Z-Score vs Percentile — Differences and Conversions

How Z-Scores and Percentiles Are Calculated

A z-score standardises a value by subtracting the mean and dividing by the standard deviation: z = (x − μ) / σ. The result tells you how far the value is from the mean in SD units. z = 0 means exactly average; z = 1 means one SD above the mean; z = −2 means two SDs below. The z-score uses the full distribution, not just ranks.

A percentile rank is calculated differently: sort all values, then count what fraction lie at or below your value of interest. If 720 out of 1,000 students scored ≤ 85 on an exam, a score of 85 is at the 72nd percentile. Percentiles describe position in terms of the empirical sample, not in terms of any theoretical distribution.

These two measures are equivalent for normally distributed data but can diverge significantly for skewed or non-normal distributions. For a right-skewed income distribution, the value at z = 2 (two SDs above the mean) could correspond to roughly the 95th percentile or to the 98th percentile depending on the skewness — the z-score formula doesn't tell you this without knowing the distribution shape.

Converting Between Z-Scores and Percentiles

For normally distributed data, z-scores and percentiles map directly to each other via the standard normal CDF (Φ): Percentile = Φ(z) × 100. For example: z = 0 → 50th percentile; z = 1 → 84.1th percentile; z = 1.645 → 95th percentile; z = 1.96 → 97.5th percentile; z = 2 → 97.7th percentile; z = 2.576 → 99.5th percentile; z = 3 → 99.87th percentile.

Going the other way: to find the z-score corresponding to a given percentile P, compute z = Φ⁻¹(P/100), where Φ⁻¹ is the inverse normal CDF (also called the probit function). A z-score calculator or statistical table gives these values instantly.

Critical limitation: this conversion assumes a normal distribution. For income data (right-skewed), the 95th percentile does not correspond to z = 1.645. Using the z-score to percentile conversion on non-normal data gives incorrect results. Always verify your distribution shape before converting.

Worked Example: Standardised Test Scores

Scenario: SAT Math scores have μ = 528, σ = 116. A student scores 760. What is their z-score and percentile?

Z-score: z = (760 − 528) / 116 = 232 / 116 = 2.0. This tells the student they scored 2 standard deviations above the mean — a score that sounds high in technical terms.

Percentile (assuming normal distribution): Φ(2.0) = 97.7th percentile — higher than 97.7% of test-takers. This is immediately understandable to a non-statistician parent or admissions officer.

Both convey the same information. In practice, the College Board reports both: a scaled score (760) and a percentile (97). The z-score is mainly used by statisticians for further analysis (comparing across different years when σ and μ change). The percentile is used in reports and college admissions materials because it requires no statistical background to interpret.

Medical Growth Charts — Why Percentiles Win

Paediatric growth charts (height and weight for age) universally use percentiles, not z-scores. A child at the 25th percentile for height is shorter than 75% of children their age — this is immediately actionable for parents and clinicians without any statistics training.

Growth charts do also include z-score lines (labelled as −2SD, −1SD, 0, +1SD, +2SD), but these are secondary. The WHO and CDC growth standards define "normal" as roughly the 3rd to 97th percentile range — equivalent to z-scores of approximately −1.88 to +1.88.

However, for classifying severe malnutrition, z-scores are preferred in research and international health settings because they can distinguish between z = −3 and z = −4 (both below the 1st percentile on standard charts). When the tails matter — extreme cases, risk stratification — z-scores provide more precision than percentiles.

Finance: Value at Risk Uses Z-Scores

In financial risk management, Value at Risk (VaR) is often expressed using z-scores. "95% daily VaR = −1.645σ" means that on 95% of days, the portfolio will not lose more than 1.645 standard deviations of its daily return distribution.

The z-score formulation is preferred here because it scales naturally with standard deviation — if σ changes (more volatile market), the VaR in dollar terms changes proportionally. Expressing VaR as a percentile (e.g., "the 5th percentile daily loss") requires re-calculating the dollar amount every time volatility changes.

This illustrates the general principle: z-scores are preferred when you need to perform further calculations (scaling by σ, combining across assets). Percentiles are preferred when you need to communicate results to a non-technical audience or when the distribution is non-normal.