The impact of estimation uncertainty on covariate effects in nonlinear models

Abstract

Covariate effects are a key consideration in model evaluation, forecasting, and policy analysis, yet their dependence on estimation uncertainty has been largely overlooked in previous work. We discuss several approaches to covariate effect evaluation in nonlinear models, examine computational and reporting issues, and illustrate the practical implications of ignoring estimation uncertainty in a simulation study and applications to educational attainment and crime. The evidence reveals that failing to consider estimation variability and relying solely on parameter point estimates may lead to nontrivial biases in covariate effects that can be exacerbated in certain settings, underscoring the pivotal role that estimation uncertainty can play in this context.