The significance level, usually written as 
It is the probability of making a Type I error:
So choosing α is really choosing how much risk of a false positive you are willing to accept.
Common Choices
Here’s how researchers typically choose α:
| α level | Interpretation | When used |
|---|---|---|
| 0.05 | Standard balance between caution and sensitivity | Most social science, education, psychology |
| 0.01 | Very strict; strong evidence required | Medical trials, safety‑critical fields |
| 0.10 | More lenient; easier to detect effects | Exploratory studies, early‑stage research |
How to Think About Choosing α
1. Consequences of a Type I Error
Ask: What happens if we falsely claim an effect exists?
- If the consequences are serious (e.g., approving a harmful drug), choose a small α like 0.01 or 0.001.
- If the consequences are minor (e.g., testing a new teaching method), α = 0.05 is usually fine.
2. Consequences of a Type II Error
Ask: What happens if we miss a real effect?
- If missing an effect is costly (e.g., failing to detect a dangerous defect), you might choose a larger α like 0.10 to increase power.
This is the classic trade‑off:
Lower α → fewer false positives but more false negatives
Higher α → more false positives but fewer false negatives
3. Field Standards
Different disciplines have different norms:
Medicine, genetics, engineering: α = 0.01 or smaller
Psychology, education, business: α = 0.05
Exploratory research: α = 0.10
Students often think α = 0.05 is a law of nature — but it’s just a convention.
4. Sample Size and Power
If your sample size is small, lowering α too much can make it nearly impossible to detect real effects.
- Small sample → α = 0.05 is usually more practical
- Large sample → you can afford α = 0.01 without losing too much power
5. Multiple Comparisons
If you run many tests, the chance of a Type I error increases.
In those cases, you might:
- Lower α
- Use Bonferroni or Holm corrections
- Control the false discovery rate
This keeps your overall error rate under control.
Fresh, Original Examples
🧪 Example 1: Medical Trial
Testing whether a new drug reduces blood pressure.
- A false positive could harm patients.
- Researchers choose α = 0.01.
📚 Example 2: Classroom Intervention
A teacher tests whether a new reading program improves scores.
- A false positive is not dangerous.
- α = 0.05 is reasonable.
🏭 Example 3: Factory Safety Sensor
A sensor detects overheating in machinery.
- Missing a real overheating event is dangerous.
- They choose α = 0.10 to reduce Type II errors.
The Big Idea
Choosing α is not about math — it’s about risk management.
You’re deciding:
- How cautious you want to be
- What kinds of mistakes matter most
- How much evidence you require before making a claim
It’s a judgment call, not a formula.