Building a Model, Building a Theory

An observable phenomenon is simply something you can see happening: the oval floods in one corner, ice melts faster in a warm room, the Moon changes shape through the month. An explanatory model is a simplified picture of the hidden machinery underneath, built so that it accounts for what you can see.

Take Mia's graph. It tells you the water reached 6 cm and lasted nine hours. That is a description, and a careful one. Now take Sam's clay. Sam can say: water soaks down through loose sand easily, but clay particles are packed so tightly that water moves through them extremely slowly. Under that corner sits a clay layer, so rainwater sinks a little way, meets the clay, and has nowhere to go but sideways and up. That is an explanation, because it names a mechanism, and the mechanism makes the flooding follow.

Here is the boundary between the two. Ask of any statement: if this were true, would the thing I observed have to happen? Mia's graph could be true of a corner that floods for any reason at all, so it does not single out a cause. Sam's clay layer, if it is really there, forces the water to pool. Description tells you what happened; a model tells you what would have to be true underneath for that to happen.

Building a model is a method, not a flash of inspiration. Four steps do the work, and the oval shows all four.

Step 1, describe the phenomenon precisely. Not 'the oval gets wet' but 'the north-east corner holds water for about nine hours after heavy rain, while ground ten metres away drains in under one hour'. A sharp description tells you exactly what your model has to account for. Step 2, propose a mechanism. Ask what could be different about that corner underneath. A buried clay layer would trap water. So would a dip in the ground, or a blocked drain. Step 3, build the model. Draw a labelled cross-section showing sand over clay, with arrows for water soaking down, stopping at the clay, and spreading sideways. Step 4, test it on a case it has not seen. Your model predicts that if you dig anywhere else that stays wet, you should find clay near the surface there too, and that a dry patch should have none.

Step 4 is the one that separates a model from a story, and it is the step most people skip. A mechanism invented to fit one observation will always fit that observation. The decision rule to carry away is this: before I believe my own model, what does it predict about a case I have not looked at yet? If you cannot answer that, you have a description wearing a model's clothes.

Your clay-layer model explains one corner of one oval. A scientific theory is what a model becomes when it grows far beyond the case it started from: it explains many different observations, it has survived repeated attempts to show it is wrong, and other scientists have checked it independently.

Follow one real model the whole way. In 1979, pathologist Robin Warren at Royal Perth Hospital kept seeing curved bacteria in samples of inflamed stomach lining. At the time, the accepted explanation for stomach ulcers was stress and excess acid, and everyone knew bacteria could not survive stomach acid. Warren and Barry Marshall proposed a different mechanism: a bacterium, now called Helicobacter pylori, lives in the stomach lining and inflames it, and that inflammation causes ulcers. That was a model. Almost nobody believed it.

So they tested it where it could fail. In 1984 Marshall deliberately drank a culture of the bacteria. He became ill with gastritis within days, the bacteria were found in his own stomach lining, and a course of antibiotics cleared both the bacteria and the illness. Other laboratories then repeated the work, patients with ulcers were cured with antibiotics rather than managed for life, and the explanation spread from one hospital to medicine worldwide. Marshall and Warren shared the Nobel Prize in 2005. The step that turned their model into a theory was not the clever idea; it was that the idea kept surviving tests run by other people on cases it had never seen.