How to perform in the new G2 SEC Science examination becomes more concrete when you can inspect a comparison and explain what it actually tests. Two results can differ because a treatment works, because the starting conditions differ, because the measurement method changes or because a small sample varies. The scientific job is to distinguish those possibilities before accepting the conclusion.
This G2 SEC Science controlled-comparison clinic provides original case files, questions, worked reasoning and fresh checks. All measurements are invented for paper-based learning. They are not results of experiments conducted by eduKate, official SEAB questions or a predicted examination paper. No home experiment is required. Use the cases to practise evaluating evidence, not to infer how marks would be allocated in an official paper.
The SEAB 2027 G2 directory lists Science combinations K223 Physics/Chemistry, K224 Physics/Biology and K225 Chemistry/Biology. Study the disciplines in your actual combination. The shared Science syllabus includes interpreting experimental data and evaluating methods. The investigation skills below transfer across combinations; a case using another discipline is optional reasoning practice, not a claim that you must study all three.
The question before the comparison
Before comparing two numbers, identify the claim. “Container A is warmer at the end” is different from “Container A cooled less”, which is different again from “Material A is the better insulator”. The first compares endpoints, the second compares changes, and the third attributes a difference to a material. Each statement asks more of the evidence.
Write four short notes before answering a difficult case: the proposed cause, the measured outcome, the other conditions and the conclusion requested. These are temporary learning supports. You do not need a large annotation system in the examination. The purpose is to see whether the comparison can support the exact claim, rather than a nearby claim that happens to sound scientific.
Vol 0036: Evidence Strength explains the difference between observation and conclusion. This volume lets you apply that distinction. Attempt each case’s questions before reading the review. Keep the original answer, because the difference between your first explanation and the corrected one identifies the next learning step.
Case one: a warmer endpoint does not identify the better material
Two students compare wraps around containers. Container A starts at 90°C and finishes at 70°C after ten minutes. Container B starts at 60°C and finishes at 45°C after ten minutes. The students use equal volumes of water and containers of the same shape. The room conditions are the same. They conclude that wrap A is better because its water is warmer at the end.
Question 1A: Which container has the warmer final temperature? Question 1B: Which has the larger temperature decrease? Question 1C: Do these readings establish which wrap better reduces cooling under comparable conditions? Question 1D: Propose the first change you would make to the comparison. Keep the questions separate; they do not all ask for the same answer.
For 1A, A is warmer at the end: 70°C rather than 45°C. For 1B, A also has the larger decrease: 20°C compared with B’s 15°C. Both statements can be true. A higher endpoint does not necessarily mean a smaller decrease, because the starting values differ. Subtract the starting and final readings before describing change.
The wrap conclusion is not established by this comparison. The containers began under different thermal conditions, so wrap and starting temperature were not separated. A fairer comparison would begin with matched water temperatures and retain comparable water volumes, containers, exposure and measurement intervals. Repeating the original unequal-start arrangement would produce more readings without removing the central ambiguity.
Do not repair the conclusion by declaring B the better wrap solely because its decrease is smaller. That still ignores the unequal starting conditions. The valid response is narrower: you can report the endpoints and decreases, but this design does not isolate the effect of the wrapping material. Recognising insufficient evidence does not require choosing the opposite conclusion.
Also preserve the quantity being measured. These readings give temperature changes, not direct numerical measurements of energy transferred. You should not write that A lost 20 joules or that B lost 15 units of heat. A calculation can be arithmetically correct and still describe the wrong physical quantity if its unit or label changes.
Fresh check: both containers now begin at 65°C. After eight minutes, A is at 56°C and B is at 50°C, with the other stated conditions matched. A has the smaller recorded temperature decrease, 9°C rather than 15°C. This is stronger evidence for a comparison under those conditions, but one trial still does not establish performance for every material thickness, container or starting temperature.
Case two: a faster change is not necessarily a larger final amount
A fictional reaction produces a measurable gas. Two trials use the same amounts of the same reactants, the same temperature and the same collection method. The intended difference is the size of the solid pieces. In Trial A the pieces are smaller. In Trial B they are larger. Treat the following as supplied data, not as instructions for carrying out a reaction.
At 0, 20, 40, 60 and 80 seconds, Trial A’s collected volumes are 0, 24, 40, 48 and 48 cm³. At the same times, Trial B’s volumes are 0, 12, 24, 36 and 44 cm³. At 100 seconds, B reaches 48 cm³ and subsequent readings remain at 48 cm³. Assume the collection apparatus is functioning as intended for this exercise.
Question 2A: Compare the average gas-collection rates during the first 20 seconds. Question 2B: Which trial reaches 24 cm³ sooner? Question 2C: Does the data show that A produces twice as much gas in total? Question 2D: Would a gas leak in only one apparatus be relevant when comparing final amounts?
During the first 20 seconds, A’s average rate is 24 ÷ 20 = 1.2 cm³/s and B’s is 12 ÷ 20 = 0.6 cm³/s. A’s average is twice B’s over that stated interval. The interval matters. Do not turn this result into a claim that A’s rate is twice B’s at every instant throughout the reaction.
A reaches 24 cm³ at 20 seconds; B reaches it at 40 seconds. This comparison uses the same outcome amount and compares time. It agrees with the early-rate evidence, but it is a different way of expressing it. In both forms, keep the common reference visible: either equal time and different volume, or equal volume and different time.
Both trials eventually record 48 cm³, so the data does not support twice the total amount for A. The early difference concerns how quickly the recorded volume increases. A learner who reads only the 20-second row may confuse progress at an intermediate time with the final outcome. Read far enough to answer the actual claim.
A leak could reduce the collected volume without necessarily reducing the amount produced by the reaction. That would make the measurement system part of the explanation for the difference. The appropriate repair would address gas collection, not automatically change the reactants. A result is evidence produced through a method, not an unfiltered view of the process.
Fresh check: two trials reach the same final volume, but one reaches half that volume earlier. State only the supported comparison. The earlier trial progresses faster to that amount under the stated conditions. Equal final amounts do not imply equal rates, and unequal early amounts do not imply unequal final amounts. Keep time and total outcome distinct.
Case three: the right improvement depends on the question
A school activity compares how quickly three insulating sleeves reduce cooling. In this fictional record, all containers begin at the same temperature. Only one reading is taken after ten minutes. The differences between sleeves are small, and the thermometer is read to the nearest whole degree. A student proposes taking twenty readings at the same instant with the same thermometer.
Question 3A: What additional information would repeated independent trials provide? Question 3B: Would repeated readings automatically give the instrument finer resolution? Question 3C: What change would help investigate the shape of cooling over time? Question 3D: Why is “collect more data” too vague as an improvement?
Repeated trials can reveal whether the observed differences recur and how much the results vary between runs. They should repeat the relevant procedure, not merely copy the same display many times. Re-reading an unchanged display does not create twenty independent cooling trials. Identify what is repeated before claiming what repetition improves.
Repeated readings do not automatically turn a whole-degree instrument into one that distinguishes tenths of a degree. A suitable instrument with finer resolution may help distinguish small differences, provided the rest of the measurement method is also adequate. Recording extra decimal places from a coarse display would invent precision rather than improve it.
To investigate the time pattern, take measurements at several defined times. More repeats at one time address a different question from more time points in one run. One describes variation at that condition; the other reveals how the outcome changes across the interval. A good evaluation names which missing information matters to the claim.
“Collect more data” could mean more trials, more temperature conditions, more materials, more time points or a wider sample. These are not interchangeable. Explain the weakness first, then choose the change that addresses it. A method can become more elaborate without becoming more informative if the additional work does not target the uncertainty.
Fresh check: a graph has measurements at 10°C, 30°C and 50°C, and you need to locate a peak suspected near 30°C. Repeating only at 10°C may improve knowledge of variation there but will not locate the peak. More closely spaced conditions around the suspected peak address that particular question. The improvement follows the information needed.
Case four: separate treatment from light
A fictional plant study investigates a treatment under two light conditions. Four groups use comparable plants, containers, soil and watering. The groups are assigned to four arrangements: untreated with lower light, treated with lower light, untreated with higher light, and treated with higher light. Each arrangement contains several plants. The measured outcome is mean height increase over the same period.
The reported mean increases are A, untreated/lower light: 4 cm; B, treated/lower light: 6 cm; C, untreated/higher light: 7 cm; and D, treated/higher light: 9 cm. This is an illustrative data set. No particular real treatment is being recommended, and the values are not evidence about a commercial product.
Question 4A: Which comparisons hold light condition fixed while changing treatment? Question 4B: Which hold treatment status fixed while changing light? Question 4C: Why is D compared with A not the cleanest estimate of treatment difference? Question 4D: What useful information is missing if only group means are reported?
A versus B compares treatment within lower light: the mean difference is 2 cm. C versus D compares treatment within higher light: the mean difference is also 2 cm. In both comparisons, the named light condition is held fixed. The purpose is not to select the largest numerical gap but to match the comparison to the cause being investigated.
A versus C compares light within untreated plants, giving 3 cm. B versus D compares light within treated plants, also giving 3 cm. These comparisons answer a different question. The same four means can support several descriptions, but each description must state which factor differs and which factor is matched.
D minus A gives 5 cm, but both treatment and light change together. It is incorrect to attribute the whole 5 cm difference only to treatment. Comparing extremes is not automatically the best test. In this case, the smaller matched comparisons provide a clearer view of the treatment difference than the largest observed contrast.
Means alone do not show the number of plants, spread of individual results, losses during the study or consistency across repeats. Without that information, avoid declaring the treatment effective for all plants under all conditions. The recorded means show the stated pattern in this example; the strength of a broader conclusion requires more evidence.
Fresh check: suppose B is 4 cm rather than 6 cm, while D remains 9 cm. The treatment difference is now 0 cm under lower light and 2 cm under higher light. Do not average away the contrast immediately. State that the observed treatment difference depends on the light condition in this record, then ask what further evidence would test whether that pattern is reproducible.
Case five: a constant offset can affect a value differently from a change
A fictional balance is known to add exactly 3 g to every reading over the range used. A container gives a reading of 53 g before material is added and 78 g afterwards. The same balance is used both times, and the offset is assumed constant. A student says that every calculated result must therefore be 3 g too high.
Question 5A: What are the corrected before and after masses? Question 5B: What mass was added? Question 5C: Does the constant offset affect the difference between the two readings? Question 5D: Would the same conclusion hold if the second reading came from a different balance with no offset?
The corrected values are 50 g and 75 g. Their difference is 25 g. The uncorrected difference is also 78 − 53 = 25 g. The same added offset cancels in this subtraction: (75 + 3) − (50 + 3) = 25. The individual readings are shifted, but this particular difference is unchanged under the stated assumptions.
This does not mean a faulty instrument is always acceptable or that all systematic errors cancel. Cancellation here depends on an equal additive offset affecting both readings. If the offset changes, if the instruments differ, or if the error scales readings rather than adding a constant, the result can behave differently. Use the actual error model given.
If the first reading is 53 g on the offset balance and the final reading is 75 g on a correct balance, subtracting the raw readings gives 22 g, which is wrong. The common-offset condition has disappeared. Correct the first reading to 50 g before comparing it with the correctly measured 75 g.
Averaging repeated raw readings would not remove the individual +3 g offset. Difference calculation and averaging are different operations. The useful lesson is not a slogan about all bias. It is to inspect how a specified measurement error enters the exact quantity being calculated. Sometimes it persists; in a particular difference, a shared additive component can cancel.
Fresh check: a temperature display adds a constant 2°C. Its readings rise from 22°C to 32°C. The corrected temperatures are 20°C and 30°C, but the increase is 10°C either way. Explain both facts. Saying the increase is 8°C would subtract the offset once from a difference where it had already cancelled.
Case six: equal means can hide different repeat patterns
Two timing methods are tested on the same repeatable event. Method A gives 18, 20 and 22 seconds. Method B gives 10, 20 and 30 seconds. For this exercise, a separate trusted reference establishes that the event takes 20 seconds. The values are invented to isolate the reasoning; three readings are not being proposed as a universal adequate sample.
Question 6A: Calculate each mean. Question 6B: Compare the spread of the recorded readings. Question 6C: Does the equal mean make the methods equally consistent in this sample? Question 6D: What changes if the trusted reference is not supplied? Distinguish closeness among repeated readings from closeness to a reference value.
Both means are 20 seconds. A’s range is 22 − 18 = 4 seconds, whereas B’s is 30 − 10 = 20 seconds. A’s readings are more closely grouped in this small record. An average compresses information; it does not preserve the full pattern of variation. Reporting only the two means would conceal that difference.
The means both match the supplied reference, but the individual readings differ in consistency. It would be misleading to say the methods produced identical evidence. A user who needs a dependable individual reading may care about variation even when the average happens to land on the reference value.
Without a trusted reference or another justified accuracy check, the sample still allows a comparison of spread. It would not establish which mean is closer to the true duration. Precision-like consistency and reference-based accuracy answer different questions. Do not invent a true value simply because two means agree with each other.
Nor should three measurements settle a permanent ranking of the methods. Further comparable trials could show whether the pattern persists and whether another feature of the procedure matters. The bounded conclusion is about the supplied record: A’s readings are less spread out, and both sample means equal the stated reference.
Fresh check: Method C gives 24, 24 and 24 seconds for the same 20-second reference event. It is perfectly consistent in those readings but displaced from the reference. More repeated 24-second readings would not by themselves repair that displacement. Identify the measurement procedure or calibration issue rather than praising consistency as if it guaranteed correctness.
Case seven: a changed procedure changes the comparison
A fictional test compares two filters using cloudy water. Filter A receives 100 ml and runs for two minutes. Filter B receives 200 ml and runs for five minutes. The student observes that the water collected from B looks clearer and concludes that B is the better filter. No numerical clarity measurement or repeated trial is supplied.
Question 7A: Name two procedure differences besides filter type. Question 7B: Explain why the conclusion is difficult to attribute only to the filter. Question 7C: Propose a comparison with a clearer outcome measure. Question 7D: Why might equal time and equal processed volume define different tests? This is paper-based method evaluation, not advice to make water safe to drink.
The amount of water and the duration differ. The visual judgement is also loosely defined. A better design would specify the claim: clarity after a fixed treatment time, clarity after processing a fixed amount, or throughput at a required clarity standard. You cannot choose meaningful controls until you know which performance question the experiment is meant to answer.
For clarity after a fixed volume has passed, use comparable starting water and equal input volumes, then measure the resulting clarity using a defined method. Record how long each filter takes rather than forcing time to be equal. For performance within a fixed time, keep the time fixed and record both processed amount and resulting clarity. These are different, legitimate comparisons.
The instruction to “change one variable” should not be applied without thinking about the outcome. Some quantities are controlled, some are deliberately changed and some are measured consequences. If filter design changes flow rate, forcing every resulting quantity to be identical could remove the performance difference you intended to investigate.
The student also needs a consistent starting mixture and a clearer measurement procedure. “Looks clearer” might be enough for a rough observation, but a repeatable rating method or suitable instrument would make comparison more specific. Do not claim that an instrument automatically solves all design weaknesses; unequal starting material would remain a problem.
Fresh check: two drying materials are compared. One absorbs more liquid after one minute, but the other eventually holds more liquid in total. Which is better? The answer depends on whether the task prioritises rapid uptake or maximum capacity. Define the intended use before turning one measured advantage into an overall ranking.
The common reasoning across the seven cases
Each case asks you to keep three things aligned: the claim, the comparison and the measurement. Case one separates endpoint from change. Case two separates rate from total amount. Case three separates repetition from resolution and coverage. Case four matches groups to a factor. Case five follows an error through a calculation. Case six checks what an average hides. Case seven defines performance before ranking designs.
These are not seven isolated tricks. They are examples of a reusable question: what else could make these results differ, and does the method allow me to separate that possibility from the proposed explanation? Sometimes the answer is another variable. Sometimes it is a measurement offset, a different interval, an unmatched starting value or a vague definition of better.
Use the structured-question process in Vol 0032 when a case feels dense. Name the task, select the evidence, identify the scientific operation and choose the answer form. The data may be long, but the particular subpart often asks for one precise comparison rather than a full account of the investigation.
Choose an improvement that resolves the named uncertainty
There is no universal improvement sentence. Repetition can examine consistency, but it does not automatically remove unequal starting conditions. A finer scale can distinguish smaller differences, but it does not make different treatment groups comparable. More time points can reveal a curve, but they do not automatically correct a constant offset.
Write the limitation and consequence before the improvement. For example: “The starting temperatures differ, so the wrap comparison also changes the initial thermal condition. Begin both trials at the same temperature.” This is stronger than “make it fair”. The sentence shows what is unfair and how the proposed change addresses it.
For a resolution problem, the logic is different: “The display changes only in whole degrees, so small differences between sleeves may not be distinguishable. Use a suitable finer-resolution measurement method.” The improvement is useful because it matches the limitation. Adding a second control variable would not answer that specific measurement issue.
Vol 0020: Experimental Questions provides the broader vocabulary for variables and measurement quality. Return there when you cannot explain the term itself. Use this clinic when you know the terms but need practice deciding which one matters in a particular question.
An independent review record
For each case, record the earliest mistake rather than only the final wrong sentence. Did you compare endpoints when the task asked for changes? Did you treat a mean as the whole data set? Did you average away a condition? Did you suggest repetition for a fixed bias? The first incorrect move usually gives a more useful repair target than the last symptom.
A learner who calculates every change correctly but overstates the conclusion needs evidence-boundary practice. A learner who identifies a control problem but cannot calculate the difference needs numerical support. A learner who knows both but answers describe when asked to explain needs command-word practice. Do not respond to all three with another full chapter of notes.
Keep the conditions of practice visible. An answer produced after a hint is guided work; a fresh answer produced later without the hint is stronger evidence of independent selection. Neither should be discarded. They show different stages of learning. Use the distinction to plan the next task rather than to label the learner.
A short mixed retest
Retest A: A container starts at 80°C and ends at 62°C; another starts at 65°C and ends at 51°C. Which has the smaller temperature decrease, and does that establish better insulation? The decreases are 18°C and 14°C, but unequal starting conditions still prevent a clean insulation comparison. Give both the arithmetic answer and its limit.
Retest B: Two processes both reach a final measured amount of 30 units. One reaches 15 units after ten seconds; the other reaches it after twenty seconds. The first reaches that intermediate amount sooner. The record does not establish a larger final amount. Do not add a mechanism unless the task provides enough subject information or asks for one.
Retest C: A ruler has a constant positive offset at both recorded endpoints of a distance measurement. Does subtracting the two readings remove that shared additive offset? Under the stated model, yes. Would dividing the readings necessarily remove it? No. The operation matters. You should be able to show the subtraction algebra rather than repeat a blanket statement about faulty instruments.
Retest D: A treatment group and comparison group have the same mean outcome, but one group contains much more variable results. Can you report identical evidence? No. The means match, but the variation differs. Whether that difference matters to a particular decision depends on the claim and the use of the measurement. Keep the full record in view.
Use the right level of certainty
Do not write “nothing can be concluded” whenever a design has a limitation. In Case one, you can still report the endpoint and temperature decrease. In Case four, you can still describe the supplied group means. A limitation restricts which explanation or generalisation is justified; it does not erase every observation.
The opposite error is treating a clear observed difference as a universal cause. Separate the statements: what happened in the record, what the comparison was designed to test, and how far the conclusion can travel. A precise answer can be confident about the first and cautious about the third without being inconsistent.
When the command asks for an explanation, use the relevant syllabus concept rather than only criticising the method. When it asks for evaluation, discuss why the evidence does or does not support the claim. The official Science syllabus includes both knowledge and investigation skills; the question determines which operation you must perform.
Return to the next learning priority
Use the Complete Science Index to repair missing concepts in your actual Science combination. Use Vol 0028: Combined-Science Mode Switching when you recognise a data structure but select the wrong discipline’s explanation. The shared comparison habit does not remove the need for accurate Physics, Chemistry or Biology knowledge.
The PSLE bridge remains Evidence Before Explanation. At G2, add control over the comparison itself: match the starting conditions, specify the outcome, inspect the measurement and keep the claim within what the record supports. For pacing and final checking, continue through Examination Craft.
A strong scientific comparison does not begin by choosing the larger number. It begins by deciding which numbers answer the same question. Identify what changed, what was matched and what was measured. Then calculate, explain or evaluate as requested. The result is an answer that can show not only a difference, but why that difference is relevant to the claim.
