Small-number volatility

Small-number volatility is the tendency for rates and percentages to change sharply when the underlying number of events is small. A change of only a few events can produce a large percentage increase or decrease. The calculation can be correct while the headline exaggerates how much information the change provides about a lasting trend.

Why small totals move so much

Percentage change divides the difference by the starting value. A small starting value makes each additional event a large share of that baseline. Population rates can show similar movement when a small community records a few more or fewer events.

The CDC's technical guidance on small health-event counts explains that rates based on small numbers can be unreliable and that suppressed values are not necessarily low. Its specific suppression threshold belongs to U.S. Cancer Statistics; it is not a universal threshold for police or fire data. The transferable point is the need to interpret small counts carefully. CDC reliability guidance.

A larger percentage change is not automatically a larger change in the number of people or events affected. Counts and percentages should be read together.

Example

Hypothetical example: A town records two qualifying burglaries in one year and four the next. The increase is two offenses. The percentage increase is (4 − 2) ÷ 2 × 100 = 100 percent.

A larger city records 200 and then 220 comparable burglaries. Its increase is 20 offenses, or 10 percent. The town has the larger percentage change, while the city has the larger numerical increase. Both statements are true, and neither alone explains the causes.

If the town has 5,000 residents in both years, its rate moves from 40 to 80 per 100,000. One additional offense would add another 20 per 100,000. The common rate scale does not make the small underlying count behave like a large sample.

Common confusion

Volatility is not proof that a change is meaningless. Each event matters, and a small count can warrant attention. The caution concerns drawing a broad trend or causal conclusion from limited observations.

A dramatic percentage is not necessarily an error. Doubling from two to four is genuinely a 100 percent increase. The problem arises when the count and baseline are hidden.

Zero is a special baseline. A change from zero to three cannot be expressed as an ordinary percentage increase because the formula divides by zero. State the change in counts rather than inventing an infinite or arbitrary percentage.

Adding useful context

Show the original counts, population, and time period beside the rate or percentage. Review several comparable periods to see whether the movement is isolated or persistent. A longer window can reduce short-term fluctuation, but it also changes the question and can conceal a recent shift; identify the window explicitly.

Check reporting coverage before attributing movement to chance or real-world conditions. BJS's LEARCAT documentation shows that agencies can contribute different numbers of months across years. A small apparent increase may reflect an added reporting month rather than a change in underlying events. LEARCAT methodology.

Where a source supplies uncertainty intervals or reliability flags, retain them. Do not manufacture a confidence interval without an appropriate statistical model. For a straightforward local chart, transparent counts, consistent definitions, and restrained wording often explain the limitation more clearly than an unsupported claim that the change is statistically significant.

Related terms

• Percentage change
• Crime rate
• Confidence interval
• Rolling average

Sources

• CDC — Suppression of Rates and Counts, June 1, 2026. Supports the general reliability concern for small counts; its cancer-specific threshold is not adopted for public safety.
• Bureau of Justice Statistics — LEARCAT Methodology, dated October 2, 2024. Supplies reporting-month and coverage context. All arithmetic examples are hypothetical and original.