Overview
SIST ISO 16269-4:2014 - "Statistical interpretation of data - Part 4: Detection and treatment of outliers" is an international standard that defines sound statistical procedures and graphical methods for identifying and accommodating outliers in measurement data. Focused primarily on univariate data, it also offers guidance for multivariate and regression contexts. The standard supports robust practice in measurement processes, quality control and data analysis by recommending validated tests, robust estimators and graphical diagnostics.
Key topics and technical requirements
- Definitions and terminology: formalizes terms such as outlier, masking, resistant and robust estimation, order statistics, quartiles and box plot.
- Data screening and graphical methods: recommended procedures for initial screening, including modified box plots, interquartile range (IQR) techniques and five-number summaries.
- Statistical tests for outliers: procedures for samples from normal, exponential and other known (and unknown) distributions; includes guidance on Cochran’s test for outlying variances and procedures for detecting single or multiple outliers.
- Robust estimation and accommodation: recommended robust estimators for location and scale (e.g., trimmed means, biweight estimators) and correction factors for scale estimation when outliers may be present.
- Multivariate and regression data: guidance on detecting outlying Y and X observations, influential points and suggestions for robust regression techniques.
- Annexes and implementation aids: Annex A presents an algorithm for the GESD outlier detection procedure; Annexes B–E provide critical-value and factor tables (exponential samples, modified box plot factors, robust estimator correction factors, Cochran’s test); Annex F provides a structured flow chart for univariate outlier detection.
Practical applications
ISO 16269-4 is applicable wherever measurement data integrity is critical:
- Manufacturing process control and product quality assurance
- Metrology and laboratory measurement analysis
- Calibration, testing and inspection
- Data cleaning in research, product development and regulatory reporting
- Statistical analysis pipelines where robust inference is required
Using standardized outlier detection improves comparability of analyses across laboratories, suppliers and regulatory bodies.
Who should use this standard
- Statisticians and data scientists implementing robust detection routines
- Quality engineers and process control specialists
- Metrologists and laboratory analysts
- Auditors and compliance professionals assessing measurement data validity
Related standards
- Other parts of ISO 16269 (Part 6: tolerance intervals; Part 7: median estimation; Part 8: prediction intervals) provide complementary statistical methods for measurement interpretation.
Keywords: SIST ISO 16269-4:2014, outliers detection, outlier treatment, robust estimation, modified box plot, Cochran test, GESD, univariate data, multivariate outliers, regression diagnostics.
