Introduction to Experimental Design
Welcome! In this chapter, we are going to look at the "blueprints" of statistics. Experimental Design is the process of planning a study so that the data we collect is as reliable and accurate as possible. Whether you are testing a new sports drink or a medical treatment, how you set up the test is just as important as the results themselves.
Think of it like building a house: if the foundation is wonky, the whole thing might fall down! By the end of these notes, you will understand how to build a "solid" experiment that avoids bias and reduces errors.
1. The Core Ingredients of an Experiment
To have a fair experiment, there are several key concepts you need to know and be able to discuss in your exam.
Control and Experimental Groups
Imagine you want to see if a new "Super Growth" fertilizer works on plants. You can't just give it to ten plants and see if they grow. Why? Because plants grow anyway! You need a baseline for comparison.
Experimental Group: The group that receives the treatment (e.g., the plants getting the "Super Growth" fertilizer).
Control Group: The group that is treated exactly the same way except they do not get the treatment (e.g., plants getting plain water).
Randomisation
This is one of the most important tools in a statistician's kit. Randomisation means using chance (like flipping a coin or using a random number generator) to assign subjects to groups. This helps ensure that the groups are as similar as possible before the experiment starts, which helps avoid bias.
Example: If a researcher hand-picked the healthiest-looking plants for the fertilizer group, the experiment wouldn't be fair. Randomly assigning them removes this human bias.
Replication
One result could be a fluke. Replication is the practice of repeating the experiment on many subjects or many times. The more subjects we have (a larger sample size), the more confident we can be that our results aren't just down to luck.
Experimental Error
Don't worry—"error" doesn't necessarily mean you did something wrong! In statistics, experimental error refers to the natural variation that occurs in data. Even if two plants are identical and get the same water, they might grow at slightly different rates. Our goal in design is to keep this "background noise" as low as possible.
Quick Review: The "Big Three"
1. Control: Something to compare against.
2. Randomisation: Use chance to keep it fair.
3. Replication: Do it many times to prove it's real.
2. Blinding: Keeping it Objective
Sometimes, what we think will happen affects the results. This is especially true in human studies.
Blind Trials
In a blind trial, the subjects do not know which group they are in. For example, in a medicine trial, one person gets the real pill and another gets a sugar pill (a placebo), but neither knows which is which. This prevents the "placebo effect," where people feel better simply because they think they took medicine.
Double-Blind Trials
This is the "Gold Standard." In a double-blind trial, neither the subjects nor the researchers interacting with them know who is getting which treatment. This prevents the researchers from accidentally giving away clues or interpreting results in a biased way.
Analogy: It’s like a "blind taste test" for soda. If you know you’re drinking the expensive brand, you might convince yourself it tastes better!
3. Improving the Design: Blocking and Pairing
If we know there is something that might mess up our results (like the age of a person or the type of soil), we can use specific designs to "cancel out" that influence.
Blocking
Blocking is when we group similar subjects together before we randomise them. We call these groups blocks. This helps reduce experimental error because we are comparing "like with like."
Example: If you are testing a running shoe, you might create "blocks" based on fitness levels (Beginner, Intermediate, Pro) because fitness will definitely affect running times.
Paired Comparisons
A paired comparison is a very clever way to reduce error. It involves matching subjects in pairs based on similar characteristics (like identical twins) and giving one treatment to one twin and the other treatment to the other twin.
Even better, you can sometimes use the same person as their own "pair."
Example: Testing two different suncreams by putting Cream A on the left arm and Cream B on the right arm of the same person. This removes almost all "person-to-person" variation!
4. Types of Experimental Design
You need to be able to distinguish between these two main setups:
Completely Randomised Design
This is the simplest setup. You take your whole group of subjects and randomly split them into groups.
Pros: Very easy to set up.
Cons: If your subjects are very different (e.g., some are very old and some are very young), the "luck of the draw" might accidentally put all the old people in one group, which ruins the fairness.
Randomised Block Design
First, you split your subjects into blocks (groups that are similar). Then, within each block, you randomly assign the treatments.
Pros: Much more accurate because it ensures each treatment is tested fairly across all types of subjects.
Cons: More complicated to organise.
Key Takeaway:
Use Blocking when you know there is a specific factor (like age, gender, or weight) that might affect your results. It helps "filter out" the effect of that factor so you can see the true effect of your treatment.
Summary Checklist for the Exam
When you are asked to discuss or critique an experiment, check for these points:
● Was it randomised? If not, point out the risk of bias.
● Was there a control group? If not, how do we know the treatment actually caused the change?
● Was it replicated? Is the sample size large enough to be reliable?
● Could it be blinded? If humans are involved, they should probably be "blind" to the treatment.
● Should they have used blocking? Identify if there's a variable (like age or species) that should have been grouped first.
Don't worry if this seems a bit wordy compared to the math-heavy chapters! In Paper 1, your ability to explain why a design is good or bad is what earns you the marks. Just keep thinking: "Is this test fair, and does it keep things consistent?"