Search papers, labs, and topics across Lattice.
This paper introduces a method for releasing network connectedness indices, which quantify assortative mixing across node attributes, while preserving edge-adjacent differential privacy. The approach addresses the high global sensitivity and composition issues inherent in standard privacy techniques by adding noise to node attributes, analytically debiasing downstream statistics, and adding a second layer of noise to protect edge presence. The authors prove consistency and asymptotic normality of their estimators and demonstrate the method's effectiveness on simulated and real-world networks.
Releasing privacy-preserving network statistics is now more practical, thanks to a new method that overcomes the limitations of standard techniques on connectedness indices.
Researchers increasingly use data on social and economic networks to study a range of social science questions, but releasing statistics derived from networks can raise significant privacy concerns. We show how to release network connectedness indices that quantify assortative mixing across node attributes under edge-adjacent differential privacy. Standard privacy techniques perform poorly in this setting both because connectedness indices have high global sensitivity and because a single node's attribute can potentially be an input to connectedness in thousands of cells, leading to poor composition. Our method, which is straightforward to apply, first adds noise to node attributes, then analytically debiases downstream statistics, and finally applies a second layer of noise to protect the presence or absence of individual edges. We prove consistency and asymptotic normality of our estimators for both discrete and continuous labels and show our method works well in simulations and on real networks with as few as 200 nodes collected by social scientists.
