LA LOI DE ZIPF SOUS LE PRISME DE L’AUTO-CORRELATION SPATIALE - LES CAS DE LA CHINE ET DE L’INDE

This paper applies a rank-size model on panel data. By introducing the spatial dimension of data, it aims in taking into account spatial autocorrelation effects on the Zipf law. We apply Debarsy and Ertur (2010) methodology in order to choose the best specification between three alternative models : the spatial autoregressive model (SAR), the spatial error model (SEM) and the general spatial model with fixed effects on panel data (SARAR). Taking into account spatial autocorrelation effects does not modify the conclusions from the a-spatial model for China the Zipf’s law holds but do modify conclusions for India : in this country the a-spatial model doesn’t confirm the Zipf’s law on the long run, while this changes when we introduce spatial autcorrelation.