ROBUST ESTIMATORS FOR GENERALIZED LINEAR MODEL IN POISSON DISTRIBUTION

• Main Author: Celline
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/49724
• Institution: Institut Teknologi Bandung

Generalized Linear Model is a generalization of classic linear regression model. Common distribution used for discrete data is Poisson distribution. However, this distribution is deficient to use in an overdispersion data, so that Negative Binomial distribution is used. The estimator used in Generalized Linear Model is Maximum Likelihood, but it is sensitive to outliers. This research discussed two new more robust estimators, M-estimator based on transformation (MT) and weighted M-estimator based on transformation (WMT). The robustness of these estimators was compared by ????????????? and absolute Deviance Residual. Furthermore, the Covid-19 data, data on many positive Covid-19 and weather data was modeled, to see whether weather condition affect many positive Covid-19 in Indonesia. Weather variables used are average temperature, average humidity and length of sunshine. After modeling, a robust estimator is obtained, ML and MT estimators for Negative Binomial. From the model, it is found that the increase in temperature, humidity and length of sunshine reduces the average of many positive Covid-19 in Indonesia. Concluded that there are factors that affect the ability of the virus to maintain an independent existence (virus viability), that is temperature and humidity. This means that, at a higher temperature and relative humidity level, the viability of the virus is faster (the virus will die faster).