Results of Applying the Personal Software Process

Author(s): Pat Ferguson, Watts S. Humphrey, Soheil Khajenoori, Susan Macke and Annette Matvya
Venue: IEEE Computer Society, Computer 30 (5)
Date: 1997


This paper summarizes the results from studies of three companies (Advanced Information Services, Inc., Motorola Paging Products Group, and Union Switch & Signal Inc.) The studies of each company include several projects with those at AIS being new development while the projects at the other two companies involve maintaining existing projects.

Data from seven projects at AIS is used to demonstrate how the PSP process causes the schedule estimating error average to drop from 394% to -10.4%. Additionally, the paper claims that use of PSP results in defects being removed earlier in the development process with defects found in acceptance testing decreasing from 0.72 defects per 1,000 lines of code to 0.17 defects per 1,000 lines of code. The collected data from Motorola claims that PSP causes reduction in integration and testing cost by 44% with removal of more than 80% of the defects before the first test. Finally, the study of US&S shows that 57 defects were found in testing and 0 defects found when the software was in use for five projects. This seems to indicate that using PSP allows for fewer errors in developed software.

While the results presented by this study are impressive in terms of reduction in the number of defects and improvements in the accuracy of estimating time needed for development, there are several factors which seem to make the study questionable. First, there is a significant lack of control data to compare the results to. Secondly, many of the studies indicate that engineers volunteered to participate in the study. This may skew the results because it seems that engineers who embrace and adopt PSP would be more likely to participate in the study. Additionally, the data presented in the results were collected by the individual engineers as they worked. It is therefore reasonable to question the accuracy of these values as the engineers may unintentionally manipulate the data in order to suggest great improvement or productivity.