Randomized Experiments and Blinding | Introduction to Statistics

Opening

You're testing a new fertilizer on your roses. Half your plants get it, half don't, and the treated ones bloom more. But you also watered them differently and gave them more sun. Can you really say the fertilizer caused the difference?

The idea

Randomized experiments with control groups and blinding isolate the explanatory variable to prove cause-and-effect.

Variables and experiments

Experiments answer questions like "Does aspirin reduce heart attack risk?" by investigating the relationship between two variables. The variable that causes change is the explanatory variable; the affected one is the response variable. In a randomized experiment, the researcher manipulates the explanatory variable's values (the treatments) and measures the response. Each object or individual measured is an experimental unit.

Vocabulary· 2 terms

Lurking variables and isolation

Suppose you want to test vitamin E's effect on disease. You ask people if they take it and find they're healthier. That doesn't prove vitamin E works—they also exercise, eat well, and don't smoke. These extra factors are lurking variables. To prove causation, you must isolate the explanatory variable so the only difference between groups is the treatment. Random assignment spreads lurking variables equally across groups, making different outcomes a direct result of the treatments.

Vocabulary· 2 terms

Worked example

A classic randomized experiment:

1
Recruit 400 people aged 50–84.
Defines the experimental units and the population of interest.
2
Randomly divide them into two groups.
Random assignment spreads lurking variables equally across groups.
3
Give one group aspirin, the other a placebo.
The aspirin and placebo are the treatments—the only planned difference.
4
Each person takes one pill daily.
Standardizes the treatment so the only difference is which pill.

Any difference in heart attack rates must come from the aspirin, not from lurking variables.

Vocabulary· 2 terms

Control and blinding

The power of suggestion can influence outcomes—believing you took a performance drug can improve times almost as much as the drug itself. To counter this, researchers use a control group given a placebo treatment that cannot influence the response. But if you know your pill is fake, the suggestion effect disappears. Blinding (or masking) keeps people from knowing who gets the active treatment and who gets the placebo. A double-blind experiment blinds both the subjects and the researchers involved with them.

Vocabulary· 2 terms

Easily confused

Looks like

Random assignment and blinding both protect experiments, so they might seem interchangeable.

Actually

Random assignment spreads lurking variables equally across groups; blinding prevents the power of suggestion from influencing outcomes.

Random assignment controls for hidden differences between people; blinding controls for psychological effects. Both are needed to isolate the explanatory variable.

Vocabulary· 2 terms

Which design element keeps participants from knowing their own treatment?

Recap

Say each point to yourself, then tap it to check.