The 68-95-99.7 Rule

Why Three Ranges Are Useful

When a distribution is approximately bell-shaped, a quick description of its spread can be more informative than a list of individual measurements. The mean gives the center, and the standard deviation describes a typical distance from that center. The 68-95-99.7 rule connects these two summaries to approximate proportions: it tells us how much of a normal distribution lies within one, two, and three standard deviations of its mean.

The rule is useful for describing central ranges, making rough estimates, and noticing values that are far from the center. It is a shortcut for a normal model, not a guarantee about every data set and not a substitute for exact probability calculations when precision matters.

Statement of the Rule

“Within \(k\) standard deviations” means from \(\mu-k\sigma\) through \(\mu+k\sigma\), including both endpoints. For example, within two standard deviations means between \(\mu-2\sigma\) and \(\mu+2\sigma\). The rule concerns proportions under a normal distribution model; for a continuous normal variable, the probability at any single endpoint is zero, so including or excluding the endpoints does not change the probability.

The percentages are approximate rounded values. More precise normal-model probabilities are about 68.27%, 95.45%, and 99.73% for the three ranges. The shorter name “68-95-99.7 rule” uses the commonly rounded figures.

A normal distribution is symmetric about its mean. Consequently, the portions outside each central range are split equally between the lower and upper tails. For instance, about 5% lies outside two standard deviations in total, or about 2.5% in each tail. This symmetry applies to a normal model; it should not be assumed for a distribution that is skewed or otherwise not approximately normal.

Why the Percentages Have These Values

The standard normal variable \(Z\) has mean 0 and standard deviation 1. If \(X\sim N(\mu,\sigma^2)\), standardizing gives \(Z=(X-\mu)/\sigma\), so the event that \(X\) is within \(k\) standard deviations of its mean is the event that \(Z\) is between \(-k\) and \(k\). Write \(\Phi(z)=P(Z\le z)\) for the standard normal cumulative probability. The central probability is therefore the difference between the cumulative probabilities at the upper and lower endpoints.

For \(k=1\), standard normal probabilities are approximately \(\Phi(1)=0.8413\) and \(\Phi(-1)=0.1587\). Thus the probability between them is \(0.8413-0.1587=0.6826\), which rounds to about 68%. For \(k=2\), the corresponding calculation is \(0.9772-0.0228=0.9544\), about 95%. For \(k=3\), it is \(0.99865-0.00135=0.99730\), about 99.7%. These calculations justify the rounded rule; they also show why it is approximate rather than a set of exact integer percentages.

Using the Rule in Original Units

The rule is stated in standard-deviation units, but most applications use the measurement’s original units. To find a central range, multiply the standard deviation by 1, 2, or 3, then add and subtract that amount from the mean. The resulting interval has the same units as the measurement. A normal model must be plausible for the variable being described before the rule is used.

Estimating Proportions and Counts

The rule can also turn a proportion into a rough count. If a normal model assigns about 68% of values to a range, then in a large group modeled by that distribution, roughly 68% of the group’s values would be expected to fall in the range. Multiplying the proportion by the group size gives an estimated count. That count is not guaranteed: actual data can differ because of sampling variation, and the rule itself is rounded.

Recognizing Values Far From the Mean

The outer range is useful as a quick reference for unusually distant values. Under a normal model, about 99.7% of values lie within three standard deviations of the mean, leaving only about 0.3% beyond that range in total. Because the distribution is symmetric, about half of that remainder lies in each tail. A value more than three standard deviations from the mean is therefore quite unusual under the model, although unusual does not mean impossible.

The calculation is a comparison with the model, not a diagnosis of why an observation occurred. A far-out value could reflect a genuine rare case, a recording error, or a model that does not describe the data well. Context and data checks matter before drawing conclusions.

When the Rule Is Appropriate

The rule is most useful when the distribution of the variable is approximately normal: roughly symmetric, single-peaked, and bell-shaped, without severe departures that make the normal model a poor description. The rule is exact only as a statement about a normal distribution, with the displayed percentages rounded. For real data, it is an approximation whose usefulness depends on whether that model is reasonable.

A normal model is often specified directly for a variable. When working with observed data instead, first consider the shape of the distribution. If the data are strongly skewed, have multiple peaks, or contain influential extreme values, the normal rule may give a misleading description. It is also important to distinguish a model statement from a guarantee about a finite sample: an approximate model proportion does not force a particular sample to contain exactly that percentage.

Common Mistake: Applying the Rule to Any Data

A common error is to see a mean and standard deviation, calculate \(\mu\pm2\sigma\), and automatically say that 95% of the observations fall in that range. The mean and standard deviation alone do not imply the 68-95-99.7 pattern. That pattern depends on a normal distribution, or on data whose shape is close enough to normal for the approximation to be useful.

The correction is to state the model and assess its suitability before applying the rule. If the normal model is plausible, describe the resulting proportion as approximate and in context. If it is not plausible, do not use this rule to claim a percentage; a different description of the data is needed.