Is a z-score of 1.5 a 1.5 percent result?

No; it is a distance in reference standard deviations.

Is a standard score always centered at zero?

No; some scales use other centers and spreads.

AI Glossary / Statistics and probability / Foundation

Standard scores

Definition

Standard scores express observations on a reference scale with a specified mean and standard deviation. A common form is the z-score, calculated by subtracting a reference mean and dividing by a positive reference standard deviation. Other standard-score scales apply a further linear transformation, such as a scale centered at fifty with standard deviation ten. A standard score describes position relative to its reference distribution; it does not automatically represent a percentile or establish that the underlying distribution is normal.

What it means

Raw values can be difficult to compare when measurements use different units or reference populations. A standard score states how far a value is from its reference center in spread units. Positive z-scores lie above the reference mean, and negative values below it. The meaning depends on the reference group and measurement. A high relative position on one measure is not necessarily the same achievement, difficulty or practical value as a high relative position on another.

How it works

Choose the reference population or dataset and document its mean, standard deviation and calculation convention. Compute z = (x minus reference mean)/reference standard deviation. To map to a target scale with center A and spread B greater than zero, calculate A + Bz. The transformation preserves ordering when the scale factor is positive. Scores calculated using a fixed external reference do not necessarily have mean zero and variance one in a new group. A constant reference variable has zero standard deviation, so the ordinary z-score formula is undefined.

Three practical examples

Example 1: A reporting system expresses a measurement relative to a documented historical reference group. Example 2: An AI evaluation compares how far a system's score lies above a benchmark mean, without calling that distance a probability. Example 3: A learning report uses a standard-score scale to communicate relative standing. Each use requires clear reference information. Changing the reference group can change the standard score even when the original measurement has not changed.

Think of it this way

Think of describing a person's position relative to the center of a crowd, using the crowd's spread as the unit of distance. The same physical position can look very different relative to a tightly packed crowd or a widely spread crowd. Standard scores work similarly: their interpretation comes from both the observed value and the chosen reference distribution.

Worked example

Suppose a reference distribution has mean seventy and standard deviation eight. A raw score of 82 has z = (82 minus 70)/8 = 1.5, meaning it is 1.5 reference standard deviations above the mean. On a reporting scale with center fifty and standard deviation ten, its standard score is 50 + 10 times 1.5 = 65. A raw score of 62 gives z = -1 and a reporting score of forty. Now consider a different measure with reference mean 100 and standard deviation five: a raw value of ninety has z = -2. Although ninety exceeds 82 numerically, it is lower relative to its own reference. This does not establish that the measures assess the same construct or that the first person has greater absolute ability. No percentile follows from these z-scores without information about the reference distribution's shape.

Illustrative case study

In this fictional case study, Rowan Learning revises a dashboard that mixes raw and standard scores. Staff initially interpret a value of 65 as 65 percent correct. The analyst labels the reference scale, shows the corresponding raw score and explains that fifty is the scale center rather than a pass mark. When the reference cohort changes, the organization records the version and assesses comparability. Managers can then distinguish a change in observed performance from a change caused by the reference used to express it.

Common misconceptions and mistakes

Common mistakes include confusing a standard score with a percentage, assuming it is automatically a percentile, and treating equal z-scores on different measures as equivalent substantive performance. Another mistake is assuming a linear transformation makes data normal. It changes center and spread but preserves the distribution's shape up to scaling and reflection where relevant. Outlying observations can affect estimated reference means and standard deviations. Comparisons also become misleading when reference populations differ in ways that matter to the purpose of the assessment.

Business use

Businesses use standard scores to place heterogeneous metrics on documented reference scales and communicate relative position. This can help with reporting, but combining scores into an index requires additional choices about weighting and meaning. Standardization alone does not make measures equally reliable, equally important or interchangeable. Retain the raw units where practical and state the reference period, population and calculation. For AI benchmarks, a high standard score relative to a weak reference group does not establish strong absolute performance or adequacy for deployment.

Check your understanding

What z-score corresponds to 82 in the example?

Answer: 1.5.

What is its score on the fifty-and-ten scale?

Answer: 65.

What does a negative z-score mean?

Answer: The value is below the reference mean.

Why might an unchanged raw score receive a different standard score later?

Answer: The reference mean, standard deviation or scoring convention may have changed.

Learn next

Learn next how linear transformations affect mean, variance and ranking. Compare standard scores and percentile ranks in skewed distributions. Then examine reference-group selection, reliability and comparability before combining several standardized measures or interpreting a change in a reported score as a change in the underlying construct.

Reference group

Sources

Sources: Howell, Statistical Glossary, https://www.uvm.edu/~statdhtx/fundamentals9/Glossary/Glossary.html; NIST, Cumulative Distribution Function of the Standard Normal Distribution, https://www.itl.nist.gov/div898/handbook/eda/section3/eda3671.htm. These primary author and institutional references describe standard-score scales and normal-model transformations. The reference-score arithmetic is original and independently checked; the dashboard case is fictional.

Previous: Standard normal random vector Next: Standard test image