Percentiles as Measures of Relative Standing

What a Percentile Tells You

A measurement by itself says how much or how many: a score of 82 points, a wait of 14 minutes, or a weight of 3.4 kilograms. A percentile gives a different kind of information. It locates a value within a distribution by describing the share of observations at or below that value. This makes percentiles useful when the question is not only “What was the measurement?” but also “How does it compare with the other measurements?”

For instance, if a score is at the 80th percentile in a reference group, it is higher than or equal to the scores of about 80% of that group. The percentile is not the percentage of questions answered correctly, nor does it say that the score is 80 points. It describes relative position in a specified distribution.

The letter \(p\) denotes the percentile level, expressed on a 0-to-100 scale. Thus, the 25th percentile is the cutoff associated with a cumulative share of about 25%; the 90th percentile is the cutoff associated with about 90%. The “th” in “90th” names a location in the distribution, not a unit of measurement.

The percentile value must be read in the context of both the variable and the reference group. Saying “the 70th percentile is 45” is incomplete unless it is clear what was measured, what units 45 uses, and which distribution was used to define the percentile. A value at the 70th percentile among one group may not occupy the same percentile in another group.

Reading Percentiles in Context

A complete interpretation identifies the cutoff, the measured variable and its units, the relevant reference group, and the share of that group at or below the cutoff. The words “about” or “approximately” are often appropriate for sample percentiles because finite data and calculation conventions can prevent the share from being exactly \(p\%\).

The phrase “at or below” matters. A percentile describes a cumulative share, not just the share with values exactly equal to the cutoff. If observations are distinct and a distribution is continuous, very few or no observations may equal a particular cutoff. In recorded data, several observations can have the same value, so some may tie at the cutoff. In that case, the share at or below it can differ from the nominal percentile level.

For an idealized continuous distribution, let \(q_p\) be its \(p\)th percentile. The cumulative probability at that cutoff is \(p/100\). Since all observations are either at or below \(q_p\), or above \(q_p\), the share above it is the complement of the share at or below it:

For example, at the 60th percentile, the share above the cutoff is approximately \(1-0.60=0.40\), or 40%. This complement calculation is a way to understand the location described by a percentile; it does not mean the cutoff splits the distribution into equal groups. Only the 50th percentile divides the cumulative share into approximately 50% at or below and 50% above, under the continuous-distribution setting.

Comparing Relative Standing

Percentiles can help compare relative positions when measurements use different scales. A higher percentile means a greater share of the relevant reference group is at or below that value. It does not necessarily mean the underlying measurement is larger in its original units, and it does not show how far apart two measurements are.

The reference group is essential to this interpretation. A percentile may be calculated among people of a particular age, students taking a particular test, or measurements collected under specified conditions. Changing the reference group can change the percentile associated with the same raw value. Always ask, “At or below this value among whom?”

The median is the middle location in an ordered distribution and is commonly associated with the 50th percentile. This connection is useful for orientation: the 50th percentile is a central location, while percentiles such as the 10th or 90th describe positions toward the lower or upper portions of the distribution. A percentile is still a location, not a measure of how spread out the data are.

Percentile Values Are Not Percentages of a Measurement

A percentile level and a measurement value answer different questions. The percentile level describes a share of observations. The percentile value is the cutoff, with units, associated with that share. If a 90th-percentile wait time is 42 minutes, then 90 is not a number of minutes, and 42 is not a percentage. The statement concerns where 42 minutes falls in the wait-time distribution.

Likewise, a percentile does not describe an individual’s probability of a future outcome. If someone is at the 90th percentile of a measurement in a reference group, the statement is about the person’s position in that group’s distribution. It does not mean there is a 90% chance of a particular future event.

Why the Exact Sample Percentile Can Vary

For a theoretical distribution, a percentile can be defined through its cumulative share. A finite sample presents an additional issue: the data consist of a limited number of observed values, and a requested percentile may fall between recorded values. Different standard calculation conventions can handle that situation differently. Tied observations add another complication because the share at or below a tied cutoff may jump past the target share.

For this reason, a percentile value calculated from a sample is not always uniquely determined until a calculation convention is specified. The interpretation remains useful: it identifies a value associated with a target cumulative share. But when an exact sample value or exact percentile rank is needed, the chosen convention and treatment of ties must be considered. The broad interpretation should not be mistaken for a promise that exactly \(p\%\) of a finite sample will lie at or below the reported value.

This distinction matters in practice. A percentile cutoff can summarize a reference distribution, support comparisons, or help describe unusually low or high values. But decisions based on a cutoff should account for the population to which the reference applies and how the percentile was obtained. A percentile from one population should not automatically be transferred to another.

Common Mistakes

• Confusing a percentile with a percent correct. The 60th percentile describes relative position in a reference distribution; it does not mean 60% of questions were answered correctly.
• Leaving out the reference group. “At the 80th percentile” is incomplete without saying among which observations or population.
• Reading the percentile level as the measurement. A 90th-percentile value of 42 minutes means the cutoff is 42 minutes, not 90 minutes.
• Assuming an exact count in a finite sample. Percentiles describe a target cumulative share, but ties and calculation conventions can affect the observed count at or below a sample cutoff.
• Comparing raw units through percentile differences. A 15-percentile-point difference describes a difference in relative standing, not a difference of 15 measurement units or 15%.