A term used in regression analysis for those observations that have an extreme value on one or more explanatory variables. The effect of such points is to force the fitted model close to the observed value of the response leading to a small residual.