Pampering the Client: Calibrating Vehicle Parts to Satisfy Customers

Abstract

We present in this paper a statistical methodology to address the following industrial problem. Car manufacturers haveto calibrate their vehicles in order to reach a level of quality which is acceptable to the customer. We consider here thespecific case of a gear-box. Our study relies on a dataset consisting of evaluations by 507 testers of 28 configurations,each described by 12 physical parameters. We suggest a procedure for selecting and calibrating the physical parameterswhich have an impact on the evaluations. Our procedure consists of two steps. We first compute the regularization pathof an L1 – penalized logistic likelihood from which we extract an increasing sequence of models. In the second step of ourprocedure, we apply the BIC criterion to select a model in the sequence obtained in the first step. We provide a simplenumerical procedure for this approach and discuss its application to the data. This article is accessible to readers with atleast an intermediate knowledge of statistics; previous exposure to logistic regression and the principles of model selectionwould be useful, although not strictly necessary.