Semiparametric residuals and analysis for a scleroderma clinical trial

Abstract

Residual analysis is useful for assessing how well the fitted regression model explains the observed data, including the impact and significance of an observation on the overall results. Such residual analysis works well when a fully parametric model is assumed. However, its usefulness in evaluating goodness of fit for semiparametric models is not well studied. In this article, we propose various definitions of residuals for a semiparametric regression model and apply them to a multi-center two-armed clinical trial for Scleroderma patients. We show we can gain interesting insights on the longitudinal data using our proposed residuals.