Exploring the Investigative World of Science

Science does not begin in a laboratory. It begins when somebody notices something and refuses to let it go. A damp patch on a wall that returns after every rain, idli batter that rises beautifully in summer but hardly at all in winter, a cycle tyre that goes soft over two days even though there is no puncture — each of these is an ordinary event, and each can be the beginning of real science if you look at it carefully enough.

An observation is what you actually notice, using your senses or an instrument. You see, hear, smell and touch (and taste only when it is food, and never in a laboratory). Instruments simply extend your senses so that you can notice more, and notice it more exactly: a metre scale, a measuring tape, a thermometer, a weighing balance, a stopwatch, a hand lens, a microscope. Your eye can tell you that one plant is taller. A scale tells you it is taller by 3.2 cm.

Both kinds of noticing have names. A qualitative observation describes something in words — the liquid turned pale blue, the smell was sharp, the surface felt rough. A quantitative observation attaches a number, along with a unit whenever the quantity has one — the temperature fell from 60 °C to 40 °C in four minutes, or seventeen of the twenty seeds sprouted, which is a plain count and needs no unit. Quantitative observations are powerful because two people sitting hundreds of kilometres apart can compare them exactly, without arguing about what slightly warmer means.

Now comes the idea that trips up almost every beginner. An inference is not an observation. An inference is the meaning you attach to what you noticed. Suppose you step outside in the morning and the road is wet. The road is wet is an observation. It rained last night is an inference — a reasonable one, but not the only one. Heavy dew, a water tanker, a burst pipe or a neighbour washing a car would leave exactly the same wet road. Milk that smells sour is an observation; the statement that it has spoiled because it was left out of the fridge contains a whole explanation you have not yet checked.

If you write the inference into your notebook as though it were the observation, everyone who reads your notes later inherits your mistake and cannot correct it, because the original evidence has been thrown away. So keep the two apart on the page. Many students simply draw a line down the middle of the notebook: on the left, only what was seen and measured; on the right, what it might mean. Add the boring details too — date, time, place, weather, which instrument was used, who did the measuring. They feel pointless while you write them and they turn out to be the most valuable part of the record a month later, when you are trying to work out why one day's readings look strange.

One last habit: record while you are observing, not from memory in the evening. Memory quietly tidies things up. It smooths out the reading that did not fit and sharpens the one that did. Tidy memories are comfortable, and they are the enemy of good science.

Curiosity is the raw material. A question is the tool that shapes it. But not every question can be settled by doing an experiment, and one of the most useful skills in this chapter is sorting your questions into the right pile before you waste a fortnight.

Some questions are matters of taste. Which mango is the tastiest? has no single correct answer, because taste genuinely differs from person to person. Some questions are already answered and only need looking up in a reliable book — What is the boiling point of pure water at sea level? Some are questions about what we should do: Should this patch of forest be cleared for a road? Science can supply facts for such a decision — how much rain the area receives, which animals live there, how much dust and noise the traffic will add — but the final choice also depends on what people value, and that part is not decided by measurement alone. And some are perfectly good scientific questions that simply cannot be tested yet with the tools we have.

What is left is the pile you can work with: testable questions. A testable question names two things clearly — the one thing you are going to change, and the thing you are going to measure. Do plants grow better with music? sounds exciting but cannot be tested as it stands, because nobody knows what better means here. Taller? Greener? More leaves? Measured after how long? Sharpen it and it becomes workable: Does playing music for one hour a day change the height of a money plant cutting over three weeks? Now anyone reading it knows what will be changed (music or no music), what will be measured (height in centimetres) and over what period (three weeks).

Here are the same three questions before and after sharpening. Which cloth is best? becomes Which of these three cloth pieces, all of the same size and each soaked with 20 g of water, loses that water fastest when hung in the same shaded place? Is hot water better for washing? becomes Does an oil stain on cotton cloth come out faster in water at 25 °C or at 50 °C, using the same soap and the same rubbing? Are LED bulbs good? becomes Does a 15 W LED bulb give a higher reading on a light meter than a 15 W filament bulb, with the meter kept at the same distance from each? In every case, vague praise words are replaced by something you can point at and measure.

A quick self-test for a good question is this: can you imagine a result that would make you answer no? If no possible result would change your mind, you are not asking a scientific question, however interesting it may be.

One more thing worth knowing early. Experiments are not the only route to knowledge. Nobody can put a star, a monsoon or a volcano into a test tube. In those fields, scientists observe very carefully over long stretches of time, compare large numbers of cases, and hunt for patterns — and they still take care to compare cases that differ in one important way, so that the pattern means something. Careful comparison of situations nature has already set up is every bit as scientific as an experiment on a bench.

Once the question is sharp, you write down your best possible answer before you begin collecting data. That written answer is a hypothesis.

A hypothesis is not a wild guess. It is a statement, based on what you already know or have already noticed, that could turn out to be either right or wrong once you test it. It is written as a sentence, not as a question, and a good one mentions both the thing you will change and the thing you will measure. Compare three attempts at a hypothesis for an investigation on seeds:

• Something happens to seeds in water. Far too vague. No result could disagree with it.
• Will soaked seeds sprout faster? This is a question again, not a hypothesis.
• Gram seeds soaked in water overnight before sowing will sprout sooner than dry gram seeds sown at the same time. This can be tested, and it says exactly what would count as a yes.

From the hypothesis you work out a prediction, usually in the shape: if I do this, then that should happen, because of this reason. For the seed example: if I sow twenty soaked seeds and twenty dry seeds of the same variety in the same soil on the same day, then more of the soaked seeds should have sprouted by day three, because water softens the seed coat and the seed needs water to start growing. The prediction is the thing you actually compare with your readings.

Two rules keep a hypothesis honest. First, it must be testable — there has to be an observation you could make that would settle it one way or the other. Second, and this surprises many students, it must be possible for it to be wrong. Consider a statement such as: there is a force acting here that no instrument can ever detect and that leaves no trace of any kind. No experiment could ever disagree with it, so it sits outside science, no matter how appealing it sounds. A statement that cannot fail cannot be tested, and a statement that cannot be tested teaches us nothing new.

Very often more than one hypothesis fits the same observation. Bread on the kitchen shelf grows fungus much faster during the rainy season. Is it the extra moisture in the air? The slightly different temperature? More fungal spores floating about? Each is a sensible hypothesis, and they cannot all be tested at once. You choose one, design a test for it, and park the others for later. That is why a single good investigation usually answers a small question rather than a huge one — small questions can actually be settled.

Finally, write the hypothesis down before you collect data, and do not quietly rewrite it afterwards so that it matches what you found. It is remarkably easy to convince yourself that you expected the result all along. Writing it first, in ink, in your notebook, is a simple protection against fooling yourself — and being hard to fool is one of the central habits of anybody who does science well.

Testing a hypothesis fairly is where most of the real thinking happens. The rule is short enough to memorise: change one thing, measure one thing, keep everything else the same.

Anything that can change during an investigation is called a variable, and it helps to give each variable a job:

• The independent variable is the one thing you deliberately change — soaked seeds against dry seeds, the length of the thread, the temperature of the water.
• The dependent variable is what you measure to see the effect — how many seeds sprouted, the time taken for ten swings, the time the sugar took to disappear.
• The controlled variables are everything you keep the same, so that none of them can be blamed for the result — same variety of seed, same amount of water, same soil, same room, same time of day, same person doing the timing.

Why only one change at a time? Suppose you set up two trays of seeds. Tray A gets more water and a sunny window; tray B gets less water and a shaded corner. Tray A does better. What caused it — the water or the light? You cannot say. Two possible causes have been tangled together, the fortnight is gone, and the experiment has told you nothing. Change the water alone, keep the light identical for both trays, and the result becomes readable.

A control setup is the setup in which the change is not made at all — the dry seeds, the plant with no fertiliser, the cloth washed in plain water. Without it, you have nothing to compare against. If nineteen of your treated seeds sprout, is that good? You cannot answer until you know how many sprouted without the treatment. The control is not a spare setup you can drop when apparatus runs short; it is the yardstick. Whenever the question is whether a treatment does anything at all, dropping the control leaves you with nothing to compare against and no conclusion you can defend. When instead you compare several values of the same factor — water at 20 °C, 30 °C, 40 °C and 50 °C, say — the setups act as comparisons for one another.

Four practical points turn a design from acceptable into good. Use enough samples. One seed proves nothing, because that single seed may have been damaged before you ever saw it; twenty seeds in each tray let the odd one wash out. Repeat each measurement. Time the swings three times rather than once. Choose a sensible range and sensible steps for whatever you are changing — if you are testing water temperature, readings at 20 °C, 30 °C, 40 °C and 50 °C show a pattern, while 30 °C and 31 °C show nothing at all. Guard against your own hopes. If you know which cup contains your favourite brand, you will judge it kindly without meaning to. Ask a friend to label the cups A and B and to keep the key until after you have written your scores.

Safety belongs inside the design, not as an afterthought. Work with an adult present when heat, flame, glass or electricity is involved. Tie back long hair and loose clothing near a flame. Never taste or directly smell an unknown chemical, and never point the mouth of a heated test tube at anyone, including yourself. Handle glassware gently, clear the table before you begin, and wash your hands afterwards. A result obtained through a burnt finger is not a better result.

An experiment is only as good as its record. The best moment to design your table is before the first reading, while your hands are still free and your mind is clear. Once the water is boiling and the stopwatch is running, you will write whatever is quickest, and quick notes are usually incomplete notes.

A useful table has one row for each trial or for each value of the thing you changed, and one column for each quantity. Put the unit in the column heading, so you write the unit once instead of forty times, and so nobody has to guess later whether 25 meant seconds or minutes. Keep a final column for remarks — that is where you note the power cut, the cloudy afternoon, the tray somebody moved, the reading you were not sure about. Here is a simple record of three timings of the same measurement:

Trial      Time for sugar to dissolve (s)
1          24
2          25
3          26
Average    25

The average is found by adding the readings and dividing by how many readings there are: 24 + 25 + 26 = 75, and 75 ÷ 3 = 25 s. An average smooths out the small, unavoidable differences between one trial and the next — your thumb on the stopwatch is never a hundredth of a second early or late in exactly the same way twice.

Every measurement is a number and a unit. A number without a unit is not a measurement; it is a rumour. Write 4.5 cm, 30 °C, 12 g, 25 s. Read your instrument properly too. When you read a measuring cylinder of water, bring your eye level with the water surface and read the bottom of the curve, because if your eye is above the level you will read a value higher than the true one, and if it is below you will read a lower one. Check that a balance shows zero before you place anything on it, and that a stopwatch has been reset. These small habits remove errors that no amount of clever analysis can fix afterwards.

Sooner or later you will get an odd reading — four values close together and a fifth far away. Do not rub it out. Circle it, note it in the remarks column, and if there is time, repeat that particular trial. Sometimes it turns out to be a mistake, such as a stopwatch started late, and you can say so honestly. Sometimes it turns out to be the most interesting thing in the whole investigation, a hint that something you assumed was constant was not constant at all. An erased reading can never do either of those jobs.

Once the numbers are in, a picture often reveals what a column of digits hides. A bar chart compares separate groups, such as soaked seeds against dry seeds. A line graph suits a quantity that changes smoothly, such as the temperature of cooling water measured every minute. Whichever you draw, label both axes with the quantity and its unit, choose a scale that spreads the data across the page, and let the pattern speak for itself rather than forcing a line through points that do not lie on one.

The conclusion is where you answer your own question using the evidence you collected — and using nothing else. A good conclusion has three parts. First, state what the readings show: more soaked seeds sprouted by day three than dry seeds, in every one of the three repeats. Second, say whether this supports the hypothesis. Third, state the limits of what you found: this was tested on one variety of gram, at room temperature, over five days.

Notice the careful wording. Say that the evidence supports the hypothesis, not that it proves it. Your investigation used the samples you had, in the conditions you had, over the time you had. It cannot speak about seeds you never tested or temperatures you never tried. Stretching a conclusion beyond the range you actually measured is one of the easiest mistakes to make and one of the hardest to defend. If you tested water at 20 °C and 50 °C, you cannot claim anything about boiling water.

There is a second trap. Two things happening together do not prove that one causes the other. A student notices that his team wins whenever he wears a particular shirt. Ice creams sell more on the days more people go swimming. In neither case does the pairing by itself show that one event caused the other. They may simply have occurred together, or a third thing may be affecting both — hot weather sends more people swimming and sells more ice cream at the same time. A fair test is designed exactly to break this trap, because you change one thing yourself and hold the rest steady.

Now the part that many students never quite believe: a negative result is still a result. Suppose the soaked seeds and the dry seeds sprout at the same time. Your hypothesis was not supported. That is not a failed experiment — it is a finding. You now know something you did not know that morning. You have ruled out one explanation, which narrows the field for everyone who comes after you. You have saved the next person the trouble of walking down the same path. And you may have exposed a weakness in the design worth reporting on its own, for example that both trays were watered so generously that the head start from soaking made no difference.

The only genuinely failed experiment is a dishonest one. Adjusting readings so that they fit the hypothesis, quietly dropping the trial that disagreed, or reporting only the half of the data that looks tidy — these turn a piece of science into a piece of fiction, and the damage lasts, because somebody else will build on it. If your result surprises you, that is a reason to check your method and repeat the work, never a reason to edit the numbers.

End a conclusion by looking forward. What would you improve if you did this again — more samples, a longer run, a better instrument, a fairer way of judging? What new question has appeared? Almost every honest investigation finishes by handing you a better question than the one you started with, and that is a sign the work was done properly, not a sign that it was incomplete.

A single result, obtained once, by one person, on one afternoon, is a starting point and nothing more. It might have come from a stopwatch pressed late, a balance that was not zeroed, an unusually warm day, or plain chance. The way science protects itself against all of these at once is by insisting that a result must be repeatable.

There are two levels of this, and they are worth separating. A result is repeatable when the same person, following the same method with the same apparatus, gets the same result again. That guards against a slip or a one-off accident. A result is reproducible when a different person, in a different place, with their own apparatus, follows your written method and reaches the same result. That is the stronger test, because it also catches a habit peculiar to you — the way you always start the stopwatch a shade early, or a thermometer of yours that reads two degrees high.

Reproducibility only becomes possible if you write the method properly, and this is why the method section of a report is not a formality. Write it so that somebody who has never met you can follow it and do exactly what you did. Give the quantities, the sizes, the times, the temperatures, the make of the seeds, the number of trials. A sentence such as take some water and heat it cannot be reproduced. Take 100 mL of water in a 250 mL beaker and heat it until the thermometer reads 50 °C can. A helpful test is to hand your method to a classmate who was not in the room and see whether they can run it without asking you a single question.

Then comes sharing and checking. In school this may mean presenting to your class or displaying at a science exhibition, where the questions from the audience are part of the process, not an ordeal to be survived. Professional scientists write up their work and send it to a journal, where other scientists in the same field read it and try to find the weaknesses before it is published. Afterwards, other groups repeat the work independently. When several separate groups, using different apparatus, reach the same result, confidence in it grows steadily. When they cannot, the claim is treated with caution, however famous the person who made it. This checking is not rudeness. It is the mechanism.

Two habits go with all of this. Keep your raw data — the original scribbled table, not just the neat final version — because if a question comes up later, only the raw numbers can answer it. And give credit to your sources, naming the book or site you took an idea or a value from, so a reader can check that too.

Finally, do not treat the stages of an investigation as a rigid staircase that you climb once. In real work they loop. An observation raises a question; the question needs a hypothesis; designing the test makes you realise the question was too broad, so you narrow it; the readings throw up something odd, which sends you back to observe more carefully; the conclusion hands you a fresh question and the cycle starts again. Science is less like a recipe and more like a conversation with nature, in which every answer you receive earns you the right to ask a better question.