Using LASSO for variable selection in exponential random graph models

Exponential Random Graph Models (ERGMs) are a powerful and flexible framework for modeling network data. A fundamental challenge in ERGM estimation is the correct specification of the (sufficient) statistics that define the model structure. This paper addresses the problem of variable selection in ERGMs by making use of LASSO, a penalized estimation technique that shrinks some parameter estimates to zero, effectively selecting relevant variables. While LASSO is well established in standard regression settings, its application to ERGMs remains less explored. Here, we demonstrate how LASSO can be employed in the ERGM framework to perform variable selection and propose a ranking procedure to assess the relevance of candidate model terms.