Theory 89/Selection & Bias/Reliability engineering
Failures Are Selected Out
The claim
Analyses of failures are biased toward components that survived long enough to fail visibly, so the riskiest designs can appear the safest.
01Mechanism
Components that fail early are removed from service and never observed; the surviving population is enriched for robust units.
02Prediction
Retrospective failure rates should understate risk for early-failing designs.
03Refutation
Failure rates matching prospective cohorts would remove the bias.