One of the main challenges in empirical software engineering today lies in the aggregation of evidence. Existing summaries often use qualitative narrative approaches or ad-hoc quantitative methods, such as box plots. With these, information important for decision makers, such as existence and magnitude of a technology's effect, is hard to obtain objectively. Meta-analysis addresses this issue by providing objective quantitative information about a set of studies; however, its usefulness for software engineering studies suffers from high heterogeneity of the studies and missing information. In this paper, we describe an approach for quantitative aggregation of controlled experiments that reduces these two problems. We demonstrate the approach by aggregating available experiments to investigate whether Perspective-Based reading (PBR) improves team effectiveness compared to alternative reading approaches. We then compare the results of our aggregation to previous summaries addressing PBR's team effectiveness. Although the findings are similar, our approach is able to provide the required quantitative information objectively. Our aggregation showed that there is no clear positive effect of PBR: Inspection teams using PBR on requirements documents are more effective when compared to ad-hoc approaches, but are less effective when compared to checklists. In addition, we found strong indicators of researcher bias.

Our Assessment: A well-performed meta-analysis showing that perspective-based reading (PBR) may be better than ad hoc reading; but it is only better than checklist-based reading for design documents not requirements documents. In addition, there was a problem with lack of independence among studies suggesting that studies by independent research groups with independent materials were less favourable to PBR than dependent studies.