Using Sensitivity Analysis to Create Simplified Economic Models for Regression Testing

Do, Rothermel

regression testing iterative development economic models software engineering

@conference{do:ssta-2008,
  title={Using Sensitivity Analysis to Create Simplified Economic Models for Regression Testing},
  author={Hyunsook Do and Gregg Rothermel},
  booktitle={International Symposium on Software Testing and Analysis},
  pages={51--61},
  year={2008},
  organization={{ACM}}
}

Economic models for software testing have been underdeveloped

Useful to simplify models as it makes data collection easier and cheaper

• Sensitivity analysis used here to eliminate parameters

Can be very time consuming to run all regression tests over large software

• E.g., office application suite of 1.8 million lines of code, test suite of 3128 tests, four days to run all tests...

Develops model of costs/benefits associated with regression testing

Experiment suite seems reasonable, caveat that programs are fairly small

• True test suites however

Four most significant factors in model are shown to be:

• Number of missed faults
• Cost associated with missed faults
• Revenue
• Cost associated with delayed fault detection feedback