Did you know that two digits in the wrong order can change a whole science conclusion, even when every formula in the spreadsheet is working correctly? A reading of 15 mm that becomes 51 mm during typing may make one observation seem unusually large. The spreadsheet faithfully calculates from the wrong entry.
A transposed number error occurs when the order of digits changes during copying or entry, such as writing 51 instead of 15. To check it properly, identify a surprising value, trace it to the original measurement record using the correct sample and time, confirm what was actually written, and correct the transcription transparently. Never change a number solely because it looks unlike the others.
What to check before drawing a graph or writing a conclusion
- Confirm the measured quantity, unit, time point and sample identifier first.
- Look for unexpectedly large or small values without assuming they are necessarily wrong.
- Compare suspicious entries with the earliest reliable source record, not with a second copy of the same spreadsheet.
- Separate a confirmed typing error from an unusual but genuine observation.
- Record the correction, its reason and the source used to justify it.
- Recalculate summaries, charts and conclusions after a confirmed correction; retain the raw audit trail where appropriate.
These steps answer a practical question: ‘Can I trust that the number I am analysing is the number that was measured?’ They do not make experiments perfect, but they prevent a common and avoidable class of mistakes.
A fully worked classroom example: seedling heights
Imagine four seedlings observed on the same day in a classroom activity. Their heights are measured in millimetres, and each is given a unique identifier A, B, C or D. This is an entirely invented example to teach checking; the measurements are not reported research.
The paper notebook has: A = 14 mm, B = 15 mm, C = 16 mm and D = 15 mm. When transferring the readings into a spreadsheet, someone enters D as 51 mm. All the other entries are copied correctly.
Step 1: calculate what the unchecked table appears to say
The spreadsheet values are 14, 15, 16 and 51 mm. The total is 14 + 15 + 16 + 51 = 96 mm. The average for four seedlings is 96 ÷ 4 = 24 mm. This arithmetic is correct for the spreadsheet values, but it does not validate that those values came from the observations.
The unchecked range is 51 − 14 = 37 mm. The one unusual entry both increases the mean and makes the apparent spread much larger. A reader may mistakenly say that the plants vary enormously in height.
Step 2: find the source record and match identities
The student returns to the signed or otherwise identifiable original notebook and checks the entry specifically labelled D at the same date and measurement time. It reads 15 mm. If the original record is legible and its identity is clear, we have evidence of a digit-order transcription mistake. If the source were unclear, we would flag the uncertainty rather than silently alter 51 to whatever looks plausible.
Step 3: correct the copied figure and recompute
The corrected dataset is 14, 15, 16 and 15 mm. The corrected total is 14 + 15 + 16 + 15 = 60 mm, so the corrected mean is 60 ÷ 4 = 15 mm. The corrected range is 16 − 14 = 2 mm. The recorded digit swap changed the reported mean by 24 − 15 = 9 mm and the apparent range by 37 − 2 = 35 mm.
Notice that the recorded measurement for D was 15 mm; a genuine 51 mm observation would have to be preserved and investigated. A number’s distance from its neighbours is a reason to check, never a licence to rewrite data.
Step 4: explain what changed in the report
A suitable note might say: ‘During transfer from the original measurement log to the summary table, the value for seedling D was entered as 51 mm instead of the recorded 15 mm. The spreadsheet entry was corrected after checking the source log. The mean changed from 24 mm to 15 mm. The original log is retained.’ This tells a reader what happened without claiming the experiment itself was rerun.
The raw observation, transcription, analysis and report are distinct stages. A correction at one stage should not be mistaken for new evidence at another.
Why a calculator cannot fix a recording mistake
A calculator can sum 14 + 15 + 16 + 51 flawlessly. It cannot know that the last value should have been 15 because the mistake occurred before arithmetic began. This is a data-provenance problem: the link between what was observed and what was analysed has been broken.
There are several ways that link can fail. A number may be assigned to the wrong student or sample, a decimal point may move, a unit may change, a column may shift, a date may be copied wrongly, or a formula may omit a row. A useful audit checks both the number and its context.
How to tell a possible error from a real outlier
An outlier is an observation noticeably different from others. It may be an error, ordinary variation or a genuinely interesting result. The same value, 51 mm, cannot be declared ‘incorrect’ just because the other seedlings are around 15 mm. That conclusion came only after an independent source established 15 mm for the matched reading.
If the notebook also read 51 mm, the learner should retain the observation, check the instrument, inspect how it was measured, and ask whether the experimental method permits that result. Perhaps seedling D is a different species, its height was measured in centimetres and mislabelled, or its record belongs to another date. These are hypotheses to test, not explanations to invent.
The United States National Institute of Standards and Technology advises investigators to understand why outlying values appeared before deciding whether to remove them. A confirmed data-entry error and a true unusual measurement call for different treatments.
A five-question diagnostic for students
- Which object or person does this row represent? Is the identifier unique and unchanged?
- What is the recorded time? Could the value belong to a later or earlier observation?
- Which unit was used in the source, and did it remain the same in the table?
- Where is the earliest dependable record, and does it independently support a corrected value?
- What calculations, graphs and claims need to be rechecked if the entry changes?
If the learner can answer these before editing, they have moved beyond simply spotting a suspicious value. They can defend the correction.
Three common mistakes to stop
Mistake one: replace every unusual number with the average
Replacing 51 with 15 because 15 ‘looks normal’ is improper. It erases the distinction between an observation and an assumption. In the worked example, the correction to 15 is justified by a source record, not by the average.
Mistake two: erase the old value without a note
A teacher or classmate who reviews the spreadsheet later may wonder how the published graph relates to the earlier worksheet. Mark the correction and its reason. For formal research, keep an unaltered copy of the original source and a traceable correction record; classroom conventions can be simpler, but honesty is the same.
Mistake three: fix the cell but leave the old chart
A pasted chart or manually typed conclusion may not update automatically. Compare the final graph labels, axis units, sample count and written mean against the corrected table. Look for any figures exported before the fix.
Progressive practice: from seeing a strange number to tracing evidence
Guided task: label the layers
Draw four boxes: observation, notebook, spreadsheet and report. Place the value ’15 mm’ in the observation record, ’15 mm’ in the notebook, ’51 mm’ in the spreadsheet and ’24 mm’ as the computed mean. Ask the student to point to the first place the evidence changed. The error appeared during the notebook-to-spreadsheet transfer.
Paired task: compare two possible explanations
Case A: the notebook says 15 mm but the table says 51 mm. Case B: both notebook and table say 51 mm. Which case supports correcting the table to 15? Only A, because A has independent documentation of the copied value. Case B requires further investigation, not a cosmetic adjustment.
Independent task: catch a new digit swap
Invented science readings of water collected in millilitres are 23, 24, 25 and 42 mL, but the original record shows 24 mL for the last sample. The unchecked total is 23 + 24 + 25 + 42 = 114 mL; mean 114 ÷ 4 = 28.5 mL. The verified corrected total is 23 + 24 + 25 + 24 = 96 mL; mean 96 ÷ 4 = 24 mL. The mean shifts by 4.5 mL. Ask the learner to produce the correction note without seeing a model.
Transfer task: a surprising number that must not be corrected
Give a different dataset in which a high value is independently confirmed by a timestamped original instrument record. Ask the learner to preserve it and explain what further experiment could test the cause. This prevents the lesson from becoming ‘remove any number that looks different’.
How teachers can design better data handling
Have learners enter sample IDs and units before recording numbers, because a value without identity is hard to verify. Use a short source-to-summary check: two randomly selected rows, one unusual row and one row used in the final calculation. The point is not to review everything forever; it is to establish the habit of verifying the chain.
Teachers can ask for both an original record and a clean summary table, with corrections marked clearly. For younger learners, a two-column check—’What I saw’ versus ‘What I typed’—is enough. For more advanced work, retain source data, processing steps and version notes.
Where pupil work contains personal information, avoid sharing identifiable datasets unnecessarily. The lesson is about data integrity, not public exposure of student records.
How parents can help at home
Ask the child to explain why they changed a number. ‘Because it was too big’ is not sufficient. ‘Because the original notebook for Sample D showed 15 mm at the same time’ is an evidence-based reason. If the notebook is missing, encourage the child to mark the value as uncertain and discuss it with the teacher rather than secretly inventing a replacement.
This habit transfers to household measurements, expenses and schedules. A digit swap in a grocery budget can also affect a total, but the check remains the same: trace the entry to an independent, reliable source and record the correction.
Four weeks to build a reliable checking routine
- Week 1: compare original and copied readings, identify sample IDs, units and times, and find two obvious transcription errors.
- Week 2: distinguish digit swaps, unit mismatches and authentic unusual measurements. Explain why each requires different evidence.
- Week 3: calculate a mean and range before and after a verified correction, then update the associated chart and conclusion.
- Week 4: inspect an unfamiliar table without prompts, document one traceable correction and explicitly preserve an unexplained outlier.
An observable improvement is when a pupil stops guessing the ‘right’ value, uses original records, states a defensible reason for editing and recomputes all affected conclusions. Accuracy is a process, not just a final number.
Limits: the original record can also be wrong
A paper notebook is not magical proof. A student may have misread an instrument before writing any number. If possible, review measurement technique, instrument resolution, labels, environmental conditions and repeated observations. When evidence is insufficient, report uncertainty honestly rather than treating the first record as infallible.
Do not retroactively conduct an imaginary replacement trial, insert an invented data point or claim a correction was verified if the raw evidence is missing. Some activities may require following a teacher’s specific lab-record protocol.
Frequently asked questions
Is a digit swap the same as rounding?
No. A digit swap changes order, such as 15 becoming 51. Rounding changes the number’s precision according to a rule. Both can affect analysis but require different checks.
Should every outlier be deleted?
No. Retain unexplained unusual values while investigating them. Only make a supported change under a stated data-handling rule.
What if a pupil used centimetres in one row and millimetres in another?
Check the unit from the original record before converting. For example, 1.5 cm equals 15 mm; the conversion should be documented, not mistaken for a newly measured value.
Does the mean always change more than the median?
Not always. In our four-value example the original ordered values 14, 15, 16, 51 mm have median (15 + 16) ÷ 2 = 15.5 mm. After correction to 14, 15, 15, 16 mm, the median is 15 mm. Which summary changes more depends on the data and type of error.
Can a spreadsheet’s automatic data validation solve this?
Validation can flag ranges, missing units and impossible formats, but plausible-looking transpositions still require source checking. A rule cannot determine what was actually observed if the source was copied incorrectly.
Verified sources and useful next reading
The NIST Engineering Statistics Handbook on detecting outliers explains why unusual values require investigation rather than automatic deletion. Compare this with repeated measurements versus independent science trials and checking chart totals. For wider study, use the eduKate Education Hub and the Clementi Mathematics learning reference.
Closing thought
A number should not have to look ordinary to be believed, and an error should not have to look dramatic to be checked. Keep the chain from observation to report intact. That is how a science result becomes something a reader can examine, reproduce and trust.