| Abstract: | Strong correlations have been observed between potential indices and the densities of spatial variables. The conventional null hypothesis of bivariate regression is inappropriate for testing their significance. A randomization test is proposed and is applied to 1975 US population data by state. The resulting relationship has a significant correlation, but its slope could occur frequently under the null hypothesis. The correlation is shown to be related to the spatial autocorrelation of densities by constructing arrangements with prescribed values of the modified Moran index. |
