PSLE Science Reality Lab Vol No.146 | “The Data Are Validated” — Does That Mean Every Value Is True and Final?

PSLE-SCI-REALITY-0146

Wait, What? A Dataset Can Be Validated and Still Contain Uncertainty, Warnings and Future Revisions

You open a science website and see a neat badge beside a table of results:

VALIDATED DATA

A classmate says, “Great. Validated means every number is true, exact and final.”

That conclusion is too strong.

In real scientific work, validation usually means that data have been checked against defined quality requirements for a particular project or purpose. The checks can include completeness, calculations, measurement quality, blanks, duplicate results, detection limits, instrument performance, sample records and other criteria. Some values may be accepted without qualification. Some may be accepted but marked with a warning or qualifier. Some may be rejected for a particular use. The exact process depends on the scientific programme.

That is very different from saying that the dataset has become perfect.

The Reality Lab habit is: when a scientific source says “validated”, ask validated against which criteria, for which purpose, with which remaining qualifiers and limits?

Quick Answer

Find out what the validation process actually checked.
Look for quality flags or qualifiers attached to individual values.
Separate data verification—checking whether records and calculations correctly represent the analytical process—from data validation—judging results against agreed data-quality criteria for the project.
Remember that a validated value can still carry measurement uncertainty or a warning.
Ask whether the data are suitable for the scientific question you want to answer. Data can be acceptable for one purpose but weak for another.
Do not treat “validated” as “unchangeable forever”. Scientific datasets can be corrected, reprocessed or reissued when better information becomes available.

The Exact Learner Job This Page Owns

This page owns one real-world evidence-transfer problem: evaluating a scientific dataset, map, monitoring table or report that carries a “validated” label.

It does not replace the existing PSLE Science owners for measurement, uncertainty, method limitations, sampling, quality control or conclusion writing. It applies those skills to a later stage in the evidence chain: after measurements have been made, how should a learner interpret the claim that the resulting data have been reviewed and validated?

Original Reality Lab Case: The River Dataset With Four Different Statuses

This is an original composite teaching case. The values and status labels are constructed for learning.

A fictional river-monitoring project measures Indicator R at four sites. After review, the public table looks like this:

Site	Reported result	Validation note
A	12.4 units	Accepted
B	14.1 units	Accepted; duplicate disagreement larger than usual
C	Estimated 4.0 units	Accepted with qualifier
D	No reported value	Rejected for this analysis because sample identity could not be confirmed

The dataset page still says Validated.

Is that contradictory? No. Validation does not require every row to receive the same treatment. A good validation process can produce different outcomes because the purpose is to preserve what the evidence actually supports.

Site A may be usable without an added warning. Site B may be usable but deserve caution because its duplicate pair was less consistent. Site C may be useful for some analyses while being marked as estimated. Site D may be excluded because the project cannot establish which physical sample the result belongs to.

The important point is that validation can add information about data quality. It is not a rubber stamp that turns every row green.

Observed, Checked, Qualified and Claimed

Layer	Example
Measured	An instrument or method produced a result.
Verified	Records, calculations and analytical information were reviewed for correctness and completeness.
Validated	The result was evaluated against project data-quality requirements and performance criteria.
Qualified	A warning or status note was attached to explain a limitation or special condition.
Claimed	A user interprets the data to answer a scientific question.

Each stage has a different job. The final scientific claim should not silently erase the information added by the earlier stages.

Verification and Validation Are Related but Not Identical

U.S. EPA guidance separates verification from validation. Verification checks that the reported results correctly represent the analytical process and that calculations and records have been reviewed. Validation then evaluates the data against agreed data-quality specifications and project performance criteria.

For a Primary learner, think of two questions:

Verification: “Did we record and process what the laboratory actually did?”
Validation: “Given the project’s quality rules, how should this result be used or qualified?”

Neither question is identical to, “Is this number the perfect truth?”

Why a Validated Value Can Still Have a Qualifier

A data qualifier is a note, code or flag that tells the user something important about the value. It might indicate that a result is estimated, that a blank contained a small amount of the target, that duplicate agreement was weaker than expected, that the value is near a reporting limit, or that some other acceptance criterion was not fully met.

EPA guidance explicitly treats qualifiers as information that helps users decide how data should be used. A qualified value is not automatically useless. The qualifier tells the reader what caution belongs with the number.

The Flag Check: A Number and Its Qualifier Travel Together

Suppose a public graph copies the value from Site C in our fictional dataset but leaves out the word estimated. The numerical value has been preserved, but the scientific meaning has changed because an important status flag was removed.

A strong learner therefore asks:

Does the dataset include a quality-status column?
Are there symbols, footnotes or letter codes beside particular results?
Does the graph reproduce those qualifiers?
Were flagged values excluded from a calculation? If so, was the rule documented?
Were qualifiers silently stripped when data were copied into another website, infographic or spreadsheet?

The value and the warning belong to the same evidence object.

The Completeness Check: Was Everything That Should Have Been Measured Actually Reported?

A dataset can contain accurate individual measurements and still be incomplete. Perhaps one sampling site was missed. Perhaps one requested parameter was not analysed. Perhaps a sensor stopped recording for half the day. Validation often checks whether the data package contains what the project expected.

Completeness matters because a missing result can change the scientific question. Ten validated samples do not magically represent the eleventh location that was never sampled.

The Criteria Check: Validated Against Which Rules?

“Validated” only has scientific meaning when there are criteria behind it.

Those criteria can include things such as detection and quantitation limits, precision, accuracy, blank contamination, calibration performance, sample handling, required metadata or other project-specific expectations. Different projects can use different criteria because they answer different scientific questions and tolerate different levels of uncertainty.

That is why a learner should not assume that two datasets carrying the same word validated went through identical checks.

The Fitness-for-Purpose Check: Useful for What Question?

A dataset can be good enough to describe a broad pattern but not good enough to decide a tiny difference.

Imagine two validated sensors. Their uncertainty is about ±2 units. Site X reports 50 units and Site Y reports 51 units. The dataset may be perfectly suitable for showing that both sites are near 50. It may be weak evidence for the stronger claim, “Y is definitely higher than X.”

USGS work on environmental-data reliability similarly emphasises both data quality and relevance to the intended use. A good dataset is not simply “good” in the abstract; it is evaluated for what someone plans to do with it.

Why Validated Data Can Be Revised Later

Scientific data management does not end the moment a validation review is completed.

A later audit may discover a transcription error. An instrument problem may be recognised after comparison with another dataset. A better calibration may become available. A station location may be corrected. A processing script may contain a bug that is later repaired. New information can justify a new version.

Revising a dataset does not automatically mean the original scientists were careless or dishonest. Good science preserves version history and explains what changed.

The Version Check: Which Release Are You Reading?

If two websites show different values for what appears to be the same scientific dataset, check the release date or version before deciding that one must be wrong.

Was one site using preliminary data?
Was one using a validated later release?
Did the dataset receive a correction?
Did the analysis method change?
Did one website cache an older version?

Version is part of provenance. It tells you which evidence object you are actually reading.

Validation Does Not Repair a Poor Sampling Design

Suppose every measurement from one small corner of a lake passes every laboratory validation check. Can the report now claim to describe the entire lake?

No. Validation can evaluate whether those measurements meet the project’s data-quality criteria. It cannot make an unrepresentative sampling design representative after the fact.

This is an important Reality Lab boundary: quality of measurement and scope of sampling are different questions.

Validation Does Not Make a Proxy Become the Thing It Represents

A satellite-derived vegetation index can be validated against reference observations and still remain a proxy rather than a direct measurement of every aspect of “plant health”. A modelled quantity can be validated for a defined task and still depend on assumptions.

Validation strengthens confidence within a defined measurement job. It does not erase the distinction between observation, proxy, model and interpretation.

What Evidence Would Strengthen a “Validated Data” Claim?

The source explains the validation procedure or links to it.
The project states its data-quality requirements.
Individual quality flags remain attached to the values they qualify.
Rejected or excluded results are documented rather than silently disappearing.
Corrections and version changes are recorded.
Sampling design, method and measurement uncertainty are available separately from the validation label.
The intended scientific uses of the dataset are stated.

What Would Weaken It?

A website uses “validated” but never explains what the word means.
Quality flags are removed when data are copied into a graph.
A qualified or estimated result is displayed as an ordinary exact value.
Rejected values disappear with no explanation.
A dataset validated for one purpose is used for a much more demanding claim without checking suitability.
A validation badge is presented as proof that sampling, causation and interpretation are all automatically correct.

Worked Case 1: The Qualified Result

A validated dataset reports “7.4 units, estimated”. A social post copies only “7.4 units” and calls it an exact measurement. The number came from the dataset, but the post weakened the evidence by dropping the qualifier.

Worked Case 2: The Missing Station

Six monitoring stations have validated measurements, but a seventh station was offline. A map fills the missing area with the nearest station’s value. The validation of the six measured stations does not validate the invented seventh observation. The interpolation or gap-filling method needs its own explanation.

Worked Case 3: The Updated Release

Dataset Version 1 reports 23.1 units. Version 2 reports 22.8 after a calibration correction. Does the revision prove Version 1 was “fake”? No. It shows that the evidence chain was updated. The user should prefer the current corrected release and keep the version history visible.

Worked Case 4: The Tiny Difference

Two validated values are 18.0 and 18.2 units, while the method uncertainty is large enough that the small difference may not be meaningful. Validation supports using the measurements according to the project’s quality rules. It does not guarantee that every tiny numerical difference represents a real physical difference.

Worked Case 5: The Wrong Big Claim

A validated dataset shows that three measured locations had low Indicator P on one morning. A headline says, “The entire river system is clean all year.” The problem is not that validation failed. The problem is that the conclusion travelled far beyond the places and times actually measured.

Tempting Reasoning That Fails

“Validated means true.” Validation means checked against defined criteria for a purpose; scientific uncertainty remains.
“Validated means exact.” Measurement uncertainty and qualifiers can remain.
“Validated means final forever.” Datasets can be corrected or reissued when better information appears.
“If a value has a qualifier, validation failed.” The qualifier may be an output of successful validation because the review identified a limitation that users need to know.
“If one value is rejected, the entire project is useless.” The effect depends on why it was rejected and what scientific question is being asked.
“Validated data can answer any question.” Fitness for purpose remains essential.

Model and Measurement Limits

Validation systems differ. One environmental programme may validate every result manually. Another may combine automated checks with human review. A small classroom investigation may use a simpler process. A satellite mission may have several processing levels and quality flags. The label alone does not reveal the full procedure.

Nor can validation test every imaginable failure. It works from known quality objectives, documented checks and available evidence. Unknown problems can still be discovered later.

That limitation is a reason to preserve provenance, versioning and uncertainty—not a reason to distrust all scientific data.

How Far Can the Conclusion Travel?

A validated dataset can provide strong evidence that the reviewed data meet defined quality requirements, with any stated exceptions or qualifiers, for the intended project use.

It does not automatically establish that:

every value is exact;
every possible error has been ruled out;
the sampling design represents a larger place or longer time;
a measured association is causal;
a proxy equals the full phenomenon it represents;
the dataset will never be corrected;
the data are suitable for every new question someone later invents.

PSLE-Style Transfer Case

A class collects temperature data at three places in a model greenhouse. The teacher checks the calculations, confirms the thermometer calibration was acceptable and marks the dataset “validated for today’s comparison”. One reading carries a note that the thermometer was moved halfway through the interval.

Question: Why should the movement note remain attached to the reading even though the dataset was validated?

Reasoned answer: Validation does not erase a condition that may affect how the reading should be interpreted. The note is part of the evidence about the measurement and helps the user judge whether it is comparable with the other readings.

Explained Practice

Practice A: A dataset is labelled validated, but one value is marked “estimated”. Can the estimate still be part of the validated dataset? Yes. Validation can retain a value with a qualifier when the project’s rules allow it and the limitation is communicated.

Practice B: A second website removes all quality flags to make a cleaner graph. What changed? The visual became simpler, but the scientific information became poorer because evidence about data quality was lost.

Practice C: Version 2 changes three values after a processing error is found. Is revision evidence that science cannot be trusted? No. Detecting, documenting and correcting errors is part of scientific quality control.

Practice D: A dataset validated for detecting changes of about 10 units is used to rank locations that differ by only 0.2 units. What should you ask? Whether the measurement quality and uncertainty are adequate for the much smaller comparison.

Delayed Independent Return: The V-A-L-I-D Check

V — Validation rules: What criteria were used?
A — Attached qualifiers: What warnings or flags travel with individual values?
L — Limits: What uncertainty, missing data or method limitations remain?
I — Intended use: What scientific job was the dataset judged suitable for?
D — Date and version: Which release are you reading, and has it been corrected since?

Parent and Tutor Teaching Guide

Create a small fictional dataset with five rows. Mark three “accepted”, one “accepted—estimated” and one “excluded—sample label uncertain”. Tell the learner that the dataset has been validated. Ask, “If validation meant every row became perfect, why would these different labels remain?”

Then ask the learner to make two graphs: one that keeps the qualifiers visible and one that strips them away. Discuss which graph helps a reader make a better scientific judgement.

Finally, change the question. First ask whether Sites A and B are broadly similar. Then ask which is definitely higher by 0.1 unit. The learner should notice that the same validated dataset can be suitable for the first question and inadequate for the second. That is fitness for purpose made concrete.

Authoritative Sources

The current PSLE Science assessment objectives require learners to interpret and analyse information, evaluate observations, information and methods, and communicate explanations and reasoning. The 2023 Primary Science syllabus also asks learners to exercise healthy scepticism, consider assumptions and uncertainty, and understand how Science is communicated. A “validated data” label is therefore not a cue to stop thinking. It is a cue to ask what scientific checking has already been done and what the data can now responsibly support.

The Quiet Return

Validation is not the moment uncertainty disappears.

It is the moment a scientific project makes a more disciplined statement about what its data are good enough to do.

When you see “validated”, do not translate it into “believe everything”. Translate it into a better question: validated how, for what, and with which limits still attached?

Small Group Tutorials