Controlling Conditions and Limiting Bias

The three kinds of variable

Before we control anything, it helps to name the moving parts. Most experiments have exactly three roles to fill:

• Independent variable, the one thing you deliberately change. (Sometimes called the manipulated variable.)
• Dependent variable, the thing you measure to see the effect. (The responding variable.)
• Controlled variables, everything else you keep exactly the same for every group, so it cannot sneak in and explain your results.

A worked example: does a new fertilizer help plants grow?

A gardener wants to know whether a new fertilizer makes tomato plants grow taller. The fertilizer is only the scenario, the skill we are practising is judging the design. Here is a first, flawed attempt:

Something is wrong here. The fertilized plant also got more sunlight. We changed two things at once, fertilizer and light, so we cannot say which one caused the extra height. The sunlight is a hidden second variable, and it makes the result impossible to trust. Each tool below fixes a different part of this problem.

Tool 1: Keep controlled variables constant

To make the test fair, every plant must get the same light, the same water, the same pot size, the same soil, and the same temperature. The only difference allowed between the groups is the fertilizer. Then, if one group grows taller, fertilizer is the only thing that could explain it.

When a procedure lets some other condition differ between the groups, that condition is called a confounding variable. Spotting confounding variables, like the sunlight above, is one of the most common CAEC inquiry tasks.

Tool 2: Use a control group for comparison

A control group is the group that does not get the treatment. It is your baseline, the "what would have happened anyway" group. Without it, you have nothing to compare against.

For the tomatoes, the control group gets everything the same except the fertilizer: same light, same water, same soil, just no fertilizer. If the fertilized plants end up taller than the control plants, and everything else was equal, the fertilizer is the believable cause. In medical studies the control group often gets a placebo (a fake treatment) so that even the expectation of being treated is the same for both groups.

Tool 3: Use enough subjects and repeat the trials

One plant per group is not enough. A single plant might be unusually strong or weak for reasons that have nothing to do with fertilizer. A large sample size, many plants in each group, lets those random differences average out, so a real effect can show through the noise.

Repeated trials do the same job over time: running the whole experiment more than once checks that the result was not a one-off fluke. Results that hold up when repeated are far more trustworthy, this is the idea behind replication in real science.

Look at how the conclusion strengthens as the sample grows:

Notice the gap between the groups is about the same in every row. What changes is your confidence: with one plant the 6 cm difference could easily be chance, but with 100 plants a steady 4 cm difference is hard to dismiss.

Tool 4: Randomize to limit bias

Bias is anything that quietly pushes the result in one direction. Suppose the gardener hand-picks the healthiest-looking seedlings for the fertilizer group, without meaning to, she has stacked the deck.

Randomization fixes this: assign each plant to a group by chance (for example, drawing numbers from a hat). Now the strong and weak seedlings are spread evenly between the groups, so neither group has a built-in head start.

In studies with people, two more guards help. In a blind study the subjects do not know which group they are in, so their expectations cannot colour the result. In a double-blind study, the researchers measuring the outcome do not know either, so they cannot unconsciously nudge the readings. Both are ways of limiting bias.

Putting it together: fixing the flawed procedure

Here is the gardener's original plan rewritten as a fair test. Compare the two side by side, this "find the flaw, then improve it" move is exactly what the CAEC asks you to do.

A fair-test checklist

When a CAEC question hands you a procedure and asks what is wrong or how to improve it, run through this list:

• One change only. Is just the independent variable different between groups, with everything else held constant?
• Control group. Is there a no-treatment group to compare against?
• Enough data. Is the sample big enough, and were the trials repeated?
• Random and blind. Were subjects assigned by chance, and were expectations kept from skewing the result?

Your turn: practice questions

For each scenario, find the weakness and say how you would fix it. Try it before you peek, the reasoning matters more than the wording.

1. A coach tests a new energy drink by giving it to the fastest five runners on the team and timing them. They run well, so he says the drink improves speed. What is the flaw?
2. A researcher compares headache relief between a new pill and no pill. The pill group is told they are getting a powerful new medicine; the other group is told nothing. What design change would limit bias?
3. A student tests whether warm water dissolves sugar faster than cold water, but she uses a big stir for the warm cup and no stir for the cold cup. Why can't she trust her result, and how should she fix the setup?