A consideration of the impact of interactions with module effects on the direct measurement of subjective software attributes

Decision support approaches for software development frequently rely on the ability of experts to measure subjective attributes consistently on an ord inal scale. Examples include. decision approaches concerning alternative so ftware architectures that are needed to optimise maintenance effort; approa ches to achieve development goals which rely on Bayesian Belief Networks; p ricing decisions for maintenance contract tendering that use direct measure ment of maintainability; and pricing decisions concerning effort estimation that require measurement of 'complexity' attributes. In addition the valid ation of prediction systems and objective indirect measures for subjective attributes (e.g. maintainability, cohesion) require that observers can dire ctly measure the attributes consistently However, intuition and some anecdo tal evidence suggest that during modular effort estimation there may be mod ule effects that lead to under estimation A Bayesian inference procedure ca n enable an assessment of whether the consistency of measurement of a modul ar attribute may be influenced by module effects. For example whether the c hance of correctly classifying a modular attribute might vary with module l ength. This study examines two data sets one taken from a cohesion experime nt and the other for a maintainability experiment. In so doing, evidence th at module length interacts with the chance of correctly ;classifying mainta inability and cohesion is inferred These interactions show that ii is neces sary for chose who undertake direct measurement of modular attributes to be made aware of the potential of unsolicited module effects to influence mea surement consistency.