An Empirical Validation of Function Point Structure and Applicability: A Replication Study

Abstract

Background: The complexity of providing accurate software size estimation and effort prediction models is well known in the software industry. Function point analysis (FPA) is currently one of the most accepted software functional size metric in the industry, but it is hardly automatable and generally requires a lengthy and costly process. Objectives: This paper reports on a family of replications carried out on a subset of the ISBSG R12 dataset to evaluate the structure and applicability of function points. The goal of this replication was to aggregate evidence about internal issues of FPA as a metric, and to confirm previous results using a different set of data. First, FPA counting was analyzed in order to determine the extent to which the base functional components (BFC) were independent of each other and thus appropriate for an additive model of size. Second, the correlation between effort and BFCs and unadjusted function points (UFP) were assessed in order to determine whether a simplified sizing metric might be appropriate to simplify effort prediction models. Methods: A subset of 72 business application projects from 2008 to 2011 was analyzed. BFCs, UFP, and effort correlation were studied. Results: The results aggregated evidence and confirmed that some BFCs of the FPA method are correlated. There is a relationship between BFCs and effort. There are correlations between UFP and inputs, enquiries, and internal files, and between BFCs and effort. Internal files and inputs are found to be correlated always, and external interface files are found to be uncorrelated with the others. A prediction model based on transactions and internal files appear to be as good as a model based on UFP. The use of some contexts attributes may improve effort prediction models. Limitations: This is an initial experiment of a research in progress. The limited size and nature of the dataset may influence the results. Conclusions: Our results might suggest an improvement in the performance of the measurement process. Simplifying FPA measurement procedure based on counting a subset of BFCs could improve measurement process efficiency and simplify prediction models.