Models which use smoothing techniques such as locally weighted regression to identify and represent possible nonlinear relationships between the explanatory and response variables as an alternative to considering polynomial terms or searching for the appropriate transformations of both response and explanatory variables.