ABSTRACT
The formal expression is the following: Similar values between neighboring observations will give rise to a low index, whereas when the difference between neighboring values is large it will produce a low value of the index. The classical linear regression model assumes normal, exogenous and spherical disturbances. However, when we observe a phenomenon in, say, n regions, non-sphericalness of residuals may arise due to the presence of spatial autocorrelation and spatial heterogeneity among the stochastic terms, in which case the optimal properties of the ordinary least squares are lost. The most widely used measure for spatial autocorrelation is based on a general measure of spatial correlation introduced by Moran and proposed by Cliff and Ord as a test statistic for the null of uncorrelation among ordinary least squares regression residuals. The notion of spatial autocorrelation among regression residuals has also been approached by introducing various measures and hypothesis test statistics.
