Averages help describe a greenhouse experiment, but the average alone may conceal how much individual observations differ. This G3 SEC Mathematics learner’s guide takes fictional percentage mass changes from a plant-tissue test and teaches students to compare datasets using mean, median, range, interquartile range, standard deviation, tables and honest statistical diagrams. The goal is not to prove a biological result from a single number. It is to decide what a set of measurements actually shows, how representative the evidence may be, and what a careful answer should say under examination conditions.
For 2027, G3 Mathematics is syllabus K310, corresponding to the earlier 4052 reference. Paper 1 has about 26 short-answer questions in 2 hours 15 minutes for 90 marks (50%). Paper 2 has 9–10 questions, also 2 hours 15 minutes and 90 marks (50%); its final question focuses on applying mathematics to a real-world scenario. Essential working matters. Consult the 2027 SEAB directory and K310 syllabus document. All figures and trials here are fictional learning material, not official exam data or agricultural advice.
Route: preserve the meaning of percentage change; distinguish sample records from a summary; compute and interpret mean and median; find quartiles and interquartile range; calculate standard deviation; evaluate outliers; compare distributions; recognise misleading axes and group sizes; plan a representative sample; work through original questions; and write an evidence-led, appropriately limited conclusion.
Start by naming the measured quantity
In the previous Biology root-uptake guide, students calculated tissue percentage mass change as (final mass minus initial mass) divided by initial mass, multiplied by 100. This guide begins after that calculation. The data now represent percentage change in mass of separate tissue pieces after a stipulated interval. A value of +5 means a 5% gain relative to that specimen’s initial mass. A negative value means a loss. Neither is a concentration or a crop yield. Define the variable before computing statistics.
Imagine two hypothetical treatments, A and B, each tested on five separate tissue pieces. The recorded percentage changes for A are 2, 4, 4, 6 and 9. The values for B are 3, 4, 5, 6 and 7. These are fabricated examples selected to make statistical reasoning visible. They are not measures of the performance of any real plant, fertiliser or growing system.
Both lists have five observations, and all are expressed in percent. Because the variables and sample sizes match, several comparisons are possible. Still, a complete analysis must ask how specimens were selected, whether treatment conditions were controlled and whether five observations are sufficient for the conclusion being proposed. Mathematics can accurately describe a biased sample. A numerically correct answer is not automatically a reliable generalisation.
Organise before calculating
Sort the observations, check that none was mistyped and verify the sign of any losses. Both sample lists are already in ascending order, but real questions may not be. A quick table with columns labelled treatment, observation and percentage mass change is often enough. If the question gives raw initial and final masses instead, calculate percentage change for each piece before comparing the treatments. Averaging raw mass changes can give a misleading picture when specimens began at different sizes.
Treat an observation as one measured specimen, not one digit. A row of five observations is a sample of size five. In this exercise there are ten observations in total, but there are two groups. Combining them indiscriminately loses the treatment comparison. If asked for the mean of treatment A, use only A’s five values. If asked for an overall mean, determine whether the sample sizes are equal; an unweighted average of group means is not generally correct when groups differ in size.
This distinction becomes particularly important in longer Paper 2 problems where a chart supplies two datasets and a later question introduces a third group. Write down how many observations belong to each group. Mark the unit beside the column heading. A correct setup often prevents the harder-looking calculation from going wrong later.
Mean describes a balance point
Add all five values for A: 2 + 4 + 4 + 6 + 9 = 25. Divide by 5 to obtain a mean of 5%. For B, 3 + 4 + 5 + 6 + 7 = 25, so its mean is also 5%. These two treatments have the same arithmetic mean in the example. A student who announces that they are equally consistent has reached beyond the statistic: the mean says nothing by itself about variation.
The mean uses every observed value, which can make it sensitive to unusual extremes. In A, the value 9 pulls the mean upward relative to several lower values. In B, the values are more evenly arranged around 5. A good interpretation says that both groups have the same sample average percentage change, then investigates how widely values are dispersed before claiming that their behaviour is similar.
This is why the task’s wording matters. “Calculate the mean” requires a numerical answer with appropriate context; “compare the results” usually requires more than one statistic. If the instruction requests a conclusion, the student should support that conclusion with what the measures actually represent. No reward is gained by giving every statistic the same verbal meaning.
Median is the middle observation
For an odd number of observations, sort the values and identify the middle one. Treatment A has median 4%; treatment B has median 5%. The middle observation is the third item in each sorted list. A useful interpretation is that half the observations are at or below the median and half at or above it, with the usual care about counting when the total number is odd. The median is a positional measure, not a sum divided by the number of values.
The difference between mean and median is informative. A’s mean is 5 but median is 4, partly because the relatively high observation of 9 affects the mean. B’s mean and median are both 5. This comparison suggests a more balanced distribution around 5 in B, but the term ‘symmetrical’ should not be asserted for every dataset with equal mean and median; additional shape evidence would be needed.
When there are an even number of values, the median is the average of the two middle ordered observations. For example, after adding 8 to B, the six values 3, 4, 5, 6, 7, 8 have median (5 + 6) ÷ 2 = 5.5. Do not take the third value simply because the original sample had five observations. A changing sample requires a fresh position calculation.
Range gives a quick but incomplete measure of spread
The range is maximum minus minimum. A’s range is 9 − 2 = 7 percentage points. B’s range is 7 − 3 = 4 percentage points. A has a larger spread between its extremes. The unit of a difference between percentages can be described as percentage points, distinguishing it from a relative percent change between two percentages. Because the original variable is percentage mass change, be clear about the language.
Range is easy to calculate but uses only two observations. Imagine replacing A’s 9 by 20; the range would grow dramatically although four of the five observations remain the same. A conclusion based only on range could overreact to one measurement error or unusual specimen. This does not make range useless. It makes it a first look that should be compared with other statistics and the raw data where available.
In an examination, a request to “state the range” does not need a lecture. Give the subtraction and answer. But when asked to compare consistency, it is reasonable to use the range alongside a more informative measure such as interquartile range or standard deviation. State the difference precisely: a smaller range means the extreme observed values are closer together; it is not proof that the treatment is biologically superior.
Quartiles show how the central part of a dataset is distributed
Quartiles divide an ordered dataset into parts. In a school-style calculation that excludes the middle observation when finding the lower and upper halves of an odd-sized list, the lower two A values are 2 and 4, so Q1 is 3. The upper two are 6 and 9, so Q3 is 7.5. The interquartile range, IQR, is Q3 − Q1 = 4.5 percentage points. For B, Q1 is (3 + 4)/2 = 3.5 and Q3 is (6 + 7)/2 = 6.5, giving IQR = 3.
The exact convention for finding quartiles from a small raw dataset may be prescribed by the question or by teaching practice. Follow the method required by the paper and show the ordered values. The meaning of IQR is stable: it describes the width of the central half of the data. An IQR of 3 says the central values span a narrower interval than an IQR of 4.5 in these two samples.
When the task requires an overall comparison, A has a larger IQR than B under this stated convention. That supports the claim that the central half of A’s observations is more spread out. It does not by itself identify the reason. Different specimens, random measurement variability or a genuinely different response pattern could be involved, and the data alone may not separate those possibilities.
Standard deviation measures spread around the mean
The G3 Mathematics syllabus includes standard deviation as a measure of spread. It is built from the differences between individual values and the mean. For A, the mean is 5. The differences are −3, −1, −1, +1 and +4. Their squares are 9, 1, 1, 1 and 16, totalling 28. Divide by the number of observations, 5, to obtain 5.6; taking the square root gives standard deviation approximately 2.37 percentage points, using the population-style formula for the set of recorded observations as presented in this school example.
For B, the differences from mean 5 are −2, −1, 0, +1 and +2. Their squares total 4 + 1 + 0 + 1 + 4 = 10; 10/5 = 2, and the square root is approximately 1.41 percentage points. B has smaller standard deviation. These calculations give more systematic evidence that B’s observations cluster more tightly around the mean than A’s.
A calculator may have several statistical modes and distinguish between population and sample standard deviation. The correct choice depends on the convention requested and the syllabus formula. Do not press a button without naming the statistic. If the question expects the defined school calculation from the five stated observations, explain that the values are being treated as the given set. Show enough working to make the logic traceable, and round at the requested final stage rather than destroying precision halfway through.
Why squaring deviations is useful
If you add signed deviations from the mean, the total is zero for every dataset by the way the mean is defined. In A, −3 − 1 − 1 + 1 + 4 = 0. Such a sum would make every dataset look perfectly balanced even when values are widely scattered. Squaring avoids cancellation and gives greater weight to large deviations. Taking a square root returns a spread measure in the same type of unit as the original variable.
Standard deviation does not say how many observations were correct. It describes the dispersion of the numbers around their mean. If a scale were systematically biased, repeated readings might have small standard deviation while all readings were inaccurate. Similarly, the two tissue treatments could have identical means and different spreads without either being proved desirable for real plants.
One clear examination sentence is enough: “Although both treatments have a mean mass change of 5%, treatment B is less variable because its standard deviation is approximately 1.41 percentage points, compared with approximately 2.37 for A.” This is an evidence-led statistical comparison. If the question also asks about reliability, discuss how many specimens were tested and how they were selected, not just the computed spread.
Compare distributions rather than declaring a winner
For the fictional A and B samples, the evidence is a small package: both means are 5%, A’s median is 4% and B’s is 5%, A has a larger range, a larger IQR under the stated convention and a larger standard deviation. These values support the conclusion that B is more tightly clustered around its mean in this small dataset. That is a statement about variation, not about guaranteed crop performance.
If the research goal were to maximise the mean percentage gain, the samples tie on the observed mean. If the goal were to minimise variation in a classroom mass-change experiment, B looks more consistent. The word “best” is incomplete until an objective has been defined. A decision rule should name what success means and acknowledge the limited number of measurements and the possibility of biological or procedural variation.
This type of analysis transfers to finance, sport, travel timing and many other real-world contexts. A low average delivery time with high variability may be less predictable than a slightly higher average with small spread. The same arithmetic becomes more useful when connected to a clear decision criterion. In the greenhouse example, the conclusion should remain a statistical interpretation unless additional biological evidence is supplied.
Histograms, box plots and cumulative frequency diagrams answer different questions
A dot diagram can display a small raw dataset clearly because every observation remains visible. A box-and-whisker plot summarises the median, quartiles and extremes, making the central spread and overall range easier to compare. A histogram can show the distribution of larger numerical datasets grouped into suitable intervals; the published G3 syllabus covers histograms with equal class intervals. Cumulative frequency diagrams can support positional questions such as estimates of medians, quartiles and percentiles.
Select a representation that suits the data and task. The five observations in A and B do not need to be hidden in a complicated histogram. A simple labelled dot plot may show them more transparently. If the examination gives a cumulative frequency curve, read the cumulative total first, locate the appropriate fraction of the number of observations and then project to the measurement axis. Do not treat a cumulative total as an ordinary frequency for that interval.
Graph choice can mislead. A bar chart of group means alone would conceal the very spread this exercise is designed to teach. Adding the actual observations or an appropriate spread display would provide a fuller picture. In a written evaluation, name what the graph shows and what it omits. A visually striking chart is not necessarily an informative one.
Beware of truncated axes and selective presentation
Imagine a graph comparing mean mass changes of 4.8% and 5.0%. If the vertical axis begins at 4.7 rather than zero, the difference may appear enormous even though it is only 0.2 percentage points. For some statistical displays a non-zero axis is defensible, but the choice must be clear and suitable for interpretation. A strong learner does not reject every truncated axis automatically; the learner asks whether the scale gives a fair impression of the magnitude.
Selective categories are another problem. Showing only the best-performing individual specimen from treatment A and comparing it with the average from B would not be a like-for-like comparison. The question “what are the units and denominators?” is often more useful than the first visual impression. If one treatment has five readings and another fifty, the number of observations should be disclosed rather than masked by identical-looking columns.
An exam answer might state: “The diagram exaggerates the apparent difference because the vertical scale starts near the measured values, and it omits variability within each group.” Each criticism identifies a precise mechanism. Do not write vaguely that a chart is “biased” without explaining how the design could lead a reader to misinterpret the underlying data.
What happens if sample sizes are unequal?
Suppose A has five observations with mean 5%, while a third treatment C has fifteen observations with mean 7%. The combined mean is not simply (5 + 7) ÷ 2 = 6%. It is the total across all twenty observations divided by twenty: (5 × 5 + 15 × 7) ÷ 20 = (25 + 105)/20 = 6.5%. This is a weighted mean. The greater number of observations in C gives that group more influence on the combined average.
If each observation is a percentage change from a separately prepared specimen, the combined mean here represents the average of those observation percentages. It is not automatically equivalent to total mass gained divided by total starting mass when specimen sizes differ. That alternative quantity uses mass-weighting and answers another question. Name the requested target before choosing a calculation.
In longer questions, the needed group total may be hidden in a frequency column. Multiply each value or class midpoint by its frequency as appropriate, sum the products, and divide by the total frequency. A candidate who understands the meaning of a weighted mean can reconstruct the method instead of memorising a mysterious formula.
Grouped tables require careful interpretation
Suppose a different fictional monitoring study presents percentage mass change in intervals, with frequencies 4 for 0 to less than 2, 7 for 2 to less than 4, and 9 for 4 to less than 6. An estimated grouped mean uses class midpoints 1, 3 and 5: (4 × 1 + 7 × 3 + 9 × 5) ÷ 20 = 70/20 = 3.5%. Because the original individual values are not provided, this is a midpoint-based estimate, not the exact mean of unknown readings.
Do not take interval endpoints as though they are the observations. When class intervals are equal, a histogram can help visualise frequencies; the area interpretation becomes important for unequal intervals in more general statistical work, but do not assume every unfamiliar display belongs to an examined task. Follow the published G3 content and the actual question directions.
When a grouped table requests standard deviation, use the formula and midpoint assumption appropriate to the given frequency table. Show the total frequency, the estimated mean and the sum of squared contributions or the equivalent calculator working. A well-organised table can reduce mistakes more effectively than memorising longer strings of calculator commands.
Sampling can distort a correct calculation
Suppose the greenhouse team measures only the five strongest-looking specimens in each bed. Their arithmetic might be flawless, yet their results would not represent the full population of specimens. This is selection bias. To assess broader conditions more fairly, define the target population and choose samples in a way that does not systematically privilege one outcome. If the task asks for a random sample, describe a method that gives eligible members a suitable chance of selection.
A representative sample does not mean that every observation must be close to the mean. Natural variability may be real and informative. Nor does increasing the sample size correct a deliberately biased selection method. One hundred specimens selected only from the sunniest corner could still tell a distorted story about the whole growing area. Sample size and sampling method are separate dimensions of evidence quality.
Mathematics exams may ask for interpretations rather than elaborate experimental designs. The disciplined answer is still concise: state what was sampled, identify how selection could bias the result and suggest a better method. Use the data to justify numerical claims; use the procedure to explain how far they may generalise. The two forms of reasoning support each other.
Outliers: investigate, do not delete
Consider A’s value 9%. It is the largest observation in the group, but high is not automatically erroneous. A legitimate biological response may vary between specimens; a recording mistake is also possible. If the task does not provide a rule for excluding observations, keep the data and make the influence visible. If an extreme value is suspected, check the source record, measurement procedure and units.
Removing the 9 gives a new mean (2 + 4 + 4 + 6)/4 = 4%. That change shows how strongly the original mean depends on the highest reading. It does not justify excluding the reading. A complete interpretation might say that the mean falls when the highest value is removed, demonstrating sensitivity, while recognising that exclusion requires evidence of error or a pre-agreed analytic rule.
This is a useful habit beyond school. People can produce attractive statistics by choosing which facts to count. Exam questions about misleading statistics reward a student who notices that changes in inclusion criteria can alter a conclusion. Mathematics provides the calculation; careful judgement asks whether the altered dataset still represents the stated question.
Make a decision under explicit constraints
Imagine that the fictional classroom must select a treatment for a follow-up investigation, not for agricultural use. It wants a mean near 5% mass gain and a standard deviation below 2 percentage points in the five observations. Based on the present example, B meets both numerical criteria: mean 5 and standard deviation about 1.41. A meets the mean condition but not the stipulated spread condition, because 2.37 is greater than 2.
The decision is conditional on the chosen rule and the observed sample. If the rule were instead “include at least one reading of 8% or more,” A would satisfy it and B would not, but that would be an unusual and weak criterion for general consistency. Maths cannot silently decide the goal. The examiner may ask you to explain why one method is preferable, in which case you should include the objective and supporting numbers in one sentence.
Also distinguish hard thresholds from rounded displays. A computed standard deviation of 2.004 may be displayed as 2.00 to three significant figures, but it does not satisfy a strict requirement of less than 2 before rounding. Preserve calculator precision when applying constraints. Then round appropriately for the reported answer. This small discipline can determine whether a proposed plan is genuinely feasible.
Write answers that an examiner can audit
When a question requests a mean, show the total and number of observations. For quartiles, write the ordered values and identify the chosen convention or positions. For standard deviation, label whether the calculation uses the formula for the provided dataset and carry sufficient precision. For an interpretation, cite the particular measures that support the comparison. A bare statement that B is “better” is weaker than saying its observed results show less variation under the specified criterion.
Units must survive every calculation. A mean mass change is in percent; a difference between two such values is conveniently described in percentage points. Standard deviation uses the same type of unit as the original variable, whereas a squared-deviation intermediate carries squared units. In final reporting, do not mix percentage change in mass with grams of mass. The two may be connected but are not the same quantity.
Under time pressure, use a short verification sequence: check the group, check the count, check the arithmetic, check the units and check the interpretation. If the task asks for evidence of sampling bias, do not spend your answer space repeating the formula for the mean. Different command words ask for different work. This ability to choose what matters is a core examination skill.
Work back from a decision to the missing evidence
Suppose a report says that treatment B is suitable for all greenhouse crops because its sample has lower standard deviation. That statement leaps from five fictional tissue measurements to a broad biological outcome. The mathematical evidence supports a narrower conclusion: in these observations, B’s percentage mass changes are less dispersed than A’s. To assess suitability across a whole crop, more representative measurements, biological outcomes, relevant conditions and expert interpretation would be needed.
This boundary links Statistics to the surrounding curriculum. The Chemistry guide defined the solution and its quantities; the Biology guide defined what a mass-change experiment can show; this guide defines how the resulting numbers should be summarised. The PSLE learning route offers an earlier foundation for separating the requested quantity from irrelevant context.
The next English chapter will test a different skill: hearing statistical claims in a spoken briefing, detecting revisions, and recording only the details the question asks for. A mathematics student who understands denominators, sample sizes and comparative language will find the listening claims easier to audit. The subjects meet through careful evidence, while each retains its own assessed method.
Original examination-style question clinic
1. Equal mean trap
Question: A and B both average 5%. Are they equally variable?
Worked solution: No. Their means coincide, but A’s standard deviation is about 2.37 and B’s about 1.41 percentage points. A statement about variation requires a spread measure rather than a mean alone. Write the comparison in words tied to the original observations.
2. Median repair
Question: Find the median of 2, 4, 4, 6 and 9.
Worked solution: The five values are ordered; the third is 4, so the median is 4%. The mean is 5%, but the question asks for the positional middle. Confusing these two measures obscures the influence of A’s highest reading.
3. New sixth observation
Question: Add 11 to A. What is the new median?
Worked solution: The six values become 2,4,4,6,9,11. The central two are 4 and 6, whose average is 5%. You must reorder after adding a value. Do not assume the previous median stays the same simply because most readings have not changed.
4. Range check
Question: Calculate the ranges of A and B.
Worked solution: A: 9−2 = 7 percentage points. B: 7−3 = 4 percentage points. The statement ‘A has larger range’ is accurate. The statement ‘A produces worse crops’ is not supported by the specified statistic.
5. First quartile
Question: Using the stated exclude-median convention, find Q1 of A.
Worked solution: The lower half excluding the middle value 4 is 2 and 4. Its midpoint is 3%. The convention matters for very small samples; if an exam specifies another method or asks for a graph reading, follow that instruction.
6. Compare IQR
Question: Find IQR for B using the stated convention.
Worked solution: Lower quartile is (3+4)/2=3.5 and upper quartile is (6+7)/2=6.5, so IQR=3 percentage points. This is the spread of the central half of B, not the difference between its maximum and minimum.
7. Deviation from mean
Question: List deviations of A from mean 5.
Worked solution: The deviations are −3, −1, −1, +1, +4. Their sum is zero, so a signed total cannot measure spread. The squared deviations total 28, which is the useful intermediate for the stated standard-deviation calculation.
8. Standard deviation A
Question: Calculate the school-formula standard deviation from A’s five observations.
Worked solution: Divide the squared-deviation total 28 by 5 to obtain 5.6; the square root is approximately 2.37 percentage points. Show whether the given set is treated as the complete dataset for this calculation, because calculator modes may offer more than one convention.
9. Standard deviation B
Question: Find the same standard deviation for B.
Worked solution: B’s deviations from 5 are −2,−1,0,+1,+2, whose squares total 10. The result is the square root of 10/5, or approximately 1.41 percentage points. It is smaller than A’s because B’s observations are closer to the common mean.
10. Translate a comparison
Question: Write one supported sentence about A and B.
Worked solution: Both groups have sample mean 5%, but B is less dispersed: its standard deviation is about 1.41 percentage points compared with A’s 2.37. The sentence tells the reader which property is being compared and the statistic supporting it.
11. Weighted combined mean
Question: Five observations average 5%, fifteen average 7%. Find the overall mean.
Worked solution: Use (5×5+15×7)/20=6.5%. The mistake 6% treats the two groups as equal-sized. The weighted calculation gives each actual observation equal contribution to the overall average of the observation values.
12. Frequency table mean
Question: Frequencies 4,7,9 occur in intervals 0–2,2–4,4–6. Estimate a grouped mean.
Worked solution: Use midpoints 1,3,5 and total frequency 20. The estimated sum is 4+21+45=70; mean estimate is 3.5%. It is an estimate because the raw values inside each interval have not been disclosed.
13. A misleading vertical axis
Question: Why can a bar graph beginning near two close means exaggerate their difference?
Worked solution: The non-zero baseline can make a small difference occupy a large fraction of the displayed axis. Check the scale and values. A non-zero axis is not always forbidden, but the reader needs a clear reference and should not infer a large absolute effect from apparent bar height.
14. Missing sample counts
Question: Two reports display equal-width bars but one is based on three readings and another on three hundred. What is missing?
Worked solution: The group sample sizes and a discussion of how the samples were drawn. Similar-looking charts do not make unequal evidence equally informative. A reliable comparison should disclose denominators and relevant variation.
15. Outlier deletion
Question: A student removes A’s value 9 because it is inconvenient. Is that sound?
Worked solution: No. An unusual reading should be investigated, not discarded without reason. Removing it changes the mean to 4%, demonstrating sensitivity, but there must be a defensible error rule or evidence for excluding the observation.
16. Biased selection
Question: Only the tallest specimens were selected for each group. Identify the limitation.
Worked solution: This selection may not represent the full population, even if the calculations are correct. A better method would define the eligible specimens and select them fairly according to the study’s aim, rather than selecting on a trait plausibly linked to the outcome.
17. Hard cutoff
Question: A condition requires standard deviation strictly less than 2. A result is 2.004. Does it pass if displayed as 2.00?
Worked solution: No. The unrounded value is greater than 2. A display rounded to three significant figures does not alter the result of a strict inequality. Apply the condition with sufficient calculation precision before presenting a rounded summary.
18. Different denominators
Question: Does the mean of the five specimen percentage gains equal total mass gained divided by total starting mass?
Worked solution: Not necessarily. If starting masses differ, averaging individual percentages gives each specimen one vote, while dividing total mass change by total initial mass weights by initial mass. Both are meaningful for different questions; name the requested statistic.
19. A scientific claim
Question: Does smaller numerical variation prove a superior nutrient treatment?
Worked solution: No. It describes consistency of the chosen measured quantity in the provided sample. Suitability needs separate biological evidence and a clear success criterion. Avoid turning a classroom statistic into a broad agricultural recommendation.
20. Paper 2 response
Question: A question asks which treatment better meets an explicit consistency criterion. What should the final sentence include?
Worked solution: State the chosen treatment, its relevant statistic, the criterion and any important limit. For instance, B has a measured standard deviation below 2 percentage points in this example, while A does not. Do not leave the selection unconnected to the evidence.
Study plan: move from arithmetic to interpretation
Start with five small datasets and calculate their mean, median and range on paper. On the next session, order the same values and identify quartiles and interquartile ranges using the stated convention. In a third session, show how deviations from the mean lead to standard deviation, then check calculator settings and numerical precision. Do not increase problem difficulty until the student can say what each measure communicates without reading a definition from notes.
The next learning task is comparison. Put two distributions side by side and ask which claim each statistic supports. Give an example in which means match but spreads differ, another in which a large outlier moves the mean, and a third where sample sizes differ. Finally, ask the learner to diagnose a graph whose axis or group count is misleading. Ask them to rewrite the conclusion with a numerical citation, a named limitation and no unsupported biological promise.
Parents can help with short questions rather than long lectures: What exactly was measured? How many observations are in the group? Did the answer use the correct denominator? Is the statistic a mean or a spread? Does the conclusion travel farther than the evidence? These habits build naturally from the Mathematics learning route and the site’s PSLE Learning Guide. The most useful progress signal is whether a student can audit a new dataset independently.