Networks of Symptoms and Exposures

We present some novel methods for analyzing and visualizing data from medical studies using methods originally developed for the study of social networks. The methods are based on spectral (eigendecomposition) properties of networks, in particular the so-called Normal spectrum. Among the many desirable properties of this spectrum is the natural handling of bipartite (2-mode) networks through negative eigenvalues, the clustering properties related to positive eigenvalues, and the relationship to the chi-squared measure of dependence in contingency tables.