Every Model Is Wrong

Your regression model from Mission 06 predicts streamflow will hit 14.2 m³/s next month. A town councillor asks the obvious follow-up: 'is that going to happen, or might it be way off?' You realize your model has never once told you how confident to be, only what to guess.

The statistician George Box's famous line, 'all models are wrong, but some are useful', isn't cynicism, it's a starting point. No model captures reality exactly. The question worth answering isn't 'is this model right?' but 'how wrong might this model be, and in which direction?'

Quantifying uncertainty means attaching an honest range, not just a single number, to every prediction, so the people using your work know how much weight to put on it.

It is like a friend who says they will arrive at 'exactly 6:00pm' versus one who says 'somewhere between 5:45 and 6:15, probably around 6:00.' The second friend sounds less precise, but they are actually giving you more honest, more useful information about what to actually expect.

Every credible scientific forecast, weather, climate, epidemiology, economics, reports uncertainty alongside its central estimate. A single number without a range is not more scientific; it's less informative, and often misleading about how much confidence is actually warranted.

Explain George Box's phrase 'all models are wrong, but some are useful' to someone hearing it for the first time.