Statistical Analyses of Scatterplots to Identify Important Factors in Large-Scale Simulations, 1: Review and Comparison of Techniques

摘要: The robustness of procedures for identifying patterns in scatterplots generated Monte Carlo sensitivity analyses is investigated. These are based on attempts to detect increasingly complex the under consideration and involve identification (i) linear relationships with correlation coefficients, (ii) monotonic rank (iii) trends central tendency as defined by means, medians Kruskal-Wallis statistic, (iv) variability variances interquartile ranges, (v) deviations from randomness chi-square statistic. following two topics related these considered a sequence example large model two-phase fluid flow: presence Type I II errors, stability results obtained independent Latin hypercube samples. Observations analysis include: errors unavoidable, 11errors can occur when inappropriate used, physical explanations should always be sought why statistical identify variables being important, important tends stable