What a Z-Score Is
A z-score (also called a standard score) is a number that expresses where a particular value falls relative to the mean of its distribution, measured in units of the distribution’s standard deviation. A z-score of 1.0 means the value is one standard deviation above the mean. A z-score of −2.0 means the value is two standard deviations below the mean. A z-score of 0 means the value sits exactly at the mean.
The formula is:
z = (x − μ) / σwhere x is the observed value, μ (mu) is the population mean, and σ (sigma) is the population standard deviation.
Z-scores convert raw values from any distribution onto a common scale, making it possible to compare values that come from distributions with different units or different spread. A test score, a weight measurement, a stock return, and a manufacturing tolerance measurement can all be expressed as z-scores and compared directly, even though their raw numbers are on completely different scales.
How Z-Scores Relate to the Normal Distribution
Z-scores are most useful when the underlying data follows a normal distribution — the symmetric, bell-shaped curve that describes many natural and measurement phenomena. Under a normal distribution, the z-score directly corresponds to a percentile: the proportion of the population that falls below the observed value.
The standard normal distribution has a mean of 0 and a standard deviation of 1. When a raw value is converted to a z-score, it is being placed on this standard normal distribution. The corresponding area under the curve to the left of the z-score is the cumulative probability, or p-value — the probability that a randomly selected member of the population would score below this value.
Three rule-of-thumb ranges define how much of a normal distribution falls within a given number of standard deviations of the mean:
- Within ±1 standard deviation (z between −1 and +1): 68.27% of the distribution
- Within ±2 standard deviations (z between −2 and +2): 95.45% of the distribution
- Within ±3 standard deviations (z between −3 and +3): 99.73% of the distribution
These are the empirical rule percentages (often rounded to 68/95/99.7 in textbooks). A z-score of +1 means the value is at approximately the 84th percentile — meaning 84.13% of the population scores below this value. A z-score of −1 corresponds to approximately the 16th percentile (15.87%).
Worked Example: Scoring Above the Mean on a Standardized Test
A standardized exam has a mean score of 1,000 and a standard deviation of 150. A student scores 1,200. Where does this score fall?
Step 1: Apply the z-score formula.
z = (1200 − 1000) / 150 = 200 / 150 = 1.3333Step 2: Find the corresponding percentile.
A z-score of 1.3333 corresponds to a cumulative probability of 0.9088, meaning the student scored above approximately 90.88% of test-takers.
The calculator confirms: z = 1.3333, p-value = 0.908789, percentile = 90.8789%.
This is a clean example of what z-scores accomplish: the raw score of 1,200 is meaningful only in context (is that out of 1,200 possible points? Out of 1,600?), but the z-score of 1.33 is immediately interpretable — it places the student in roughly the 91st percentile regardless of the scale.
The Inverse Problem: Finding a Score from a Percentile
The inverse z-score problem runs in the other direction: given a target percentile, find the z-score (and, from it, the raw value) corresponding to that position.
Common z-score benchmarks for percentiles:
| Percentile | Z-Score |
|---|---|
| 50th | 0.0000 |
| 84th | 0.9945 |
| 90th | 1.2816 |
| 95th | 1.6449 |
| 97.7th | 1.9954 |
| 99th | 2.3263 |
For the exam example above: if a student wants to score at the 95th percentile, the target z-score is 1.6449. Converting back to the raw score: x = μ + z × σ = 1000 + 1.6449 × 150 = 1000 + 246.7 ≈ 1,247.
Notice that the 97.7th percentile corresponds to z ≈ 2.00. This aligns with the 95.45% within ±2 rule: the upper tail beyond +2 accounts for about 2.275% of the distribution, so z = +2.00 sits at roughly the 97.7th percentile.
Common Applications of Z-Scores
Comparing Scores Across Different Tests
Suppose two students each take different versions of a college admissions exam, with different means and standard deviations. Student A scores 620 on a scale with mean 500 and standard deviation 100; student B scores 27 on a scale with mean 21 and standard deviation 5.
Student A: z = (620 − 500) / 100 = 1.2000, percentile ≈ 88.5%
Student B: z = (27 − 21) / 5 = 1.2000, percentile ≈ 88.5%
The raw scores are on completely different scales, but the z-scores reveal equivalent relative performance: both students scored at the same percentile rank. This cross-test comparison would be impossible with raw scores alone.
Quality Control in Manufacturing
A factory produces bolts with a target diameter of 10 mm and a process standard deviation of 0.05 mm. A bolt measures 10.12 mm. The z-score is (10.12 − 10) / 0.05 = 2.4. The cumulative probability at z = 2.4 is approximately the 99.2nd percentile — this bolt is larger than about 99.2% of expected output, flagging it as a potential outlier that may fall outside specification limits.
Medical Reference Ranges
Height, weight, blood pressure, lab values, and many other clinical measurements are reported with z-scores (sometimes called standard deviation scores or SDS) in pediatric and research medicine. A child whose height falls at z = −2.0 is at the 2.3rd percentile for height-for-age — two standard deviations below the mean — which many growth charts flag as the threshold for clinical evaluation. Z-scores allow comparison across age groups where the mean and standard deviation differ substantially.
Finance and Risk Management
In quantitative finance, z-scores appear in risk management as a measure of how far an observed return is from the historical mean in units of standard deviation. The Altman Z-score is a specific multi-variable model for predicting corporate bankruptcy. More broadly, the concept of “how many standard deviations away from normal” underlies Value at Risk (VaR) calculations and hypothesis testing in econometrics.
How to Read a Z-Score Table (and When to Use the Calculator Instead)
Traditional statistics courses teach z-score lookups using a standard normal distribution table — a printed matrix of cumulative probabilities for z-values from roughly −3.4 to +3.4, in steps of 0.01. The table gives the probability P(Z ≤ z) — the area under the standard normal curve to the left of the given z-value.
To use a z-table:
- Calculate the z-score from the formula.
- Find the row corresponding to the first decimal place of z (e.g., for z = 1.33, find the row 1.3).
- Find the column corresponding to the second decimal place (e.g., column 0.03).
- The cell value is the cumulative probability — the percentile expressed as a decimal.
For z = 1.33, a standard table gives approximately 0.9082. The calculator computes 0.9088 because it uses a high-precision numerical approximation (accurate to within 5 × 10⁻⁷), while most printed tables round to 4 decimal places and lose some precision at intermediate values.
For everyday use, a calculator is more accurate and eliminates table-reading errors. The table is primarily useful in exam settings where electronic tools are not permitted.
Limits of Z-Score Analysis
Z-scores assume normality. The connection between z-scores and percentiles holds precisely only for normally distributed data. If the underlying distribution is skewed, bimodal, or heavy-tailed, the percentile calculation from a z-score will be incorrect — sometimes dramatically so. Before applying z-score analysis, consider whether the data is approximately normal (using a histogram, Q-Q plot, or a formal normality test).
Z-scores require knowing μ and σ. The formula uses the population mean and population standard deviation, which are often unknown in practice. Using a sample mean (x̄) and sample standard deviation (s) in place of μ and σ is technically a t-score, not a z-score. For large samples (n ≥ 30), the distinction is small; for small samples, use the t-distribution instead.
Outliers affect the sample standard deviation. If the dataset contains extreme values, the sample standard deviation will be inflated, compressing z-scores for all other values toward zero and understating how unusual those values actually are.
Z-scores are relative, not absolute. A z-score tells you where a value sits within its own distribution but says nothing about whether that distribution is appropriate, whether the data was collected correctly, or whether the comparison is meaningful.
Frequently Asked Questions
What is a “good” z-score? There is no universally good or bad z-score — it depends entirely on what is being measured and which direction is favorable. A z-score of +2 on a test (top 2.3% of scores) is typically excellent. A z-score of +2 on a biomarker that is harmful at high levels (like certain inflammatory markers) may indicate elevated risk. Context determines interpretation.
How is a z-score different from a percentile? A percentile is a direct statement of rank: “at the 80th percentile” means scoring above 80% of the population. A z-score is a statement of relative distance from the mean in standard deviation units. The two are related: for normally distributed data, each z-score corresponds to a unique percentile, and the calculator performs this conversion automatically. For non-normal data, the relationship between z-scores and percentiles depends on the actual distribution.
What is a z-score used for in hypothesis testing? In hypothesis testing, a z-statistic (computed from sample data) is compared to a critical z-value to decide whether to reject the null hypothesis. For a two-sided test at the 5% significance level, the critical z-value is approximately ±1.96 (the 2.5th and 97.5th percentiles). If the test’s z-statistic exceeds 1.96 in absolute value, the result is statistically significant at the 5% level. This is one of the most common uses of z-scores in applied statistics.
What is the difference between a z-score and a T-score (in psychology)? In psychological testing, T-scores (capital T) are a standardized scale with a mean of 50 and a standard deviation of 10, used to report many personality and cognitive test results. They are derived from z-scores by the transformation T = 50 + 10z. A T-score of 60 corresponds to a z-score of +1.0 (about the 84th percentile). This avoids negative numbers and decimal points in clinical reporting. The “T-score” in this context is different from the t-statistic in inferential statistics.
Can z-scores be negative? Yes. A negative z-score simply means the value is below the mean. A z-score of −1.5 means the value is 1.5 standard deviations below the mean, at approximately the 6.7th percentile. There is no mathematical restriction on the sign or magnitude of a z-score.