Could enough samples be more important than better designs for computer experiments?

A study was conducted to compare fifteen approaches to improve Latin hypercube designs for computer experiments, based on simulation tests and statistical analyses ANOVA. Kriging models were employed to approximate twenty test functions. Validation at 5000 or 10,000 points was conducted to find prediction errors. The results show that there are statistically significant differences between the approximate results of employing different designs, but more often the difference is not significant. In most cases, the number of runs or the sample size has stronger impact on the accuracy than do different designs. When the dimension is low, a small size increment can often reduce more error than do "better designs". To get the desired precision by one-stage method, enough samples may be needed regardless what design is used. Sample size determination may need much more attention for computer experiments.