Z Score Calculator

What a z score measures

A raw score on its own tells you little. An 85 on a biology test sounds strong until you learn the class average was 88, and it sounds weak until you learn the average was 70. What a z score adds is position: it converts any value into a distance from the mean, measured in standard deviations instead of points, percent or milliliters. A z of 1.63 means the value sits 1.63 standard deviations above the mean, whatever the original units were.

Because the unit cancels out, z scores put different scales on one ruler. Test points, bottle volumes, machine part widths and annual rainfall all become the same currency once each is standardized. That is why teachers curve grades with them and why factories set tolerance limits with them. Penn State's intro statistics course calls this a standardized score for that reason: subtract the mean, divide by the standard deviation, compare.

Three facts hold for every z score. A value equal to the mean has z = 0. Values above the mean get positive z scores, values below it get negative ones. And a z of 2 is not twice as good as a z of 1 in any probability sense: the curve's tail thins out faster than that, which is the whole point of the percentile column below.

The formula and what each part means

The formula is z = (x - mu) / sigma. Read it in two steps. The top, x - mu, is the distance from the value to the mean in raw units. The bottom, sigma, is the standard deviation of the population in the same units. Divide one by the other and the units cancel, leaving a count of standard deviations.

With Greek symbols unavailable, the calculator above uses plain words: score, mean, standard deviation. If you hold sample data instead of population values, the same shape applies with the sample mean x-bar and the sample standard deviation s: z = (x - x-bar) / s. The reading does not change. If you need the mean or the standard deviation first, the average calculator and the standard deviation calculator work them out.

Run the exam case through it. The top is 85 - 72 = 13 points. The bottom is 8 points. So 13 / 8 = 1.625, which rounds to 1.63. Two students on two different tests can both be 13 points above the mean, yet sit at z = 1.63 on one exam and z = 0.52 on another, because the spread differs. The spread is the denominator, and it decides everything.

Worked example: one exam score

Take a midterm with a class mean of 72 and a standard deviation of 8. Three students ask where they stand.

The first scored 85. The top is 85 - 72 = 13, so 13 / 8 = 1.625, a z of 1.63. On a normal curve that is the 94.8th percentile: about 94.8 percent of the class scores at or below 85 and 5.2 percent score above. The second scored 68, giving (68 - 72) / 8 = -0.50, the 30.9th percentile. The third scored exactly 72, the mean, so z = 0 and the percentile is 50.

Notice student B. A 68 out of 100 reads as a D-plus at a glance, but in this class it beats 30.9 percent of the room, and letter-grade cutoffs set by the curve would put it near a C. The z score and the percentage correct answer different questions, and only the first one accounts for how hard the test was. For grading practice built on this idea, see what grading on a curve means.

Negative z scores: a bottling line example

A negative z score is not an error and not a bad number. It means one thing: the value sits below the mean. The sign tells direction, the size tells distance.

A juice plant fills bottles to a target of 500 ml, and the filling machine has a standard deviation of 5 ml. A quality check pulls one bottle holding 493 ml. The top is 493 - 500 = -7 ml, and -7 / 5 = -1.40, so that bottle sits 1.40 standard deviations under target, in the bottom 8.1 percent of fills. About 91.9 percent of bottles receive more product.

Negative z scores carry real weight on spec limits. If the label declares 490 ml as the fill floor, bottles at or below it run (490 - 500) / 5 = -2.00, and 2.3 percent of production fails the label check. That single number, 2.3 percent, is what the plant manager sees, because a z of -2 sits at the edge where small shifts turn into visible complaints. Run your own fill target with the calculator above and watch how a 1 ml change in the mean moves the failure share.

From z score to percentile

The percentile comes from the area under the standard normal curve to the left of z. NIST's Engineering Statistics Handbook gives the standard normal distribution in full; the practical version is that the cumulative area is looked up, not derived, and every statistics textbook prints the table.

Printed tables round z to two decimals, so the exam z of 1.625 is read at 1.63, and the table entry reads 0.9484. Multiply by 100 and 94.84 percent of values fall at or below, which rounds to the 94.8th percentile. The calculator above skips the rounding step and evaluates the area at the exact z, which is why its percentile can differ from a paper table by a tenth of a point.

Two anchors are worth memorizing. The 95th percentile sits at z = 1.645, the cutoff behind one-sided 95 percent checks. And z = 1.96, just under 2, is the two-sided 95 percent marker used for margins of error in polls. The 1.63 from the exam example lands between the 90th and 95th percentiles, close to the 95th but short of it.

The empirical rule: 68, 95, 99.7

For a normal distribution you do not need a table for rough work. The empirical rule says about 68 percent of values fall within 1 standard deviation of the mean, 95 percent within 2, and 99.7 percent within 3. NIST lists the precise figures as 68.27, 95.45 and 99.73 percent for the normal model.

On the exam scale, the mean plus and minus 1 SD spans 64 to 80, since 72 + 8 = 80 and 72 - 8 = 64. So roughly 68 percent of the class lands between 64 and 80, and student A's 85 already sits outside the fat middle. The mean plus and minus 2 SDs spans 56 to 88, holding about 95 percent of scores, and student A's 85 sits inside that band but near its upper rail.

The rule also gives a fast sanity check on any z the calculator returns. A z between -1 and 1 is ordinary, a z beyond 2 is the top or bottom 2.5 percent, and a z beyond 3 happens about 3 times in 1000 under a normal model. If a data set hands you z scores past 3 on a regular basis, the data is heavier in the tails than a normal curve, and the percentile column stops being trustworthy.

When a z score misleads

The z score has two parts, and they fail separately. The distance part, how many SDs from the mean, holds for any data set with a defined spread. The percentile part adds an assumption: that the data follows a normal curve. For skewed data like household incomes, house prices or machine downtime, the z can be honest while the percentile is badly off, and ranks or percentiles counted straight from sorted data serve better.

Small samples trip the other end. With five values, the standard deviation itself wobbles, so a z built on it inherits the wobble; treat the percentile as a rough guide until a few dozen observations back it. A standard deviation of zero breaks the formula outright: every value equals the mean, division by zero has no meaning, and the calculator above says so instead of returning a number.

Comparing z scores across groups also has a catch: they standardize within one distribution. A z of 1.63 in a class of 25 is not the same achievement as a z of 1.63 on a state exam of 40,000 takers. Different test, different mean, different SD: on a placement test with mean 1050 and SD 120, a score of 1180 gives (1180 - 1050) / 120 = 1.08, the 86.1th percentile, a full rank lower than the exam student with the same raw distance. Scaled scores exist to do this translation on purpose. IQ scales sit at mean 100 with SD 15, so 100 + 15 x 1.63 = 124.45, an IQ of about 124; T scores sit at mean 50 with SD 10, so 50 + 10 x 1.63 = 66.3. For measuring spread before standardizing, the percent error calculator covers the close-to-expected question from the other side.