Investigating training effects on software reviews: a controlled experiment

The software review/inspection task is a labour and time intensive activity. Naturally, any activity aimed at improving the performance of inspectors would be deemed favourable to both practitioners as well as to researchers. This study is motivated by previous work by Chowdhury and Land, on the effects of inspector training on inspection performance. There have been few studies in this area. One classic controlled experiment, consisting of 86 subjects was conducted. We manipulated one independent variable (training). The control group undertook no training, the other 3 treatments were process training, process training with practice and process training with worked examples. The results show practice and worked examples proceeding process training, were both very promising training approaches. They did not affect false positive identification. However, their relative benefits were less clear. We also pinpointed a few possible areas for future research from this work.