The Mahalanobis-Taguchi system (MTS) is a new collection of methods for diagnosis and forecasting using multivariate data. A number of case studies using MTS have been published. However, there remain some unsolved problems about MTS. One of these problems is a countermeasure against multicollinearity. This paper deals with this subject. Firstly, we point out the theoretical defect of MTA which is a countermeasure applied in MTS. Secondly, we propose two distances in place of MTA. Furthermore, we pay attention to the numerical calculation in MTS and the bias of Mahalanobis distance for future item from the "normal" group.
In Taguchi method, several signal-to-noise ratios are utilized in order to evaluate the performance of system. In this paper, statistical tests of the equality of signal-to-noise ratios for two kinds of binary input-output systems having the same objective function are discussed. It is shown that the problem is equivalent to the test of the equality for two odds ratios between two 2×2 contingency tables. The problem is formulated through the log-linear model for three-way contingency table. As a result, simple procedures whose sizes are close to the nominal level are exhibited.