7 Septiembre



Conferencia magistral

Robust estimation and outlier detection with spatial correlation coefficients

The most widely used measure of spatial autocorrelation, introduced by Moran (1950), is based on the calculation of the covariance between a variable and its spatially lagged values. Such a measure provides a good summary and suggests reliable hypothesis testing procedures when the data-generating process is stationary over space and the variable is, at least approximately, Gaussian. However, in many empirical situations, data are characterized by the presence of non-normal distributions which may take the form of skewness and/or kurtosis deviations from normality, or the presence of a large number of outliers. The Moran test is not robust to the violation of the normality assumption, providing scope to introduce some alternative hypothesis-testing procedures. In this paper, we suggest a robust version of the spatially lagged value and, making use of it, we then derive a class of alternative robust measures of spatial correlation. Furthermore, we present the robust version of the local Moran test. We study the influence funtion of the alternative measures and, through a series of Monte Carlo experiments, we study their power compared to the traditional Moran test,. Finally, we present two illustrative examples, respectively based on regional and on point data

Giuseppe Arbia


Sesiones arbitradas


Presentación de trabajos/ premio joven investigador senior




Conferencia magistral

Guillermo Avellán


Foro: Reactivación Económica del país