The case against cross-over designs in software engineering

Abstract

We are often encouraged to follow experimental procedures in undertaking software engineering studies, however we should not do so blindly as often assumptions are made as part of that process that software engineering methods artefacts and processes breach. One such example is the use of crossover designs. We consider the case where there are period by treatment interactions, (i.e where the treatments are non-commutative) and demonstrate the hazards in using a cross-over design in these cases.