The effectiveness of Liu-Estimator in predicting poverty levels in Indonesia: Comparative study and application of simulation

Several logistic regression studies have frequently encountered problems where the data contains multicollinearity. This condition negatively impacts parameter estimation. The Least Absolute Shrinkage and Selection Operator (LASSO) and the Liu Estimator are methods that can be employed to eliminate multicollinearity among independent variables. The objective of this research is to evaluate the effectiveness of the Liu Estimator in removing multicollinearity compared to the Maximum Likelihood Estimator (MLE) and LASSO, using simulation data for sample sizes of n = 25, 50, and 75 in a binary logistic model with 6 independent variables and a multicollinearity level of 0.99 between independent variables. The method with the smallest MSE and AIC values is used for comparison. The results of this research indicate that the Liu Estimator method is more effective than MLE and LASSO in addressing multicollinearity issues, as it produces lower MSE and AIC values for all sample sizes studied. The application of the Liu Estimator to poverty levels in Indonesia reveals that population density, the Human Development Index (HDI), average years of schooling, per capita expenditure, literacy rate, and life expectancy significantly influence the poverty rate in Indonesia.