Skip to lesson

11 / 12 · Make a fair comparison

What does a small p-value actually say?

Read a p-value as a question under an assumption, not the probability that a claim is true.

Builds on Better results, or different people?

Go to practice ↓

A question to keep in mind

If p is about 0.04, does that mean the new method has a 96% chance of being better?

No. You cannot subtract p from 1 to get the probability that the new method is better. A p-value asks about possible test results while treating a starting model as true. It does not give probabilities to the competing explanations.

For example, assume a model in which the methods have no real difference, together with the test's other assumptions. Under a fixed test rule, how likely is a test statistic this extreme or more extreme? That probability is the p-value. 'More extreme' is defined by the chosen test, not by changing the rule after seeing the result.

In the example below, count all runs that are at least as extreme, not just exact copies of the observation. A small p-value signals an unusual result under the assumed model. It does not tell you the effect's size, its practical importance, or whether another explanation is correct.

Work through an example

All simulated runs1000 runs
At least as extreme40 runs
Invented teaching counts for a no-difference model and a fixed test rule; no simulation was run here. The ratio illustrates an approximation, not an exact p-value or a real study result.
  1. Fix the no-difference model, assumptions, and what counts as 'at least as extreme' before looking at the result.
  2. In these invented teaching counts, 40 of 1,000 runs meet that rule. Every run uses the same starting model; 40 is a count of qualifying results, not a count of false hypotheses.
  3. 40/1000 = 4%, so the simulation gives a rough p-value of 0.04. This calculation keeps the starting model assumed.

Your turn

0 / 3
Practice 1Not checked

Of 1,000 simulations under the stated model, 40 are at least as extreme. What percentage is that?

Your answer
Read solution · does not award completion
  • Simulation percentage: 4 %

It is 4%. This approximates a probability of the test result under the model, not a probability of the model itself.

Practice 2Not checked

Which interpretation of a small p-value is appropriate?

Your answer
Read solution · does not award completion

The chosen test result is unusual under the assumed model.

Do not reverse 'probability of a result under a model' into 'probability of a model given a result'. It is the conditional-probability distinction from lesson 3.

Practice 3Not checked

A very large experiment finds a tiny improvement with a small p-value. What else is needed before deciding whether it is useful?

Your answer
Read solution · does not award completion

The effect size, uncertainty, study design, and practical costs or benefits.

Read the full result: how big, how uncertain, measured how, and useful for what. A threshold alone cannot answer those questions.

Bring it back to your own work

Complete this sentence: 'A small p-value does not tell me ____. To decide what it means here, I would also ask ____.'

Saved only on this browser0 / 6000