Designing a valid experiment

A hypothesis with a reason

In Year 7 you wrote predictions. At Stage 5 a hypothesis should be a testable statement about the relationship between the independent and dependent variables, with a scientific reason drawn from your background research:

If [IV changes in this way], then [DV will change in this way], because [scientific reason].

For example: "If pots are covered with organic mulch, they will lose less water than bare soil, because mulch shades the soil and slows the air moving over it, reducing evaporation." A hypothesis must be able to be shown wrong by your results. If no possible result could disagree with it, it isn't testable. Results that agree with a hypothesis support it; they never prove it, because a later test might still disagree.

Control groups and controlled variables

These two terms sound alike but mean different things:

• Controlled variables are all the other factors kept the same for every group (see Year 7, 1E).
• A control group (or experimental control) is a group that does not get the treatment at all, such as pots with no mulch, or seedlings watered with plain water. It shows what happens without the independent variable, so you can see how big the effect really is.

In medical trials the control group is often given a placebo, a dummy treatment that looks the same as the real one, because people can feel better just from believing they are being treated.

Choosing the values of the independent variable

If your independent variable is a number (concentration, temperature, depth), choose its values carefully:

• Use at least five values so you can see the shape of the relationship, and include zero if it makes sense (it acts as your control).
• Spread them across a wide enough range to see the whole effect, and space them evenly.
• Run a pilot test to find the interesting range before the real experiment.

Sample size and replicates

Living things vary. Two seedlings from the same packet, in the same soil, can differ in height by several centimetres. If you test one plant per group, a difference might just be luck. Using several individuals per group (replicates) and calculating the mean lets the real effect show through the natural variation.

How many is enough? In a school project, aim for at least 3 replicates per value for physical experiments with little variation, and 10 or more per group for living things or people. If you can't afford many replicates, test fewer values of the independent variable rather than skimping on repeats.

Randomisation and blind testing

Hidden differences between groups can creep in without you noticing. If all the fertilised seedlings sit near the window, they also get more light, and you can't tell which caused the extra growth. This is a confounding variable. The fix is randomisation: assign pots to positions by drawing numbers or using a random number generator, and swap positions regularly.

People are also a source of bias. If you know which plants were fertilised, you might measure them a little more generously without meaning to. In a blind test the person measuring doesn't know which group each sample belongs to. In a double-blind trial, neither the participants nor the researchers know who got the real treatment until the end.

Making it quantitative

Qualitative observations (colour, smell, "more bubbles") are useful in a pilot test, but your main data should be numbers you can graph and compare. Most qualitative questions can be turned into quantitative ones:

Choose equipment to suit the size of what you're measuring: a trundle wheel for metres, a ruler for centimetres, vernier calipers for tenths of a millimetre; a balance reading to 0.1 g or 0.01 g rather than kitchen scales; a measuring cylinder rather than a beaker for volume. Page 8F explains how the instrument affects your uncertainty.