Glossary term
Z-Score
A standardized value that expresses how many standard deviations an observation lies from a mean.
Definition
metricA z-score is a dimensionless standardized value equal to the distance between an observation and a reference mean divided by the reference standard deviation.
Z-scores put values from a distribution onto a common scale, making outliers, process variation, normalized residuals, measurement deviations, and probabilistic thresholds easier to compare. Their interpretation depends on the distribution, reference population, standard deviation estimate, and whether the data are independent and stationary.
A z-score standardizes a value relative to a mean and standard deviation:
where x is the observation, mu is the reference mean, and sigma is the reference standard deviation. A positive z-score means the value is above the mean; a negative z-score means it is below the mean.
Engineering use
Z-scores are used in quality control, tolerance analysis, process monitoring, measurement validation, anomaly detection, reliability studies, machine-learning preprocessing, and comparison of variables with different units. They allow engineers to ask whether a deviation is small relative to expected variation or large enough to deserve investigation.
The method is most direct when the reference distribution is stable and approximately normal. In skewed, multimodal, censored, drifting, or heavy-tailed data, a z-score can still be useful but may not correspond to familiar normal-probability thresholds. The chosen mean and standard deviation may come from design assumptions, historical production data, a calibration population, or a current sample, and those choices change interpretation.
Common mistakes
A common mistake is treating any large z-score as a confirmed fault. It may indicate a real anomaly, a changed operating regime, a bad reference population, sensor drift, non-normal data, or a poor uncertainty model. Another mistake is mixing population and sample standard deviations without stating the basis. A strong statistical review states the reference dataset, distribution assumption, estimation method, sample size, outlier policy, units before standardization, and decision threshold.