Over the last decades spatial econometrics models have represented a common tool for measuring spillover effects across different geographical entities (counties, provinces, regions or nations). Unfortunately, no one has considered that when these entities share common borders but obey to different institutional settings, ignoring this feature may induce misleading conclusions. In fact, under these circumstances, and if institutions do play a role, we expect to find spatial effects mainly within" entities belonging to the same institutional setting, while the "between" effect across different institutional settings should be attenuated or totally absent, even if the entities share a common border. In this case, relying only on geographical proximity will then produce biased estimates, due to the composition of two distinct effects. To avoid these problems, we derive a methodology that partitions the standard contiguity matrix into within and between contiguity matrices, allowing to separately estimate these spatial correlation coefficients and to easily test for the existence of institutional constraints. In our empirical analysis we apply this methodology to Italian Local Health Authority expenditures, using spatial panel techniques. Results show a strong and significant spatial coefficient only for the within effect, thus confirming the importance and validity of our approach.