersion after fitting is not near the expected values, then the data may be over dispersed. One of the causes of overdispersion is outlier. Outlier is a data which is unusual with respect to the group of data in which it is found. In this paper, data were simulated based on poison model using SPSS and first analysed to see whether the estimated parameters is unbiased of the fixed parameters. Thereafter, two different values of outliers, 10's and 20's were introduced to different percentages of the generated data and then analysed using the STATA package to observe the effect of the outliers being introduced on the data for small, moderate and large samples. The data simulated were replicated 300 times for all categories. The averages of the results were computed. The results showed that the higher the percentage of outliers the more the over-dispersion occurs in the models and the larger the sample size
the less the over-dispersion.