A new approach to testing spatial autocorrelation by Moran's I statistics

Abstract

Moran's I statistics, both global and local, are arguably the most widely-used methods for testing spatial autocorrelation in areal data. Statistical testing by the I statistics generally depends on their asymptotical normality, which is known to be assumable when the number of zones in a study region is sufficiently large and the target variable, X, satisfies given conditions. This study examines potential impacts that varying probability distributions of the target variable X has on the normality of the I statistics through intensive Monte Carlo simulation. A new simulation-based approach to testing spatial autocorrelation for discrete target variables is also proposed.