Test a model at the extremes

The idea

Extreme-condition tests ask what a model predicts under boundary inputs. They can reveal hidden impossibilities that ordinary-looking data do not expose. A model of work should not complete positive output with no available workers unless it explicitly includes an automated process.

Worked example

A fictional lending model predicts ten checkouts each day plus three per volunteer. At zero volunteers, the equation still predicts ten checkouts, even though the library has no self-service system. Staff discover that the intercept was copied from a statistical fit without an operational explanation. They distinguish a prediction within observed staffing levels from a causal simulation intended to evaluate closures. The same formula need not be appropriate for both purposes.

Use it carefully

An extreme may lie outside the model’s intended range, so failure there does not always make a local prediction useless. State the model’s purpose and boundaries. Conversely, a simulation used for policy should respect relevant physical and accounting constraints. Test zeros, capacity limits, and very large inputs, then document which responses are assumptions and which are supported by observations.

RETRIEVE. REASON. REMEMBER.

Knowledge check.

Choose one answer for each question, then check your reasoning.