Summary: We focus on variable selection and propose a variable selection criterion based on the extended quasi-likelihood for joint generalized linear models with structured dispersions. Its performance is investigated through simulation studies and a real data application.