Software effort estimation accuracy: A comparative study of estimations based on software sizing and development methods

The number of project failures and those projects completed over cost and over schedule has been a significant issue for software project managers. Among the many reasons for failure, inaccuracy in software estimation---the basis for project bidding, budgeting, planning, and probability estimates---has been identified as a root cause of a high percentage of failures. Poor estimates have not only led projects to exceed budget and go over schedule but also, in many cases, to be terminated entirely. The ability to accurately estimate software development projects changes as newer methodologies replace old ones. Research in this area has been sporadic and there has been little research into the root cause of estimation inaccuracy. The purpose of this study was to determine if there was a difference in software effort-estimation accuracy between the development methodologies of waterfall and incremental methodologies, and the newer agile development methodology. Specifically, the impact of using source line of code (SLOC), function point (FP), or story-point sizing methods were explored across three common development methodologies of waterfall, incremental, and agile. The International Software Benchmarking Standards Group (ISBSG) database was analyzed using Statistical Package for the Social Sciences (SPSS) 15.0 software and examined the relationship between the independent variables (IVs) (estimation parameters) and the dependent variable (DV) estimate accuracy of software projects. Though the hypotheses were not able to be tested, the research question was explored by testing the independent variables through one-way ANOVAs. This resulted in no difference in effort accuracy for the two sizing methods SLOC and FP while there were not enough statistical samples of story-point to determine effect. Similarly, there was no difference in effort accuracy for the two development methods waterfall and agile while there were not enough statistical samples of incremental to determine effect. While limitations existed using the ISBSG database, it is recommended that the database becomes a common tool for future research.