Comparing Crime Rates Using Mean, Median and Standard Deviation

One unusually high crime rate can change the story told about a whole group of areas. Four places might have fairly similar rates, yet a fifth can pull the average far above what most residents see locally. To understand that difference, we need more than a ranked list. The mean, median and standard deviation each answer a different question about the same numbers.

Start With Five Hypothetical Crime Rates

Suppose we select five areas from a larger set of comparable areas. Each rate covers the same year, uses the same offence categories and measures recorded offences per 1,000 residents. Our initial dataset is:

10, 12, 14, 16, 18 offences per 1,000 residents.

These are invented figures for learning the calculations. They do not describe actual places. Each observation represents one area's rate, and each area receives the same weight in our calculations.

To test the effect of an unusually high value, we will replace 18 with 48. The revised dataset becomes:

10, 12, 14, 16, 48 offences per 1,000 residents.

Only one observation changes. That makes it easier to see which summaries respond and how much they move. We are comparing two hypothetical versions of the dataset, rather than claiming that crime increased in a real area.

Understand Why Rates Differ From Counts

A crime count tells you how many offences were recorded. A crime rate relates that count to a population. For a rate per 1,000 residents, the calculation is:

Crime rate = (recorded offences ÷ resident population) × 1,000.

An area with 200 recorded offences and 10,000 residents has a rate of 20 per 1,000. Another area with 600 recorded offences and 60,000 residents has a rate of 10 per 1,000. The second area has three times as many offences, yet half the rate.

This population adjustment makes areas of different sizes easier to compare. It does not make every comparison fair automatically. A resident population may not capture the number of commuters or visitors spending time in an area. Recorded offences also depend on reporting and recording practices.

A rate of 20 offences per 1,000 residents does not necessarily mean that 20 different residents experienced an offence. One person may experience several offences. Some offences may involve businesses or visitors. The numerator counts offences under the dataset's rules, rather than necessarily counting distinct resident victims.

Calculate the Mean and Watch It Move

The mean is the sum of the observations divided by their number. For our first dataset, the total is:

10 + 12 + 14 + 16 + 18 = 70.

There are five observations, giving:

Mean = 70 ÷ 5 = 14 offences per 1,000 residents.

After replacing 18 with 48, the total becomes 100:

Revised mean = (10 + 12 + 14 + 16 + 48) ÷ 5 = 20.

The mean rises from 14 to 20, even though four of the five rates stay unchanged. Increasing one rate by 30 increases the mean by 30 ÷ 5, or 6.

The revised mean now exceeds the rates in four of the five areas. It remains a correct arithmetic summary, but it gives a different impression from the experience represented by most observations. High values pull the mean upward because every value contributes to the total.

Find the Median and Check the Middle

The median is the middle observation after sorting the values. With five observations, its position is:

Middle position = (5 + 1) ÷ 2 = 3.

The third value in 10, 12, 14, 16, 18 is 14. The third value in 10, 12, 14, 16, 48 is also 14. Our change therefore leaves the median unchanged.

This is why the median can be helpful when a dataset contains unusually high or low observations. Moving the largest value farther upward does not change the middle position here.

That does not make the median immune to changes. If enough observations move, or if the middle values change, the median can move too. For an even number of observations, calculate it by averaging the two middle values.

Read Crime Rankings With the Underlying Rates in View

Real-world crime rankings provide a context for asking these statistical questions. Before comparing ranked rates, check that the offence definitions and reporting periods are consistent. A full-year rate for one set of offences cannot be treated as directly equivalent to a six-month rate for a different set.

Population estimates and geographic boundaries also deserve attention. A change in the population denominator or the area covered can alter a rate even when the recorded offence count stays the same.

A ranking alone does not describe variation across areas. Both of our hypothetical datasets put the five observations in the same order. Yet the highest rate sits only 2 above the next highest rate in the first dataset, compared with 32 in the revised dataset.

First place therefore tells us nothing about the size of the lead. To describe those gaps across the full dataset, we need a measure of spread.

Calculate Sample Standard Deviation Step by Step

Standard deviation describes how spread out observations are around their mean. We will use the sample formula because our five areas are being treated as a sample from a larger collection:

s = √[Σ(x_i − x̄)² ÷ (n − 1)].

Here, x_i represents an individual rate, x̄ is the sample mean, and n is the number of observations. The symbol Σ means to add the specified quantities.

For the revised dataset, follow these steps:

1. Find the mean. The rates total 100, giving a mean of 20.
2. Subtract the mean from each rate. The deviations are −10, −8, −6, −4 and 28.
3. Square each deviation and add the results. This gives 100 + 64 + 36 + 16 + 784 = 1,000.
4. Divide by n − 1. With five observations, the sample variance is 1,000 ÷ 4 = 250.
5. Take the square root. The sample standard deviation is √250, approximately 15.81 offences per 1,000 residents.

Squaring prevents positive and negative deviations from cancelling. It also gives greater weight to larger departures from the mean. The rate of 48 contributes 784 to the squared-deviation total of 1,000, which explains much of the revised dataset's spread.

Compare the Spread Before and After the High Value

The original dataset has a mean of 14. Subtracting that mean from 10, 12, 14, 16 and 18 gives deviations of −4, −2, 0, 2 and 4.

Squaring and adding gives:

16 + 4 + 0 + 4 + 16 = 40.

The original sample variance is 40 ÷ 4 = 10. Its sample standard deviation is therefore √10, approximately 3.16 offences per 1,000 residents.

Replacing 18 with 48 raises the sample standard deviation from about 3.16 to 15.81. That is a fivefold increase. The mean rises from 14 to 20, while the median stays at 14.

Each result contributes something useful. The mean captures the increase in the total of the area rates. The median shows that the middle observation stays fixed. The standard deviation shows that the observations become much more dispersed.

Keep the Formula and Units Straight

Sample variance uses squared units. In our revised example, 250 is measured in squared units of offences per 1,000 residents. Taking the square root returns standard deviation to the original rate units, making 15.81 easier to interpret alongside the observations.

The denominator also matters. If these five areas were the entire population we wanted to describe, the population standard deviation would divide the squared-deviation total by 5. That would give √200, approximately 14.14. Our sample calculation divides by 4 and gives 15.81.

Using n − 1 accounts for estimating the mean from the sample when estimating population variance. It does not correct an unrepresentative selection of areas. A sample chosen only from unusually busy city centres would still need careful interpretation.

Standard deviation is also not the literal average absolute distance from the mean. It comes from squared deviations and a square root. Treating it as a measure of spread avoids confusing it with a different statistic.

Choose a Summary That Fits Your Question

If your question concerns the middle area in this dataset, the median of 14 is a useful answer. If your question concerns the arithmetic average of the five area rates, use the mean of 20. If you want to communicate how widely those rates differ, report the sample standard deviation of 15.81 alongside the centre.

Be careful with another distinction: an average of area rates is not generally the combined rate for all residents. Our calculation gives each area equal weight. A combined rate requires adding the offence counts, adding the populations, and dividing those totals before multiplying by 1,000.

Avoid assuming that most rates must fall within one standard deviation of the mean. Familiar percentage rules depend on distribution shape. Five observations with one high value provide little basis for assuming a bell-shaped distribution.

Let the Middle and the Spread Tell Their Parts

A clear report of our revised example could read: “Across five sampled areas, the mean crime rate was 20 and the median was 14 offences per 1,000 residents. The sample standard deviation was approximately 15.81. Four rates ranged from 10 to 16, while the fifth was 48.”

That description gives readers more than a single average or rank. It shows where the middle sits, how far the values spread, and which observation drives the difference. With consistent offence definitions, reporting periods and population measures, those three statistics support a much clearer comparison.