Reliability of function point counts

An important characteristic of any software is its size. Frequently used metrics for measuring the size of software are source lines of code (SLOC) and function points. Lines of code are easy to count and seem unambiguous (but different definitions can cause variations of 500%). Function points are normally counted by certified professionals, which may introduce differences between counters. In this article we survey existing literature on inter-rater reliability that classically involves recounts, but recounts are hardly possible in practice. We propose multiple methods to test for differences between raters that do not involve recounts. In a case study of 311 projects and 58143 function points from a large institution we determined that function point counts are a reliable base. Using our proposed method, we did not find statistical evidence for systematic differences between counters, and recounts were not necessary for that. So in this organization, the function point counts are a reliable data source for IT management.