Head-to-head
Percentile Rank vs Z-Score
Both numbers answer where a value sits in a group, and they answer it in different currencies. A percentile rank is measured in people: it says what share of the group scored below you. A z-score is measured in standard deviations: it says how far the value is from the mean, in units of spread.
Last updated
The percentile rank is the more robust of the two because it needs nothing but an ordering. It works on a bimodal distribution, a skewed one, or a set of exam marks with no assumption about shape at all. Its weakness is resolution: it discards the distances between values, so a group clustered at 70 and 71 produces the same percentile spread as one spread evenly from 40 to 100.
A z-score keeps those distances, which is what makes it useful for combining measurements on different scales — a height and a weight, or two tests marked out of different totals. It is also the input to everything downstream: confidence intervals, p-values and the normal table all take a z-score.
The catch is the assumption. A z-score is only interpretable against a shape, usually the normal curve, and converting it to a percentile quietly imports that assumption. In a skewed distribution the z-score is still arithmetically correct while the percentile you derive from it is not.
Percentile rank
How To Calculate A Percentile Rank
A percentile rank tells you what fraction of a group sits below a value. For 12th out of 250 there are 238 below, and the two standard conventions give 0.9558232931726908 and 0.952 — answers that differ by less than one person, yet must be kept distinct from the percentage score.
Z-score
How To Calculate A Z-Score
A z-score restates a value as how many standard deviations it sits from the mean. That rescaling removes the original units, which is exactly what lets you compare things that were never commensurable.
How They Differ
| Attribute | Percentile rank | Z-score |
|---|---|---|
| Measured in | People below you, out of 100 | Standard deviations from the mean |
| What it needs | The rank, or the whole distribution | The mean and the standard deviation |
| Assumes a shape | No — it is a count | Yes, once you turn it into a probability |
| Keeps distances between values | No, only the ordering | Yes |
| Typical range | 1 to 99 | Roughly −3 to +3 |
| Combines different scales | No | Yes, that is its main use |
| Sensitivity to one outlier | Barely moves — one place at most | Large, because the mean and the spread both move |
| Feeds into | Reporting and ranking | Confidence intervals, p-values, the normal table |
The Same Numbers, Both Ways
A test score of 78 against a mean of 70 and a standard deviation of 8, in a cohort of 500.
- Score
- 78
- Mean
- 70
- Standard deviation
- 8
- Cohort size
- 500
| Line | Percentile rank | Z-score |
|---|---|---|
| Result for a score of 78 | 84.13% | 1.00 |
| Test takers below | 421 | Not stated by the score |
| At the 90th percentile | 80.3 marks | z = 1.2816 |
| Distance from the mean | 8 marks | 1 standard deviation |
| Needs the normal curve | Only for this figure | Yes, to become a percentile |
| If the distribution is skewed | Still correct | Misleading |
A score of 78 sits 8 marks above the mean, which is a z-score of 1.00 and — assuming a normal curve — the 84.13th percentile, with about 421 of the 500 test takers below it. The z-score carries the distance; the percentile carries the crowd. Reaching the 90th percentile would need 80.3 marks.
When To Use Each
Percentile rank
- You are reporting a ranking to people who think in shares of a group rather than in standard deviations.
- The distribution is skewed or has no known shape, and any normal-based figure would be fiction.
- You only have ranks, not the underlying values.
Z-score
- You need to compare values measured on different scales, such as two exams with different totals.
- You are feeding a confidence interval, a hypothesis test or a normal-table lookup.
- The distribution is plausibly normal and the mean and spread are both known.
Common Questions
Is a z-score of 1 the same as the 84th percentile?
Only if the distribution is normal. On a normal curve z = 1 corresponds to 84.13%, and that is where the familiar numbers come from. In a skewed distribution a z-score of 1 can sit at the 70th percentile or the 90th.
Which should I report?
Report the percentile if the audience is deciding who clears a cut-off, and the z-score if you are combining measurements or testing a hypothesis. Many score reports give both, precisely because they answer different questions.
Can a percentile rank be 0 or 100?
Not on the usual definition, which counts half the ties: the lowest score in a group of 500 lands near 0.1%, not 0. Some textbooks use a definition that does produce 0 and 100, which is why two percentile figures for the same score can differ by a point.
Why is my z-score negative?
Because the value is below the mean. A negative z-score is not bad, it is below average: −1 means one standard deviation below, which on a normal curve is the 15.87th percentile.
What if the standard deviation is zero?
Then every value is identical and the z-score has no value, because the calculation divides by zero. Check before computing — a constant column in a dataset is more common than it sounds.
More Comparisons
- Average vs Median
The average uses every value; the median uses the middle one. On five incomes where one is an outlier, the average moves 8.65 and the median moves 1.5 — here is why that matters.
- Geometric Mean vs Weighted Average
Returns of +50%, −50% and +20% average +6.67% arithmetically but compound to −3.45% a year. The geometric mean follows the money; a weighted average follows the weights you chose. On the same three years they disagree by ten points.
- Exponential vs Logistic Growth
Starting at 100 with a 20% growth rate, exponential growth reaches 619 by period ten and 146,977 by period forty. Logistic growth towards a capacity of 1,000 reaches 451 and then 997 — same start, different world.
Sources
- NIST/SEMATECH e-Handbook of Statistical Methods — National Institute of Standards and Technology
- Standard score — definition and properties — Wikipedia
- Percentile — definitions and the tie-handling problem — Wikipedia
Both methods sit in Math.