β = P(Type II Error) = P(Fail to Reject H₀ | H₀ False)
Type I Error: Rejecting a true null hypothesis (false positive)
Type II Error: Failing to reject a false null hypothesis (false negative)
Alpha (α): Probability of Type I error; significance level of test
Beta (β): Probability of Type II error
Power: 1 - β = P(correctly rejecting false H₀)
| Reality | H₀ True | H₀ False |
|---|---|---|
| Decision | ||
| Reject H₀ | Type I Error (α) | Correct (1-β) |
| Fail to Reject H₀ | Correct (1-α) | Type II Error (β) |
H₀: Patient does not have the disease
H₁: Patient has the disease
Understanding the consequences of each error type
Type I Error: Reject H₀ when it's true
Consequence: Test says patient has disease, but patient doesn't
Also called: False Positive
Type II Error: Fail to reject H₀ when it's false
Consequence: Test says patient doesn't have disease, but patient does
Also called: False Negative
Type I Error consequence: Unnecessary treatment, anxiety, costs
Type II Error consequence: Missed diagnosis, delayed treatment, potential harm
For serious diseases: Type II error may be more severe
For screening tests: May accept higher Type I error to minimize Type II
Type I Error: False positive - patient told they have disease when they don't
Type II Error: False negative - patient told they don't have disease when they do
• Error Definition: Type I = Reject true H₀, Type II = Fail to reject false H₀
• Context Matters: Severity depends on application domain
• Consequence Analysis: Evaluate impact of each error type
Given: α = 0.05, Power = 0.80
Power = 1 - β
0.80 = 1 - β
β = 1 - 0.80 = 0.20
α = P(Type I error) = 0.05
Expected Type I errors = 1000 × 0.05 = 50
β = P(Type II error) = 0.20
Expected Type II errors = 1000 × 0.20 = 200
α + (1-α) = 0.05 + 0.95 = 1 ✓
β + (1-β) = 0.20 + 0.80 = 1 ✓
On average, 5% of true null hypotheses will be rejected (Type I)
On average, 20% of false null hypotheses will not be rejected (Type II)
Probability of Type II error (β) = 0.20
Expected Type I errors in 1000 tests = 50
Expected Type II errors in 1000 tests = 200
• Power-Beta Relationship: Power = 1 - β
• Expected Values: Expected errors = n × probability
• Probability Complement: α + (1-α) = 1 and β + (1-β) = 1
Initial: α = 0.05, β = 0.30, Power = 0.70
Goal: α = 0.01
Decreasing α makes rejection region smaller
This increases acceptance region
More difficult to reject H₀, even when it's false
When α decreases, β increases
New β will be greater than 0.30
Maybe β ≈ 0.45-0.50
Power = 1 - β
If β increases, power decreases
New power will be less than 0.70
Fewer false alarms (Type I errors)
But more missed defects (Type II errors)
Lower probability of detecting true problems
To maintain power while reducing α:
Increase sample size
Increase effect size
Reduce variability
When α decreases from 0.05 to 0.01:
β will increase (from 0.30 to approximately 0.45-0.50)
Power will decrease (from 0.70 to approximately 0.55-0.50)
• Alpha-Beta Trade-off: α ↓ ⇒ β ↑ (inverse relationship)
• Power-Affect: Power = 1 - β, so β ↑ ⇒ Power ↓
• Compensation Strategies: Increase sample size to maintain power
Initial: n₁ = 25, Power₁ = 0.70, β₁ = 0.30
New: n₂ = 100
Standard error = σ/√n
As n increases, SE decreases
Sample means become more precise
With smaller SE, easier to detect true differences
β will decrease (fewer Type II errors)
Maybe β₂ ≈ 0.10-0.15
Power = 1 - β
If β decreases, power increases
Power₂ will be greater than 0.70
Maybe Power₂ ≈ 0.85-0.90
Power increases approximately with √n
Going from n = 25 to n = 100 is 4× increase
This corresponds to 2× improvement in precision
Larger samples detect smaller effects
Higher probability of finding true relationships
But more expensive and time-consuming
With n = 100, power will increase significantly (to approximately 0.85-0.90)
β will decrease significantly (to approximately 0.10-0.15)
Power increases with sample size, improving ability to detect true effects
• Sample Size-Power: Larger n ⇒ Higher power ⇒ Lower β
• Standard Error: SE = σ/√n, so precision increases with √n
• Cost-Benefit: Consider trade-off between power and resources
H₀: Drug is ineffective
H₁: Drug is effective
Given: α ≤ 0.01, β = 0.20
Power = 1 - β = 1 - 0.20 = 0.80
Type I Error: Approve ineffective drug (α ≤ 0.01)
Type II Error: Reject effective drug (β = 0.20)
Type I concern: Taking ineffective medication (low risk due to α ≤ 0.01)
Type II concern: Missing out on beneficial treatment (higher risk due to β = 0.20)
Type I cost: Legal liability, reputation damage
Type II cost: Lost revenue, R&D investment lost
Low α protects public from harmful drugs
Higher β may delay beneficial treatments
Regulatory agencies must balance both risks
Power = 0.80
Type I Error: Approve ineffective drug (probability ≤ 0.01)
Type II Error: Reject effective drug (probability = 0.20)
• Regulatory Standards: α set by regulatory requirements
• Risk Assessment: Consider consequences for all stakeholders
• Balance Principle: Regulators often prioritize Type I error control