Normal distribution and the empirical rule

The normal distribution is a probability model for continuous measurements whose values tend to cluster around a mean and taper into two tails. Its familiar bell shape is often called the Gaussian curve. It is a model rather than a guarantee: real observations may only resemble the curve.

The empirical rule provides a quick way to estimate probability from a mean and standard deviation. Roughly 68% of values lie within one standard deviation of the mean, 95% lie within two, and 99.7% lie within three. These percentages are most reliable when the bell-shaped model is appropriate.

What the bell curve represents

The mean marks the centre of a normal distribution, while the standard deviation measures typical distance from that centre. A small standard deviation creates a narrow, tall curve; a larger one creates a broader, flatter curve. The area under the entire curve equals 1, representing all possible outcomes.

For example, suppose the heights of a hypothetical group of Australian adults have a mean of 172 centimetres and a standard deviation of 9 centimetres. The curve would be centred at 172, and most heights would be concentrated around that value. Because the distribution is symmetric, equally sized portions of the data appear on either side of the mean.

How the empirical rule estimates probability

The rule divides the curve into practical bands. Approximately 34% of values lie between the mean and one standard deviation above or below it, giving 68% within one standard deviation. Another 13.5% lies in each band between one and two standard deviations, while about 2.35% lies in each band between two and three.

Consider students with hypothetical marks averaging 70 and a standard deviation of 8. The interval from 54 to 86 represents the mean plus or minus two standard deviations, so the empirical rule estimates that about 95% of marks fall in that range. Extending the interval to 46 through 94 captures roughly 99.7%. These are estimates based on the assumed shape of the marks distribution, not exact guarantees.

Turning raw values into standard deviations

A z-score expresses how far a value sits from the mean in units of standard deviation:

[ z=\frac{x-\mu}{\sigma} ]

Here, (x) is the value, (\mu) is the mean and (\sigma) is the standard deviation. If a student scores 86 on the earlier example, then (z=(86-70)/8=2). The score is therefore two standard deviations above the mean and belongs near the boundary of the central 95% interval.

Standardisation also makes different measurements comparable. A result in dollars, centimetres or marks can be analysed in the same way once its z-score is known. This is useful when combining information from markets, laboratories, schools and weather records.

When the approximation is appropriate

A normal model works best for continuous data that is roughly symmetric, unimodal and concentrated around one centre. A histogram can provide an initial check: it should have one peak, balanced tails and no extreme gaps or isolated clusters. Sample size alone cannot repair a strongly skewed distribution, so transformations or a different model may be needed.

Australian Bureau of Statistics data illustrates the issue clearly. Household income figures are often right-skewed because a small number of households have incomes far above the main cluster. Likewise, daily rainfall in Sydney is usually lumpy and highly skewed, making the empirical rule unsuitable for the raw observations. Averages calculated from larger samples may be closer to normal, even when individual measurements are not.

A reliable calculation workflow

Several checks help prevent common errors when applying the empirical rule:

• Check the distribution: Examine the data’s shape and decide whether a normal approximation is reasonable.
• Keep units consistent: Ensure the value, mean and standard deviation use the same units, such as dollars and cents or kilograms.
• Calculate the mean and spread: Use the appropriate standard deviation for the data and state whether it is a population or sample value.
• Convert each target into a z-score: Subtract the mean and divide by the standard deviation.
• Match the z-score to an area: Use one, two or three standard deviations as the main guide, then state the result as approximate.

Intermediate percentages may need small adjustments when a z-score falls between whole standard deviations. For example, about 16% of a normal distribution lies below one standard deviation under the mean. A z-score of (-1) therefore corresponds to roughly the 16th percentile. Rounding should happen only at the final stage so that earlier calculations remain accurate.

Applying the rule to an Australian market dataset

Imagine a student records prices for 500-gram packets of apples at Coles and Woolworths supermarkets in Sydney, Melbourne, Brisbane and Perth. Suppose the selected, non-promotional prices have a mean of $4.80 and a standard deviation of $0.60. The central 95% interval would run from $3.60 to $6.00 because these endpoints are two standard deviations from the mean.

A price of $5.40 has a z-score of 1, so about 84% of comparable prices would be at or below that amount. A price of $3.60 has a z-score of (-2), placing it near the lower 2.5% boundary. The result gives a useful estimate for budgeting or comparing stores, although promotions, regional differences and differences between apple varieties can affect the actual distribution.

The same caution applies when using Australian Government, retail or Bureau of Meteorology data. Read the definitions, check the units and consider whether outliers or seasonal effects make the bell curve inappropriate. The practical takeaway is to standardise first, identify the relevant band under the curve and report an approximate probability while naming the assumptions behind it.