G3 SEC English, Mathematics and Science | Learner’s Guide Vol 0160. A school tests whether an earlier shuttle-arrival alert improves punctuality. The reported figures seem to say that the alert makes results worse. But the morning and afternoon figures each point the other way. The difference is not a calculation mistake: the two groups contain different proportions of easier and harder journeys. This original EMS case shows how to find the correct denominator, explain the apparent contradiction and recommend a better test.
This is a fictional learning case with invented route observations and an invented alert system. No real Singapore transport operator, school or travel timetable is being described. It is also not a claim that students sit one official combined EMS paper. The aim is to transfer English source reading, G3 Mathematics percentages and Science experimental reasoning into a defensible decision.
1. Keep the official 2027 examinations straight
The SEAB G3 SEC 2027 school-candidate directory, last updated 26 October 2026, identifies English Language as K300 and Mathematics as K310. It also lists Science (Physics, Chemistry) K326, Science (Physics, Biology) K327 and Science (Chemistry, Biology) K328. Students sit the relevant individual examinations, not a new combined English–Mathematics–Science paper.
The K300 English syllabus includes comprehension, inference, synthesis and audience-aware written communication. The K310 Mathematics syllabus includes statistics, probability, applications and interpretation of data in real-world contexts. The G3 Science syllabus requires students to interpret observations, recognise experimental limitations and evaluate conclusions.
English Paper 1 Writing is 1 hour 50 minutes, 70 marks, weighted 35 per cent; Mathematics has two 2-hour-15-minute papers of 90 marks each, weighted 50 per cent each. The combined Science syllabus has Paper 1, the relevant two theory papers and Paper 5 practical, each with its published timings and weightings. This workshop develops shared skills; its fictional scenario and mathematical extension are not official question forecasts.
2. The fictional school decision
A school has an early-morning shuttle and an afternoon shuttle. A staff member proposes sending an arrival reminder to some passengers before each journey. The committee wants to know whether the alert is worth introducing more widely. An early briefing states: “Punctuality was worse with the reminder, so the alert should be abandoned.”
The claim sounds decisive because it refers to a measurable result. Yet the briefing does not explain that morning trips tend to be more punctual than afternoon trips. It also does not say how many journeys with and without alerts occurred in each time period. Both omissions matter because the denominators are different.
The committee asks three questions. English: exactly what does “worse” refer to, and how strong is the evidence? Mathematics: were the rates calculated using the correct numbers? Science: does this observational comparison establish that the alert caused a change? A good decision needs all three answers.
3. Define on-time before examining a percentage
In this fictional packet, a journey is counted as on time if it arrives no later than its scheduled arrival time. Any positive lateness, even one minute, counts as late. This is a deliberately strict teaching definition, not a claim about a real operator’s service standard. The same definition applies to every row of the table.
Without that definition, a statement such as “80 per cent on time” is incomplete. Some organisations may use a grace window, but this example does not. A student who silently changes the threshold midway through a comparison is no longer comparing like with like.
Write the outcome definition at the top of the calculation. If a later question changes it to “within five minutes”, the results cannot simply be reused because the raw arrival times would be needed to reclassify journeys. This is how English interpretation and Science measurement meet quantitative reasoning.
4. The original four-row data table
Morning journeys with alert: 10 journeys observed, 9 on time, 1 late. Morning journeys without alert: 30 journeys observed, 24 on time, 6 late. Afternoon journeys with alert: 10 journeys observed, 4 on time, 6 late. Afternoon journeys without alert: 10 journeys observed, 3 on time, 7 late. All figures are invented for this learning case.
The rows are not equally large. The morning-without-alert group contains 30 journeys while each other row contains 10. If a student gives every row the same weight without noticing the counts, the resulting overall rate may describe a different comparison from the one originally recorded.
The data show counts, not individual passengers’ reasons for being late. They do not tell us which trips had heavy traffic, rain, equipment difficulties or other influences. Those missing variables will matter later when evaluating possible causes. First, though, calculate what the four rows actually say.
5. Percentages need the correct denominator
For any row, on-time percentage = on-time journeys ÷ observed journeys × 100. Morning with alert: 9/10 × 100 = 90 per cent. Morning without alert: 24/30 × 100 = 80 per cent. The difference is 10 percentage points in favour of the alert within the morning group.
For the afternoon group, with alert: 4/10 × 100 = 40 per cent. Without alert: 3/10 × 100 = 30 per cent. Again the recorded difference is 10 percentage points in favour of the alert within that time group.
Do not divide morning on-time counts by the total of all journeys in the packet, because that answers a different probability question. The word within matters. Each group rate describes the successes among journeys belonging to that row, with that row’s count as its denominator.
6. Now combine the alert group correctly
With alerts, the data contain 10 morning journeys and 10 afternoon journeys, for 20 journeys. Of those, 9 morning journeys and 4 afternoon journeys were on time, for 13 on-time journeys. The combined alert on-time rate is 13/20 × 100 = 65 per cent.
Without alerts, the data contain 30 morning journeys and 10 afternoon journeys, for 40 journeys. Of those, 24 morning and 3 afternoon were on time, for 27 on-time journeys. The combined no-alert rate is 27/40 × 100 = 67.5 per cent.
The overall rates therefore look worse with the alert: 65 versus 67.5 per cent, a gap of 2.5 percentage points. Yet each time group separately showed a 10-percentage-point advantage with the alert. No arithmetic contradiction has occurred. The groups being combined have different compositions.
7. Why the combined conclusion reverses
Morning journeys have higher punctuality under both alert conditions: 90 or 80 per cent, compared with 40 or 30 per cent in the afternoon. The no-alert sample contains 75 per cent morning journeys (30 out of 40), while the alert sample contains only 50 per cent morning journeys (10 out of 20).
The no-alert combined average therefore receives more weight from the easier, higher-punctuality morning group. The alert combined average receives a larger relative contribution from the harder afternoon group. The aggregate comparison partly reflects time-of-day composition, rather than only the potential relationship between alerts and punctuality.
This is an accessible example of a phenomenon sometimes called a reversal of aggregate and subgroup trends. Students do not need to memorise that name to reason well. They need to ask whether one group was more heavily represented on one side, and whether the success rates differ greatly across those groups.
8. Reconstruct the combined rates as weighted averages
The combined alert rate can be written as 0.5 × 90 per cent + 0.5 × 40 per cent = 65 per cent, because alert observations are evenly split between morning and afternoon. The combined no-alert rate is 0.75 × 80 per cent + 0.25 × 30 per cent = 67.5 per cent.
Those weights are fractions of the observed sample. They are not arbitrary importance scores. The weighted average is simply another way to express the original count calculation. It makes the composition difference visible without introducing a new equation.
If you average 90 and 40 and compare it with the simple average of 80 and 30, you obtain 65 versus 55 per cent. That is a useful standardised comparison using equal morning–afternoon weights, but it is not the actual unadjusted no-alert rate. Label clearly whether your result is observed or standardised.
9. Compare both alert conditions on a common mix
Suppose a fair reporting comparison assigns equal importance to morning and afternoon. Under those common weights, alert performance is 0.5 × 90 + 0.5 × 40 = 65 per cent. No-alert performance is 0.5 × 80 + 0.5 × 30 = 55 per cent. The group-adjusted difference is 10 percentage points in favour of the alert.
Alternatively, use the observed overall mix of 40 morning and 20 afternoon journeys, or two-thirds morning and one-third afternoon. The alert-standardised rate becomes (2/3 × 90) + (1/3 × 40) = 73.3 per cent to one decimal. The no-alert-standardised rate becomes (2/3 × 80) + (1/3 × 30) = 63.3 per cent. Again the difference is 10 percentage points.
These comparisons answer a conditional mathematical question: what would the rates look like under a common mixture of the two recorded time groups? They do not by themselves prove a causal effect because the alert and no-alert journeys were not necessarily comparable in every other relevant respect.
10. The 10 percentage points are not 10 per cent relative improvement
A change from 80 per cent to 90 per cent is a rise of 10 percentage points. As a relative percentage increase compared with 80 per cent, it is (90 − 80)/80 × 100 = 12.5 per cent. These are different quantities.
In the afternoon, a rise from 30 to 40 per cent is also 10 percentage points, but its relative increase is (40 − 30)/30 × 100 ≈ 33.3 per cent. The same absolute rate difference can be a different proportion of the original rate.
Use percentage points to compare proportions directly. Use relative percentages only when the question asks for the size of the increase relative to the original value. Precision in this language prevents a confident but misleading headline.
11. Why the observational result is not proof of causation
An alert might influence whether a passenger leaves earlier. It might also be sent preferentially on difficult days, or the afternoon group might have different traffic conditions. The packet does not describe random assignment, identical journey routes or controlled weather conditions. Therefore the association cannot establish that the alert caused the observed within-group difference.
A stronger trial would define the outcome in advance, balance comparable journeys, assign the alert under an appropriate fair procedure, and record other relevant variables. Even then, the result would be an estimate supported by the study’s design and limits, not a guarantee for every future journey.
The careful conclusion is: within the observed morning and afternoon strata, alert journeys had higher on-time rates, while the unadjusted combined rate was lower because of unequal group composition. That statement accurately reflects the supplied numbers without claiming a causal effect the data cannot establish.
12. A small sample is an additional limitation
The morning alert group has only 10 observations; the afternoon alert group also has 10. One journey changes either row’s on-time rate by 10 percentage points. That should make a reader cautious about interpreting a 10-point gap as stable across future journeys.
The no-alert morning group is larger, with 30 observations, so each journey contributes about 3.33 percentage points to its row’s rate. Different sample sizes affect how sensitive the observed proportions are to the outcome of one more journey. This does not automatically make a larger sample unbiased, but it does influence random variability.
Do not invent a confidence interval unless your course or task supplies an appropriate method and assumptions. For the present SEC-level workshop, a plain statement that the samples are small and uneven is informative, honest and fully supported by the packet.
13. One additional journey can change the headline
Suppose the next observed alert afternoon journey arrives on time. That row would become 5 on-time out of 11, not 5 out of 10. The revised alert total would be 14 on-time out of 21 journeys, or about 66.7 per cent. This remains slightly below 67.5 per cent but is closer than before.
If instead that additional journey is late, the alert total becomes 13/21 ≈ 61.9 per cent. A single outcome therefore changes the combined rate visibly. This sensitivity is a reminder that one week’s snapshot is a weak foundation for a definitive long-term policy.
The calculation does not imply that either new observation is more likely. It simply tests the effect of one more data point. This is a useful habit when interpreting percentages drawn from small denominators.
14. Separate punctuality frequency from delay severity
On-time rate asks how often journeys meet the defined threshold. It does not say how late the failures are. A service could have a high on-time percentage but a few very long delays. Another could have more one-minute delays but rarely strand passengers for twenty minutes. Those are different user experiences.
Consider an independent fictional example of twelve late journeys with lateness values of 1, 1, 1, 2, 2, 2, 3, 3, 3, 8, 8 and 25 minutes. The total positive lateness is 59 minutes. The mean among these late journeys is 59/12 ≈ 4.92 minutes. The median is halfway between the sixth value (2) and seventh value (3), so 2.5 minutes.
The largest delay is 25 minutes. That outlier lifts the mean above the median and may matter strongly to families who have fixed commitments. These twelve delays belong to this separate exercise, not the original four-row alert data. Do not merge the datasets as if they record the same journeys.
15. Mean, median and maximum answer different questions
The mean late delay spreads the total lateness across the twelve late journeys. The median locates the middle of the ordered list, making it less sensitive to one unusually large delay. The maximum reports the worst observed case. None is a complete account by itself.
If a school asks, “How much time do delayed passengers lose on average among late journeys?” the mean is relevant. If it asks, “What lateness is typical among those late journeys?” the median may give a useful complementary answer. If it asks about safeguarding a tightly timed transfer, the largest observed delay and the tail of the distribution matter.
A reliable report names the population behind each measure. The 4.92-minute mean is for late journeys only. It must not be described as the mean lateness across all journeys unless the on-time journeys are also included with an appropriate zero or early-arrival convention.
16. A negative or zero outcome needs a definition
Because this workshop defines on-time as arriving no later than schedule, a journey arriving early is not assigned negative “late minutes” in the separate positive-delay calculation. We have defined lateness as max(actual arrival − scheduled arrival, 0). On-time and early journeys therefore contribute zero positive delay.
A different analysis might study signed schedule deviation, where an early arrival has a negative value. In that setting, negative early times can cancel positive late times in an average, potentially hiding a pattern of both very early and very late arrivals. The statistic then answers a different question.
Before averaging anything, tell the reader whether the values are signed deviations, positive lateness only or absolute differences. An attractive mean without a clear underlying variable is difficult to interpret responsibly.
17. Reliability is not identical to individual fairness
A single combined percentage can conceal whether the afternoon group consistently experiences poorer service. In the fictional packet, both afternoon on-time rates are much lower than the corresponding morning rates. The group difference persists regardless of the alert status shown in the table.
That does not prove the afternoon staff work less carefully or that passengers behave differently. The groups may face distinct road, weather or scheduling conditions that were not measured. It does indicate a reason to study that part of the service specifically, rather than allowing the larger morning sample to dominate the overall headline.
In a school recommendation, avoid blaming people without evidence. State the observed subgroup gap, then propose a targeted investigation of conditions that might explain it. This protects both scientific validity and respectful communication.
18. How to write the headline faithfully
A misleading headline would say: “Alerts made shuttle punctuality worse.” It treats an observational aggregate as a causal finding and ignores the contrary within-group comparisons. A more accurate version is: “Overall on-time rates were 65% with alerts and 67.5% without, but alert journeys had higher rates within both morning and afternoon groups.”
That headline is longer, but each clause earns its place. The first reports the unadjusted result; the second explains why an unqualified conclusion would be incomplete. A short follow-up can explain that the alert observations contained a greater afternoon share.
Another careful wording is: “The results are inconclusive about whether alerts cause improvement because the groups differ in composition and may differ in other ways.” This does not erase the observed pattern. It gives the pattern its proper evidential strength.
19. A four-paragraph decision note
Decision question: Should the fictional school expand an arrival-alert system? Observation: Within morning journeys, the recorded on-time rates were 90% with alerts versus 80% without; within afternoon journeys, they were 40% versus 30%. Aggregate: The combined rates were 65% with alerts and 67.5% without because the mixes of morning and afternoon journeys differed.
Limitation: The small, uneven observational samples cannot establish a causal benefit; additional unmeasured factors may differ. Recommendation: Run a more balanced, suitably designed comparison with the same punctuality definition, record journey conditions and examine both on-time rates and the length of delays before changing the policy.
The prose is not a substitute for a real school transport policy. It is an original model of how an SEC learner can combine accurate source reading, numerical results and scientific caution without hiding a difficult result or overstating certainty.
20. Design a better comparison
The first improvement is to observe comparable morning and afternoon journeys under both alert conditions. An appropriate plan might balance sample counts within each stratum and specify the on-time definition before any data are collected. This avoids creating a headline whose direction depends largely on which time group dominated the sample.
Next, record relevant context such as route, planned departure time, road conditions and how alerts were assigned. If alert recipients were selected based on expected delays, the comparison would require extra care because the groups were not formed independently of the outcome risk.
A study can be improved without becoming perfect. The useful habit is to name one weak link and propose a check that addresses it. “Collect more data” is vague; “collect matched morning and afternoon journeys under both conditions using the same punctuality rule” is a defensible improvement.
21. What the calculation does not establish
The packet does not show that alerts save money, improve safety, reduce average commute duration or guarantee a certain arrival time. Those outcomes would need separate measurements. A student should not import desirable benefits into a recommendation merely because the alert sounds like a sensible idea.
Similarly, the aggregate reversal is not evidence that the measurements were fabricated or incorrect. It can occur naturally when rates from unequal groups are combined. Check arithmetic first, group mix second, and possible causal explanations last.
Responsible reasoning means keeping separate the claim, its numerical support, the assumptions behind it and the consequences of a future decision. The strongest conclusion may identify exactly what remains unknown.
22. An independent audit of every rate
Add all the observed journeys: 10 + 30 + 10 + 10 = 60. Add the on-time journeys: 9 + 24 + 4 + 3 = 40. Across the complete fictional dataset, the overall on-time rate is 40/60 ≈ 66.7 per cent. This combines both alert conditions, so it answers neither of the two alert-specific questions by itself.
Total morning journeys are 10 + 30 = 40, with 9 + 24 = 33 on time, or 82.5 per cent. Total afternoon journeys are 10 + 10 = 20, with 4 + 3 = 7 on time, or 35 per cent. Those figures reveal the strong time-of-day difference that drives the aggregate reversal.
The sums also check one another: 33 + 7 = 40 on-time journeys, and 40 + 20 = 60 journeys. If these reconciliation checks failed, the row table, arithmetic or transcription would need correction before any interpretation could be trusted.
23. Thirty original questions with worked answers
The tasks use the fictional shuttle packet unless a separate data set is expressly named. They progress from counting and rates to group adjustments, delay severity and defensible conclusions. These are original learning exercises, not SEAB questions or an official marking grid.
Worked question 1
Task. Morning with alert: how many observed? Model answer. 10 journeys.
Why. Use the row count, not the overall sample. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 2
Task. Morning with alert: on-time rate? Model answer. 90%.
Why. 9/10 × 100. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 3
Task. Morning without alert: on-time rate? Model answer. 80%.
Why. 24/30 × 100. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 4
Task. Afternoon with alert: on-time rate? Model answer. 40%.
Why. 4/10 × 100. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 5
Task. Afternoon without alert: on-time rate? Model answer. 30%.
Why. 3/10 × 100. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 6
Task. Morning within-group gap in percentage points? Model answer. 10 percentage points.
Why. 90 − 80. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 7
Task. Afternoon within-group gap in percentage points? Model answer. 10 percentage points.
Why. 40 − 30. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 8
Task. Total alert journeys? Model answer. 20.
Why. 10 morning plus 10 afternoon. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 9
Task. On-time alert journeys? Model answer. 13.
Why. 9 + 4. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 10
Task. Overall alert on-time rate? Model answer. 65%.
Why. 13/20. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 11
Task. Total no-alert journeys? Model answer. 40.
Why. 30 morning plus 10 afternoon. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 12
Task. On-time no-alert journeys? Model answer. 27.
Why. 24 + 3. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 13
Task. Overall no-alert on-time rate? Model answer. 67.5%.
Why. 27/40. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 14
Task. Unadjusted difference, alert minus no alert? Model answer. −2.5 percentage points.
Why. 65 − 67.5. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 15
Task. Morning proportion of alert journeys? Model answer. 50%.
Why. 10/20. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 16
Task. Morning proportion of no-alert journeys? Model answer. 75%.
Why. 30/40. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 17
Task. Equal-mix standardised alert rate? Model answer. 65%.
Why. Half of 90 plus half of 40. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 18
Task. Equal-mix standardised no-alert rate? Model answer. 55%.
Why. Half of 80 plus half of 30. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 19
Task. Equal-mix adjusted gap? Model answer. +10 percentage points.
Why. 65 − 55. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 20
Task. Overall observed morning on-time rate? Model answer. 82.5%.
Why. (9+24)/(10+30). Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 21
Task. Overall observed afternoon on-time rate? Model answer. 35%.
Why. (4+3)/(10+10). Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 22
Task. Overall on-time rate across all 60 journeys? Model answer. About 66.7%.
Why. 40/60. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 23
Task. Relative increase from 80% to 90%? Model answer. 12.5%.
Why. (90−80)/80 × 100. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 24
Task. Relative increase from 30% to 40%? Model answer. About 33.3%.
Why. (40−30)/30 × 100. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 25
Task. One extra on-time alert-afternoon journey: new alert total? Model answer. 14/21, about 66.7%.
Why. The denominator rises by one as well as the success count. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 26
Task. One extra late alert-afternoon journey: new alert rate? Model answer. 13/21, about 61.9%.
Why. Only the total count rises. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 27
Task. Separate delay list: total positive late minutes? Model answer. 59 minutes.
Why. 3×1 + 3×2 + 3×3 + 2×8 + 25. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 28
Task. Separate delay list: mean among late journeys? Model answer. About 4.92 minutes.
Why. 59/12. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 29
Task. Separate delay list: median? Model answer. 2.5 minutes.
Why. Average the sixth and seventh sorted values, 2 and 3. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
Worked question 30
Task. Does the packet establish that alerts caused improved punctuality? Model answer. No.
Why. Uneven observational groups and other possible influences remain. Write down the relevant population, the denominator and the meaning of the statistic before stating a policy conclusion. Avoid replacing an observed association with a claim of causation.
24. A five-stage error clinic
First, check which denominator belongs to the question. A morning percentage uses morning journeys, not all journeys. Second, check whether the comparison uses common weights. Unadjusted totals from differently composed samples cannot automatically be interpreted as a treatment effect.
Third, distinguish percentage points from relative percentages. Fourth, distinguish on-time frequency from the severity of delays. Fifth, separate what happened in the fictional observations from what would happen if the school implemented alerts everywhere.
These steps can be applied to test scores, club attendance, science investigations and resource-allocation decisions. They are a useful transfer beyond transport because the fundamental error is the same: a strong-looking number conceals a weaker comparison.
25. A week-long study route
Day one: read the four-row packet and identify every subgroup and total. Day two: calculate the four within-group percentages by hand. Day three: calculate both combined rates and explain why they point the other way. Day four: rebuild the result as a weighted average.
Day five: compare using a common morning–afternoon mix, explaining that standardisation is a conditional arithmetic comparison, not proof of causation. Day six: analyse the independent delay list, distinguishing mean, median and maximum. Day seven: write a short committee note that reports observation, limitation and a justified next investigation.
A good sign of readiness is the ability to explain the apparent reversal without quoting a complicated formula. If a learner can show how the composition changed and what still cannot be concluded, the underlying reasoning is becoming secure.
26. Connect to the existing learning owners
This case builds on Volume 0156, the canteen queue trial, by moving beyond simple averages and experimental controls to an advanced subgroup-composition trap. It also extends the earlier Mathematics modelling route and the source-independence workshop in Volume 0133 without replacing their separate purposes.
The core subject owners are the Secondary English Learning Hub, Mathematics Hub and Science Hub. The G3 SEC Learner Guide provides the wider examination path, while the PSLE Learning Guide supports earlier reading and numerical foundations.
For the authoritative examination specifications, consult the official K300 English, K310 Mathematics and K326/K327/K328 Science syllabuses. This original advanced workshop is not an additional SEC paper.
27. The stronger conclusion is the one that knows its limits
The observed aggregate alert rate of 65 per cent is numerically lower than the 67.5 per cent no-alert rate. That statement is true. It is also true that alert journeys had higher recorded on-time rates in both morning and afternoon groups. The two facts coexist because the proportions of morning and afternoon trips differ between the groups being combined.
A useful report preserves that tension instead of picking the statistic that suits a preferred conclusion. The school can design a better test, collect balanced observations and investigate delay severity before making a wider decision. It cannot responsibly claim the alerts definitely caused either improvement or harm from this packet alone.
That is the examination skill worth keeping: identify the relevant population, count the correct outcomes, compare on a shared basis and write a conclusion no stronger than its evidence. The arithmetic becomes more powerful when its limitations remain visible.
