Abstract

We propose the parametric bootstrap for an undirected network model with covariates. The joint maximum likelihood estimator (JMLE) of the network models is asymptotically biased due to the well-known incidental parameter problem. We show that the bootstrap distribution is consistent for the distribution of JMLE. Importantly, the bootstrap mimics the asymptotic bias. Based on this result, we propose to construct confidence sets by standard bootstrap percentile methods without correction for the bias.