03 · What You Need to Know
How to Plan Duplicate Measurements and Interpret Their Results
What Is a Duplicate Measurement?
A duplicate measurement involves obtaining two measurements of the same intended quantity, object, sample, or phenomenon under specified conditions.
Depending on the research context, duplicates may involve repeated readings from an instrument, separate measurements of the same specimen, independent scoring of the same performance, or repeated recording of information.
The term should be defined carefully because different forms of duplication answer different methodological questions.
For example, measuring the same specimen twice using one instrument may help evaluate repeatability. Having two observers independently rate the same performance may help evaluate agreement between raters.
Entering the same source value twice may identify transcription discrepancies, but it does not provide two independent measurements of the underlying phenomenon.
What Are the Different Types of Duplicate Measurements?
| Type |
Description |
Principal purpose |
| Technical replicate |
Repeated measurement of the same sample or material under defined conditions. |
Assess technical variation. |
| Repeat measurement |
Repeated observation of the same quantity or characteristic. |
Evaluate repeatability or short-term variation. |
| Independent duplicate rating |
Two raters assess the same material independently. |
Evaluate rater-related differences. |
| Duplicate data entry |
The same source information is entered independently more than once. |
Detect transcription errors. |
| Repeated participant assessment |
A participant completes the same or comparable measurement at another occasion. |
Evaluate stability or change, depending on the design. |
These procedures should not be treated as interchangeable. Repeating an instrument reading does not assess the same sources of variation as repeating an assessment several weeks later.
Why Might Duplicate Measurements Improve Research Data Quality?
Repeated measurements can help estimate the variability associated with a measurement process.
They may reveal inconsistent instrument readings, transcription mistakes, rater disagreements, or unusual variation that warrants investigation.
For example, if two measurements of a stable physical quantity differ substantially under nominally identical conditions, the discrepancy may indicate instrument instability, procedural inconsistency, or an unexpected change in the measured object.
Duplicate measurements can also contribute to estimating measurement precision and assessing whether a procedure is suitable for the study's intended distinctions.
However, the benefit depends on the sources of error represented by the repeated observations.
What Is the Difference Between Repeatability and Reproducibility?
In measurement science, repeatability generally concerns precision under a defined set of repeatability conditions, such as the same measurement procedure, operator, equipment, location, and a short time interval.
Reproducibility concerns precision under specified reproducibility conditions, which may involve differences in locations, operators, or measurement systems.
These terms should be used with reference to the conditions being compared.
For example, taking two readings using the same sensor within one minute may provide information about repeatability. Comparing measurements obtained by different laboratories may address a broader reproducibility question.
Neither property alone establishes that the measurements are free from systematic bias.
Which Variables Are Most Suitable for Duplicate Measurement?
Duplication is especially worth considering for variables that are critical to the research question and vulnerable to meaningful measurement error.
Examples may include primary outcome measurements, manually scored assessments, sensitive instrument readings, and variables requiring subjective classification.
However, a variable's importance is not sufficient justification for duplication if repeating it provides little additional information.
For example, independently verifying a manually entered achievement score may be useful, whereas repeatedly extracting the same automatically stored score may offer little benefit if the principal risk is an incorrect scoring algorithm.
The decision should follow the broader principles for identifying variables requiring strong quality controls.
Should Every Measurement Be Duplicated?
No. Universal duplication may increase cost, participant burden, equipment use, and collection time without proportionate improvement in data quality.
Some measurements are already generated through highly controlled procedures, while others are difficult or inappropriate to repeat.
For example, asking participants to complete the same psychological questionnaire twice within minutes may introduce recall, fatigue, or response effects rather than provide an independent measurement.
Similarly, repeated observation of an evolving social interaction may capture genuine changes rather than measurement error.
A risk-based approach selects the measurements for which duplication can answer a meaningful quality question.
How Should Duplicate Measurements Be Selected?
The protocol should specify whether all eligible observations or a defined subset will receive duplicate measurement.
Where only a subset is duplicated, selection may be random, systematic, risk-based, or otherwise justified by the research objective.
Random selection can help assess measurement performance across the observed population, provided the sampling procedure is appropriate.
Risk-based selection may concentrate effort on measurements with greater consequences or known vulnerabilities.
However, selecting duplicates only after identifying unusual results may provide a biased picture of overall measurement precision.
The selection method and its limitations should be documented before the principal collection begins where feasible.
Should Duplicate Measurements Be Independent?
Independence may be important when the objective is to evaluate variation between observers or separate measurement attempts.
For example, two raters should generally score the same performance independently before discussing their judgments if the purpose is to estimate inter-rater agreement.
Likewise, duplicate data entry is more informative when the second entry is not simply copied from the first.
However, technical replicates are often correlated because they share the same specimen, equipment, or preparation process.
Researchers should specify the degree of independence intended and avoid assuming that repeated observations represent independent experimental units.
How Can You Evaluate Differences Between Duplicate Measurements?
For numerical measurements, a simple starting point is the difference between the two readings.
Across many duplicate pairs, researchers may examine the mean difference, standard deviation of differences, patterns across the measurement range, and appropriate limits of agreement.
For categorical measurements, exact agreement or suitable agreement coefficients may be more relevant.
The appropriate analysis depends on the purpose and measurement scale.
Can Duplicate Measurements Be Averaged?
Sometimes. Averaging repeated numerical measurements can reduce random error when the measurements represent the same underlying quantity and relevant assumptions are satisfied.
However, averaging is not always appropriate.
If the measured quantity changes between observations, the average may not represent the intended measurement occasion.
If one reading contains a documented technical failure, blindly averaging it with a valid reading may introduce error.
Similarly, averaging ordinal categories may be difficult to justify without an appropriate measurement model.
The protocol should specify whether repeated readings will be averaged, retained separately, reconciled, or analyzed using a model that accounts for repeated observations.
Researchers should not select whichever reading produces the more favorable substantive result.
Do Duplicate Measurements Increase Sample Size?
They increase the number of recorded observations, but not necessarily the number of independent experimental or sampling units.
For example, measuring the same specimen twice does not create two independent specimens.
Likewise, scoring the same student's presentation twice does not create an additional student.
Treating technical replicates as independent observations can underestimate uncertainty and produce misleading statistical conclusions. This problem is often described as pseudoreplication.
Researchers should identify the unit of analysis and account for dependence among repeated observations.
The statistical treatment should be consistent with the study design and intended estimand.
What If Duplicate Measurements Disagree?
Disagreement should trigger an appropriate investigation rather than automatic deletion of one value.
The research team may examine instrument status, recording procedures, environmental conditions, rater decisions, or possible changes in the measured object.
Where the cause is identifiable, an authorized correction or repeat measurement may be justified.
Where the cause remains uncertain, the discrepancy may need to be retained and reflected in the analysis or reported as a measurement limitation.
Acceptance thresholds should be justified before results are inspected when feasible.
The procedure should distinguish a genuine discrepancy from a technical error and preserve an appropriate audit trail.
How Should Discrepancy Thresholds Be Established?
A threshold defines the degree of disagreement that requires review or corrective action.
Its value should reflect measurement precision, technical specifications, substantive requirements, and the consequences of error.
For example, a difference of one unit may be negligible for one instrument but unacceptable for another.
Researchers should avoid selecting an arbitrary percentage threshold without considering the measurement scale and intended use.
Where differences are expected to vary with measurement magnitude, absolute thresholds may be insufficient and proportional or model-based approaches may be more appropriate.
The rationale for the threshold should be documented in the protocol or measurement SOP.
Can Duplicate Measurements Detect Systematic Bias?
Not necessarily. If both measurements share the same systematic error, they may agree closely while remaining inaccurate.
For example, a miscalibrated sensor may produce nearly identical repeated readings that are consistently offset from the true quantity.
Similarly, two raters trained to apply an invalid scoring rule may agree while producing misleading scores.
Duplicate measurement should therefore be complemented by relevant instrument validation, calibration, or external reference comparisons when necessary.
The procedures for research equipment calibration address a different aspect of measurement quality.
How Do Duplicate Ratings Relate to Inter-Rater Agreement?
When two independent raters assess the same performance, their ratings can be used to evaluate inter-rater agreement.
For example, two assessors may independently score a student's written explanation using a common rubric.
Agreement statistics can help determine whether the scoring procedure produces sufficiently comparable results.
However, duplicate rating is not automatically equivalent to technical replication. Human judgment may involve different sources of variation and requires appropriate measurement models.
Researchers should distinguish duplicate ratings intended for quality assurance from multiple ratings deliberately combined to produce a final score.
The considerations for testing inter-rater agreement are particularly relevant when several observers contribute to the same outcome.
Are Duplicate Measurements Appropriate for Qualitative Research?
They may be, but their purpose should be methodologically coherent.
For example, independently checking a transcript against an audio recording can identify transcription errors. Repeated observation may also provide additional contextual evidence about a phenomenon.
However, asking two researchers to produce identical interpretive themes is not necessarily appropriate for reflexive qualitative analysis.
Similarly, repeating an interview may generate a different account because participants' understanding or circumstances have changed, not because the first interview was inaccurate.
Qualitative quality assurance should reflect the methodology's assumptions about interpretation, context, and evidence rather than mechanically import measurement procedures from quantitative research.
What Should the Research Protocol Specify?
A useful protocol should explain the purpose of duplicate measurement and how the procedure will be implemented.
| Protocol element |
What to document |
| Purpose |
Whether duplication assesses precision, agreement, transcription accuracy, or another quality property. |
| Selection |
Which variables and observations will be duplicated. |
| Frequency |
How often duplicate measurements will be obtained. |
| Procedure |
Equipment, personnel, timing, and measurement conditions. |
| Independence |
Whether measurements or ratings must be obtained independently. |
| Acceptance criteria |
What discrepancies require investigation. |
| Corrective action |
How unacceptable differences will be evaluated and resolved. |
| Analysis |
How duplicates will be combined, modeled, or retained. |
| Documentation |
How source values, discrepancies, and corrections will be preserved. |
The degree of detail should reflect the consequences of measurement error. A short protocol statement may be sufficient for simple verification, while complex measurement systems may require a separate SOP.
When Should Duplicate Measurements Be Collected?
Timing depends on the property being assessed.
Measurements taken immediately after one another may be useful for evaluating short-term repeatability, provided the underlying quantity remains stable.
Measurements collected across different sessions may reveal longer-term variability but may also include genuine changes in the measured phenomenon.
For example, repeating a laboratory measurement of a stable reference material differs from reassessing a student's knowledge after several weeks of instruction.
Researchers should distinguish measurement error from substantive change and specify the relevant time interval accordingly.
How Should Duplicate Measurements Be Monitored During Collection?
The team may periodically review discrepancy rates, mean differences, variability, and unusual patterns across collectors, instruments, or sites.
Monitoring can help identify changes in measurement performance before they affect a substantial portion of the dataset.
For example, increasing disagreement between duplicate instrument readings may indicate equipment deterioration or inconsistent procedures.
However, an observed change in discrepancy patterns may also reflect changes in the types of cases being measured.
Investigations should therefore consider the measurement context rather than automatically attribute every change to equipment or personnel failure.
How Much Duplication Is Necessary?
The appropriate amount depends on the importance of the measurement, expected variability, desired precision, and cost of repetition.
Duplicating every observation may be justified for some small, high-consequence measurement tasks. In larger studies, a planned subset may provide sufficient information about measurement performance.
Researchers should also consider whether alternative controls, such as automated validation, calibration, or independent source verification, would address the principal risks more effectively.
The objective is proportionate quality assurance, not duplication for its own sake.
The broader issue of balancing quality control with research feasibility should inform the decision.
Watch Out
Do not treat technical replicates as independent participants or specimens. Repeated measurements of the same unit generally share sources of variation and must be handled according to the study's sampling and measurement structure. Counting duplicates as independent observations can produce artificially narrow uncertainty estimates.
