Z-Score & Standard Normal Calculator

Z-Score Guide

What Is a Z-Score?

A z-score (standard score) measures how many standard deviations a value is from the mean. Formula: z = (x − μ) ÷ σ. Where x = individual value, μ = population mean, σ = population standard deviation. A z-score of 0 means exactly average. Positive z: above average. Negative z: below average. Example: exam mean = 65, SD = 12. A score of 77 → z = (77−65)/12 = 1.0. A score of 41 → z = (41−65)/12 = −2.0 (two standard deviations below average).

What's the key thing to understand about Z-Scores and Probability?

Z-scores link to probabilities via the standard normal distribution (mean = 0, SD = 1). A z-score of 1.0 corresponds to the 84th percentile — 84% of values fall below this point. Z = 2.0 → 97.7th percentile. Z = −1.0 → 15.9th percentile. Z = 1.96 → 97.5th percentile (the 5% significance level boundary in two-tailed tests). These probabilities come from tables (z-tables) or can be calculated from the cumulative normal distribution function.

What should I know about Applications of Z-Scores?

Standardised testing: converting raw exam scores to z-scores allows direct comparison across tests with different means and SDs. IQ: mean 100, SD 15 → IQ 130 has z = (130−100)/15 = 2.0, placing it at the 97.7th percentile. Medical: a patient's blood test result is compared to the reference range (typically mean ± 2 SD) using z-scores. Finance: z-scores compare asset returns to their historical average, adjusted for volatility. Quality control: z-scores determine whether a production batch is within acceptable tolerance limits, flagging unusual measurements that fall too many standard deviations from the expected average.

What's the difference between Z-Score and T-Score?

Z-scores apply when population mean and SD are known (or sample size is large, n > 30). T-scores apply when the population SD is unknown and estimated from a sample (small n). For large samples, z and t are nearly identical. The t-distribution has heavier tails than the normal distribution — accounting for additional uncertainty from estimating SD from a small sample. At n = 30, t-critical values are only slightly larger than z-critical values. At n = 10, the difference is significant. This is why small clinical trials and other small-sample studies routinely use the t-distribution rather than the standard normal distribution for their statistical tests.

What do I need to know about Converting Between Raw Score and Z-Score?

The z-score formula z = (x−μ)/σ converts any raw value into standard deviations from the mean, and it rearranges just as easily in the other direction: x = μ + zσ. This reverse calculation answers 'what raw score corresponds to the 90th percentile?' — look up the z-score for the 90th percentile (z≈1.28), then compute x = μ + 1.28σ. For a dataset with mean 100 and standard deviation 15 (a typical IQ-test scale), a z-score of 1.28 corresponds to x = 100+1.28×15 = 119.2. This two-way conversion is exactly how percentile charts and standardised test score tables are built, and it's the reason z-scores are described as the universal translator between any two differently-scaled normal distributions — converting both datasets to z-scores lets you compare a score of 85/100 on one test with 650/800 on another, directly.

What should I know about the Empirical Rule (68-95-99.7)?

For any approximately normal distribution, about 68% of values fall within 1 standard deviation of the mean (z between −1 and +1), about 95% within 2 standard dedeviations, and about 99.7% within 3. This gives a fast sanity check without consulting a full z-table: a z-score of exactly 2 corresponds to roughly the 97.5th percentile (since 95% is split evenly, leaving 2.5% in each tail). A z-score beyond ±3 is rare enough (0.3% combined) that in quality control it's often flagged as an outlier worth investigating. The empirical rule is why standard deviation is such a useful single number to summarise spread — once you know the mean and standard deviation of a normal dataset, you effectively know the whole shape of the distribution without needing every data point.

What do I need to know about Comparing Scores from Different Distributions Using Z-Scores?

One of the most practical uses of z-scores is comparing values that come from different distributions with different means and spreads — something that's otherwise meaningless to compare directly. If a student scores 85 on a test with mean 75 and standard deviation 5, their z-score is (85-75)/5 = 2. If the same student scores 70 on a different test with mean 60 and standard deviation 20, their z-score is (70-60)/20 = 0.5. Even though 85 is a higher raw number than 70, the z-scores reveal the second performance was actually weaker relative to its own test's typical spread, since a z-score of 2 is a much rarer, stronger result than a z-score of 0.5. This "standardizing" is exactly why standardized tests, IQ scores, and growth charts convert raw measurements into z-scores or percentile ranks before comparing individuals against a reference population.

Worked Example: Calculating and Interpreting a Z-Score

A class of students sits a test with a mean score of 70 and a standard deviation of 8. One student scores 82. Their z-score is (82 − 70) ÷ 8 = 12 ÷ 8 = 1.5. A z-score of 1.5 means the student scored 1.5 standard deviations above the mean. Looking this up against the standard normal distribution, a z-score of 1.5 corresponds to a cumulative probability of about 0.9332 — so the student scored higher than roughly 93.3% of the class, placing them in the top 7% or so.

This table is just the cumulative standard normal distribution read off at common z-values — it's the same lookup a z-table or statistical function performs, and it's what lets you convert a raw z-score into a percentile ranking without needing calculus.

Worked Example: Comparing Scores Across Two Different Tests

Z-scores are especially useful for comparing performance across two tests that use different scales. Suppose a student scores 78 on a maths test with mean 65 and standard deviation 10, and 68 on an English test with mean 60 and standard deviation 6. On the raw scores alone, the maths result looks higher — but converting both to z-scores tells a different story. Maths z-score: (78 − 65) ÷ 10 = 1.3. English z-score: (68 − 60) ÷ 6 ≈ 1.33. The two performances are almost identical relative to their own class, even though the raw scores (78 vs 68) look quite different — the English test simply has a tighter spread of results, so the same relative standing produces a lower raw score.

Worked Example: Z-Score From Summary Statistics on a Different Scale

A company's customer satisfaction survey has a mean score of 72 and a standard deviation of 9 (scale 0-100). A particular branch scores 88. Its z-score is (88 - 72) / 9 = 16 / 9 ≈ 1.78. Using the standard normal distribution, a z-score of 1.78 corresponds to roughly the 96.2nd percentile, meaning this branch outperforms about 96% of branches assuming scores are approximately normally distributed. If a second branch scores 54, its z-score is (54 - 72) / 9 = -18/9 = -2.00, which sits at about the 2.3rd percentile -- a result worth flagging for a performance review, since only roughly 2 in 100 branches would be expected to score this low by chance alone. Converting raw scores to z-scores this way lets managers compare performance across metrics with very different natural scales (survey points, dollars, response times) on one common, interpretable axis.

Modified Z-Scores for Outlier Detection

The standard z-score uses the mean and standard deviation, both of which are themselves sensitive to extreme values -- a single huge outlier can inflate the standard deviation enough to mask its own deviation. For robust outlier detection, statisticians often use the modified z-score, which replaces the mean with the median and the standard deviation with the median absolute deviation (MAD): M = 0.6745 × (x - median) / MAD. The constant 0.6745 rescales MAD so that it behaves like a standard deviation for normally distributed data. A common screening rule flags any data point with |M| > 3.5 as a likely outlier. This approach is preferred over the ordinary z-score in quality control, fraud detection, and sensor-data cleaning, precisely because medians and MAD do not shift much when one or two extreme points are present -- the method stays reliable in the presence of the very anomalies it is meant to detect.