PNSQC 2013 - Douglas Hoffman
summary
Argues common software metrics are misleading: measurement is hard, metrics invite gaming, and math is often misapplied. Shows concrete examples: defect density halved by doubling the codebase as a manipulation, the industry rule 110 Fortran lines per function point as an averaged surrogate with wide variance, and a case where an 18‑month project hits the predicted bug‑find curve every week (suggesting fabrication). Recommends using metrics as observational signals, personal feedback, and building explicit models that link measures to the attributes they intend to represent.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 0:30Topic
- 1:40Superstitions
- 3:37Measurement Myths
- 5:37Models
- 7:50Metric
- 8:50Measurement
- 13:25Function Points
- 17:00Measuring
- 20:05Observation vs Control
- 28:05MTTF
- 33:24Estimates
- 37:53Unreasonable Expectations
- 48:58Nominal Scale
- 51:13Ordinal Scale
- 52:53Intervals
- 53:33Ratio Data
- 55:28FMEA
related talks
49mMetrics for Architects
01h 24mPNSQC 2014 - Dr. Richard Turner
18mBad Metrics and Very Bad Decisions
01h 00mMeasure, Then Build
41mComputer Science, Software Engineering, and the Scientific Method
37mDantotsu Radical Software Quality Improvement
49mPragmatic Approach to Architecture Metrics
44mBehavioral Software Engineering