Bootstrap Estimation of Confidence Intervals of Multiple Regression Model Parameters in the Presence of Multicollinearity Using Principal Component Analysis

Abstract

Multicollinearity in multiple regression can result in biased parameter estimators and increase the risk of accepting the null hypothesis of the regression model as an insignificant variable. This study aims to determine the use of the Principal Component Analysis method in overcoming multicollinearity problems that occur in Facebook metric data and to implement the use of the PCA (Principal Component Analysis) method with Bootstrap in overcoming multicollinearity problems that occur in Facebook metric data. One method that can be used to overcome multicollinearity is principal component regression analysis. Principal component regression will produce a point estimate. To measure the accuracy of the point estimation, the bootstrap method can be used, which generates confidence intervals by resampling the data with returns. The results of this study indicate that the problem of multicollinearity in Facebook metric data can be resolved using Principal component analysis and point estimation and classical confidence intervals in principal component regression are not significantly different from the results of estimated means and bootstrap confidence intervals.