Overview
ASTM E178-21: Standard Practice for Dealing With Outlying Observations provides guidance for identifying and handling outlier data points in samples. Published by ASTM International, this practice addresses the detection of outliers in experimental datasets-primarily those consisting of replicate measurements or observations assumed to be from random samples. The standard outlines statistical methods recognized internationally for determining whether a data point is significantly different from the remainder of a dataset. The importance of correctly managing outliers is emphasized, as they can result from either valid but extreme natural variability or experimental error.
Key Topics
- Definitions and Terminology: Clarifies key terms such as "outlier," "order statistic," and "null hypothesis" in the context of quality and statistics, referencing ASTM E456 for consistency.
- Statistical Significance Testing: Details established statistical tests, including methods using sample mean and standard deviation, to determine if extreme values should be considered outlying observations.
- Recommended Procedures: Provides step-by-step guidance on treating outliers, including physical investigation and multiple statistical tests (e.g., Grubbs' test, Dixon's criteria, Tietjen-Moore tests).
- Handling Outliers in Practice:
- When a known physical cause is identified, the observation may be discarded.
- If no cause is known, formal statistical testing is advised before excluding any data.
- When outliers are retained, documentation should clearly note how they were addressed.
- Significance Levels and Critical Values: Introduces the concept of significance levels (e.g., 1%, 5%, 10%) and the use of critical values, acknowledging that most outlier criteria assume an underlying normal distribution.
- Recursive and Multiple Outlier Testing: Covers recursive testing methods for identifying multiple outliers on one or both ends of a sample.
- Alternative Criteria: Discusses the use of skewness and kurtosis as additional statistical tools for assessing sample normality and outlier detection.
Applications
The practical application of ASTM E178-21 spans various industries and fields where data integrity and reliability are critical:
- Laboratory Testing: Ensures accurate measurement reporting by identifying and scrutinizing suspicious results in chemical, materials, or environmental labs.
- Quality Control in Manufacturing: Facilitates root cause analysis and process improvement by distinguishing between random variation and genuine process anomalies.
- Scientific Research: Guides researchers in deciding whether to retain, adjust, or exclude anomalous data points, maintaining scientific rigor in fields such as biology, physics, or engineering.
- Data Analysis and Statistical Auditing: Provides a framework for analysts to report and handle data outliers transparently in compliance, regulatory reporting, and peer-reviewed studies.
- Process Optimization: Assists organizations in segregating suspect test units for further investigation, which may inform corrective actions or further research.
Related Standards
- ASTM E456 – Terminology Relating to Quality and Statistics: Offers standardized definitions relevant to statistical practices and quality analysis, referenced throughout E178-21.
- ASTM E2586 – Practice for Calculating and Using Basic Statistics: Provides foundational procedures for statistical calculations, which underpin the statistical significance tests used in ASTM E178-21.
- ISO/IEC Guide 98-3 – Uncertainty of Measurement: Though not directly referenced, this international guideline is relevant for laboratories concerned with uncertainty and data quality.
By implementing ASTM E178-21, organizations can improve the validity of their data analyses, enhance consistent decision-making, and align with recognized best practices for outlier detection and management. This standard is essential for industries reliant on statistically sound methodologies to ensure product quality, safety, and compliance.