Chimeric generation of null models for multivariate analysis of network data

In this study three topics about social networks analysis converge: - Proper characterisation of multidimensional networks as relational tabular multivariate data that can be analysed through multivariate models. An innovation here is the focus in the dimensionality of the edges as the core feature of multidimensional (multi-layer) networks. - When multivariate models are applied to network data, it is possible to interpret the results as evidence of spillover effects. This interpretation suffers of a generalized problem of confounding between agency effects and structural and topological (homophily) effects. Part of remedy involves adoption of neutral models for null hypothesis testing. - Techniques of procedural formation of networks are often inadequate to represent complex multidimensional neutral models, since neutral models are developed to understand the effects of general homophily in multidimenstional networks. Chimera networks, that set a fixed number of nodes and then proceed to attach layers, is discussed as a not procedural technique for neutral models formation. Chimeric formation does not require knowledge of tensor algebra. Being coded as a relational database of two tables, it allows to suppress with relative ease the feature tested in the neutral model. For these reasons, we believe that, as a tool, they are particularly useful for multidimensional social networks analysis.