The study of homophily in networks usually involves investigating whether nodes have a tendency to make more direct connections with similar nodes than what would happen in a random network. However, the study of homophily between nodes that are not directly connected is also of interest. Remarkably, even though there is a vast literature on long-range degree homophily, just a few works have investigated other properties, but only considering linear relations among scalar attributes of the nodes. In this work, we put forth a novel measure of long-range homophily that can deal with non-scalar and non-categorical node attributes and enables the analysis of nonlinear relations. This measure can be viewed as an extension of the usual Newman assortativity, as it is based on the comparison of the similarity of node attributes in a given network to that expected in a non homophilic network where connections are randomly made while preserving node degrees. We show the application of the new measure in a few real-world examples, some of which exhibit a non-trivial long-range homophilic behavior.