A STATISTICAL INFERENCE ANALYSIS ON CRIME RATES IN PENINSULAR MALAYSIA USING GEOGRAPHICAL WEIGHTED REGRESSION

Abstract

Geographical Weighted Regression (GWR) is used to improve decisionmaking in spatial analysis. Instead of the Ordinary Least Square (OLS) regression method that gives a single estimated parameter, the GWR method can provide unique estimated parameters in each location. This study aims to conduct a formal statistical inferential framework on the violent crime rate using the GWR. This analysis discovers the geographical distribution and pattern of criminal cases in Peninsular Malaysia using the average crime rates from 2000-2009, with focus on on violent crime. The comparison of OLS regression, known as Multiple Linear Regression (MLR) with the GWR method, was done to show that GWR was the best model. The GWR output suggests that about 30% of districts showed a significant correlation between violent crime and non-citizen rates. These findings contradict the result from the MLR model, also known global model. The global model could not create any other connection to explain the lack of parameter-location correspondence. Finally, the importance of local relationships in crime studies is necessary to understand the actual crime rate.