Experimental evaluation of agreement among programmers in applying the rules of cohesion

The cohesion or strength of a component of a software system is an indicato r of its maintainability. The most popular way-as evidenced from coverage i n textbooks-of determining the cohesion of a component is a set of rules de veloped by Stevens, Myers, Constantine acid Yourdon in the early 1970s, Usi ng Stevens et al.'s approach, a component is assigned one of seven levels o f cohesion. This paper presents the results of an experiment analysing thes e rules of cohesion, The experiment, using fifteen computer science graduat e students as subjects, was conducted to assess whether Stevens et al.'s ru les were objective, i.e,, whether there is a better-than-chance agreement i n the cohesion levels assigned by different programmers, The data, though p reliminary due to the small sample size, indicate that there is a significa nt variation in the cohesion levels assigned by them even though the subjec ts were assessed to have understood the concepts well. This decoupling betw een the understanding of the concepts of the scale and of the use of the sc ale in proper fashion is intriguing and deserves further study, The results also raise questions about the precision of the material taught in the sof tware engineering curriculum, Copyright (C) 1999 John Wiley & Sons, Ltd.