Empirical Rule Calculator (68-95-99.7)

Turn a mean and standard deviation into the three intervals that hold roughly 68%, 95%, and 99.7% of normally distributed data — the fastest way to judge what counts as typical and what counts as unusual.

Before You Calculate

  • The rule assumes your data is at least approximately normal: one peak, roughly symmetric, no heavy tails.
  • Use the mean and standard deviation of the same dataset — mixing a historical mean with this month's spread produces meaningless ranges.
  • If you only have raw data, compute both statistics first; our standard deviation calculator handles that step.

What the Tool Returns

μ ± 1σ: The interval covering about 68% of observations — the everyday, expected range.

μ ± 2σ: The interval covering about 95% — values outside it deserve a second look.

μ ± 3σ: The interval covering about 99.7% — values outside it are rare enough to treat as anomalies.

What the Empirical Rule Says

For a normal (bell-shaped) distribution, the share of observations falling within a fixed number of standard deviations of the mean is always the same, regardless of what is being measured:

P(μ − 1σ ≤ X ≤ μ + 1σ) ≈ 68%

P(μ − 2σ ≤ X ≤ μ + 2σ) ≈ 95%

P(μ − 3σ ≤ X ≤ μ + 3σ) ≈ 99.7%

Where:

  • μ = the mean (center) of the distribution
  • σ = the standard deviation (spread)
  • X = a randomly chosen observation

This universality is what makes the rule so useful: heights in centimeters, reaction times in milliseconds, and portfolio returns in percent all obey the same three percentages once you standardize by their own μ and σ.

Why 68, 95, and 99.7?

The three percentages are areas under the standard normal curve. Integrating the normal density between z = −1 and z = +1 gives 0.6827; between ±2 it gives 0.9545; and between ±3 it gives 0.9973. The familiar 68-95-99.7 statement simply rounds these to memorable values.

The rounding hides one subtlety worth knowing: an interval that captures exactly 95% of a normal distribution actually extends to ±1.96 standard deviations, not ±2. That is why confidence intervals are built with 1.96 while quick mental estimates use 2 — the two conventions differ by less than half a percentage point of coverage.

Checking Whether the Rule Applies to Your Data

The empirical rule is only as good as the normality assumption behind it. Quick checks before relying on the ranges:

  • Shape: A histogram should show a single peak with roughly symmetric tails. Income, house prices, and waiting times are typically right-skewed and break the rule.
  • Bounds: If μ − 2σ lands below a hard physical limit (a negative time, a negative count), the data cannot be normal near that tail.
  • Outlier frequency: Under normality only about 0.3% of values fall beyond 3σ. Seeing them regularly signals heavy tails.

When normality fails, Chebyshev's inequality is the distribution-free fallback: at least 75% of any dataset lies within 2 standard deviations of the mean and at least 88.9% within 3 — weaker guarantees, but ones that hold for every distribution.

Real Applications

The rule converts abstract spread into actionable thresholds.

  • Manufacturing: Control charts flag measurements beyond μ ± 3σ as evidence a process has drifted — the origin of the phrase "three-sigma limits."
  • Medicine: Laboratory reference ranges are often set at μ ± 2σ of a healthy population, so about 5% of healthy patients fall outside them by design.
  • Education: Standardized test reports translate a score into "within one standard deviation of average" language for parents.
  • Finance: Risk teams describe daily returns beyond two standard deviations as notable moves and beyond three as tail events.

Worked Example: Exam Scores

Final exam scores in a large statistics course are approximately normal with mean μ = 75 and standard deviation σ = 6. Applying the rule:

  1. One standard deviation: 75 − 6 = 69 and 75 + 6 = 81, so about 68% of students scored between 69 and 81.
  2. Two standard deviations: 75 − 12 = 63 and 75 + 12 = 87, so about 95% scored between 63 and 87.
  3. Three standard deviations: 75 − 18 = 57 and 75 + 18 = 93, so about 99.7% scored between 57 and 93.

The intervals immediately classify individual results. A score of 90 sits (90 − 75) / 6 = 2.5 standard deviations above the mean — outside the 95% band but inside the 99.7% band, putting it roughly in the top 1% of the class. A score of 55 lies below μ − 3σ = 57, an outcome expected for fewer than 2 students in a thousand, which might prompt a check for a recording error or a student who needs support.

Entering mean 75 and standard deviation 6 into the calculator returns exactly these three ranges: 69 to 81, 63 to 87, and 57 to 93.

Frequently Asked Questions

Does the empirical rule work for non-normal data?

Not reliably. The 68-95-99.7 percentages are properties of the normal curve specifically. For skewed or heavy-tailed data the actual coverage can differ substantially. Chebyshev's inequality provides distribution-free minimums instead: at least 75% of values within 2 standard deviations and at least 88.9% within 3, for any distribution whatsoever.

Are the percentages 68, 95, and 99.7 exact?

They are rounded. The exact areas under the normal curve are 68.27% within one standard deviation, 95.45% within two, and 99.73% within three. For an interval holding exactly 95%, you would use plus or minus 1.96 standard deviations, which is the constant used in confidence intervals.

Should I use the population or sample standard deviation?

Use whichever correctly describes your data. If you computed sigma from a sample, the sample standard deviation (dividing by n - 1) is the right input, and the resulting ranges are estimates subject to sampling error. With small samples the ranges can shift noticeably from sample to sample.

How is the empirical rule related to z-scores?

A z-score counts how many standard deviations an observation sits from the mean, so the rule is a statement about z-scores: about 68% of normal data has |z| < 1, 95% has |z| < 2, and 99.7% has |z| < 3. Computing a z-score and comparing it to these thresholds is the same operation as checking which interval the value falls in.

Can I use the rule to detect outliers?

Yes, it is one of the oldest outlier heuristics. Values beyond 3 standard deviations occur only about 3 times in 1000 under normality, so they warrant investigation. Be careful in two situations: small samples, where extreme values distort the mean and standard deviation themselves, and non-normal data, where 3-sigma events may be perfectly ordinary.