Identifying the important factors in simulation models with many factors

Simulation models may have many parameters and input variables (together called factors), while only a few factors are really important (parsimony principle). For such models this paper presents an effective and efficient screening technique to identify and estimate those important factors. The technique extends the classical binary search technique to situations with more than a single important factor. The technique uses a low-order polynomial approximation to the input/output behavior of the simulation model. This approximation may account for interactions among factors. The technique is demonstrated by applying it to a complicated ecological simulation that models the increase of temperatures worldwide.