Compare Late-Arrival Data When the Rule Changes

Did you know that a school report can show a dramatic fall in late arrivals even when the pupils are arriving at almost the same times? The explanation may be hidden in an apparently harmless change to the meaning of the word late. What the register counts depends on its rule, not just on what happened at the classroom door.

A late-arrival threshold is the recorded amount of time after an agreed starting point that triggers a late classification. To compare periods fairly, use the same threshold and the same denominator for both, or clearly report that the definitions differ. A reduction from five recorded late arrivals to two is not evidence of improvement if the first period counted arrivals more than five minutes late and the second counted only arrivals more than ten minutes late.

A simple diagnostic: ask what was counted before looking at the percentage

Try this conversation with a pupil: “Our class had five late arrivals last week but only two this week. Did punctuality improve?” A plausible first answer is yes. Next reveal that the first week used more than five minutes and the second week used more than ten minutes as the cutoff. Ask whether the comparison still has the same meaning. A thoughtful student should now ask for the underlying arrival times, rather than defending the original conclusion.

That change of mind is the learning target. The arithmetic may be correct at every step, but the question being answered may have changed. The core aim is not to distrust all school data; it is to learn which facts must remain stable before a trend can mean what we say it means.

Why definitions change the answer

A classification is a decision rule applied to observations. The observation might be “arrived seven minutes after the agreed start”; the classification might be “late” or “not recorded as late”. When the rule changes from more than five to more than ten minutes, that same seven-minute arrival changes category without anyone arriving earlier.

Schools can have legitimate reasons to change a practice, including improving clarity or recording procedures. But a revised rule and improved behaviour are different events. Comparisons should distinguish the administration of the indicator from the behaviour the indicator is supposed to describe. A reporter who hides that distinction gives a false impression of precision.

Worked example: two invented weeks of arrival times

These numbers are invented solely to teach the method. They are ten arrival events in each fictional week, not real pupils, a school policy or an actual eduKateSG result. All delays below are in whole minutes after the same agreed start time. A zero means an on-time arrival. Neither list contains missing observations. The first reporting system flags a delay greater than 5 minutes; a revised system flags a delay greater than 10 minutes.

Week A, ten recorded delays (minutes): 0, 2, 4, 6, 7, 9, 11, 12, 0, 5.
Week B, ten recorded delays (minutes): 0, 1, 6, 8, 9, 10, 11, 15, 0, 4.
Original Week A count at >5 minutes: 6, 7, 9, 11 and 12 qualify, giving 5 of 10 = 50%.
Revised Week B count at >10 minutes: 11 and 15 qualify, giving 2 of 10 = 20%.
Consistent >5-minute rule for both weeks: Week A has 5/10 = 50%, but Week B has 6/10 = 60% because 6, 8, 9, 10, 11 and 15 qualify.
Consistent >10-minute rule for both weeks: both Week A and Week B have 2/10 = 20%.

The original mixed-rule report reads as a fall from 50% to 20%—a change of 30 percentage points. Yet harmonising the rule to more than five minutes gives 50% then 60%, a rise of 10 percentage points, while harmonising to more than ten minutes gives 20% in both weeks. Those comparisons are not contradictory calculations. They are different classifications of the same invented underlying arrival times.

Notice that the two weeks also contain only ten arrival events each. Even with a stable rule, one week cannot establish a lasting behavioural trend. The first duty is to calculate correctly; the second is to explain the limits of the evidence.

One word at the boundary can change the register

The phrases more than five minutes and five minutes or more are not the same. For a tiny fictional set of delays 0, 5, 5 and 6 minutes, a rule of greater than five classifies only the six-minute event as late: 1/4 = 25%. A rule of at least five classifies the two five-minute events as well: 3/4 = 75%. The cutoff is not a decorative footnote. It controls the numerator.

Ask students to underline comparison words before counting: over, at least, after, on or before. Then translate the words into a precise test. If times were recorded only to the nearest minute, the measurement and rounding convention also matter near a boundary. State that limitation rather than pretending to know hidden seconds.

What to do if individual timestamps are unavailable

Sometimes a learner receives only weekly totals, not the underlying observations. In that case, an old count under a five-minute rule cannot generally be recalculated into a count under a ten-minute rule. You need to know how many recorded arrivals were in the interval that changes classification. Do not guess that half of the late arrivals exceeded ten minutes; no such split follows from the total.

A responsible chart can display the figures as two separately labelled series, draw a visible vertical divider where the definition changed, and annotate that the apparent trend is not directly comparable. If source data are available under legitimate access permissions, recode both periods to the same rule while keeping the original reported figures intact. Records should be protected, and personally identifiable attendance data should not be copied into an unsanctioned classroom spreadsheet.

Progressive practice: from guided counting to independent judgement

Round 1 — count with the rule printed above the data

Give the learner the invented Week A delays and the sentence “late means more than five minutes”. Ask for three outputs: the qualifying observations, the numerator, and the fraction of all recorded arrivals. The worked answer is 6, 7, 9, 11, 12; five observations; 5/10 = 50%. The student should point to the delay of exactly five and explain why it is excluded.

Round 2 — change the rule but not the observations

Keep Week A unchanged and change the rule to more than ten minutes. The learner should find 11 and 12, so 2/10 = 20%. Ask: “Did any pupil arrive earlier after we changed the rule?” No. The classification moved; the observed arrivals did not. A learner who can explain this without copying the earlier arithmetic has begun to understand the mechanism.

Round 3 — compare two weeks independently

Give both lists but hide the answer. Require a two-column table that applies the same rule to both weeks. Then ask for a plain-language conclusion containing the words according to the same cutoff. A valid sentence is: “Using more than five minutes for both fictional weeks, the share of these recorded arrivals classified late is 50% in Week A and 60% in Week B.” It is not valid to turn this into a causal claim about a new school programme.

Round 4 — transfer to an unfamiliar situation

Replace arrival times with an independent task: a reading log initially counts a session only if it lasts more than fifteen minutes, then later counts sessions of at least ten minutes. Ask whether the new count alone proves that learners read more. The principled answer is no: first reconcile the definition and the eligible sessions. Transfer matters more than recognising the original school-arrival story.

Practical steps for pupils, parents and teachers

For learners: write a one-line definition under every graph or table; record thresholds and comparison symbols; keep units in minutes; calculate numerators and denominators separately.
For teachers: include a “what changed in the rule?” question before asking for an interpretation; make data definitions part of the rubric, and show one near-boundary example.
For parents: ask which arrivals were counted, whether the same people or events were eligible in each period, and whether a policy or recording method changed.
For school project groups: document the data source, measurement precision, missing records, start-time reference, cutoff, reporting period and any coding change.
For everyone: do not publish identifiable arrival records. A small fictional practice set teaches the reasoning without exposing a real child.

How to observe real learning progress

Start with a four-item diagnostic. Give one fixed-threshold count, one inclusive/exclusive boundary, one changed-definition comparison and one case where the source observations are missing. Record not only whether the answer is right, but also whether the pupil names the rule, denominator and information limits. Repeat with an unfamiliar dataset after an interval. Stronger performance means fewer unsupported comparisons and a clearer explanation of the numerator.

A simple mastery check is for the learner to write: “These two published percentages may be correct within their own reporting rules, but they cannot be treated as a time trend until we confirm the definitions match.” That sentence shows understanding which is useful in mathematics, science, geography and daily life. It is not a guarantee of any examination grade.

What the original data-quality guidance can and cannot tell us

The US National Center for Education Statistics guidance on attendance data explains why consistent reporting supports meaningful comparisons. It is a reference about data quality, not a statement of current Singapore school punctuality policy. A school’s actual attendance definitions and privacy requirements must be checked from the responsible local source; they cannot be inferred from the invented thresholds in this article.

The broader lesson agrees with the UNESCO Institute for Statistics explanation of methods and standards: consistent classifications and documented measurement methods are essential when comparing indicators. We use those principles here as a way to make classroom thinking more reliable, not to claim that an international reporting framework determines any school’s late-arrival rules.

Frequently asked questions

What if both weeks have the same number of students?

That removes one possible source of mismatch, but it does not repair a changed cutoff or different observation opportunities. Ten pupils are not the same as ten arrival events unless the task defines one arrival per pupil, and repeated days must be counted consistently.

Can a school legitimately change its late-arrival rule?

Yes, a school may have reasons to revise a rule. The issue here is not whether a change is allowed; it is whether numbers calculated under different definitions can be presented as evidence of a behavioural change without qualification.

Is late exactly five minutes under a ‘more than five’ rule?

No. “More than five” excludes exactly five, whereas “five or more” includes it. If recording is rounded, the underlying precision may limit classification at the boundary.

What if a report includes only percentages?

Ask for the total eligible arrivals and exact definition. A percentage without its denominator hides how many events contributed, and an unreported threshold can make a trend misleading.

Do these invented results tell us anything about the effectiveness of a real school programme?

No. They demonstrate how a change in definitions alters a count. They do not evaluate any named school, child or programme, and they cannot establish a cause of observed behaviour.

The core aim: make the meaning travel with the number

A well-written report should let the next reader reconstruct what was counted and why. When a pupil learns to ask what late, complete, passed or improved actually means, the pupil is doing something bigger than checking a percentage. They are learning not to confuse a changed measuring rule with a changed world.

The Core Aim of Education | How to Compare School Late-Arrival Records When the Definition of Late Changes