Model misspecification in stochastic-actor oriented models

V. Amati${}^a$

${}^a$University of Milano-Bicocca

V. Amati

Stochastic actor-oriented models (SAOMs) [1] are a widely used framework for analyzing panel network data, which involves repeated observations of a network at discrete time points. The specification of SAOMs is guided by substantive theories and hypotheses regarding the mechanisms that lead to changes in networks and behaviors over time. Similar to standard regression models [2], misspecification in SAOMs can occur in several ways: relevant effects may be omitted from the model, irrelevant effects may be incorrectly included, or the functional form of an effect may be inadequately specified. Each of these scenarios can distort parameter estimates, undermine the validity of inferences, and result in misleading substantive conclusions.

Despite its practical significance, model misspecification in SAOMs has received limited attention in the literature. We aim to address this gap by examining multiple misspecification scenarios within the SAOM framework. By drawing on diagnostic approaches from the theory of generalized linear models and econometrics, we investigate how established tests and common model-checking practices can be adapted to detect misspecification in SAOMs. We evaluate the performance of these diagnostic tools across various types of misspecifications and data conditions, assessing their sensitivity, reliability, and practical utility

Keywords: Model misspecification, Panel network data, Stochastic actor-oriented model