Materials Science · Measuring and checking material claims

Comparable results and independent reproduction

Investigate a disagreement between laboratories by matching the quantity and procedure, checking raw records, comparing the difference with its uncertainty, and choosing what an independent experiment should repeat.

Useful background: Units, loads, stress and strain; Processing, structure and useful properties; Energy, stability and environmental context. Each worked case explains the quantities it uses.

Back to the chapter

After reading, you can

Explain why a common unit, common material label and common method name do not fully define a comparison.
Trace an apparent strength difference to an area-entry error and distinguish recalculation from a new physical test.
Use supplied standard uncertainties to evaluate a declared compatibility criterion for two matched mean results.
Choose between independent analysis, fresh specimen testing and independent preparation to test a specific remaining explanation.

Two numbers can carry different procedures

Comparable measurements address compatible quantities with suitable definitions, units and conditions. Different laboratories may implement a method differently. Agreement requires examining those details as well as each result’s uncertainty.

Start with the measurand: the quantity intended to be measured. A report of “strength, 300 MPa” leaves several questions open. Is it a proof stress defined by a specified permanent-strain offset, the maximum engineering tensile stress, or a stress calculated at fracture? All can use MPa while describing different endpoints. “Stiffness and strength answer different questions” distinguishes the properties behind those endpoints.

For the investigation below, tensile strength S means maximum engineering tensile stress: S = F_max/A₀, where F_max is the largest tensile force in the recorded test and A₀ is the measured original cross-sectional area. With force in N and area in mm², the quotient is in N/mm² = MPa. A₀ is not the smaller area of a neck after deformation. “Applied force and stress over an area” develops why force alone cannot make this comparison.

Next specify the material state and procedure: grade, batch and processing history; specimen direction and preparation; dimensions and how they were measured; temperature, loading rate and endpoint extraction. A longitudinal coupon and a transverse coupon need not have the same response. Different preparation or loading conditions can change the physical response, while different area entries can change the reported value even when the stored force is unchanged. A procedure needs enough detail to distinguish these possibilities.

The International Vocabulary of Metrology (VIM) uses “metrological comparability” in a narrower technical sense: same-kind quantities with results traceable to the same reference. Traceability uses a documented calibration chain, with uncertainty contributions. That connection does not require close numerical values or rule out mistakes. In this lesson, making a comparison meaningful also means matching the quantity and conditions to the question; metrological compatibility will then ask whether a specified difference fits the uncertainty.

An apparent material difference can begin in the calculation

Consider original invented teaching data from two laboratories testing six distinct tensile coupons. For this case, the records establish matched stock, treatment, direction, preparation, temperature, loading rate and maximum-engineering-stress endpoint. Each coupon has measured original width 2.00 mm and thickness 2.00 mm, giving A₀ = 4.00 mm². The displayed dimensions and forces are chosen central values; measurement uncertainty is considered separately below. The case is not a named alloy, a real interlaboratory dataset or a complete tensile-testing procedure.

Laboratory A uses the measured areas. Laboratory B enters the drawing’s nominal thickness of 1.80 mm, giving 2.00 × 1.80 = 3.60 mm², although its retained pre-test dimensional record says 2.00 mm. The nominal drawing value is not the required measured input for this stipulated calculation. B2 therefore reports 1180/3.60 = 327.78 MPa instead of 1180/4.00 = 295 MPa. The smaller denominator raises the reported stress by the factor 4.00/3.60 = 1.111… .

The three A strengths average 300 MPa. The original B values average 327.78 MPa, making B appear 27.78 MPa higher. Recalculation with the retained measured areas gives a B mean of 295 MPa, now 5 MPa lower than A. Keep the original report, the dimensional evidence and the corrected calculation together so the revision can be followed.

This correction has used existing physical observations. It has not created another specimen test or independently reproduced the experimental finding. Nor does correcting one error establish that every other input is sound. Check the force record, the area-measurement method, the endpoint and the remaining uncertainty budget rather than simply declaring the laboratories equal.

Original illustrative coupon records. Every coupon is distinct; all measured original areas are 4.00 mm². Stresses are rounded for display, and means below use unrounded calculations.
Coupon	F_max (N)	Area entered (mm²)	Initially reported S (MPa)	S with measured area (MPa)
A1	1180	4.00	295	295
A2	1200	4.00	300	300
A3	1220	4.00	305	305
B1	1160	3.60	322.22	290
B2	1180	3.60	327.78	295
B3	1200	3.60	333.33	300

Original illustrative tensile-record correction. The same saved B2 maximum force of 1180 newtons is divided first by incorrectly entered nominal area 3.60 square millimetres, giving 327.78 megapascals, and then by measured original area 4.00 square millimetres, giving 295 megapascals. Below, a linear horizontal axis from 280 to 340 megapascals deliberately excludes zero. A filled blue circle marks the A mean of 300 megapascals with a one-standard-uncertainty bar from 296 to 304; a violet square marks the corrected B mean of 295 with a bar from 292 to 298. The erroneous B mean is shown as a hollow orange circle at 327.78 without a valid uncertainty bar. Labels and different shapes supplement the colours. Each mean represents three distinct coupons, not repeated rupture of one coupon. The complete corrected-mean uncertainty budgets are stipulated and uncorrelated; these bars are not confidence intervals. The original mean difference is plus 27.78 megapascals and the corrected B minus A difference is minus 5 megapascals. Correcting the saved calculation is not a new physical test.

Ask whether the corrected difference fits its uncertainty

Different destructive coupons have different individual strengths. For this comparison, stipulate that the two laboratories estimate the same specified mean response of matched coupons from the stated stock, direction, preparation and test conditions. Their corrected estimates are m_A = 300 MPa and m_B = 295 MPa. This shared target is an assumption to justify from the experimental design, not something established by the unit MPa.

Suppose documented complete budgets give standard uncertainties of the mean results u_A = 4 MPa and u_B = 3 MPa. A standard uncertainty expresses uncertainty on the scale of an estimated standard deviation. These supplied budgets already include appropriate repeat and between-coupon variation, dimensional and calibration contributions, and other relevant model contributions for the specified mean. The three strengths alone do not establish those budgets. Do not add a further scatter-of-the-mean term on top of an already included contribution.

Define the signed difference d = m_B − m_A = −5 MPa. If the uncertainties of the two estimates are uncorrelated, its standard uncertainty is u_d = √(u_A² + u_B²) = √(4² + 3²) = 5 MPa. Squaring retains units of MPa², and taking the square root returns MPa. The two uncertainty contributions add to the variance of the difference even though the central values are subtracted.

Choose and state a comparison rule before using it: here call the results compatible when |d| < 2u_d. The chosen multiple 2 gives a limit of 10 MPa; |−5| = 5 MPa is smaller. This is metrological compatibility under the stated assumptions and criterion. It is not exact equality, a probability that either report is correct, or a guarantee that a component meets a strength requirement. The multiplier alone does not establish 95% coverage.

Shared uncertainty contributions can correlate the estimates. For a difference, a common additive offset cancels; other shared effects require their actual model. Uncorrelated combination must therefore be justified, not inferred from different laboratory names. The worked calculation deliberately stipulates uncorrelated complete budgets. “Calibration, uncertainty and repeatability” explains correction, calibration and uncertainty before this two-result comparison.

If a difference fails the criterion, investigate a changed quantity, a calculation or procedure problem, and an incomplete uncertainty assessment. Do not enlarge an uncertainty just to make reports agree. Conversely, passing the criterion can coexist with a shared mistake. Compatibility of the difference and correctness of the absolute values answer different questions.

Independent repetition tests transfer

Reproducing a finding with separately prepared specimens or different laboratories can reveal hidden dependencies. A failure to reproduce invites investigation of material state, protocol, analysis and sampling. It does not by itself identify which result is wrong or establish misconduct.

State what was actually repeated. An independent analyst can receive the same saved force and dimension records and implement the calculation separately. Agreement checks arithmetic, units and interpretation of those records. Both analysts still inherit any wrong physical reading or unrecorded preparation change. Re-running the same software on the same file checks an even narrower part of the process.

Fresh physical tests use new specimens and observations. For a tensile test taken to rupture, the same coupon cannot supply another intact-specimen test. A common preparer can make matched coupon sets and distribute distinct coupons to the two laboratories. This checks transfer of measurement and procedure more broadly than reanalysis while preparation remains shared. An appropriate stable reference can check aspects of an instrument, but it does not replace testing the actual specimen workflow.

If each laboratory prepares its own new matched coupons from documented stock, preparation becomes another part of the workflow being tested. Machining, surface condition, specimen orientation or treatment can then expose dependencies that common preparation concealed. “Independent” still needs a description: the laboratories may share a drawing, calibration reference or analysis recipe, including a common error.

VIM repeatability conditions hold the procedure, operator, measuring system, location and operating conditions the same over a short period, using the same or similar objects. Reproducibility conditions involve changes such as location, operator or measuring system; specify what changes and what stays fixed. Usage of “reproducibility” and “replicability” differs across fields, so describing the physical and analytical work is more informative than relying on either label alone.

A tight cluster of same-session results does not test all day-to-day, handling or laboratory effects. Changing several things at once may reveal that a result does not transfer, but usually leaves the cause ambiguous. Use a subsequent comparison that changes a suspected factor while preserving the conditions needed for the intended comparison.

Three different investigations. These are options for specific questions, not mandatory stages or a complete experimental design.
Work repeated	What it can test	Important shared dependency
New analyst, same saved records	Calculation and interpretation transfer	Original physical observations and preparation
New matched coupons, common preparation, two test laboratories	Measurement and procedure transfer	Common preparation and any common recipe
Each laboratory prepares and tests new matched coupons	Transfer including preparation workflow	Common stock definition, instructions or references may remain

Three separate original investigation routes use arrows branching from a common input to two independent operations. There is no quantitative scale and the rows are options, not sequential stages. In route A the same saved records branch to analyst A and analyst B using separate calculations: the original physical observations and preparation remain shared. In route B one common preparer makes new distinct matched coupon sets; set A goes to test laboratory A and set B to test laboratory B. Physical observations and measurement implementation can now differ, while preparation stays shared. In route C matched documented stock branches to laboratory A preparing and testing new set C, and laboratory B preparing and testing new set D. Preparation transfer is additionally tested, while common instructions, stock definitions and references can remain shared. The branches depict common versus separately repeated work, not a complete experimental design or proof of statistical independence. A ruptured tensile coupon cannot be retested intact, and agreement does not rule out a shared error.

Use the pattern of agreement to choose the next check

Suppose independent analysts agree when using common raw records. New coupon sets made by one preparer also give compatible mean results in both test laboratories. When each laboratory prepares its own coupons, however, the results separate. Preparation is now a useful lead: the first comparisons did not expose a large analysis-only or test-laboratory-only difference under those particular conditions. They did not prove those influences are absent under every condition.

Inspect the preparation records for a difference that predicts an observable effect: direction, actual dimensions, surface finish, thermal history or damage. Then arrange a crossed comparison in which each preparer supplies distinct matched coupons to each test laboratory, with identities and conditions retained. If the shift follows preparation at both test laboratories, a preparation-linked explanation is strengthened. If it follows test laboratory regardless of preparer, measurement implementation remains a lead. More complicated patterns may indicate interactions or unresolved specimen variation; one comparison does not identify a microscopic mechanism.

This is a way to separate competing explanations, not permission to change material histories silently or average unlike results. A chosen next check should predict different observations under the alternatives. Record the intended quantity, the changed factor and the decision criterion before seeing the new results. “Specimens, dimensions and test conditions” supplies preparation and boundary-condition questions; “From a property claim to a measurement” connects a claim to an observation that can challenge it.

A common method name can leave important implementation choices. In a historical compression interlaboratory investigation, blanks from one plate were distributed to laboratories that machined their own specimens and returned electronic test traces with setup information. Implementations differed, including fixtures, strain measurement and end lubrication. An observed association between lubrication and reported yield strength did not by itself establish the cause: a limited finite-element friction analysis did not explain that discrepancy. This historical compression example is separate from the invented tensile table. It shows why records and competing explanations matter; it does not supply a current testing prescription.

Original hypothetical crossed follow-up means, in MPa, developed in the worked example. Every cell uses a distinct matched coupon set. Distinguishability of the 20 MPa preparation shift is supplied by an independently justified uncertainty assessment; this small table alone does not establish it.
Preparer	Test laboratory A	Test laboratory B
P	300	302
Q	280	282

Worked examples

Reconcile the reports before assigning a material cause

Two laboratories obtain different strengths for nominally identical stock. What should be compared first? Use the six-coupon records above to decide what the first investigation actually resolves.

Match grade, batch and treatment history.
Compare specimen direction, dimensions, preparation and test definition.
Inspect calibration, raw records and uncertainty.
Here the matched material and test-condition records leave a concrete calculation mismatch: B used nominal area 3.60 mm² rather than the measured original 4.00 mm². Recompute B1, B2 and B3 as 290, 295 and 300 MPa.
The B mean becomes (290 + 295 + 300)/3 = 295 MPa rather than 327.78 MPa. The difference B − A changes from +27.78 MPa to −5 MPa without another physical test.
With supplied uncorrelated complete standard uncertainties 4 and 3 MPa, u_d = 5 MPa. The corrected absolute difference 5 MPa passes the chosen strict criterion |d| < 2u_d = 10 MPa. Report the correction and the limited compatibility conclusion, preserving the original records.

Reconcile the measured systems and procedures before interpreting the difference as a material contradiction. In this case a documented area-entry correction removes most of the apparent discrepancy; the corrected means are compatible under the stipulated criterion. Independent experimental reproduction remains a separate task.

Overlapping individual intervals can mislead

In a new matched comparison, m_A = 300 MPa with u_A = 4 MPa and m_B = 312 MPa with u_B = 3 MPa. The complete budgets are uncorrelated. Does the same |d| < 2u_d criterion pass just because the individual ±2u intervals overlap?

The individual intervals are 300 ± 8 → [292, 308] MPa and 312 ± 6 → [306, 318] MPa. They overlap from 306 to 308 MPa.
Evaluate the specified difference instead: d = 312 − 300 = 12 MPa; u_d = √(16 + 9) = 5 MPa. The declared comparison limit is 2u_d = 10 MPa.
Because 12 is not smaller than 10, the criterion fails. Interval overlap compares the separation to the sum of individual half-widths, here 8 + 6 = 14 MPa, which is a different rule.
Investigate quantities, conditions, records and budgets. The result does not identify a faulty laboratory, prove misconduct or establish which estimate is closer to the intended value. No automatic confidence level is assigned to these intervals.

No. The intervals overlap, but the absolute difference of 12 MPa exceeds this comparison’s 10 MPa limit. Use the uncertainty of the difference and the declared rule.

Follow the shift through a new physical comparison

Independent analysts agree on common records, but laboratories using their own preparation workflows disagree. In the crossed follow-up table, preparation P gives means 300 MPa at test laboratory A and 302 MPa at B; preparation Q gives 280 MPa at A and 282 MPa at B. Each cell comes from a distinct matched coupon set, and an independently justified uncertainty assessment makes the 20 MPa preparation shift distinguishable. All other stated test conditions are held matched. Which explanation does the pattern favour?

Within test laboratory A, preparation Q is 280 − 300 = −20 MPa relative to P. Within test laboratory B, it is 282 − 302 = −20 MPa. The direction and size of the shift follow preparation in both laboratories.
Within either preparation, laboratory B is 2 MPa higher than A: 302 − 300 = 2 and 282 − 280 = 2. This table does not establish whether that smaller contrast is distinguishable; do not make that claim without its uncertainty.
The pattern favours a preparation-linked effect over an explanation depending only on test laboratory. Inspect actual preparation differences and choose a targeted follow-up, such as matching a documented orientation or finish while keeping the remaining protocol fixed.
Do not infer that a named defect caused the shift from mean values alone. A preparation label can bundle dimensions, damage, orientation or history, and uncontrolled specimen variation could also matter.

The distinguishable 20 MPa shift follows preparer in both test laboratories. Preparation is the next discriminating lead, while the underlying material or measurement mechanism needs further evidence.

Try the reasoning

Does using the same unit guarantee comparability? Two reports both say 300 MPa, but one is maximum engineering tensile stress and the other a specified offset proof stress. What must be fixed before claiming reproduction?
Show the answer and reasoning
No. Match the intended endpoint and the material/test conditions before comparing the numbers as estimates of the same quantity.
Both endpoints have dimensions of stress and can coincidentally share a value. Maximum engineering stress uses the largest force divided by original area; proof stress uses a specified permanent-strain convention. Repeating either calculation cannot turn one endpoint into the other.
Why use separately prepared specimens? An analyst independently recalculates a saved tensile record and agrees with the first analyst. Has specimen preparation now been independently tested?
Show the answer and reasoning
Separately prepared specimens examine preparation-dependent effects. Agreement on a saved record has not tested that stage.
The analysts share the original physical test and specimen history. They can reveal an area-entry or unit error, but cannot discover every unrecorded preparation change from identical inputs. Use new documented matched specimens for a physical comparison, and state which preparation steps are repeated independently.
A new coupon’s stored maximum force is 1500 N. Its measured original area is 5.00 mm², but the report used nominal area 4.50 mm². Calculate the corrected strength and the factor by which the original calculation exceeded it. Does this correction supply another experimental test?
Show the answer and reasoning
Corrected S = 300 MPa; the original 333.33 MPa is larger by 5.00/4.50 = 1.111… . No new physical test has occurred.
Use 1500/5.00 = 300 N/mm². The same force divided by 4.50 gives 333.33 MPa. Their ratio is the inverse area ratio. Correct the known input mistake with an audit trail and reassess the budget; do not treat the erroneous value as unexplained random scatter or count the reanalysis as a new coupon.
Two matched estimates of the same specified mean are 250 MPa and 258 MPa, with complete uncorrelated standard uncertainties 3 MPa and 4 MPa. Use |d| < 2u_d. Explain the conclusion and one thing it does not establish.
Show the answer and reasoning
d = 8 MPa, u_d = 5 MPa and 8 < 10, so this declared compatibility criterion passes.
The root-sum-of-squares calculation gives √(3² + 4²) = 5 MPa. Passing concerns the difference under the target and budget assumptions; it does not prove equality, eliminate shared mistakes or certify a component’s strength. These budgets are already for the reported mean results, so do not add a duplicate repeatability term.
Two reported values have the form m_A = a + c and m_B = b + c, where c is exactly the same uncertain additive calibration offset in both. Write their difference. Does agreement of the difference establish that c is correct?
Show the answer and reasoning
m_B − m_A = (b + c) − (a + c) = b − a. The common offset cancels in the difference, but may still bias both absolute values.
The c terms are correlated, so treating their contributions as independent would give the wrong uncertainty of the difference. Cancellation here depends on exactly the same additive influence, not merely using the same brand of instrument. A shared multiplicative factor or different sensitivities require the actual model. Agreement alone does not verify the absolute calibration.
Both laboratories agree when analysing one shared record and when testing new common-preparation coupon sets. They disagree when each prepares its own coupons. Choose a useful next investigation and explain why simply averaging the discrepant means is inadequate.
Show the answer and reasoning
Check documented preparation differences, then cross preparer with test laboratory using distinct matched new coupon sets and retained identities.
A shift that follows preparer across test laboratories supports a preparation-linked explanation; one that follows test laboratory regardless of preparer supports a measurement-implementation lead. Averaging can conceal unlike quantities or histories and tests neither explanation. Preserve uncertainty and specimen-variation limits; the pattern does not by itself name a physical mechanism.

What to carry forward

Match the measurand, material state and implemented procedure before interpreting a numerical difference.
A documented correction of saved records is useful evidence, but is different from a new physical test.
Compatibility uses a declared criterion and uncertainty of the difference, including relevant correlations; it does not prove absolute correctness.
Describe what was independently repeated and what remained shared, then choose a follow-up whose outcomes distinguish the remaining explanations.

References

Teaching resources and research supporting this lesson’s explanations and examples.

International Vocabulary of Metrology (VIM3)Joint Committee for Guides in Metrology (JCGM)International metrology vocabulary, actual definition and notes 2.47International Vocabulary of Metrology · 2.47 Metrological compatibility of measurement results International Vocabulary of Metrology · 2.46 Metrological comparability of measurement results International Vocabulary of Metrology · 2.41 Metrological traceability International Vocabulary of Metrology · 2.20 Repeatability condition of measurement International Vocabulary of Metrology · 2.24 Reproducibility condition of measurementjcgm.bipm.org
National Institute of Standards and Technology (NIST)3 sources
Stress-Strain CurvesDavid Roylance · MITUniversity materials-mechanics module, 2001; selected pages 1–4 · PDFweb.mit.eduRead source