Penalty analysis offers a method to consider the individual effects of JAR ratings on Overall Liking (OAL), but does not provide a way to assess the impacts of simultaneous changes in JAR ratings on Overall Liking. Standard multiple regression is of limited use in this situation because of its strong assumptions of linearity. A form of non-parametric regression, which we will refer to as “tree-based” regression, removes that assumption and allows you to determine the combinations of the JAR ratings that have the strongest impact on Overall Liking. There are wide variety of “tree-based” regressions packages available, such as CART, MARS, KnowledgeSeeger, and SPSS AnswerTree, as well as free implementations such as part in R. This example will use MARS (multivariate adaptive regression splines) as its example [1]. This is commercial software, sold by Salford Systems (http://www.salfordsystems.com/) [2].