Identifying need for new factors in EVOP
Abstract
Evolutionary operation (EVOP) was first introduced by George Box (1957) and was successfully applied usually for limited time periods, in a number of operations. Until recently, EVOP was run in a manual fashion. EVOP teams had daily or weekly meetings to determine whether the EVOP should shift to a new phase by changing the levels of the factors in the current EVOP and/or by changing to a new set of factors. Recent development (Holmes and Holmes, 2002) of an automated EVOP approach has required that a replacement be found for the periodic team meetings to handle the phase changes in EVOP. This article describes a method for determining when new factors need to be incorporated in to the EVOP design. The approach we describe caitalizess on the time series aspect of EVOP. The method capitalizes on the time series aspects of EVOP. The method compares, at each treatment, a time order-independent variance of the response variable with one that is time order dependent. If these two estimates are significantly different, then factors other than those included in the current EVOP may be causing changes in the response variable. Thus if this condition occurs, one should investigate to see whether other factors should be included in the experiment or stabilized to eliminate confounding Examples of the proposed method are given.