Type I and Type II Errors
A Type I error rejects a true null hypothesis, a false positive; a Type II error keeps a false null, a false negative. This guide explains both with courtroom and medical examples, shows the decision table, and proves with numbers that raising the sample size shrinks both errors at once.
Every hypothesis test can go wrong in exactly two ways: it can cry wolf, or it can miss the wolf. A Type I error cries wolf — it rejects a null hypothesis that is actually true. A Type II error misses the wolf — it keeps a null hypothesis that is actually false. This guide names both errors, shows them in a decision table, and proves with numbers that a bigger sample shrinks the second error without raising the first.
The Two Errors, Defined
A Type I error is a false positive: the test rejects the null hypothesis H₀ when H₀ is true. Its probability is α, the significance level you choose — usually 0.05. So α = P(Type I error). A Type I error is the error you control directly: set α = 0.05 and you accept a 5% false-positive rate.
A Type II error is a false negative: the test keeps the null hypothesis H₀ when H₀ is false — the effect is real, but the test misses it. Its probability is β. Power is the complement: power = 1 − β, the probability of detecting a real effect. So β = P(Type II error) and power = 1 − β.
The Decision Table
Every test ends in one of four cells: two correct decisions and two errors.
| H₀ is true | H₀ is false | |
|---|---|---|
| Reject H₀ | Type I error (probability α) | Correct decision (power, 1 − β) |
| Keep H₀ | Correct decision (probability 1 − α) | Type II error (probability β) |
Read it across: if H₀ is true, you either commit a Type I error (probability α) or decide correctly (probability 1 − α). If H₀ is false, you either decide correctly (power, 1 − β) or commit a Type II error (probability β). The what is a p-value guide explains α = P(Type I error) in more depth.
The Courtroom Example
Think of a trial. The null hypothesis is “the defendant is innocent”. A Type I error convicts an innocent person — a false positive. A Type II error frees a guilty person — a false negative. The justice system sets its α very low (“beyond reasonable doubt”) because convicting the innocent is treated as the worse error: it would rather free some guilty people than convict an innocent one.
The Medical Example
Think of a screening test for a disease. The null hypothesis is “the patient is healthy”. A Type I error is a false positive: the test says the patient is ill when they are healthy — worrying, but a follow-up test can correct it. A Type II error is a missed diagnosis: the test says the patient is healthy when they are ill — and the disease goes untreated. Here the Type II error is usually the worse one, so screening tests are designed to keep β low even at the cost of more false positives.
The Tradeoff: Lowering α Raises β
At a fixed sample size, the two errors trade off against each other. Lower α — make it harder to reject H₀ — and you cut the false-positive rate, but you also make it harder to detect a real effect, so β rises. Raise α and the reverse happens: fewer missed effects, more false alarms. You cannot shrink both errors by moving α alone.
Increase n to Shrink Both
There is one lever that shrinks β while α stays fixed: the sample size. A bigger sample gives the test more information, so it detects real effects more reliably without loosening the false-positive rate. The table below proves it with numbers — a two-sample t-test, effect size d = 0.5, α = 0.05, two-sided, exact noncentral-t power. The only thing that changes is n.
| n per group | Power (1 − β) | β = P(Type II error) | α = P(Type I error) |
|---|---|---|---|
| 16 | 0.2777 | 0.7223 | 0.05 |
| 32 | 0.5036 | 0.4964 | 0.05 |
| 64 | 0.8015 | 0.1985 | 0.05 |
At n = 16 per group the test misses a real medium effect 72% of the time (β = 0.7223). At n = 64 it misses it only 20% of the time (β = 0.1985) — and α is still 0.05 throughout. That is the proof: raising n shrinks β while α stays fixed. Work out the power for your own study with the statistical power calculator, or find the sample size you need with the sample size calculator.
The Three Levers
You control three things, and they are not equal. α is set by you before the test — 0.05 is the convention, lower when a false positive is costly. β is set by the sample size and the effect size — it falls as n rises. And n is the lever that improves both sides of the table at once: it shrinks β without touching α. When a study is underpowered, the honest fix is more data, not a looser α.
One more thing the table cannot show: a one-tailed test changes the balance too, because it puts all of α in one direction. The one-tailed vs two-tailed test guide explains when that is justified — and when it is p-hacking. And the hypothesis testing explained guide puts the whole framework together.
Frequently Asked Questions
Can you make both errors at once?
No. A single test makes at most one error: it either rejects H₀ or keeps it, and H₀ is either true or false. If H₀ is true and you reject it, that is a Type I error. If H₀ is false and you keep it, that is a Type II error. The two errors are mutually exclusive for one test — but across many tests you will make both kinds, at rates α and β.
Which error is worse?
It depends on the cost. Convicting the innocent (Type I) is usually worse than freeing the guilty (Type II), so trials set α very low. Missing a diagnosis (Type II) is usually worse than a false alarm (Type I), so screening tests keep β low. There is no universal answer — the worse error is the one whose consequences you cannot afford.
How do I reduce both errors?
Increase the sample size. At a fixed n, lowering α raises β and vice versa — the tradeoff is unavoidable. Raising n is the only lever that shrinks β while α stays fixed: for a two-sample test with d = 0.5 and α = 0.05, β falls from 0.7223 at n = 16 per group to 0.1985 at n = 64. The sample size calculator tells you the n your study needs.
Check Tools
Work out the numbers on this page with the site’s free calculators:
Statistical Power Calculator
Power analysis for t-tests — see how n changes power and β.
Try the statistical power calculator →
Sample Size Calculator
Find the n your study needs for the power you want.
Try the sample size calculator →
P-Value Calculator from Z-Score
One-tailed and two-tailed p-values for any z-score, with the shaded curve.
Explore All Statistical Tools
See every calculator on the site in one place.
Check Articles
Go deeper with these guides:
One-Tailed vs Two-Tailed Test
How the α split changes the p-value — and when a one-tailed test is p-hacking.
Hypothesis Testing Explained
The full testing framework: null and alternative hypotheses, p-values and decisions.
What Is a P-Value?
What the p-value really means — and why α = P(Type I error).