Notes from the Author
- Creator: A. Gunes Koru
- Date: February 23, 2007
- Contact: gkoru AT umbc DOT edu Phone: +1 (410) 455 8843
- A. Gunes Koru, Dongsong Zhang, and Hongfang Liu, “Modeling the Effect of Size on Defect Proneness for Open-Source Software”, Predictive Models in Software Engineering Workshop, PROMISE 2007, May 20th 2007, Minneapolis, Minnesota, US.
- Abstract: Quality is becoming increasingly important with the continuous adoption of open-source software. Previous research has found that there is generally a positive relationship between module size and defect proneness. Therefore, in open-source software development, it is important to monitor module size and understand its impact on defect proneness. However, traditional approaches to quality modeling, which measure specific system snapshots and obtain future defect counts, are not well suited because open-source modules usually evolve and their size changes over time. In this study, we used Cox proportional hazards modeling with recurrent events to study the effect of class size on defect-proneness in the Mozilla product. We found that the effect of size was significant, and we quantified this effect on defect proneness.
- The full paper can be downloaded from A. Gunes Koru’s Websitehttp://umbc.edu/~gkoru by following the Publications link or from the Web site of PROMISE 2007.
- This data set is used to create a conditional Cox Proportional Hazards Model
# id: A numeric identification assigned to each separate C++ class(Note that the id’s do not increment from the first to the last data row)
# start: A time infinitesimally greater than the time of the modification that created this observation (practically, modification time). When a class is introduced to a system, a new observation is entered with start=0
# end: Either the time of the next modification, or the end of the observation period, or the time of deletion, whichever comes first.
# event: event is set to 1 if a defect fix takes place at the time represented by ‘end’, or 0 otherwise. A class deletion is handled easily by entering a final observation whose event is set to 1 if the class is deleted for corrective maintenance, or 0 otherwise.
# size: It is a time-dependent covariate and its column carries the number of source Lines of Code of the C++ classes at time ‘start’. Blank and comment lines are not counted.
# state: Initially set to 0, and it becomes 1 after the class experiences an event, and remains at 1 thereafter.
See above section “Relevant Information” for further reference.
A. Gunes Koru, Dongsong Zhang, and Hongfang Liu. 2007. Modeling the Effect of Size on Defect Proneness for Open-Source Software. In Proceedings of the Third International Workshop on Predictor Models in Software Engineering (PROMISE ‘07). IEEE Computer Society, Washington, DC, USA, 10-. DOI=10.1109/PROMISE.2007.9http://dx.doi.org/10.1109/PROMISE.2007.9