PSLE Science Reality Lab Vol No.064 | “The Data Were Adjusted” — What Was Changed, and Why?

PSLE-SCI-REALITY-0064

Wait, What? Sometimes scientists change a number because they are trying to stop the measuring system from changing the science.

A climate graph on a science website says the data have been adjusted or homogenised. The word can sound suspicious. If a thermometer recorded 28.4°C, why would anyone later change the value?

Now imagine the thermometer station was moved from an open grass field to a rooftop beside an air-conditioning exhaust. Or an old instrument was replaced by a new one that reads consistently 0.3°C higher. Or the observer changed the time of day at which the daily reading was recorded. A long data record can suddenly jump even when the wider environment did not change by the same amount.

Scientists sometimes adjust a dataset because the measurement system changed and created a shift that is not the scientific phenomenon they are trying to study. The goal is not to make the result look nicer. The goal is to separate real-world change from a known change in how the world was measured.

But an adjustment is not automatically correct just because a scientist made it. A strong reader asks what changed, how the problem was detected, what evidence supports the correction, whether the raw record is preserved, and how sensitive the conclusion is to the adjustment.

Quick Answer

“Adjusted data” means the originally recorded observations have been processed to account for a documented or statistically detected non-target influence such as an instrument replacement, station move, observing-practice change or other measurement discontinuity. Do not treat adjustment as either automatic evidence of manipulation or automatic evidence of correctness. Ask for the raw observations, metadata about what changed, the method used to detect the shift, comparison evidence, and whether the scientific conclusion survives reasonable alternative corrections.

Reality Lab rule: Before judging an adjusted number, identify the unwanted change the adjustment is trying to remove.

What This Guide Teaches—and What It Does Not

This Reality Lab owns one job: how a Primary 5/6 learner should evaluate a real-world scientific dataset that has been adjusted for known or suspected measurement-system changes.

It does not teach the full mathematics of climate homogenisation, calibration or statistical quality control. It also does not decide whether every particular public dataset is correct. The transferable skill is simpler and more powerful: separate changes in reality from changes in the way reality was observed.

eduKate already owns the core PSLE micro-skills for measurement error, comparable measurements and method flaws. This page applies them to a modern scientific data product.

The Original Reality Lab Case: The School Weather Station Move

Imagine a fictional school has recorded afternoon temperature for ten years. For the first six years, the thermometer is mounted over grass in an open garden. During renovation, the station is moved to a concrete rooftop beside a wall.

The next month, the recorded afternoon temperatures are suddenly about 0.8°C higher than nearby weather stations. The school newspaper says:

“Our campus warmed almost one degree overnight.”

What are the plausible explanations?

The local environment really became warmer.
The rooftop location measures a different micro-environment from the former grass site.
The thermometer was changed or installed differently.
The observation time changed.
More than one of these happened together.

If neighbouring stations do not show the same sudden jump, and the jump begins exactly when the school station moves, that is evidence that the measurement change contributed to the break in the record.

Raw Observation, Metadata, Adjustment, Conclusion

Layer	Example in the teaching case
Raw observation	The thermometer recorded 31.2°C.
Metadata	The station had just moved from grass to a rooftop.
Comparison evidence	Nearby stations did not show the same jump.
Adjustment	The historical series is corrected to reduce the artificial break associated with the move.
Conclusion	The corrected record is used to study longer-term temperature change.

These layers should not be collapsed. The raw reading is part of the historical evidence. The adjusted value is a processed estimate intended to make observations taken under changed conditions more comparable.

Why Long Scientific Records Can Develop False Jumps

NOAA explains that weather-station records can contain non-climatic shifts caused by changes in thermometers, station location, observation practices and the surroundings of a station. These changes matter because climate science often asks how temperatures changed over decades. A measurement-system break can imitate a climate change if it is not recognised.

Think of a race measured with two rulers. The first ruler is accurate. Halfway through, someone swaps it for a ruler whose centimetres are slightly longer. The runners did not suddenly shrink. The measuring system changed.

Adjustment Does Not Mean “Choose the Answer You Want”

A scientifically defensible adjustment should have a reason that exists independently of the preferred conclusion. That reason can come from station history, instrument records, known calibration changes, comparisons with neighbouring observations or a documented processing method.

If a correction is made only because one value looks inconvenient, that is weak science. If a correction follows a transparent method that is applied consistently to all qualifying cases, and the raw data and metadata remain available, the adjustment is much more auditable.

A Powerful Clue: Compare With Neighbours

Suppose Station A suddenly jumps 1°C in 2007. Nearby Stations B, C and D do not. Station A’s history says its thermometer was replaced that month.

That pattern makes an instrument-related shift more plausible than a real regional temperature jump. If all four nearby stations rise together during the same weather event, the shared change is more consistent with a real environmental signal.

NOAA’s pairwise homogenisation approach uses comparisons among stations to help identify unusual breaks that do not match surrounding records. The full method is advanced, but the Primary Science reasoning is familiar: compare the suspect result with independent evidence under similar conditions.

Adjustments Can Move Values Up or Down

A common misunderstanding is that adjustments always push results in one direction. A correction method should follow the evidence, not a preferred sign. Depending on the measurement change, some values may be adjusted upward, some downward and some not at all.

The important question is therefore not “Was the number changed?” but “What measurement discontinuity was identified, and does the correction reduce that discontinuity in a justified way?”

Raw Data Should Not Disappear

Good scientific data stewardship keeps provenance: what was originally recorded, what processing steps were applied, which version of the dataset was used and why a value changed.

NOAA maintains station-history metadata through its Historical Observing Metadata Repository. That kind of metadata matters because a number without its observation history can be much harder to interpret.

An adjusted dataset should therefore be thought of as a new evidence layer connected to the raw observations, not as permission to erase the original record.

The Adjustment Audit

What was originally measured? Find the raw observation if available.
What changed in the measurement system? Instrument, location, time, method, processing or environment?
How was the break detected? Metadata, calibration record, neighbouring measurements, statistical pattern or several together?
What exactly was adjusted? One observation, a period of the record, a baseline, a grid or a derived series?
Was the method applied consistently? A rule that changes depending on whether the result is convenient is weak.
Are raw and adjusted versions traceable?
Does the main conclusion depend completely on one uncertain correction?

Worked Case 1: The New Thermometer

A science club replaces a thermometer. For one week, it operates the old and new instruments side by side. The new thermometer reads about 0.4°C higher on almost every comparison.

If the club wants one continuous long-term series, it now has direct evidence of an instrument offset. A documented adjustment may be justified. The strongest record would preserve the paired comparison, the original readings and the correction method.

Worked Case 2: The Sudden Regional Heatwave

Five nearby stations all jump by about 2°C during the same week. One student says, “We should adjust the jump away because it is unusual.”

That would be poor reasoning. Unusual does not mean erroneous. A shared rise across multiple independent stations may be evidence of a real heat event. Adjustment should target measurement artefacts, not erase genuine extremes.

Worked Case 3: The Station Move Without Metadata

A temperature record shows a sudden shift, but there is no surviving note about a station move. Neighbouring stations do not show the shift. Can scientists still investigate?

Yes, but the evidence is weaker than if the station history were documented. Statistical comparisons can identify a likely discontinuity, but the cause may remain uncertain. The conclusion should preserve that uncertainty rather than inventing a definite story.

Worked Case 4: A Correction That Changes the Conclusion

A fictional experiment reports that Product A performs 10% better than Product B. Later, the lab discovers that one measuring instrument systematically overstated Product A by 12%. After correction, Product A no longer performs better.

The correct response is not to defend the old headline. The conclusion must follow the corrected evidence. This connects directly to Reality Lab Vol No.026 on scientific corrections.

What Would Strengthen an Adjustment?

clear station or instrument metadata;
side-by-side calibration measurements;
agreement among several independent comparison methods;
a correction algorithm published in advance or documented transparently;
raw values remaining accessible;
sensitivity tests showing whether different reasonable corrections change the conclusion;
neighbouring observations that support the identified break.

What Would Weaken It?

no explanation of what was changed;
no preserved raw record;
a correction applied only after seeing whether it helps the preferred result;
different rules applied inconsistently to similar cases;
a claimed instrument change with no supporting metadata or comparison evidence;
a conclusion that changes dramatically under several equally reasonable correction methods.

PSLE Science Transfer: Did the System Change, or Did the Measuring Tool Change?

Suppose a student measures the mass of the same object every day. On Day 5, the balance is accidentally reset so that zero is 2 g too high. Every later reading becomes 2 g higher even though the object did not change.

The correct question is not “Why did the object suddenly gain 2 g?” It is “Did anything about the measuring system change?”

This is the same reasoning used in long real-world records. eduKate’s guide on measurement errors that shift results owns the core micro-skill.

Tempting Reasoning That Fails

“The data were adjusted, so they are fake.” A documented correction can improve comparability when the measurement system changed.
“Scientists adjusted it, so it must be correct.” The correction still needs evidence and a transparent method.
“Raw data are always the truth.” Raw readings preserve what the instrument recorded, but the instrument itself can contain known biases or discontinuities.
“Any sudden jump should be removed.” Real events can produce sudden changes.
“An adjustment should always make the graph smoother.” Scientific correction is not beautification. A genuine extreme should remain if the evidence supports it.

Practice 1: The Moved Rain Gauge

A rain gauge is moved from an open field to a spot partly sheltered by a roof edge. Rainfall readings suddenly fall compared with nearby stations. What should you investigate before concluding the town became drier?

Answer: Investigate the station move, exposure conditions and comparison with nearby gauges. The lower readings may reflect the new measuring environment rather than a real drop in rainfall.

Practice 2: All Stations Change Together

Six weather stations all record a strong temperature increase during the same afternoon. One station has not moved or changed instruments. Should the rise automatically be adjusted away?

Answer: No. Agreement across independent stations is evidence that the rise may be a real environmental event.

Practice 3: Raw and Adjusted Disagree

A chart provides both raw and adjusted values. Which should you hide?

Answer: Neither should be hidden. They answer different questions: the raw series preserves what was recorded; the adjusted series applies a documented correction intended to improve comparability. A good explanation shows their relationship.

Delayed Independent Return

The next time you see words such as adjusted, corrected, bias-corrected, homogenised or calibrated, do not decide immediately whether the data are trustworthy. Ask three questions:

What unwanted measurement change was identified?
What independent evidence supports the correction?
Can I trace the adjusted result back to the original observations?

If you can answer those, you are evaluating the adjustment instead of reacting to the word.

Teaching Guide for Parents and Tutors

Use two rulers in a simple thought experiment. Tell the learner that the first ruler is correct but the second begins at the 1 cm mark even though someone labels it zero. Ask what happens to every later measurement. Then ask whether the object changed or the measurement system changed.

Next, show a short fictional data series with a sudden jump exactly when the ruler changes. Ask the learner to keep two explanations alive: real change and measurement-system change. Only after adding comparison evidence should they decide which is more plausible.

The weak link to watch is binary thinking: “raw equals honest” versus “adjusted equals dishonest”, or the reverse. Replace that with evidence questions about provenance, mechanism and validation.

Authoritative Sources

The Quiet Return

Scientific records are not made outside history. Instruments age. Stations move. methods change. People improve how they observe.

The strongest dataset is not the one that pretends none of that happened. It is the one that preserves what happened, documents what changed and shows why any correction was justified.

Small Group Tutorials