Pendekatan Winsor pada Analisis Regresi dengan Pencilan

Abstract

Regression analysis is one of the statistical methods to estimate the relationship between two or more variables. Ordinary least squares is generally used to estimate parameters in linear regression, but it is susceptible to outliers. Winsor approach can be used to solve regression with outliers problem, especially outliers in response variables. To find out the effectiveness of this method on handling the outliers, we utilized simulation study. Simulation was applied in various sample sizes and outlier proportions. The result showed that Winsor 3 method was the best on handling outliers, particularly on medium sample size. Winsor 3 is a method that implements iterative principal on residual standard error and uses tuning constant values by 1.345. This method was applied to the real data about the height of hills againts the climbing time as well. The method which gave the best result based on R2adj value is Winsor 3 method. This result is supported by the study using resampling technique in the real data.