Software development effort estimation should be adjusted if requirements are added or changed. The results of a questionnaire on effort estimation present what actually happens in practice. All 52 respondents had previous experience with effort estimation.
The mean magnitude relative error (MMRE) in estimation is 29 percent, with the top four reasons for error involving requirements. Table 7 is most revealing. No re-estimation was done in 27 percent of the projects, despite these projects having to deal with change requests. Twenty-three percent of the projects, which also had to deal with change requests, performed re-estimation; this did not impact project planning. As the authors emphasize, under such circumstances, comparing estimated and actual effort makes no sense. The majority of respondents perceived their estimates to be good; however, several explained that management deliberately restricted estimates, so the estimation error would have been less without management interference. The authors conclude that the reported MMRE of 29 percent is not an actual assessment of accuracy, and re-estimation and re-planning should take place in projects exposed to change.
A quarter of the projects were reported as agile. Since agile methods such as Scrum emphasize re-estimation and re-planning, it is unfortunate that the analysis of Table 7 does not distinguish between agile and nonagile projects. Despite this criticism, I strongly recommend this paper to those managing or estimating software development effort and those researching the software development process.