Model validation

Model validation evaluates whether a model is adequate for a stated scientific purpose.

Independent information

When possible, evaluate predictions against data not used for calibration. This may involve a held-out time period, another experiment, another population or cross-validation.

What to assess

Validation can examine predictive error, calibration of predicted probabilities, qualitative biological behaviour and whether conclusions remain stable under plausible uncertainty.

Purpose matters

A model can be useful for one task and inadequate for another. Validation should therefore be tied to the intended biological question rather than treated as a universal certificate of correctness.