Pemodelan Statistika Atas Dasar Sebaran T-Student

Abstract

In this paper we have investigated, by means of simulation, properties of estimates of regression parameters based on t distribution. The best estimate was obtained by comparing the mean square errors resulted from fitting t distribution with various degrees of freedoms. We used simulated data generated from symmetric distributions including uniform, normal, logistic, Laplace, Student-t, and Tukey 's Symetric lambda distribution. C?'e also used the model to analyze several real data. The results showed that, in general, the best estimate was given by the t distribution with 5 degrees of Peedom. However, we have not studied the distributions of the estimates which is essential for inference.