Influence of single observations on the choice of the penalty parameter in ridge regression

摘要: Penalized regression methods, such as ridge regression, heavily rely on the choice of a tuning, or penalty, parameter, which is often computed via cross-validation. Discrepancies in value penalty parameter may lead to substantial differences coefficient estimates and predictions. In this paper, we investigate effect single observations optimal tuning showing how presence influential points can dramatically change it. We distinguish between "expanders" "shrinkers", based their model complexity. Our approach supplies visual exploratory tool identify points, naturally implementable for high-dimensional data where traditional approaches usually fail. Applications real examples, both low- high-dimensional, simulation study are presented.