Collecting Data

Observational Studies vs. Experiments

Distinguishing observational studies from experiments and identifying confounding variables.

Intermediate15 min lesson3 min readUpdated August 12, 2026Author not yet attributed

Prerequisites

Sampling Methods

What a Study's Design Allows You to Conclude

A school notices that students who eat breakfast tend to score higher on tests than students who don't. Before reading on: since nobody assigned which students eat breakfast, could some other factor — something that affects both breakfast habits and test scores — be responsible for this pattern instead?

Definition — Observational Studies, Experiments, and Confounding

An observational study measures variables as they naturally occur, without assigning any treatment. An experiment deliberately assigns subjects to different treatments (ideally at random) and compares the outcomes. A confounding variable affects the outcome and is tangled up with the variable of interest, making it impossible to isolate that variable's own effect. Observational studies generally support only association, not causation; a well-designed randomized experiment can support a causal conclusion.

Worked Example — Identifying Confounding in an Observational Study

The breakfast/test-score observation is purely observational — students chose their own breakfast habits. A plausible confounding variable is family stability or income: students from more resourced households might be more likely to eat breakfast regularly and simultaneously have more academic support, tutoring, or a quieter place to study. Because breakfast wasn't randomly assigned, this study cannot separate breakfast's own effect from these other tangled-up factors — the honest conclusion is association only.

Worked Example — Turning It Into an Experiment

Researchers instead recruit student volunteers and randomly assign half to eat a provided breakfast each morning for a month, the other half to skip it, then compare test performance at the end. Because breakfast was randomly assigned rather than self-selected, random assignment tends to balance other factors (family background, study habits, and more) evenly between the two groups — isolating breakfast's own effect and making a causal conclusion possible in a way the observational study alone could not support.

Tip

Ask one diagnostic question before trusting any causal claim: was the variable of interest randomly assigned by researchers, or did it happen naturally on its own? Only the first supports a causal conclusion.

Common Mistakes

Assuming a large, carefully measured observational study can establish causation if the association is strong enough.
No matter how large or precisely measured, an observational study can't rule out confounding without random assignment — strength of association and justification for causation are separate questions.

Key Takeaways

Observational studies measure variables as they occur; experiments deliberately assign treatments, ideally at random.
A confounding variable is tangled up with the variable of interest's effect on the outcome, and can only be ruled out reliably through random assignment.
Observational studies generally establish only association; well-designed randomized experiments can support causation.

Summary

This closes the unit on collecting data. The next unit turns to probability itself — the mathematical foundation behind how likely different outcomes are.

Track your progress

← Sampling Methods Back to Collecting Data