Distributions of centrality on networks (Q2195685)

This paper studies agent centrality in a large random network, where links of the network are generated independently and the link probabilities are governed by a matrix \(\bar{A}\) with element \(\bar{A}_{ij}\) being the probability of having an edge \(ij\). For networks with \(n\) agents, suppose \(A(n)\) is a sequence of random networks that has non-vanishing spectral gap and has large enough eigenvalues. For \(\varepsilon>0\), it is shown that with probability at least \(1-\varepsilon\) the matrix \(A(n)\) has a unique largest eigenvalue \(\lambda_1(n)\) and the eigenvector centralities \(v(A(n))\) satisfy \(\|v(A(n))-v(\bar{A}(n))\|_2\le \varepsilon\) for all large \(n\). Moreover, suppose \(\phi(n)\) is a sequence of constants such that \(\limsup_n\phi(n)\bar{\lambda}_1(n)<1\). It is shown that with probability at least \(1-\varepsilon\) the vectors of Katz-Bonacich centralities satisfy \(\|c(A(n),\phi(n))-c(\bar{A}(n),\phi(n))\|_2\le\varepsilon\sqrt{n}\) for all large \(n\). These results are applied to relate homophily and inequality of outcomes, the influence of link probabilities on homophily, as well as relative inequality of outcome called Lorenz dominance. Some discussions about the segregation and geography in economy behavior through the developed framework are presented.