Advanced Techniques for the Application of Meta-Analysis in Non-Normal Contexts

With the increased interest in application of all kinds of software systems, understanding and controlling software development processes have gained increasing importance. One of the main goals, in this sense, is to build robust descriptive statistical models that would characterize them in the best possible way. However, individual experimental studies that form the basis of these models often produce different results. Consequently, the resulting models tend to differ substantially. One way to make sense of the vast number of accumulated study findings is to apply meta-analysis to the results of individual studies. This study is focused on meta-analytical methods for combining regression coefficients from different statistical models describing software systems. Two metaanalytical methods are described – one for situations when only correlation coefficients from individual studies are available, and the other when the whole datasets are available. Although some limitations exist when applying these methods in practice, they can provide useful experience about behavior of software systems.