✦ScienexySCIENCE KNOWLEDGE LIBRARY
topic · SCI-CHE-005-131

Outlier Detection and Rejection

Outlier detection identifies observations that differ unusually from the pattern of a dataset. Statistical procedures can help assess unusual analytical results, but rejection requires a justified rule and investigation of possible experimental causes.

Explanation

An outlier is an observation that is unusually distant from other observations in a dataset or inconsistent with an assumed statistical model. In analytical chemistry, an unusual result may arise from a transcription mistake, contamination, instrument malfunction, sample handling problem, or genuine natural variation. Outlier detection methods include graphical inspection, standardized residuals, the Grubbs test for certain approximately normal datasets, Dixon's Q-test for suitable small samples, and other methods designed for particular distributions and experimental designs. Detection does not automatically justify rejection. The analyst should investigate the measurement process, document the reason for any exclusion, and apply a predefined or scientifically justified procedure. Repeatedly testing observations until an inconvenient result disappears can seriously distort statistical conclusions.

Variables

SymbolNameDescription

xᵢ

Individual observation

An analytical measurement being evaluated.

x̄

Sample mean

The average of the observations.

s

Sample standard deviation

The estimated spread of the observations.

zᵢ

Standardized deviation

The deviation of an observation from the mean expressed in standard deviation units.

Q₁

First quartile

The lower quartile of the ordered dataset.

Q₃

Third quartile

The upper quartile of the ordered dataset.

IQR

Interquartile range

The difference between the third and first quartiles.

G

Grubbs statistic

A statistic used to assess a single suspected outlier under specified assumptions.

Q

Dixon statistic

A statistic comparing a suspected extreme gap with the total range.

Example or application

A worked example has not yet been added.

Definition

An outlier is an observation that deviates markedly from the general pattern of a dataset or from a specified statistical model. Outlier detection is the process of identifying and evaluating such observations.

Purpose

To identify potentially unusual analytical measurements, investigate their causes, and determine whether they should remain in the dataset under an appropriate statistical and quality-control procedure.

Importance

A single unusual observation can strongly influence the mean, standard deviation, regression model, and statistical conclusions, especially in small datasets. Proper evaluation protects analytical conclusions from both genuine errors and unjustified data removal.

Key Concepts

  • Unusual observation

  • Statistical outlier

  • Measurement error

  • Influential observation

  • Grubbs test

  • Dixon Q-test

  • Standardized residual

  • Interquartile range

  • Robust statistics

  • Predefined rejection criteria

  • Data integrity

  • Investigation and documentation

Principle

An observation is evaluated against an explicitly defined statistical model or decision rule. Its extremeness is assessed relative to the expected variability and sample size, while the experimental record is examined for a plausible assignable cause.

Working Process

  • Review the raw measurements and laboratory records for transcription or calculation errors.

  • Plot the observations using a suitable graph such as a dot plot or box plot.

  • Check for possible instrument, sample preparation, contamination, or procedural problems.

  • Select an outlier test appropriate for the sample size, distribution, and experimental design.

  • Apply the test using its stated significance level and assumptions.

  • Distinguish a statistically flagged observation from a measurement with a documented assignable cause.

  • Retain or reject the observation according to a justified, documented procedure.

  • Report the test used, decision criteria, and any excluded data.

Equations

  • Standardized deviation: zᵢ = (xᵢ − x̄) / s

  • Interquartile range: IQR = Q₃ − Q₁

  • Common box-plot lower fence: Q₁ − 1.5 × IQR

  • Common box-plot upper fence: Q₃ + 1.5 × IQR

  • Grubbs statistic for a single suspected outlier: G = max|xᵢ − x̄| / s

  • Dixon Q statistic for an appropriate ordered small dataset: Q = gap / range

Units

  • Original analytical measurement unit for observations and quartiles

  • Standardized deviation: dimensionless

  • IQR: same unit as the original measurement

  • Grubbs statistic: dimensionless

  • Dixon Q statistic: dimensionless

Examples

  • A series of titration results contains one value substantially higher than the others. The analyst checks the burette record and calculations before applying an appropriate statistical test.

  • A box plot identifies a measurement beyond the common 1.5 IQR fence, flagging it for investigation rather than automatically proving that it is invalid.

  • A Grubbs test may be used to evaluate one suspected extreme observation in a dataset that meets the test's assumptions.

Applications

  • Evaluating replicate analytical measurements

  • Reviewing calibration datasets

  • Investigating quality-control failures

  • Checking experimental research data

  • Assessing regression residuals

  • Supporting laboratory data review and validation

Advantages

  • Helps identify unusual measurements that may affect analytical conclusions.

  • Encourages investigation of laboratory and instrument problems.

  • Provides objective methods for evaluating extreme observations.

  • Improves transparency when data review decisions are documented.

Limitations

  • Different outlier tests have different assumptions and sample-size requirements.

  • A genuine extreme observation may be incorrectly flagged.

  • Some tests are designed for only one suspected outlier.

  • Multiple outliers can distort tests that assume a single unusual observation.

  • Statistical detection alone cannot establish the physical cause of an unusual result.

Common Mistakes

  • Rejecting a measurement simply because it differs from the mean.

  • Using an outlier test without checking its distributional assumptions.

  • Applying a single-outlier test when several unusual observations may be present.

  • Removing data without recording the reason and procedure.

  • Repeating tests until an observation can be rejected.

  • Confusing an influential observation with a proven measurement error.

Prerequisites

  • Mean and standard deviation

  • Normal distribution

  • Variance

  • Statistical hypothesis testing

  • Basic laboratory quality control

Learning Objectives

  • Define an outlier in analytical data.

  • Explain possible experimental causes of unusual observations.

  • Describe graphical and statistical outlier detection methods.

  • Explain the basic purpose of Grubbs and Dixon tests.

  • Distinguish identifying an outlier from justifying its rejection.

  • Describe appropriate documentation of data exclusion.

Practice Questions

  • What is an outlier in analytical chemistry?

  • List possible experimental causes of an unusual measurement.

  • How does a box plot help identify potential outliers?

  • What is the purpose of the Grubbs test?

  • Why should an observation not be rejected solely because it differs from the mean?

  • Why is documentation important when excluding an analytical result?

Key Takeaway

Outlier detection is an investigation, not an automatic deletion process. An unusual measurement should be assessed using appropriate statistical methods and laboratory evidence before any rejection decision.

References

TitleTypeUrl

Miller and Miller, Statistics and Chemometrics for Analytical Chemistry

textbook

Harris, Quantitative Chemical Analysis

textbook

NIST/SEMATECH e-Handbook of Statistical Methods

technical reference

https://www.itl.nist.gov/div898/handbook/

Sources and verification

Sources are shown as provided in the project data. Not every source has yet been independently verified. Please report errors or missing citations.

  • Source verification pending; check an authoritative textbook, university, standards body, or peer-reviewed source.

Suggest a correction or source · More in Analytical Chemistry