A student improves after a teaching change. Did the change cause the improvement? Sometimes that is a reasonable conclusion. Sometimes the timing is real but the cause is uncertain. Mathematics learning is influenced by school teaching, tuition, independent practice, topic mix, prior knowledge, sleep, confidence, assessment design, time pressure and ordinary variation. A before-and-after mark can tell us that two events happened in sequence; it cannot, by itself, separate all those causes.
This guide is about maths formative assessment, instructional decision-making and causal caution in everyday Mathematics teaching. It is not a research-methods textbook and it does not ask parents or tutors to run formal experiments on children. Its aim is more practical: when a new explanation, practice routine, tutor, class format or study method is introduced, what evidence should make us more or less confident that it helped?
The answer matters because good decisions depend on correct attribution. If we credit the wrong change, we may preserve something unnecessary. If we blame the wrong cause for a decline, we may discard a useful approach. If we declare success from one favourable test, we may accelerate before learning is stable. Causal humility is not indecision; it is disciplined decision-making under ordinary educational uncertainty.
This article follows Did the Student Actually Learn?. That page asks whether learning changed. This one asks the next question: if learning did change, how confidently can we connect that change to a particular teaching action?
1. Sequence is necessary, but it is not sufficient
For a teaching change to cause an improvement, the change must obviously occur before the improvement. That makes timing useful. But “after” does not automatically mean “because of”. A student may begin a new tutor in August and score higher in September while also completing school revision, recovering from illness, meeting a more favourable paper and independently increasing practice.
The correct everyday statement is often layered. “The score improved after the intervention” is a temporal fact. “The exact error targeted by the intervention reduced” adds mechanism evidence. “The improvement remained on delayed and changed tasks” adds learning evidence. “The same pattern appeared repeatedly while other conditions were reasonably stable” strengthens the case further. Each layer increases confidence without pretending that a home or tuition setting has the control of a formal trial.
This distinction is consistent with how rigorous education research treats causal claims. The What Works Clearinghouse uses research-design standards because intervention and comparison conditions, baseline differences, attrition and confounding factors affect whether an outcome difference can credibly be attributed to an intervention. Everyday teaching operates at a different scale, but the intellectual habit transfers: ask what else changed.
2. A six-level evidence ladder for teaching changes
- Timing: performance changed after the teaching change.
- Target match: the specific mathematical behaviour the change was designed to improve also changed.
- Retention: the change remained after time passed.
- Transfer: the change appeared in fresh or different-looking tasks.
- Replication: the pattern appeared more than once, not only on one favourable assessment.
- Alternative explanations checked: major competing explanations such as paper difficulty, extra support or a different topic mix were considered.
This ladder does not produce a numerical probability. It disciplines language. At level one, say “improvement followed the change”. At higher levels, it becomes more reasonable to say “the change appears to have contributed”. Strong causal language should be reserved for evidence capable of carrying it.
3. Match the outcome to the intervention
The cleanest classroom evidence begins by asking what the teaching change was actually supposed to change. If a tutor introduces a sign-checking routine, the primary outcome is not the whole examination score. It is whether sign errors reduce and whether the learner catches them independently. If a new mixed-practice routine is introduced to improve method selection, test method selection in unfamiliar mixed work. If a pre-teaching approach is meant to reduce first-encounter overload, examine school-entry behaviour and later independent use.
Broad scores can move for many reasons. Narrow mechanism measures are easier to interpret. Once the targeted behaviour changes, then ask whether it contributes to wider performance. This order prevents an intervention from receiving credit for outcomes it was never designed to affect.
It also makes failure informative. If the target behaviour does not change, the teacher can revise the intervention quickly rather than waiting for a term-end score to announce that something went wrong.
4. Establish a baseline before changing everything
A baseline is simply a reasonable picture of what was happening before the change. In ordinary tutoring, that can be two or three recent samples, a marked school paper, one fresh task and a record of prompt dependence. It does not need to be elaborate. Its value is that memory becomes unreliable after improvement: once a student is doing better, everyone tends to remember the earlier state more dramatically or more vaguely than it really was.
Baseline equivalence is a formal issue in research design, and the What Works Clearinghouse treats it carefully when comparing groups. A family is not creating comparison groups, but the same principle offers a warning: comparisons are more trustworthy when the conditions being compared are reasonably similar.
If the “before” sample is a difficult prelim paper and the “after” sample is a short topical quiz, the difference is hard to attribute. If both samples test similar skills under comparable conditions and the targeted error pattern changes, the evidence is cleaner.
5. Change fewer things when you want to learn what worked
Parents understandably react to a poor result by changing everything at once: new tutor, new schedule, more worksheets, less CCA, daily correction, extra online videos and a stricter bedtime. The student may improve, but attribution becomes impossible. More importantly, the family will not know which changes are worth sustaining.
When the situation is not urgent, prefer a smaller number of deliberate changes. Define the problem, choose a mechanism, apply the smallest plausible intervention and retest. If several changes are necessary, document them rather than pretending one deserves all the credit.
This is not an argument for withholding help in the name of clean data. Educational welfare comes first. If a learner urgently needs several supports, provide them. The appropriate conclusion is simply broader: “the package of changes was followed by improvement; we cannot isolate the contribution of each component.”
6. Regression toward a student’s usual range can fool us after extreme results
An unusually poor test often triggers an intervention. The next result is frequently better even when nothing fundamental changes, because extreme observations contain ordinary variation and are often followed by observations closer to the student’s typical range. Researchers refer to this general statistical phenomenon as regression toward the mean. In everyday teaching, it is one reason a single terrible result should not be used as both the diagnosis and the baseline proof of success.
Suppose a usually 70-level student scores 42 after illness and a very difficult paper. A new revision routine begins, and the next test is 68. The routine may be useful, but the return toward the learner’s previous range would likely have occurred to some degree anyway. Stronger attribution asks whether the specific behaviours targeted by the routine changed and whether those changes persist across later tasks.
The practical safeguard is simple: use more than one pre-change and post-change observation where possible, and inspect mechanism evidence rather than only the most dramatic numbers.
7. Paper difficulty and topic mix are competing explanations
Two Mathematics assessments are rarely identical samples. One may contain more routine algebra; another may emphasise unfamiliar geometry and problem solving. One may award accessible method marks; another may place a larger fraction of marks in multi-step questions. If a student’s score rises, compare the papers before attributing the movement to teaching.
This does not require a sophisticated difficulty model. Ask whether the same capabilities were tested, whether unfamiliarity changed, whether time demand was similar and whether the student’s previously weak areas actually appeared. Then inspect overlapping question types.
A targeted intervention is most convincing when the targeted behaviour improves on questions that genuinely require it, not when the later assessment simply avoids the old weakness.
8. Improvement should appear where the mechanism predicts
A useful causal story has a mechanism. If spaced retrieval is introduced, we expect better delayed access to previously learned material. If worked examples are faded, we expect greater independent method construction. If representation training is added, we expect better translation between words, diagrams and equations. If checking routines are taught, we expect more self-detected errors and fewer repeated slips.
When the observed improvement occurs somewhere unrelated to the mechanism, remain cautious. A new checking routine should not automatically receive credit for a major improvement in conceptual trigonometry. A new algebra explanation should not be assumed to cause better geometry timing unless a plausible connection is visible.
Mechanism matching is one of the most practical tools available to parents and tutors because it does not require a control group. It asks whether the change happened in the place the intervention touched.
9. Forty everyday attribution problems in Mathematics
1. New tutor begins after a poor examination
Observed sequence. The next school test rises sharply. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The tutor may have helped, but recovery from an unusually poor paper, school revision and topic mix are competing explanations. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Check the exact errors identified at intake, prompt dependence, delayed retention and performance on new questions. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Say the new tutoring was followed by improvement; stronger attribution waits for repeated target-matched evidence. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
2. A sign-checking routine is introduced
Observed sequence. Overall marks rise modestly and sign errors almost disappear. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The narrow error change matches the intervention mechanism closely. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Count sign errors before and after across several comparable tasks, including one timed set. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. It is reasonable to attribute the reduction in sign errors more confidently than the entire score increase. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
3. Mixed practice replaces topical blocks
Observed sequence. Practice feels harder, but mixed-paper method selection improves over several weeks. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The observed change matches the intended mechanism: selecting methods without chapter labels. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Compare method-choice accuracy on unseen mixed sets before and after. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Credit the routine cautiously for selection improvement if retention and transfer persist; do not claim it caused every score movement. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
4. Student starts sleeping more before tests
Observed sequence. Marks and attention improve during the same period. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Sleep may contribute, but teaching, practice and paper variation are also changing. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Track error timing, late-paper deterioration and several assessments rather than one. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Treat the change as part of a plausible support package unless cleaner evidence isolates it. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
5. New workbook begins and scores rise
Observed sequence. The workbook is more structured and the student completes far more questions. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Improvement may come from increased practice volume, better sequencing, topic familiarity or the workbook itself. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Use unseen questions from another source and test the same skills after a delay. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. If gains transfer beyond the workbook, confidence rises that learning rather than source familiarity changed. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
6. Past-year papers begin weekly
Observed sequence. Exam scores improve rapidly. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Paper familiarity, pacing, retrieval and learning can all contribute. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Separate gains in time completion, old error types and unseen-paper transfer. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Attribute specific gains to rehearsal where evidence fits; avoid treating all improvement as conceptual growth. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
7. A new algebra explanation is introduced
Observed sequence. The student suddenly stops cancelling additive terms. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The targeted misconception changed in the predicted place. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Use contrasting valid and invalid cancellation examples weeks later. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. If the discrimination persists across topics, attribution to the explanation is reasonably strong in everyday teaching terms. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
8. A class moves from six students to three
Observed sequence. The learner receives more individual questioning and marks later improve. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Smaller group size may alter visibility and feedback, but tutor, content, peers and schedule may also differ. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Track which intermediate behaviours changed: prompts, corrections, speaking, independent starts. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Credit the smaller setting only through mechanisms actually observed rather than assuming class size alone caused the result. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
9. Tuition moves earlier in the week
Observed sequence. Homework quality improves and school lessons feel easier. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Timing may reduce fatigue or improve pre-exposure, but other factors can coexist. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Compare similar weeks and look for the predicted change in readiness rather than only test scores. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Use cautious language unless the pattern repeats with reasonably stable conditions. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
10. Pre-teaching starts before school topics
Observed sequence. The student participates more confidently when the topic appears in school. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The timing and target match the intervention, but participation can also change with familiarity and confidence. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Check later independent performance after school teaching and after a delay. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. The strongest claim is that pre-teaching appears to improve first-encounter access when that pattern repeats. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
11. Daily revision replaces weekend cramming
Observed sequence. Immediate practice scores fall slightly but delayed recall improves. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Distributed exposure plausibly matches the retention change. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Use delayed retrieval on the same content and keep total study time reasonably comparable. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Attribute improved retention cautiously to spacing if the effect replicates across topics. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
12. A calculator lesson is added
Observed sequence. Calculator errors drop, total marks rise. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The narrow error reduction directly matches the intervention. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Audit input errors on several tasks and check estimation behaviour. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Be confident about calculator-control improvement, modest about its share of the total score gain. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
13. A student receives a new formula sheet
Observed sequence. Routine accuracy improves immediately. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The external support may improve performance without increasing retrieval or understanding. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Remove the sheet later and use a changed problem. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Do not attribute immediate improvement to learning until capability survives support removal. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
14. A retrieval quiz is added at lesson start
Observed sequence. Old topics are accessed more readily in mixed work. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The mechanism predicts improved retrieval, though lesson attention and review time also change. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Track previously forgotten items and delayed mixed questions. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. If old knowledge becomes more available across weeks, the quiz routine is a plausible contributor. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
15. Worked examples are faded systematically
Observed sequence. Independent starts initially worsen, then improve. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Removing support changes the measurement condition and may build independent control. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Record prompt levels and fresh-question starts over time. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Credit the fading process if independence rises while accuracy and transfer recover. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
16. A student changes school teacher
Observed sequence. Marks improve during the term. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Curriculum pacing, classmates, maturation and assessment conditions also changed. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Look for a teaching mechanism that matches the change and use multiple assessments. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Avoid strong causal claims from the transition alone. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
17. A parent stops sitting beside homework
Observed sequence. Homework scores fall, later independence improves. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Support removal can reduce immediate performance while increasing autonomous control. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Track help required, fresh-task starts and school-test transfer. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Judge the change by independence, not the first homework percentage. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
18. An error log is introduced
Observed sequence. Repeated mistakes become easier for the student to name, and some decline. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The log may support awareness and checking, but ordinary practice continues simultaneously. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Test whether logged error families decline more than unlogged incidental slips. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Attribute self-monitoring gains to the routine when the link is visible and repeated. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
19. A teacher adds more visual representations
Observed sequence. Word-problem entry improves. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Representations plausibly support modelling, but question selection may also have changed. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Use new contexts and ask students to construct rather than copy representations. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Confidence rises if translation improves outside the original examples. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
20. Student begins peer explanation
Observed sequence. Conceptual explanations improve and some errors decline. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Explaining may deepen processing, but the learner also gets more practice and peer exposure. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Compare individual independent work before and after, especially on concepts explained to peers. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Use the mechanism-matched changes as evidence without claiming the entire grade change. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
21. A difficult paper is followed by an easier one
Observed sequence. Score jumps twenty points. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Paper difficulty is an obvious alternative explanation. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Compare overlapping skills, error types and independent changed tasks. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Do not use the raw jump as evidence that a teaching change caused equivalent learning. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
22. A very poor outlier triggers emergency revision
Observed sequence. The next mark returns to the student’s previous range. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Regression toward usual performance and the intervention can both contribute. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Use several pre-change results and later assessments, not the outlier alone. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Treat the rebound cautiously until target behaviours improve beyond the old baseline. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
23. A very strong outlier is followed by a lower score
Observed sequence. A new method was introduced between the two tests, so the decline looks like failure. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The first score may have been unusually favourable, and the second paper may be harder. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Compare mechanism-specific evidence and longer trends. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Do not abandon the new method solely because performance returned toward the learner’s usual range. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
24. The student studies twice as long and changes technique simultaneously
Observed sequence. Scores improve. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Dose and method are confounded; either or both may matter. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Where possible, examine whether the new technique changes its targeted behaviour beyond what more time would predict. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. State that the combined change helped unless later evidence separates components. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
25. School starts revising the same topic as tuition
Observed sequence. Performance improves quickly. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Two instructional sources now overlap. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Use fresh transfer tasks and document the overlap. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Avoid assigning exclusive credit to either source; the combined exposure may be the relevant cause. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
26. A new tutor also changes class size and materials
Observed sequence. Marks improve over two months. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Several intervention components changed together. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Track intermediate mechanisms such as feedback frequency, prompt dependence and representation quality. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Treat the package as the unit of change unless components can be separated ethically and naturally. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
27. Student confidence improves before scores
Observed sequence. The learner attempts more questions and abandons fewer. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Confidence may enable more engagement, while successful learning may also create confidence. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Track direction over time: does increased attempting precede more completed reasoning and later score stability? The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Use reciprocal language when cause likely runs both ways. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
28. Scores improve before confidence
Observed sequence. Written work is stronger but the learner still feels uncertain. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Learning can precede subjective belief. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Show concrete evidence of independent success and monitor whether confidence later catches up. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Do not assume low confidence caused the earlier weakness simply because it remains. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
29. Teacher starts giving more feedback
Observed sequence. Marks do not change. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Quantity alone may not alter uptake, and opportunity cost may reduce practice time. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Measure whether students act on feedback and whether targeted errors change. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Judge the change by feedback uptake, not comment volume. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
30. Teacher gives less but more targeted feedback
Observed sequence. Scripts contain fewer comments; retest performance improves. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Prioritisation may increase actionability, though task selection may also differ. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Track uptake on the targeted error families. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Attribute improvement to feedback design cautiously when later independent behaviour changes in the predicted direction. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
31. Student changes calculator model
Observed sequence. Speed improves and input errors initially rise. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The tool change affects execution conditions rather than conceptual knowledge. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Separate calculator-control outcomes from mathematical reasoning. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Do not interpret the temporary score movement as broad learning gain or loss. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
32. A formula is memorised using flashcards
Observed sequence. Recall becomes fast, but application remains weak. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The intervention achieved its direct target—retrieval—but not selection or transfer. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Test conditions and method choice in mixed problems. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Credit the flashcards for recall only; add different teaching for application. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
33. A diagnostic test identifies a fraction gap and it is repaired
Observed sequence. Later algebraic-fraction work improves. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The dependency mechanism is plausible and target-linked. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Confirm ordinary fraction stability, then check algebraic transfer on unseen questions. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Attribution is stronger when the repaired prerequisite and dependent topic improve in the predicted sequence. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
34. A diagnostic test identifies a gap that was not causing the current error
Observed sequence. Time is spent repairing it, but the original task still fails. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The gap was real but not the active bottleneck. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Return to the reasoning chain and probe representation, selection and execution. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Do not confuse finding a weakness with proving that weakness caused the present failure. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
35. A student begins using a checking checklist
Observed sequence. Self-corrections rise, but first-pass errors are initially unchanged. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The intervention targets detection before it necessarily reduces generation of errors. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Measure caught errors separately from initial errors. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Credit the checklist for detection first; later ask whether repeated detection changes the underlying habit. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
36. The same intervention works for one student but not another
Observed sequence. Two learners receive identical teaching; one improves substantially. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Prior knowledge, misconceptions, motivation and fit can moderate response. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Compare learner states rather than assuming universal effectiveness. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Treat intervention effectiveness as conditional on the student and context. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
37. A new approach works in one topic but not another
Observed sequence. Representation training helps ratios but not geometry. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The mechanism may be topic-specific or the implementation may differ. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Test whether the relevant representation skill is actually used in each topic. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Avoid generalising from one successful chapter to all Mathematics. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
38. Marks improve during a less stressful month
Observed sequence. Study routine stayed similar while CCA and family demands eased. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Reduced load may improve attention and sleep, confounding attribution to teaching. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Document major context changes and compare again when normal load returns. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Use the broader context in the causal story rather than attributing everything to instruction. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
39. A student starts asking more questions
Observed sequence. Understanding and scores later improve. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. Questioning may drive learning, but growing understanding may also make better questions possible. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Track whether questions precede repaired misconceptions and independent transfer. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Treat questioning and learning as a reinforcing loop unless evidence supports a one-way claim. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
40. An AI homework tool is introduced
Observed sequence. Completion increases and answers become more accurate. This establishes timing, not cause. The first discipline is to keep the statement descriptive before moving to explanation. A causal claim becomes stronger only when the later evidence matches what the change was designed to affect.
Alternative explanations. The tool may support learning or may simply supply more external solution structure. Naming alternatives does not mean the intervention failed. It means the evidence has more than one possible story. Good instructional judgement keeps plausible alternatives visible long enough to test them.
Mechanism check. Remove the tool on delayed changed tasks and inspect independent reasoning. The best everyday check looks close to the proposed mechanism and uses fresh work. Broad marks are useful downstream, but a narrow target measure often tells us sooner whether the change did the job it was given.
Decision language. Attribute learning only to what survives beyond tool access; attribute completion improvements separately. Use language proportional to evidence: “followed by”, “associated with”, “appears to have contributed”, or, when the pattern is especially well supported, “the evidence is consistent with this change producing the targeted improvement”. Avoid turning one before-and-after comparison into certainty.
10. A safe N-of-one teaching inquiry
A teacher or tutor can learn from one student’s data without pretending to run a clinical trial. Define one problem, one intended change and one observable target. Keep ordinary necessary support in place. Gather a small baseline. Introduce the change. Retest the target on fresh tasks, after a delay where relevant. Record major contextual changes. Then decide whether to keep, refine or replace the approach.
The method should never require withholding instruction that the student needs simply to create a comparison. Educational welfare comes first. The purpose is disciplined reflection, not experimental purity.
A useful example is a recurring sign error. Baseline: six sign errors across three comparable sets. Change: explicit sign-boundary cue plus final line check. Target: sign errors and self-corrections, not the entire Mathematics grade. Post-change: repeat comparable sets and one timed mixed paper. If sign errors fall and self-correction rises, the mechanism has evidence. The grade may later improve, but attribution should begin with the behaviour the intervention touched.
11. When a comparison is available, use it carefully
Sometimes naturally occurring comparisons help. The same student may have two similar skills, one receiving a new routine earlier than the other. Two classes may use different resources. A term may contain repeated comparable assessments. These observations can strengthen reasoning, but they are not automatically controlled experiments because groups, teachers, topics and timing can differ.
Formal research standards exist precisely because comparison quality matters. The current What Works Clearinghouse handbook discusses randomized controlled trials, quasi-experimental designs, regression discontinuity and single-case designs, alongside threats such as baseline differences and confounding factors. A tutor does not need to reproduce these designs, but should understand the lesson: a comparison is valuable only to the extent that competing explanations are controlled or acknowledged.
12. What to write in a parent progress update
Strong progress language separates observation from interpretation. “The student’s last three mixed sets contained one sign error compared with seven across the three baseline sets” is observation. “The new sign-checking routine appears to be helping” is interpretation. “The routine caused the entire examination improvement” would require much more evidence.
A five-line update can be enough: what changed; what intervention was used; what target behaviour moved; what other factors changed; what will be retested next. This makes the reasoning inspectable and protects everyone from overclaiming.
Parents usually do not need statistical language. They need accurate language.
13. When to stop a change
An intervention should not continue forever merely because it sounded sensible. Stop or redesign it when the target behaviour does not move after a reasonable implementation period, when the learner becomes more dependent, when the opportunity cost is too high, when another mechanism better explains the problem, or when the target has become secure and support can fade.
Implementation quality matters before judging effectiveness. A good method used inconsistently may appear ineffective. Conversely, a poor fit delivered intensively can consume large amounts of time. Record enough of what actually happened to distinguish the idea from its implementation.
The desired end state is usually less intervention, not more. Mathematics support has done its job when the learner carries more of the system independently.
14. Three-student tuition and causal caution
Small-group tuition creates many rapid adjustments: a different question, an extra diagram, a prompt withdrawn, a misconception corrected, an explanation reframed. Because changes happen frequently, attribution can become casual. “That explanation fixed it” may be true, but the better habit is to retest the exact decision later.
The three-student format can help because the tutor has enough visibility to observe intermediate mechanisms: who needed which prompt, which error changed, whether the student transferred the method and whether support reduced. That makes progress more explainable than a system that records only attendance and scores.
The point is not to produce research papers from tutorials. It is to make teaching self-correcting.
15. Research and evidence notes
The What Works Clearinghouse Procedures and Standards Handbook explains the standards used to assess causal evidence in education research. Its current Version 5.0 framework distinguishes research designs and examines issues including baseline equivalence and confounding. These formal standards are far beyond ordinary parent or tutor monitoring, but they provide the intellectual boundary for strong causal claims.
The WWC standards resources also define confounding factors as factors that can make it impossible to tell whether an intervention or something else is responsible for outcome differences. This page translates that caution into everyday Mathematics decisions without pretending that ordinary teaching data meet research-study standards.
For instructional data use, the What Works Clearinghouse guide Using Student Achievement Data to Support Instructional Decision Making recommends treating data as part of an ongoing cycle of instructional improvement. That cycle—observe, adapt, retest—is the practical backbone here.
16. Frequently asked questions
If marks rise after starting tuition, can I say tuition caused the rise?
You can confidently say the rise followed the start of tuition. Stronger attribution requires more: the targeted mathematical behaviours should change, gains should persist and transfer, and obvious competing changes should be considered. In ordinary life, “appears to have contributed” is often the most accurate wording.
Do I need a control group to decide whether a teaching method is useful?
No. Parents and tutors make practical decisions from repeated within-student evidence, mechanism match and professional judgement. A control group is relevant to stronger research-level causal claims, not a prerequisite for every classroom adjustment.
What is the easiest way to improve attribution?
Define the target before the intervention. If you know exactly what the change is meant to improve, you can measure that behaviour rather than waiting for a broad grade.
Why are before-and-after scores weak by themselves?
Because time passes and many other things change: teaching, practice, topic mix, difficulty, maturity, stress and ordinary variation. A before-and-after result is informative, but several causal stories can fit it.
What if several changes are necessary at once?
Make them. Student welfare is more important than analytic neatness. Record the package and avoid pretending one component deserves all the credit. Later, components can sometimes be simplified as the learner stabilises.
Can a method be useful even if the score does not rise yet?
Yes. If it produces the intended intermediate change—better retention, cleaner representation, fewer prompts, stronger checking—that is meaningful evidence. The broader score may lag or be affected by other new demands.
Can a score rise even if the method is not useful?
Yes. Easier papers, favourable topic mix, ordinary variation, extra study time or other teaching changes can produce a rise. That is why mechanism-matched evidence matters.
17. The shortest useful summary
When a student improves after a teaching change, begin with the true statement: the improvement followed the change. Then ask whether the targeted behaviour changed, whether the change survived time and transfer, whether it repeated, and what else changed at the same time.
The more those pieces align, the more confident the practical conclusion can become. When they do not align, keep the language modest and gather the next useful observation. Educational evidence does not need to be perfect to be useful, but it should be honest about what it can carry.
For the prior step—deciding whether the student actually learned—use Score Change, Retention and Transfer in Mathematics. For diagnosing what failed, use Foundation Problem or Method Problem?. For feedback as an intervention, use Mathematics Feedback Uptake.