Percentiles and Z-Scores: Where a Number Sits in the Data

An Average Tells You the Middle, Not Where You Stand

Knowing that the class average was 70 says nothing about whether your 78 was unremarkable or the best mark in the room. That depends on how spread out the other marks were, and spread is exactly what an average throws away.

Two tools answer the question properly. A percentile says what share of the data you are above. A z-score says how many standard deviations you are from the mean. They are different answers to the same question, and each is right in different circumstances.

Percentiles: What Share You Beat

The nth percentile is the value below which n% of the data falls. The median is simply the 50th percentile.

Take twelve exam marks: 52, 58, 61, 63, 67, 70, 71, 74, 78, 82, 85, 91.

A mark of 78 has eight of the twelve below it, so it sits at the 66.7th percentile — better than two thirds of the group.

The common misreading is worth stating plainly, because it runs in exactly the wrong direction:

A percentile is a position, not a score. Being in the 90th percentile does not mean 90% correct. You could score 32% on a brutal paper and still be in the 90th percentile if almost everyone else scored lower. Child growth charts, exam results and salary bands all work this way, and all three get misread the same way.

Quartiles and the Interquartile Range

Quartiles are the percentiles that cut the data into four. Sort the values, find the median, then find the median of each half:

Q1 (25th percentile): 62
Q2 (the median): 70.5
Q3 (75th percentile): 80

The interquartile range is Q3 − Q1, here 18. It describes the spread of the middle half of the data and ignores the tails entirely, which makes it far more robust than the full range — one freak value cannot move it.

The IQR also gives the standard test for an outlier. Anything below Q1 − 1.5 × IQR or above Q3 + 1.5 × IQR is conventionally flagged:

Here that is anything below 35 or above 107. Nothing in the set qualifies, so there are no outliers by this rule.

Z-Scores: How Many Standard Deviations Out

z = (value − mean) ÷ standard deviation

A z-score of 0 is exactly average. Positive is above, negative below, and the number is the distance measured in standard deviations rather than in marks or pounds or seconds.

With a mean of 70 and a standard deviation of 10: a score of 85 gives z = (85 − 70) ÷ 10 = +1.5. A score of 50 gives −2.0.

That change of units is the whole point. A raw mark only means something inside its own test; a z-score means the same thing everywhere.

Turning a Z-Score Into a Percentile

If the data follows a normal distribution, each z-score corresponds to a fixed percentile:

z = −2.0 → 2.3rd percentile
z = −1.0 → 15.9th
z = 0 → 50th
z = +1.0 → 84.1st
z = +1.5 → 93.3rd
z = +2.0 → 97.7th
z = +3.0 → 99.9th

This is the empirical rule, usually quoted as 68–95–99.7: about 68.3% of values fall within one standard deviation of the mean, 95.4% within two, and 99.7% within three.

It is worth noticing how fast the tails thin out. Moving from z = 1 to z = 2 takes you from beating 84% of the data to beating 98%; the same one-standard-deviation step is worth four times less at the top than it was in the middle.

The Reason Z-Scores Exist: Comparing Unlike Things

Here is where the raw numbers actively mislead.

You score 78 in maths, where the class mean is 70 with a standard deviation of 10. You score 65 in English, where the mean is 55 with a standard deviation of 5.

The maths mark is higher. The English result is far better:

Maths: z = (78 − 70) ÷ 10 = +0.80, around the 79th percentile
English: z = (65 − 55) ÷ 5 = +2.00, around the 98th percentile

The English paper had a tighter spread, so being ten marks clear of the mean was a much rarer achievement than being eight clear on a paper where marks were scattered. Comparing 78 against 65 gets this exactly backwards.

The same logic applies outside exams: comparing your 5K time to your deadlift, or a shop's sales against a region with different variability, needs a common scale before the comparison means anything.

When the Bell Curve Assumption Breaks

Every z-to-percentile figure above assumes a normal distribution. Plenty of real data is not normal, and income is the standard example: it is right-skewed, with a floor at zero and a long tail of high earners.

On skewed data a z-score of +1 is not the 84th percentile, and the empirical rule does not hold. The mean is dragged by the tail, the standard deviation inflates with it, and both inputs to the z-score are distorted.

For skewed data, use percentiles directly. They make no assumption about shape at all — they only count what is above and below, which is why the Office for National Statistics reports earnings by percentile rather than by standard deviations from the mean.

Sample or Population, Again

The z-score needs a standard deviation, and which one you use depends on what your numbers represent. Divide by n when you have the entire group, and by n − 1 when they are a sample standing in for something wider.

For the twelve marks above, the population standard deviation is 11.13 and the sample figure is 11.62 — about 4% apart on twelve values, and less on larger sets.

If those twelve are the whole class, use 11.13. If they are twelve pupils sampled from a year group, use 11.62.

Conclusion

An average locates the middle; percentiles and z-scores locate you. Use a percentile when the data is skewed or you simply want to know what share you are above, since it assumes nothing about the shape. Use a z-score when you need to compare measurements taken on different scales, which is the one thing a raw mark can never do.

Paste your figures into a Statistics Calculator for the mean, median and quartiles, and a Standard Deviation Calculator for the spread the z-score depends on.

Try our Statistics Calculator

Try our Standard Deviation Calculator