Empirical Rule Overview
The empirical rule, also known as the 68-95-99.7 rule, states that for a normal (bell-shaped) distribution:
- 68% of data falls within 1 standard deviation (σ) of the mean (μ)
- 95% of data falls within 2 standard deviations of the mean
- 99.7% of data falls within 3 standard deviations of the mean
This rule applies only to approximately normal distributions.
The 68-95-99.7 Rule
Formulas
$$\text{68% of data: } \mu \pm 1\sigma$$ $$\text{95% of data: } \mu \pm 2\sigma$$ $$\text{99.7% of data: } \mu \pm 3\sigma$$
Visual Representation
68% 95% 99.7%
|----|----|---|
|
__|__
/ \
99.7%: |-----|-----|-----|
μ-3σ μ-σ μ μ+σ μ+3σ
95%: |--------|--------|
68%: |-----|-----|
Empirical Rule for Ungrouped Data
Example 1: Test Scores
A class of 200 students took a test with:
- Mean (μ) = 75
- Standard Deviation (σ) = 8
Estimate the number of students in each range.
Solution:
Within 1 Standard Deviation (μ ± σ):
Range: 75 ± 8 = [67, 83]
Expected percentage: 68%
Expected count: 200 × 0.68 = 136 students
Approximately 136 students scored between 67 and 83.
Within 2 Standard Deviations (μ ± 2σ):
Range: 75 ± 16 = [59, 91]
Expected percentage: 95%
Expected count: 200 × 0.95 = 190 students
Approximately 190 students scored between 59 and 91.
Within 3 Standard Deviations (μ ± 3σ):
Range: 75 ± 24 = [51, 99]
Expected percentage: 99.7%
Expected count: 200 × 0.997 ≈ 199 students
Approximately 199 students scored between 51 and 99.
Example 2: Heights Distribution
Heights of adult males are normally distributed with:
- Mean (μ) = 70 inches
- Standard Deviation (σ) = 2.5 inches
Find the percentage of men with heights in various ranges.
Solution:
Height between 65 and 75 inches (μ ± 2σ):
Range: 70 ± 5 = [65, 75]
Percentage: 95%
Height between 67.5 and 72.5 inches (μ ± 1σ):
Range: 70 ± 2.5 = [67.5, 72.5]
Percentage: 68%
Height less than 67.5 inches:
This is below μ - 1σ
Percentage: (100% - 68%) / 2 = 16%
Height between 72.5 and 75 inches:
From μ + 1σ to μ + 2σ
Percentage: (95% - 68%) / 2 = 13.5%
Empirical Rule for Grouped Data
For grouped data that is approximately normally distributed:
- Calculate the mean (μ) and standard deviation (σ)
- Determine the class intervals corresponding to μ ± 1σ, μ ± 2σ, μ ± 3σ
- Count the frequency in each interval
- Compare with expected percentages
Example: Grouped Data
| Class Interval | Frequency |
|---|---|
| 50-60 | 3 |
| 60-70 | 12 |
| 70-80 | 25 |
| 80-90 | 32 |
| 90-100 | 22 |
| 100-110 | 5 |
| 110-120 | 1 |
N = 100, μ = 80, σ = 10
Verify the empirical rule.
Solution:
Within 1σ (70-90):
Frequency: 25 + 32 = 57
Percentage: 57% (close to 68%)
Within 2σ (60-100):
Frequency: 12 + 25 + 32 + 22 = 91
Percentage: 91% (close to 95%)
Within 3σ (50-110):
Frequency: 3 + 12 + 25 + 32 + 22 + 5 = 99
Percentage: 99% (close to 99.7%)
The data approximately follows the empirical rule, confirming it’s approximately normally distributed.
Detailed Distribution Breakdown
| Range | Percentage | Count (if n=100) |
|---|---|---|
| μ - 3σ to μ - 2σ | 2.15% | 2-3 |
| μ - 2σ to μ - 1σ | 13.5% | 13-14 |
| μ - 1σ to μ | 34% | 34 |
| μ to μ + 1σ | 34% | 34 |
| μ + 1σ to μ + 2σ | 13.5% | 13-14 |
| μ + 2σ to μ + 3σ | 2.15% | 2-3 |
Outlier Detection Using Empirical Rule
Values beyond μ ± 3σ can be considered extreme outliers:
- Beyond μ + 3σ: Extremely high values (0.15% chance)
- Beyond μ - 3σ: Extremely low values (0.15% chance)
This provides a rule-of-thumb for identifying unusual observations.
Conditions for Using the Empirical Rule
- Data must be approximately normally distributed
- Data should be roughly bell-shaped
- Mean and median should be approximately equal
- Data should not be heavily skewed
Applications
- Quality Control: Identifying defects outside acceptable ranges
- Risk Assessment: Determining probability of extreme events
- Academic Performance: Grading and benchmarking
- Manufacturing: Process capability analysis
- Investment Analysis: Portfolio risk assessment
- Medical Testing: Reference ranges for health metrics
Limitation
The empirical rule is specifically for normal distributions. For non-normal distributions, use Chebyshev’s theorem instead, which provides more conservative estimates valid for any distribution.