Computing Pseudolikelihood Estimators for Exponential-Family Random Graph Models

The reputation of the maximum pseudolikelihood estimator (MPLE) for Exponential Random Graph Models (ERGM) has undergone a drastic change over past 30 years. While first receiving broad support, mainly due to its computational feasibility and lack alternatives, general opinions started with introduction approximate likelihood (MLE) methods that became practicable increasing computing power MCMC methods. Previous comparison studies appear yield contradicting results regarding preference these two point estimators; however, there is consensus prevailing method obtain an MPLE’s standard error by inverse Hessian matrix generally underestimates errors. We propose replacing approximation Godambe in confidence intervals appropriate coverage rates that, addition, enables examining model degeneracy. Our also provide empirical evidence asymptotic normality MPLE under certain conditions.