Six seedlings receive compost and six receive ordinary growing medium. After fourteen days, one compost seedling has died and the label on one comparison pot is missing. A class calculates two attractive averages from the ten complete height readings and declares that compost increased growth. Yet the original experiment had twelve units, not ten, and the two missing outcomes mean different things. The gold standard of school scientific missing outcome data analysis is to preserve the full investigation, distinguish genuine results from unavailable measurements, and avoid claims that the remaining data cannot support.
Quick answer: Record the originally assigned experimental units, their measurements and missing-outcome reasons. Distinguish valid zeros, missing observations, invalid readings and outcomes that are undefined. Ask whether absence might be related to the treatment or result. Calculate transparent summaries of available data, examine how uncertainty affects interpretation and preserve original records. Never invent missing measurements or exclude inconvenient cases to manufacture a positive finding.
This problem is different from an anomalous measured value or a wrong digit entered into a spreadsheet. A missing outcome is evidence that the planned observation is unavailable. Sometimes the original reading can be recovered; other times the loss cannot be repaired. Schools can teach the fundamental reasoning without requiring young learners to perform advanced statistical missing-data modelling.
Why a complete-case average can mislead
An average describes the numbers used to calculate it. If five plants per group have valid height gains, two group means describe those ten plants. They do not automatically describe every seedling originally assigned. This matters when plants disappear for reasons related to growth, survival or the treatment itself. An experiment that loses its weakest performers may seem to improve simply because its poorest outcomes are no longer represented.
A student should ask which units the original research question covered, how many outcomes were planned, how many were collected and why others are missing. Merely having the same number of complete rows in each group does not guarantee comparability. A dead plant, an unlabeled plant, an invalid balance reading and an observation omitted because of rain represent different circumstances.
Good reporting can be concise: ‘Six seedlings were allocated to each condition; valid height gains were available for five per group. One treated seedling died and one control pot lost its identifying label.’ That sentence protects the meaning of the later arithmetic.
Four statuses that should not share one blank symbol
A valid zero is measured under the intended method—for example, a five-minute butterfly survey during which no butterflies were detected. Missing means the reading is unavailable or was never made. Invalid means a value was recorded but cannot be trusted under a specified quality requirement, such as a balance failing calibration. Undefined means the original measurement is no longer meaningful for that unit, as with the final standing height of a seedling that died.
Most spreadsheet tools ignore empty cells when averaging and include numeric zeros. Their arithmetic is correct given their inputs, but a blank mistakenly used in place of an observed zero can inflate the mean. A zero used in place of a missing measurement can lower it. The software cannot infer whether an observer skipped the site, saw nothing or simply forgot to enter a result.
Give students a status column rather than asking them to force every case into one number. The data table can retain the actual reason, an associated observation time and any later evidence-based correction. Such record design reduces confusion before mathematical analysis begins.
When missingness can be related to the result
Suppose the least healthy treated seedlings are most likely to die before the final measuring day. Their missing height gains are connected to the treatment experience. Calculating only surviving plants can overstate the average performance of that treatment across all originally assigned seedlings. This possibility is sometimes called survivor or attrition bias, depending on the design.
Contrast a lost label. If the label became unreadable through an accidental event unrelated to growth, the absence may be less directly related to the height outcome. But it should not be assumed harmless without examining when, where and how the problem occurred. A method that loses labels mostly on one type of pot or one growing condition can create a systematic issue.
Advanced research distinguishes several missing-data mechanisms and may use formal modelling. Primary students need not memorise those terms. Ask simply: ‘Could a plant with a particular outcome be more likely to disappear from our measured table?’ If so, the observed subset may differ meaningfully from the intended sample.
Worked twelve-seedling experiment
Consider a fictional classroom study with twelve independent pots: six compost-assigned and six comparison pots. The teacher establishes a starting-height measurement and a fourteen-day follow-up. Five treated seedlings yield valid gains of 3.8, 3.5, 3.9, 3.6 and 3.7 centimetres. The sixth treated seedling dies before the final height measurement. Five controls yield 3.1, 3.0, 3.3, 3.2 and 3.4 centimetres, while the sixth control has an unreadable label so that a matching gain cannot be established.
These figures are invented teaching data, not real horticultural findings. They demonstrate why the original allocation needs to stay visible in the table. If the dead treated plant and the unlabeled control were simply deleted, a reader would see two balanced five-plant groups, but not the actual twelve-unit experiment. Important biological and record-keeping facts would disappear.
The study should preserve each pot identifier, original assignment, starting measurement, end status and explanation for any missing value. Students should also describe the relevant measurement conditions. A death cannot be treated automatically as the same kind of missing result as an unreadable ID, because survival itself may be important to the investigation.
| Pot | Original group | Height gain | Recorded outcome |
|---|---|---|---|
| C1 | Compost | 3.8 cm | Valid |
| C2 | Compost | 3.5 cm | Valid |
| C3 | Compost | 3.9 cm | Valid |
| C4 | Compost | 3.6 cm | Valid |
| C5 | Compost | 3.7 cm | Valid |
| C6 | Compost | — | Seedling died |
| B1 | Base medium | 3.1 cm | Valid |
| B2 | Base medium | 3.0 cm | Valid |
| B3 | Base medium | 3.3 cm | Valid |
| B4 | Base medium | 3.2 cm | Valid |
| B5 | Base medium | 3.4 cm | Valid |
| B6 | Base medium | — | Label lost |
Calculate correctly and qualify the result
The five observed compost gains total 3.8 + 3.5 + 3.9 + 3.6 + 3.7 = 18.5 cm. The complete-case mean is 18.5 ÷ 5 = 3.70 cm. The five recorded control gains total 3.1 + 3.0 + 3.3 + 3.2 + 3.4 = 16.0 cm, so their complete-case mean is 16.0 ÷ 5 = 3.20 cm. The observed difference is 0.50 cm.
This arithmetic describes the measured plants correctly. It does not establish a treatment effect of 0.50 cm for every plant originally assigned. The compost group’s missing outcome is a death, potentially relevant to treatment performance; the control group’s missing outcome arose from lost identity. Their absence can influence interpretation in different ways.
An appropriately worded conclusion is: ‘Among the five seedlings with valid height-gain measurements in each group, the compost group’s mean was 0.50 cm greater. Because one compost seedling died and one comparison plant’s gain was unavailable, these means do not establish the average result across all twelve originally allocated units.’ That is clear Science, not indecision.
Why filling blanks with invented numbers is risky
A pupil may propose assigning zero height gain to the dead plant. Another may fill it with the mean of the five survivors. Neither is an observed final-height measurement. Different codings could be appropriate for a carefully defined advanced outcome, but they cannot be inserted into a raw class table as if they were observed facts. Death may call for a separate survival outcome rather than a fake height reading.
One can teach sensitivity by marking hypothetical values explicitly. If the missing control gain had been 2.8 cm, the six-plant control mean would be (16.0 + 2.8) ÷ 6 ≈ 3.13 cm. If it had been 4.0 cm, the mean would be 3.33 cm. Both are imagined scenarios used to show how a conclusion depends on the unknown value; neither reconstructs the lost measurement.
An estimate should always be identifiable as an estimate, with its assumptions stated. Advanced modelling does not grant a licence to invent precision. A classroom can learn the important habit: preserve the gap, explore implications honestly and propose a better study rather than quietly inserting the number everyone hopes to see.
Survival and growth are different scientific outcomes
Height gain among surviving plants answers one question: how much did those plants grow? Survival answers another: how many originally assigned plants remained alive by the end of the defined period? A treatment can have different effects on these two dimensions. A high growth average among survivors does not automatically compensate for a greater number of deaths, nor does a single death establish that the treatment is harmful.
In the fictional case, five out of six compost seedlings survived. Six control plants may be alive, but only five gains are reliably identified. Record those distinct facts. If a future investigation aims to compare overall treatment performance, the teacher can specify both survival and height gain as outcomes from the start. Students should not redefine what counts after seeing which treatment looks better.
Conceptual clarity matters more than choosing a single convenient metric. Science can sometimes produce several relevant outcomes that cannot fairly be compressed into one number without a justified model. Learning to say what each measure represents prepares students to read research claims critically.
Preserve group assignment even when outcomes are missing
Random allocation helps create a fair initial comparison. A plant remains part of its originally assigned group even when its result becomes unavailable. Quietly transferring an underperforming plant to the other group after treatment begins breaks the experimental record and can bias the apparent difference.
Likewise, replacing a dead seedling with a healthy one halfway through fourteen days does not create an equivalent original unit. The new plant has a different observation history and perhaps a different length of exposure. It could become part of a separately designed follow-up, but should not be presented as the lost plant.
Keep a simple unit-level audit: pot identifier, assignment, starting condition, observation dates, measured values, result status and reason for absence. A class can do this on paper without professional software. The objective is that a later reader can reconstruct exactly which plants the calculation describes.
Missing results caused by measurement methods
Some gaps arise from instruments rather than organisms. A digital balance might malfunction during one series, a thermometer may display an unreadable value, or a clock might stop during a timed observation. Record the circumstances and assess whether any available measurement remains valid for the intended purpose. Do not replace an invalid result with zero simply because it is numeric.
The cause may itself be systematic. If the balance fails only during trials with a fan operating, the missing readings may concentrate in the windy condition. Averaging the remaining observations can underrepresent precisely the circumstances the study set out to test. More repetitions using the same malfunctioning arrangement will not automatically solve the problem.
An appropriate next step may be a suitable instrument check, supervised replacement, clearer timing protocol or a separate valid new trial. The teacher should explain how the change addresses the diagnosed weakness. A generic phrase such as ‘repeat for accuracy’ misses the causal reasoning.
Lost identities and transcription errors
An unreadable label is different from a biological failure. If another trustworthy record, such as a dated notebook or controlled position map, can establish which pot it belonged to, the missing value may be recoverable. A correction should identify its source and date, preserve the original uncertainty and regenerate affected calculations.
If no reliable matching evidence exists, the proper result is missing. Students should not identify a pot by guessing which height seems to fit the treatment prediction. That would turn a plausible story into supposedly observed data and make the experiment less trustworthy.
A table copied into a spreadsheet can also lose an actual zero or omit a valid row. Reconcile with the original field sheet rather than treat every blank as a new scientific event. The Gold Standard of Observation Record Reconciliation addresses source-to-summary consistency; the present article focuses on evidence that never became available in the intended outcome.
Sampling and detection still matter
A garden survey may contain a complete table while remaining biased because the class selected only convenient locations. Conversely, a well-designed sample may contain legitimate missing observations because weather prevented access to particular sites. Reporting the achieved sample makes it possible to consider how the omissions affect representativeness.
Likewise, an observer may conduct a valid five-minute search and record zero butterflies even when hidden butterflies were present. That is a detection limitation, not missing data. The distinction between ‘not seen’ and ‘not measured’ matters. A classroom should avoid importing the terminology of one problem into another simply because both produce small numbers.
Good teaching therefore places missingness within the larger evidence chain: sample selected, treatment or observation planned, instrument or observer used, data recorded, analysis performed and inference made. Repair whichever link actually failed.
Planning prevents predictable missingness
Before starting the investigation, agree what will count as a valid observation, how sample IDs will be maintained, and what happens if the defined outcome cannot be measured. For a plant study, plan how to record germination, death, missed measurements and pot labels. For fieldwork, prepare safe stop rules and document whether inaccessible sites can be replaced under a preplanned method.
A short pilot can reveal that labels detach when watering, the spreadsheet lacks a missing-data status or the measuring scale is unsuitable. Improving the protocol before collecting the main data is more reliable than trying to manufacture a correction afterward.
The teacher should make it safe for learners to report missing outcomes honestly. If pupils fear being marked down solely because an instrument failed, they may be tempted to invent a number. Reward transparent records, thoughtful explanations and appropriate scientific follow-up rather than an artificial 100% completed table.
A practical learner progression
Repair: Show one garden square inspected with zero sightings and another never inspected. Ask pupils which yields a real zero. Then compare a spreadsheet average containing zero with one where the entire measurement is unavailable. The objective is understanding before terminology.
Stabilisation: Use the twelve-seedling table. Ask learners to calculate the two complete-case means, count all original units and explain why the two missing outcomes matter differently. A strong answer explicitly limits the conclusion to measured plants while identifying survival as a distinct concern.
Extension: Invite advanced learners to design a follow-up study that records survival and height gain separately, uses robust IDs, and tests assumptions about missing values. Discuss attrition after random allocation without suggesting that sophisticated statistical formulas are mandatory for younger students.
Teaching sequence and examination communication
A ninety-minute lesson can begin with ten minutes distinguishing zero and missing data, followed by fifteen minutes reconstructing the sample. Allow twenty minutes for the worked arithmetic, fifteen minutes for the two missingness reasons, fifteen minutes for hypothetical scenarios, ten minutes for writing a qualified conclusion and five minutes for transfer to a new experiment. This sequence is illustrative rather than MOE-prescribed.
In school answers, the learner should not write a long generic list of every possible mistake. A concise statement tied to the actual evidence is stronger: ‘One treated plant died, so the growth mean of surviving plants may not represent all assigned treatment plants.’ The precise consequence is what demonstrates understanding.
Parents can support this habit by asking ‘How many samples were there at the beginning, how many at the end, and why did any disappear?’ rather than encouraging their child to fill every blank. The goal is evidence literacy and scientific honesty.
Break-tests: can the learner defend the denominator?
If a quadrat was observed and no insect was found, should it be recorded as zero? Yes, if that is the valid outcome under the method. If it was not visited because of unsafe weather, should it be zero? No. A student who understands both examples can distinguish non-detection from non-measurement.
If one treated seedling dies and the surviving plants grow well, is the survivor mean sufficient to describe the whole treatment? Not automatically. If both groups lose exactly one measurement, does that prove attrition was harmless? No; reasons may differ. If a pupil inserts group means into blank cells, can they still claim those are observed data? No.
Another break-test concerns chronological integrity. If the missing value is later recovered from an original notebook, the corrected table and resulting graph should record the source of the correction. If no source exists, the missingness should remain explicit. A professional-looking graph is never an excuse to fabricate evidence.
Frequently asked questions
Is a blank the same as zero? No. Zero can be observed; missing means the measurement is unavailable. Can I average complete rows? Yes, provided the average is labelled as describing the measured subset and the missingness is discussed.
Should a dead plant be assigned zero growth? Not automatically; survival and final height gain are different outcomes. Does random assignment solve attrition? It helps the initial comparison but does not prevent missing outcomes from biasing the achieved sample.
Can I estimate a missing value? Advanced analyses can make estimates under assumptions, but estimates are not observed measurements. Should I repeat everything? Not necessarily. First identify whether the cause was a source transcription error, design flaw, instrument problem or scientifically meaningful event.
What is a good school conclusion? State the observed means, the actual denominators, why values were unavailable and how that limits the interpretation. How can parents help? Ask pupils to trace the graph back to observations and explain blank cells without guessing.
An original-to-final sample audit a class can reuse
A teacher can create a one-page reconciliation worksheet that keeps the planned experiment visible. List every sample ID before treatment. Add the original group, the date of each observation, a status field and the most specific missingness reason supported by evidence. At reporting time, count planned units, observed units, legitimate exclusions and outcomes that changed form, such as death. The worksheet is not an opportunity to relabel plants after seeing the results. It is an evidence chain connecting the experimental design to the denominator used in each calculation.
A useful check is to compare the planned list with the final graph. In the fictional study there should be twelve assigned pot IDs, ten usable height gains, one death and one unmatched pot. A graph with only ten anonymous dots may be mathematically accurate but omits information required to interpret what happened. The report can retain the two excluded or undefined outcomes visibly in an accompanying status table. Readers should not infer that the original experiment contained only ten plants.
Why the percentage with an outcome can be more revealing than the mean
Measurement coverage can be described without treating every absence as identical. In the worked case, usable height gain is available for five of six originally assigned pots in each condition, or approximately 83.3% of pots per group. The matching percentages do not establish matching reasons for absence. The compost group lost a living experimental unit through death. The base group retained an apparently living plant whose before-and-after identity was lost. The denominator is the same, but the scientific implication differs.
Suppose a different study observes 100% of the treatment group and only 50% of controls because the controls were harder to reach. A simple comparison of available means might appear impressive, yet the achieved control sample may no longer represent the intended comparison. Ask why the observations are missing and whether those reasons are associated with group or outcome. Coverage percentages help readers notice incompleteness; they do not by themselves remove its bias.
Worked follow-up question: when a missing value is later recovered
Suppose the teacher finds a dated position map and original notebook establishing that the unlabeled control pot B6 gained 2.9 cm. That is a new piece of verified historical evidence, not an invented measurement. The full six-pot base-group total becomes 16.0 + 2.9 = 18.9 cm, giving a base mean of 18.9 ÷ 6 = 3.15 cm. The five measured compost survivors still average 3.70 cm, and one compost plant still died. The arithmetic can be updated, but the comparison’s biological limitation remains.
A good correction log states which value was recovered, where the evidence came from, when it was found and what summary changed. Do not announce that the treatment effect is now exactly 0.55 cm across all plants; that number compares five measured compost survivors with six base plants. The valid denominators still differ. The recovery fixes an identification gap but does not solve the issue of how to interpret survival alongside growth.
A cross-topic transfer: an unfinished filtration investigation
A Secondary Science group tests how effectively two filter materials remove particles from comparable water samples. One filter becomes blocked and no final filtrate volume is obtained. A blank in the resulting table cannot automatically mean zero millilitres of clean filtrate. Did the process yield no filtrate, did the apparatus stop functioning or was the measurement lost? Those scenarios describe different physical and evidentiary events. The group should preserve the event and the agreed outcome rule rather than delete the difficult filter.
The same reasoning applies to electrical circuit trials interrupted by a loose connection, survey quadrats missed because of unsafe weather and measurements taken with instruments that later fail a quality check. A useful transfer question asks whether the missing result could be related to the condition under study. The goal is for pupils to recognise missingness in unfamiliar contexts instead of memorising that a blank seedling height must not become zero.
Checklist for an honest investigation conclusion
Before presenting an average or graph, ask: Which units were included by design? Which produced valid measurements? Which produced meaningful alternative outcomes? Which were not measured, and why? Could the pattern of missing results be connected with the treatment, sampling or apparatus? Does the calculation’s denominator match the text explaining it? Has any estimate been labelled as an estimate? Those questions are short enough for a schoolwork review but powerful enough to prevent several common distortions.
A strong final sentence is not simply ‘More experiments are needed.’ It says what new evidence could repair the uncertainty. The next study might record survival as a separate outcome, use more durable labels or include additional independent pots under the same safe observation period. This makes the limitations productive: they lead to a better investigation rather than an invented answer.
Sources and connected Gold Standard reading
Singapore’s MOE Primary Science Teaching and Learning Syllabus emphasises gathering, recording, comparing and analysing observations and data. The advanced missing-data ideas here extend those habits for research literacy; they are not presented as a compulsory Primary-school statistics syllabus.
Read Random Assignment to Treatment Groups for allocation, Control Group Design for fair reference conditions, Negative Findings Interpretation for conclusions where an effect is not detected, and Observation Record Reconciliation for transcription and table errors.
Final principle: Science does not become more convincing when missing outcomes disappear from the spreadsheet. A trustworthy report shows both what happened and what could not be measured, and keeps its conclusions within the evidence.