90th percentile response time

The 90th percentile response time is a duration at or below which approximately 90 percent of the included response times fall, using a stated calculation method. It describes the slower portion of the distribution without being the single longest response. Its meaning depends on the interval, eligible records, and percentile convention.

What it tells a reader

An average can conceal a substantial group of longer responses. The 90th percentile helps show a threshold reached by most included responses while leaving roughly the slowest tenth above it. It does not explain why those longer responses occurred or establish a guarantee for the next call.

NIST's statistical handbook explains that percentiles use ordered observations and that several accepted interpolation methods exist. With small datasets, different methods can return noticeably different values. A label such as “90th percentile” should therefore be accompanied by a documented method. NIST percentile guidance.

The response interval must be defined separately. Washington's RCW 35.103.020, for example, defines response time in that chapter from the start of travel to arrival, after turnout. That is a named local legal convention, not the only interval used in public-safety reports. Washington definitions.

Example

Hypothetical example: Ten eligible responses have durations of 3, 4, 4, 5, 5, 6, 6, 7, 9, and 15 minutes, already sorted from shortest to longest.

Using the nearest-rank method, multiply 10 observations by 0.90 and round up to the next whole rank. The result is rank 9, so the 90th percentile is 9 minutes. Nine of the ten responses were at or below that duration; one took longer.

An interpolating method can produce a different answer between neighboring observations. Neither result should be compared with another agency's number without confirming the convention. The example names nearest rank specifically rather than suggesting every software package calculates percentiles that way.

Common confusion

The 90th percentile is not the average of the slowest ten percent. It is a threshold in an ordered distribution. Averaging selected long responses is a different statistic.

It is not a 90 percent compliance rate. Compliance asks what share met a predetermined target. A percentile instead finds a duration at a selected position in the observed data.

Component percentiles cannot generally be added. The response with a long turnout may have a short drive, while another has the reverse. Adding separate 90th percentiles of turnout and travel does not preserve which durations belonged to the same response. Calculate each response's total first, then find the percentile of those totals.

Define the included records

Specify the agency, period, service, priority, and unit or incident counting rule. A first-arriving-unit measure differs from a measure covering every assigned unit. Canceled responses and records without valid arrival times also need explicit treatment.

Report the number of included observations and the number excluded for missing or invalid timestamps. Excluding difficult or incomplete records can change the statistic, especially if those records are systematically slower.

Finally, avoid comparing unlike periods or populations. A percentile for all calls and a percentile for urgent medical responses describe different groups. The number becomes useful when its interval and eligibility rules are as visible as its minutes and seconds.

Related terms

• Percentile
• Response time
• Median
• Total response time

Sources

• NIST/SEMATECH — e-Handbook of Statistical Methods, Percentiles. Explains ordered data and multiple accepted percentile methods.
• Washington State Legislature — RCW 35.103.020, checked November 2, 2026. Provides a specifically scoped response/turnout interval example. Nearest-rank arithmetic and the component-percentile explanation are original.