# How science knows: observation, models, measurement and testing: Observation, hypothesis and experiment

Source page: https://www.materialinsights.org/learn/science/how-science-knows/p2/

[All courses](https://www.materialinsights.org/learn/) / [Understanding Science](https://www.materialinsights.org/learn/science/) / [Module 1: How science knows: observation, models, measurement and testing](https://www.materialinsights.org/learn/science/how-science-knows/) Part 2 of 6

## Observation, hypothesis and experiment

The classic cycle runs from noticing something to guessing why, then to a test that could prove the guess wrong. Real research is messier than the textbook steps, but the logic underneath is steady: ideas earn trust by surviving fair tests. Here are the pieces and two famous cases in which they worked.

### 2.1 A good hypothesis is a guess that can fail

Your phone battery drains faster on some days. You notice it, wonder why, and guess: maybe it is the weak signal in the lift. That sequence of noticing, asking and guessing is how most science starts. The guess is a hypothesis. It becomes useful only when it says something you could check and be wrong about.

Compare two guesses. 'The battery drains faster when the phone searches for signal' tells you what to do: switch to flight mode in the weak area and measure the percentage lost over an hour. 'The battery has a mood' tells you nothing, because no result could contradict it. A scientific hypothesis must stick its neck out. It must forbid something. If every possible result fits, the idea explains nothing.

This is the sharp idea of the philosopher Karl Popper in the 1930s. We can never prove a general claim by piling up cases, since the next case could break it. A thousand white swans do not prove all swans are white, but one black swan, seen in Australia in 1697, ended the belief. So science works by trying hard to knock ideas down and trusting the ones that stay up. The first principle is about asymmetry: failure is final, success is only provisional.

The tidy list called the scientific method, which runs from observing to hypothesising, predicting, testing and concluding, is a useful picture of this logic. Working scientists do not always go in that order. Some begin with a chance finding, such as Fleming's mould-spoiled dish in 1928, and some with a pattern in data that nobody expected. Many ideas come from imagination and hunches. What is not optional is the test at the end. Where an idea came from does not matter. Whether it survives does.
 Common misreading

#### 'Science follows five fixed steps'

The five-step list is a way to describe the logic of testing, not a recipe that every project follows. Researchers loop back, change their question halfway and often stumble on things they were not looking for. Penicillin, X-rays and the cosmic microwave background all began as surprises. What stays fixed is that a claim must meet a test it could fail.

- Hewitt, Lyons, Suchocki and Yeh, Conceptual Integrated Science, 2nd edition (Pearson, 2013) · ch. 1, 'About Science'

### 2.2 A fair experiment changes one thing at a time and keeps a control

A farmer wants to know whether a new seed variety gives more grain. She plants it in one field and the old seed in another, and finds the new one yields more. Has she shown the seed is better? Not yet. The fields may differ in soil, water or sun. The experiment mixes up two differences, so she cannot tell which one caused the result.

The cure is to change one factor, the independent variable, and measure the effect on another, the dependent variable, while holding everything else the same. The group that gets the change is the test group. The group that does not is the control, which shows what would have happened anyway. In the farm case, she should split each field into strips and plant the two seeds in alternating strips. Soil differences then fall on both seeds equally.

Several habits guard against being fooled. Randomise which plot gets which treatment, since choosing by hand tends to put the 'good' treatment on the best ground. Use enough samples so that luck averages out. Where people are involved, keep them blind to which treatment they got, because expectation alone changes how people feel. Where measurements are made by people, keep the measurers blind as well. Finally, repeat the experiment, and let others repeat it. A result that only one team can get is not yet a result.

Why do these rules work? Every real measurement has some random scatter, and many things vary at once. Randomising and replicating turn the unknown differences into scatter that averages away, and a control isolates the factor you changed. You can then say that, beyond luck, the change caused the difference.
 Worked example

#### Seed trial on paper

Ten alternating strips are planted, five with each seed. New seed yields in tonnes per hectare: 4.1, 3.8, 4.4, 4.0, 3.9 (mean 4.04, calculated). Old seed: 3.6, 3.7, 3.9, 3.5, 3.8 (mean 3.70, calculated). The gap is 0.34 t/ha, about 9% (calculated). Scatter inside each group is about 0.2 t/ha, so the gap is larger than the scatter but only moderately. A careful report would say that the new seed probably helps, then call for a second season. A claim of 'at least 10% more grain' has not yet passed.

- Hewitt, Lyons, Suchocki and Yeh, Conceptual Integrated Science, 2nd edition (Pearson, 2013) · ch. 1, 'About Science'

### 2.3 Two cases: the 1854 cholera outbreak and the 1919 eclipse

On 31 August 1854 cholera struck the Soho district of London. The common belief was that it spread through foul air. The physician John Snow doubted this and listed where the dead had lived; his famous map came later. The deaths clustered around a public water pump in Broad Street. He asked what the survivors had in common, found that a nearby brewery's workers drank beer and their own well water and mostly escaped, and persuaded the authorities to remove the pump handle. The outbreak was already fading, so the removal did not prove his case alone. The stronger evidence was a later natural experiment: houses supplied by one water company, drawing from a sewage-polluted stretch of the Thames, had far more deaths than houses supplied by another, which drew upstream. The hypothesis, that the disease travels in water, predicted exactly that. Air did not.

The second case shows a prediction that could have failed. Einstein's general relativity of 1915 said that gravity bends the path of light. For starlight grazing the Sun, it predicted a deflection of 1.75 seconds of arc, twice the value that an older Newtonian argument gave. Stars near the Sun can be seen only during a total eclipse. On 29 May 1919 expeditions to Principe, off West Africa, and Sobral in Brazil photographed the star field around the eclipsed Sun and compared it with night photographs of the same stars taken months before or after. Sobral's best plates gave 1.98 and Principe's 1.61 seconds of arc, both near the larger value. Modern radio-telescope measurements have since confirmed the prediction to better than 0.1%.

Both cases share a pattern. A bold idea gave a specific, checkable prediction, a fair comparison was possible, and the evidence told against the rival. Notice too what was not needed: neither Snow nor the eclipse teams could see the mechanism directly. Filippo Pacini saw the cholera bacterium in 1854, but it was firmly tied to the disease only when Robert Koch isolated it in 1883.

Process flowchart How a hypothesis earned trust: the 1919 eclipse Inputs **Hypothesis**gravity bends light Process steps 1 **Prediction**

star positions near the Sun shift by about 1.75 arcseconds
 2 **Test**

photograph the stars during a total eclipse and compare with night plates
 3 **Compare**

measured shifts between roughly 1.6 and 2 arcseconds on the best plates
 Outputs **Outcome**the larger prediction preferred; later radio tests agree to 0.1% or better

Original eclipse uncertainties were large. The case rests on many later checks.
 Source: Gilmore and Tausch-Pebody, The 1919 eclipse results which verified General Relativity and their later detractors: a story re-told (2020), https://arxiv.org/abs/2010.13744  Case

#### A prediction that could have embarrassed the theory

Newton's picture and Einstein's gave different numbers for the same eclipse, roughly 0.9 and 1.75 seconds of arc. A tiny angle: 1.75 seconds is about the width of a 20 mm 1-rupee coin seen from 2.5 km away (calculated). The teams needed careful corrections for the Sun heating the mirrors that fed the telescopes. The experiment was hard, and honest scientists argued afterwards about how clean the 1919 data were. Later, better measurements settled the question.

- Hewitt, Lyons, Suchocki and Yeh, Conceptual Integrated Science, 2nd edition (Pearson, 2013) · ch. 1, 'About Science'
- Dyson, Eddington and Davidson, A determination of the deflection of light by the Sun's gravitational field, Philosophical Transactions of the Royal Society A 220 (1920) · report of the 1919 eclipse expeditions
- Snow, On the Mode of Communication of Cholera, 2nd edition (1855) · the Broad Street outbreak and the water company comparison
- [Gilmore and Tausch-Pebody, The 1919 eclipse results which verified General Relativity and their later detractors: a story re-told (2020)](https://arxiv.org/abs/2010.13744) · web
- [History.co.uk, 1854 Broad Street cholera outbreak](https://www.history.co.uk/articles/1854-broad-street-cholera-outbreak) · web
- [UCLA Department of Epidemiology, John Snow site](https://epi-snow.ph.ucla.edu/) · web

### See it in the table
[H₂O Water](https://www.materialinsights.org/commodities/water/)

### Check your understanding

Which hypothesis is most scientific?
 Only the nitrogen claim states a specific result that a trial could contradict. The others fit any outcome, so they explain nothing.

In a trial of a fertiliser, what is the purpose of the control plots?
 A control gives the baseline. Without it you cannot tell whether a better yield came from the fertiliser, the weather or the soil.

John Snow's map of cholera deaths was strong evidence mainly because
 The pump map and the later comparison of water companies tested a clear prediction. The outbreak was already ending, so the handle removal alone proved little, and the germ was not tied to the disease until 1883.

[← Part 1: What science is and where it came from](https://www.materialinsights.org/learn/science/how-science-knows/p1/)[Mark read · Part 3: Facts, models, laws and theories →](https://www.materialinsights.org/learn/science/how-science-knows/p3/)

## Other sections

- [Module overview](https://www.materialinsights.org/learn/science/how-science-knows/index.md)
- [What science is and where it came from](https://www.materialinsights.org/learn/science/how-science-knows/p1/index.md)
- [Facts, models, laws and theories](https://www.materialinsights.org/learn/science/how-science-knows/p3/index.md)
- [Measurement, units and powers of ten](https://www.materialinsights.org/learn/science/how-science-knows/p4/index.md)
- [Uncertainty: how sure can a measurement be](https://www.materialinsights.org/learn/science/how-science-knows/p5/index.md)
- [Limits of science, and science with technology and society](https://www.materialinsights.org/learn/science/how-science-knows/p6/index.md)
- [Key terms, quiz and sources](https://www.materialinsights.org/learn/science/how-science-knows/review/index.md)
