M-estimators for Models with a Mix of Discrete and Continuous Parameters

A variety of parametric models are specified by a mix of discrete parameters, which take values from a countable set, and continuous parameters, which take values from a continuous space. However, the asymptotic properties of the parameter estimators are not well understood in the literature. In this paper, we consider the general framework of M-estimation and derive the asymptotic properties of the M-estimators of both discrete and continuous parameters. In particular, we show that the M-estimators are consistent and the continuous parameters are asymptotically normal. We also extend a large deviation principle from models with only discrete parameters to models with discrete and continuous parameters. The finite-sample properties are assessed by a simulation study, and for illustration, we perform a break-point analysis for the clinical outcomes of COVID-19 patients.