Abstract
Although numerous studies on Web effort estimation have been carried out to date, there is no consensus on what constitutes the best effort estimation technique to be used by Web companies. It seems that not only the effort estimation technique itself can influence the accuracy of predictions, but also the characteristics of the data set used (e.g., skewness, collinearity; Shepperd & Kadoda, 2001). Therefore, it is often necessary to compare different effort estimation techniques, looking for those that provide the best estimation accuracy for the data set being employed. With this in mind, the use of graphical aids such as boxplots is not always enough to assess the existence of significant differences between effort prediction models. The same applies to measures of prediction accuracy such as the mean magnitude of relative error (MMRE), median magnitude of relative error (MdMRE), and prediction at level l (Pred[25]). Other techniques, which correspond to the group of statistical significance tests, need to be employed to check if the different residuals obtained for each of the effort estimation techniques compared come from the same population. This chapter details how to use such techniques and how their results should be interpreted.