Week 9: What does cancellation cost?

People who leave a queue before being served disappear from its wait-time records — this lecture asks what that does to every number the course has computed so far.

19 April 2027

Every average, median and outlier calculation so far has used one dataset: tickets that were eventually served. This lecture asks about the tickets that weren’t — the people who took a number and left before it was called — and what leaving them out does to everything measured before.

Outline#

• reneging: the specific term for leaving a queue before being served, distinct from never joining it at all
• survivorship bias applied to a queue: if you only measure the wait times of people who stayed, you are conditioning on reaching service; this group may differ from everyone who arrived, but the direction and size of the difference require evidence
• why cancellation looks like good news in the Office’s own records (fewer people being served slowly) while being a sign the queue is failing the people it lost
• reconnecting to week 2: an average computed only over survivors is a second, quieter way an average can hide who a system is actually failing
• what it would take to notice reneging at all — the Office would need to count tickets issued, not just tickets served, and it currently doesn’t

Office of Everything file — synthetic case material. This semester’s reported “average wait: 4 minutes” was computed from served tickets only. A separate reconciliation finds that 15% of issued tickets left without service. Every issued ticket is accounted for as served or departed; no ticket remains waiting at the end of this reporting period. Departure times were not recorded.

The Queue Lab asks you to work out what that 15% gap implies about the served-only average — and what remains unknown without departure times.