An empirical investigation into the effect of slice types on slice-based cohesion metrics

ContextThere is a debate about whether end slice or metric slice is preferable for computing slice-based cohesion metrics. However, up till now, there is no consensus about this issue. ObjectiveWe aim to investigate the relationship between end-slice-based and metric-slice-based cohesion metrics and then determine which type of slice is preferable for computing slice-based cohesion metrics. MethodWe used forty widely used open-source software systems to conduct the study. First, we compute the baseline values for end-slice-based and metric-slice-based cohesion metrics. Then, we investigate their relationships with module size. Finally, we employ correlation analysis and principal component analysis to analyze the relationships between end-slice-based and metric-slice-based cohesion metrics. ResultsEnd-slice-based and metric-slice-based cohesion metrics have similar baseline metric values. Furthermore, they exhibit a similar negative correlation with module size. In particular, the results from correlation analysis and principal component analysis reveal that they essentially measure the same cohesion dimensions. ConclusionFrom the viewpoint of metric values, there is little difference between end-slice-based and metric-slice-based cohesion metrics. We hence recommend choosing end slice for computing slice-based cohesion metrics in practice, as extra cost involved in data collection could be avoided.