The detection of social positions and roles is one of the key tasks in the structural analysis of complex social systems. Traditionally, structural analysis methods are able to identify local and global roles and positions in networks with one or more types of relations. However, the classical way to deal with a multi-layer network consists on either (a) individually identifying positions in each layer at a time, then studying their relationships or (b) flattening the social network into a mono-relational graph and apply some of the existing blockmodeling techniques. While both previous approaches are useful to understand the structure of the social network, none of them gives us information about positions defined by the collection of individuals' relations at each of the different layers and the overall structure of the multi-layer graph. The leading cause is the information lost during the post-aggregation of single-layer positions or because of the graph flattening transformation. We extend the traditional indirect approaches for blockmodeling to the analysis of social roles and positions in multi-layer networks. Our method, based on recent blockmodeling developments, substitutes the traditional relational matrices by an extended version. In the new socio-matrix the relations between two actors are substituted by relations between actors and sets of actors, resulting in two-mode data where each row represents an actor, and each column a subsetting of the network population. In second term, we substitute the node- to-node measures by multi-layer measures, as they can always be computed as function of a graph vertex actor and a set of vertices. Finally, positions are computed by clustering the rows of the extended socio-matrix, while roles are computed by comparing the distributions of their values. Differently from both traditional approaches, this new computation does not consider only the regular intra-layer information, but also the inter-layer one: all the relation types are considered together, not one at a time and not after merging them. As an example, we could identify as positions actors acting as a bridge of information between disconnected actors in a layer by simply calculating a binary measure that ``is 1 if an actor is an inter-layer bridge that connects two disconnected actors in one of the layers, and 0 otherwise''. We used our framework and the new set of extended measures, to find positions and roles of interest in several multi-layer social networks representing socio-technical communities, where users communicate using different technological channels. The partitions of actors identify meaningful inter-layer positions not captured by other indirect or direct approaches. We also relate these partitions with the users that are influenced by them, that is, the columns of the extended socio-matrix. Finally, we plan to apply this new method to other social networks represented by different graph structures, like hypergraphs, where the comparison between actors and sets of actors avoids loosing information about the network structure – using, for example, the hyperlinks as target sets of actors for the columns.