Two groups of seedlings are assigned at random to a compost treatment and a standard growing medium. The treatment group’s final plants are taller, so a pupil declares that compost caused the difference. However, the treatment plants were already taller before the experiment began. The gold standard of school scientific baseline balance assessment is to check what was different between groups before the treatment, distinguish starting differences from changes during the experiment, and report a conclusion that the design and measurements actually support.
Quick answer: Record meaningful pre-treatment characteristics before outcomes are known; compare group starting conditions and the allocation method; distinguish random imbalance from deliberate selection bias; calculate both baseline and outcome or change measures where scientifically suitable; look for material confounders; preserve original assignments; and avoid declaring a treatment effect from final group means alone. Baseline balance improves interpretation but does not prove that all unmeasured characteristics are identical.
Contents: from group starting conditions to a defensible conclusion
- Why final results are not enough
- What a baseline measure is
- Random assignment and imbalance
- Worked twelve-seedling comparison
- Baseline, final height and height gain calculations
- How the scientific conclusion changes
- Other pre-existing differences
- When to repair the design
- Baseline timing and measurement validity
- Missing and unusable baseline values
- Three learner pathways
- A 90-minute lesson
- Break-tests
- Frequently asked questions
- Sources and connected articles
1. The treatment effect is not simply the final height difference
If one seedling finishes at 12 centimetres and another at 10 centimetres, the difference is two centimetres at that time. It is not necessarily two centimetres of extra growth caused by the treatment. One plant might have begun the study two centimetres taller. The first scientific step is to identify the starting condition and the precise change that was intended to be measured.
This distinction is especially important in school projects where plants are selected after germination. Seedlings may already differ in height, leaf count and development. If a teacher deliberately gives the largest plants to one treatment, the groups have been selected in a way that can make the later comparison unfair. Even a genuinely random allocation can produce groups that are not perfectly matched in a small sample. Students need to tell those causes apart before explaining the graph.
A baseline comparison is a check on the starting state, not permission to erase inconvenient numbers. The observed data should be preserved. If the initial design was poor, the report must state the limitation or use a properly designed future investigation. Relabelling groups after seeing the result would damage the very evidence the baseline measurements were intended to protect.
2. A baseline must be measured before the factor acts
A baseline is a pre-treatment measurement relevant to the intended outcome or potential competing explanation. For plant growth it might be starting height or developmental stage. For a cooling demonstration it might be starting water temperature. For a rainfall observation it might be the initial amount accumulated in the gauge before the chosen measurement period. The relevant baseline depends on the physical question.
Timing is essential. Measuring the ‘initial’ height after the compost treatment has already been applied for a week does not automatically provide a true pre-treatment baseline. Some of the treatment’s influence may already have occurred. Similarly, if one group is measured at 09:00 and the other at 14:00 under changing conditions, the difference may include time-related effects. Record when and how the baseline was actually observed.
Not everything measurable belongs in the baseline table. Identify variables with plausible relevance to outcomes or fairness. In a plant study, initial height, plant variety and pot conditions may matter. The colour of the student’s notebook usually does not. More columns cannot compensate for poor understanding of what the investigation is testing.
An ordinary-language definition
Students can learn: The baseline is where each experimental unit stood before the chosen change could influence the outcome. A strong pupil can state both why a baseline matters and which particular difference it helps reveal. That reasoning is worth more than copying ‘initial reading’ into every worksheet.
3. Random allocation does not guarantee perfect starting groups
Random assignment reduces the chance that a teacher deliberately allocates healthier seedlings to the favoured treatment. But chance can still place more tall plants in one group, especially when there are only six plants in each. Small-sample imbalance is therefore not automatic proof that the randomisation was fraudulent or invalid. It is relevant information about the observed comparison.
Students should distinguish selection bias, in which someone chooses group membership according to relevant pre-existing traits, from chance imbalance, which can remain after an honest random allocation. The first may compromise causal comparison systematically; the second creates uncertainty and calls for careful interpretation. The proper response is not to regenerate random assignments repeatedly until an attractive arrangement appears after inspecting the outcomes.
If strong baseline differences are known in advance, a more careful planned allocation—for example, randomly allocating within starting-height categories—may be useful. Such a design must be specified and executed transparently before treatment and outcomes. A post-hoc adjustment that moves weak plants between groups after their results appear is not a fair replacement for planning.
The related article on Scientific Random Assignment to Treatment Groups addresses allocation methods. The present topic concerns the evidence students should inspect after assignment but before interpreting the outcomes, and why a randomised label alone is not a sufficient conclusion.
4. Worked example: twelve seedlings, unequal starting heights
Consider a fictional experiment with twelve independently prepared pots. Six are assigned the compost treatment and six the base-medium comparison under a documented chance-based method. Each plant’s height is recorded before the treatment and at the end of a fourteen-day interval. The following values are invented instructional data, not observations from a real Singapore school or a claim about the effect of compost.
| Pot | Assigned group | Initial height | Height gain over 14 days | Final height |
|---|---|---|---|---|
| C1 | Control | 6.0 cm | 2.8 cm | 8.8 cm |
| C2 | Control | 6.5 cm | 3.1 cm | 9.6 cm |
| C3 | Control | 6.2 cm | 2.9 cm | 9.1 cm |
| C4 | Control | 6.8 cm | 3.2 cm | 10.0 cm |
| C5 | Control | 6.3 cm | 3.0 cm | 9.3 cm |
| C6 | Control | 6.2 cm | 3.0 cm | 9.2 cm |
| T1 | Compost | 7.2 cm | 3.0 cm | 10.2 cm |
| T2 | Compost | 6.9 cm | 2.9 cm | 9.8 cm |
| T3 | Compost | 7.4 cm | 3.2 cm | 10.6 cm |
| T4 | Compost | 6.8 cm | 3.1 cm | 9.9 cm |
| T5 | Compost | 7.1 cm | 3.0 cm | 10.1 cm |
| T6 | Compost | 7.0 cm | 3.1 cm | 10.1 cm |
All heights and gains are displayed in centimetres to one decimal place. The six control plants began at 6.0, 6.5, 6.2, 6.8, 6.3 and 6.2 centimetres. The six treated plants began at 7.2, 6.9, 7.4, 6.8, 7.1 and 7.0 centimetres. The treated group was somewhat taller at baseline, even though the allocation procedure was described as random. We should not ignore that simply because the treatment group also finishes taller.
The table enables a clear comparison of initial state, change and final state. We should preserve all three, rather than pick only the quantity that makes the experimental treatment look effective. A teacher can ask students to predict which comparison will be most relevant to the question about growth during the treatment period.
5. Worked calculations: the apparent advantage largely predates treatment
The control group’s initial heights total 38.0 cm, giving a mean starting height of 38.0 ÷ 6 ≈ 6.33 cm. The compost group’s initial heights total 42.4 cm, giving a mean of 42.4 ÷ 6 ≈ 7.07 cm. The mean baseline difference is therefore about 0.73 cm in favour of the treatment group, before the measured growth period.
The control plants’ recorded height gains are 2.8, 3.1, 2.9, 3.2, 3.0 and 3.0 centimetres, totalling 18.0 centimetres. Their mean gain is 18.0 ÷ 6 = 3.00 cm. The treatment gains total 3.0 + 2.9 + 3.2 + 3.1 + 3.0 + 3.1 = 18.3 centimetres. Their mean gain is 18.3 ÷ 6 = 3.05 cm.
The difference in mean gain during the fortnight is thus only 0.05 cm. Meanwhile, mean final heights are approximately 9.33 cm for the control group and 10.12 cm for the compost group, a final difference of about 0.78 cm. That 0.78 cm comprises roughly 0.73 cm of baseline imbalance and 0.05 cm of observed difference in mean gain in the example.
This decomposition is an arithmetic description, not a formal causal-effect estimate with established statistical confidence. It demonstrates why final height alone can create a misleading narrative. In a small class experiment with natural variation and finite ruler resolution, a 0.05 cm difference in mean gain may not be practically distinguishable from measurement uncertainty and ordinary variability.
6. What a defensible conclusion now sounds like
A weak conclusion says, ‘Compost made plants 0.78 centimetres taller.’ The final difference is real within the fictional table, but most of it existed before treatment. A stronger conclusion is: ‘The compost group finished about 0.78 cm taller on average, but it began about 0.73 cm taller. Mean recorded growth over fourteen days differed by only about 0.05 cm. This small study did not provide clear evidence of a substantial additional height gain from compost under these conditions.’
That statement is confident where the data are clear and cautious where they are not. It does not assert that the compost had no effect anywhere. It also does not identify a nutrient mechanism that the study never measured. A different plant species, growing environment, duration or experiment might yield a different result. The task is to explain what this test showed, not predict universal agriculture.
An ordinary mean-change comparison is useful for teaching, but it does not resolve every design weakness. If the plant groups differ in maturity, soil moisture or health, their natural growth rates could also differ. The measured change may be affected by those factors, and a careful study would evaluate them according to a justified method. Advanced statistical adjustments can have assumptions and should not be presented as a magic repair for a poor classroom design.
7. Baseline differences can hide in other conditions
Starting height is one relevant factor, not the only one. Seedlings may differ in species, leaf count, root development and previous light or water exposure. Pots may have different volumes, drainage or positions. Some differences can be controlled by choosing comparable materials; others may be reduced through appropriate chance-based assignment, while remaining limitations should be recorded.
A teacher who assigns all treated pots beside a sunny window creates a possible confounder after allocation. Even if initial heights were identical, future growth could differ because light exposure differed. Baseline balance and follow-up treatment control therefore address different parts of the design. Students should not conclude that one balanced starting measurement makes the entire test fair.
A school cooling experiment offers a parallel. If one cup begins at 80°C and another at 70°C, comparing their final temperatures is not enough to isolate cooling behaviour. Students may compare appropriately defined temperature drops and control relevant cup and environmental properties. The particular physical system can matter; a numerical ‘change’ comparison must still follow a scientifically meaningful model.
There is no value in listing hundreds of implausible confounders. The learner should name a specific relevant factor and how it might influence the outcome. ‘One group started taller’ is meaningful. ‘Maybe something went wrong’ does not explain the scientific problem.
8. What to do when groups are badly imbalanced
If a large starting difference is discovered before treatment begins, the teacher can consider whether the design needs a properly documented change consistent with the intended assignment method. It may be sensible to recruit more suitable independent units, improve eligibility criteria or use a preplanned blocked randomisation in a new study. The process should be transparent and should not choose an arrangement because it promises the desired result.
If the treatment has already begun, silently reassigning plants can destroy the evidence. Preserve the actual original allocation and measurements. A student may still analyse the recorded gains descriptively and explain the pre-treatment imbalance. A new carefully planned experiment may provide stronger evidence, but it does not rewrite what the earlier one observed.
Do not automatically delete the tallest or smallest plant solely to make group means match. Exclusion requires a scientific reason defined or justified under the investigation’s rules, not cosmetic symmetry. An unusual but valid baseline may show genuine natural variation and be part of the evidence.
Where the same treatment is assigned to a whole tray of seedlings, recognise that the tray is the independent treatment unit, not every individual plant. No amount of baseline balancing among leaves can create independent replication that was not present in the treatment design.
9. Measure baseline the same way for both groups
A baseline table can contain measurement bias. Suppose one student measures treated plants from the pot rim while another measures control plants from the soil surface. The resulting starting-height difference could partly reflect different definitions, not real plant development. Establish the same reference point, instrument, units and timing for all relevant groups.
Record if the instrument or procedure changes. A ruler’s resolution, the angle used to view the scale and the definition of the plant’s highest point can affect readings. Baseline checks should reveal whether the initial comparison is meaningful, not simply whether a number is present in every cell. An observation that fails the agreed protocol needs appropriate annotation or review.
If a baseline is taken after the treatment already influenced the plant, that reading may not be a true pre-treatment value. Label it accurately rather than describing it as ‘before treatment’. The history matters because the inference depends on when each measurement was made. A date alone may be insufficient if environmental conditions change quickly.
10. A missing baseline is not automatically zero
Suppose pot T6 has a valid final height but its starting label is lost. The student cannot compute its individual height gain reliably from the original source. Replacing the baseline with zero would produce an absurd gain. Replacing it with the group mean would create an estimated value, not an observation. The responsible response is to check reliable original records and mark the baseline unavailable if it cannot be recovered.
Missing baselines can also undermine a comparison if loss is associated with the group or starting condition. The class should preserve which units were originally assigned and which measurements were recorded. A complete-case calculation may still be descriptive of the measured subset, but its sample and limitations must be explicit.
The related Missing Outcome Data Analysis article concentrates on unavailable follow-up outcomes. The same integrity principles apply here, while the scientific problem is different: without the starting state, we may not know whether a final result represents growth during treatment or a pre-existing difference.
11. Repair, stabilise and extend: three learning pathways
Repair: the student mistakes ‘taller at the end’ for ‘grew more’
Show two pictures of plants at day 0 and day 14. One starts taller but grows the same amount as the other. Ask which grew more, not merely which finishes taller. Use a simple number line or stacked height segments to make baseline and change visible. Let the learner explain orally before introducing a data table.
Once the concept is secure, add a treatment and control label. The learner should be able to say, ‘We must compare how much the plants changed, because they did not start the same.’ Avoid teaching a generic formula without the physical reason for using it.
Stabilise: the arithmetic is correct but the interpretation is too strong
Give students the twelve-pot table and ask for the mean starting heights, mean final heights and mean gains. Discuss why the final difference of roughly 0.78 cm differs so much from the gain difference of roughly 0.05 cm. Require a short scientific conclusion that names the actual metric and time.
If a student writes ‘random means all starting conditions are equal’, introduce a small set of randomly drawn cards showing uneven groups. The point is to clarify what chance allocation does and does not guarantee, not to memorise a phrase about fairness.
Extend: assess relevance, imbalance and design trade-offs
More advanced learners can identify baseline variables with plausible causal significance, explain how chance imbalance may occur in small samples and propose a preplanned matched or blocked allocation for a future investigation. They should recognise the difference between describing baseline imbalance and proving that a specific confounder caused the result.
Challenge pupils to decide whether a mean gain is always the right analysis. For some physical processes the relevant outcome may be rate, proportional change, survival or another measure, and the scientific model controls what the calculation means. Mathematical sophistication should support the research question, not replace it.
12. A practical 90-minute lesson
| Time | Learning move | Evidence of understanding |
|---|---|---|
| 0–10 minutes | Compare pictured start and end states. | Pupil distinguishes final height from height gained. |
| 10–25 minutes | Identify relevant baseline conditions. | Learner names when and why they should be measured. |
| 25–45 minutes | Calculate group initial, final and gain means. | Correct arithmetic with units and denominators. |
| 45–60 minutes | Explain the 0.73 cm starting gap and 0.05 cm gain gap. | Learner avoids falsely attributing final difference to treatment. |
| 60–75 minutes | Evaluate random allocation and other confounders. | Recognises chance imbalance and nonrandom bias. |
| 75–85 minutes | Write an evidence-based conclusion. | States the result with appropriate limits. |
| 85–90 minutes | Transfer to a cooling investigation. | Applies baseline reasoning in a new topic. |
The teaching plan is illustrative and not an MOE-prescribed sequence. The fictional numbers should remain identified as fictional. Any real experiment requires appropriate school approval, safe procedures and honest recording. Students can learn the essential concepts from prepared classroom data when growing plants is impractical.
13. Break-tests that reveal a hidden starting advantage
- The compost group is taller at day 14 but was equally taller at day 0. Does that establish a treatment effect?
- The original allocation was genuinely random, but groups have different mean starting heights. Is the experiment necessarily fraudulent?
- The teacher assigns all tallest seedlings to compost before treatment. Can pupils identify deliberate allocation bias?
- One group is measured from soil level and the other from the pot rim. Are their baselines comparable?
- A pot has no initial measurement. Is it scientifically valid to enter zero?
- One treatment is applied to a shared tray of six plants. How many independent treatment assignments exist?
- The group gain means differ by 0.05 cm with substantial reading uncertainty. Can the pupil avoid false precision?
- A student removes the tallest plant solely to make the baseline means equal. Is that a justified data correction?
A robust answer explains what the evidence does and does not settle. Baseline assessment is not about forcing every comparison to look identical. It is about discovering whether the reported treatment contrast might be influenced by where each group began.
14. Frequently asked questions
What is a baseline in a school Science experiment?
A relevant measurement taken before the tested factor could influence the result, such as starting plant height or starting water temperature.
Is final height difference the same as difference in growth?
No. Final heights can differ because the plants began at different heights. Compare an appropriate change measure when the question concerns growth during a period.
Does random assignment guarantee identical baselines?
No. Chance imbalance can remain, especially in small groups. Random assignment reduces deliberate selection but does not make all characteristics equal.
Should groups be re-randomised until their means match?
Do not repeatedly choose an allocation because it creates the preferred observed result. An appropriately preplanned balanced assignment may be useful before treatment begins.
Does a matching baseline guarantee a fair test?
No. Other factors can differ during treatment, and unmeasured characteristics may remain imbalanced.
Can I subtract starting height from final height for every growth study?
Height gain is often useful, but the scientific question and valid measurement method determine the appropriate outcome. Growth can also involve survival, biomass or rates.
What if a baseline measurement is missing?
Check trustworthy original records. If the value cannot be recovered, mark it missing and explain which comparisons are limited. Never invent it as zero.
What should a strong conclusion say?
Report the actual baseline and outcome differences, distinguish final values from changes and avoid treating a small sample as universal proof of causation.
15. Sources and connected reading
Singapore’s MOE syllabus information and the Primary Science Teaching and Learning Syllabus provide the school inquiry context. This article extends fair-comparison principles into baseline-balance reasoning; it is not a claim that every Primary pupil must study formal clinical-trial methods or statistical adjustment.
Related eduKate Gold Standard articles answer different steps: Random Assignment to Treatment Groups concerns allocation; Control Group Design concerns choosing a relevant reference; Confounding Variables concerns alternative explanations; and Missing Outcome Data Analysis concerns unavailable results. Baseline balance specifically checks the state of the groups before treatment can explain their later differences.
Takeaway: An experimental group can look better at the end because it was ahead at the beginning. Good Science makes the starting line visible, calculates the right outcome and allows evidence—not appearances—to determine what treatment may have changed.