Systematic and Random Errors

Introduction

When analytical results disagree with expectation, the pattern of disagreement provides clues. Some errors shift almost every result in one direction. Others make results scatter around an average. A third category consists of blunders such as using the wrong sample label. Recognising the pattern helps choose a useful check rather than simply repeating the same flawed procedure.

Core explanation

A systematic error, or bias, affects results in a recurring way. A balance that reads 0.020 g too high, a pipette whose actual delivery differs from its assumed volume, or incomplete extraction of the analyte can shift an entire set. Some biases are approximately constant in absolute size; others scale with analyte amount. A blank signal may add a constant apparent concentration, while a 90% recovery creates a proportional low bias. Repetition under identical conditions often preserves these errors.

Random error creates unpredictable variation in direction and size among replicate results. Slight variation in endpoint judgment, electrical noise and small differences in positioning a flask meniscus can contribute. Averaging independent repetitions can reduce the uncertainty of the mean from random scatter, roughly as 1/√n under suitable assumptions. It does not remove a persistent bias. Nor does every observed difference count as random: a slow drift with time may indicate temperature change or detector instability and should be investigated as a patterned effect.

Gross errors are mistakes outside the intended measurement process: transposing digits, using the wrong calibration file, spilling part of a sample or confusing bottles. They should be investigated through records and checks rather than hidden by averaging. One surprising result is not automatically a blunder. It may reflect a genuinely heterogeneous sample or unrecognised chemistry, so exclusion needs a documented reason or justified statistical test.

Different controls reveal different issues. Reagent and procedural blanks detect contamination that adds signal. Certified reference materials test recovery and calibration against an accepted value in a similar matrix. A matrix spike can expose suppression or incomplete extraction. Independent preparations test preparation precision; repeated instrument readings mainly test instrument repeatability. Calibration checks across a run reveal drift. Comparing methods based on different principles can reveal a bias common to one method but not another.

Errors may combine. Suppose a method loses 5% of analyte during extraction and also has a variable endpoint reading. The mean can be biased low and the repeats scattered. A complete uncertainty evaluation considers both contributions and the sampling stage.

Step-by-step reasoning

1. Look for a consistent shift from a suitable reference, a broad spread, a time trend or an isolated anomaly. 2. Match the pattern to possible preparation, calibration, instrument or sampling causes. 3. Run an appropriate blank, reference material, spike or independent replicate. 4. Correct a known bias only with a validated correction and its uncertainty. 5. Document gross mistakes and rerun affected samples rather than silently deleting values.

Visual explanation

Plot measured concentration against measurement number. A horizontal cluster consistently below the reference line suggests bias. A broad cloud around the line suggests random scatter. A rising sequence suggests drift. One isolated distant point suggests a possible gross mistake or unusual sample. These are diagnostic patterns, not automatic statistical verdicts.

Real-world analogy

A clock that is always six minutes slow has systematic error. A clock whose second hand wobbles unpredictably has random variability. Writing down the wrong date is a gross mistake. Averaging many readings from the slow clock gives a precise but still late time, just as repeated biased assays remain biased.

Real-world example

A laboratory's calcium results for a reference material are consistently 3% low. Repeating the instrument reading of the same vial gives excellent agreement. The analyst checks preparation recovery and discovers a precipitate left in the digestion vessel. The problem is incomplete transfer, so adding more instrument replicates could never correct the low result.

Why?

Why does a procedural blank help identify one kind of systematic error? It experiences reagents and operations without intentional analyte. Any measured analyte-like response reveals contamination or background that can add to samples. A blank cannot by itself reveal analyte lost during extraction, which requires a different check such as recovery or reference-material analysis.

Common misconception

“Every deviation from the mean is experimental error” treats natural sample heterogeneity as if it were instrument noise. Another mistake is assuming that multiplying the number of replicates fixes calibration bias. Replicates improve knowledge of scatter but do not make the central value true.

Worked example

Four measurements of a 10.00 mg L⁻¹ reference yield 9.41, 9.39, 9.40 and 9.42 mg L⁻¹. Their mean is 9.405 mg L⁻¹, about 5.95% low relative to the reference. Their range is only 0.03 mg L⁻¹, so the readings are tightly grouped. This is evidence for a shared low bias rather than random variation large enough to explain the difference. The analyst should inspect calibration and recovery before releasing related sample results.

Quick check

1. A detector response increases steadily during a day. Should this simply be described as random error? Answer: No. A directional time trend suggests drift, a patterned source of bias for samples measured at different times. Periodic calibration checks can reveal and quantify it.

Exam focus

State an error source and the direction or pattern it would cause. Match each check to the problem it can detect: blank for contamination, reference material for trueness, replicates for spread, and chronological controls for drift. Distinguish a justified correction from deleting an inconvenient result.

Advanced insight

An error can be systematic for one study yet vary between studies. If an analyst always uses one miscalibrated pipette, its effect is systematic within that data set. Across laboratories using different pipettes, the bias may appear as between-laboratory variability. Error classification therefore depends partly on the scope and conditions of repeated measurements.

Summary

Systematic effects shift results consistently, random effects create scatter, and gross mistakes arise from discrete procedural failures. Their patterns suggest different investigations. Repetition addresses random uncertainty, while blanks, references, spikes and independent checks are needed to detect bias. A defensible result documents both diagnosis and any correction.

Practice questions

1. What pattern would a constant reagent contamination create at low analyte concentrations? Answer: It would add an approximately constant positive signal, producing a particularly large relative high bias for low-concentration samples. A procedural blank can reveal it.

2. Why can a matrix spike help when a clean calibration standard does not? Answer: The spike experiences the sample matrix and preparation. Poor recovery can reveal matrix suppression or analyte loss that clean standards do not show.

3. A single result is far from its replicates. What should be done before rejecting it? Answer: Review raw records, preparation steps, labels and instrument behaviour, and consider genuine sample variation. Reject only with documented cause or a justified statistical rule, not merely because it is inconvenient.