Spatial, Cultural, and Ecological Autocorrelation in U.S. Regional Data

Positive autocorrelation implies that proximate observations take on similar values. “Proximate” can be defined in many different dimensions. In a cross-section of U.S. regions, it can be defined using physical distance, cultural similarity, ecological similarity, or using frequency and intensity of interaction, such as migration or commuting relationships. Autocorrelation of regression residuals presents well-known problems in least-squares estimation, but autocorrelation also provides useful information for exploratory data analysis and model specification. The paper shows that autocorrelation is widespread in U.S. regional data.