Confounders

mental-model

A confounder is a difference between groups that can impersonate the effect of a treatment. Keeping conditions comparable—and using random assignment when hidden differences cannot be measured—makes the treatment the clearest remaining explanation.

Suppose one ship’s sailors get less scurvy than another’s. The tempting conclusion is that the tested treatment worked—but perhaps that ship simply served different food, enforced a different sleep schedule, or had a captain who ran a healthier routine.

The rival cause hiding between the groups

A comparison can support a causal claim only when the groups are comparable. If treatment and control groups also differ in diet, leadership, sleep, or routine, each difference becomes a rival explanation for the outcome. The treatment receives credit for a bundle of changes it may not have caused.

Measured differences can sometimes be controlled directly. Unknown ones are harder: you cannot adjust for what you never thought to record. Randomization tackles that problem by assigning participants without reference to their characteristics, so hidden differences are more likely to be distributed across groups instead of lining up systematically with the treatment.

Where it shows up

Two ships, many treatments

Comparing sailors ship by ship does not isolate one intervention. Food, captain, sleep schedule, and daily routine all travel with the ship, so the apparent treatment effect could belong to any of them.

Random assignment balances; it does not clone

Randomization does not make every person or group identical, especially in a small experiment. It reduces systematic imbalance; chance differences can remain. And when entire ships receive different conditions, ship-level differences are still entangled unless treatments are assigned across enough comparable units.

Write the rival-explanations list first

Before accepting a comparison, make two columns for the groups and list every condition that differs besides the proposed cause—diet, leadership, timing, routine, selection, or setting. Then ask which differences were measured, which were controlled, and whether random assignment could prevent the unknown ones from clustering on one side.

Episodes that teach this

The Randomness of Science - Why Randomised Controlled Trials (RCTs) Work - FutureIQ 4,109 views
The transcript warns that sailors on different ships might differ in food, captain, sleep schedule, or routine, any of which could affect scurvy outcomes.