Are there different convergence local behaviors hidden under the regional level?

The large literature about economic convergence explores widely different ways of measure convergence as well as sophisticates and improves the econometrical techniques. Contradictory the number of contributions that analyze how relevant is the data spatial desegregation is reduced. Our hypothesis is that the spatial unit of analysis definition could affect significantly the convergence conclusions. The assumptions of the Neoclassical models of economic growth might not to be affected by the spatial desegregation but New Economic Geography framework pointed out the core-periphery behaviors and if the spatial unit is too much aggregated it could make diffuse this processes. To study it in this paper a multilevel perspective is proposed. That allows differentiating possible different local behaviors inside regions. It is applied to the U.S. economy: States and counties. The results show high intra class correlation indexes that indicate that there exists a significant variance inside of the States. A general behavior of convergence is observed coherent with previous studies but we can indentify some States that present internal behaviors of divergence or significant changes in the speed of convergence.