Did you know? A student can calculate the variance of a small dataset correctly and still use the wrong denominator for an H1 Mathematics question. The difference is not a button on the calculator. It is whether the values are the full population being described or a sample used to estimate something larger.
For parents looking for a JC1 H1 Mathematics tutor in Bukit Timah, summarised data questions reveal whether a learner understands what the symbols mean. The exam may give only n, Σx, Σx² or even Σ(x−a) and Σ(x−a)². A good solution reconstructs the mean and unbiased variance without pretending the raw observations are known.
Quick answer: for a random sample of n greater than one, the sample mean is x̄ = Σx/n. The conventional unbiased sample variance estimator is s² = [Σx² − (Σx)²/n]/(n−1). For d = x−a, use x̄ = a + Σd/n and s² = [Σd² − (Σd)²/n]/(n−1).
Quick reading route: first decision · raw data · summarised data · shifted sums · merged samples · FAQs.
Population versus sample: the first decision
The population is the entire group you intend to describe. A sample is a subset chosen to obtain information about the population. The unknown population mean and variance are parameters; the computed sample mean and sample variance are statistics.
If five values form the entire population under discussion, the descriptive variance uses division by five after calculating the squared deviations from their mean. If those same values form a sample used to estimate a larger population’s variance, the usual unbiased sample-variance formula uses division by four.
These results are not competing answers to the same question. They answer different mathematical questions. For an exam that specifically asks for an unbiased estimate of the population variance, a tutor should train the student to recognise the n−1 denominator immediately.
- n: number of observations.
- Σx: the sum of the observations.
- Σx²: the sum of the individually squared observations.
- (Σx)²: the square of their sum, which is generally different from Σx².
- x̄: sample mean.
- Σ(x−x̄)²: sum of squared deviations about the sample mean.
- s²: the usual unbiased estimator of population variance.
- s: positive square root of s², expressed in the original measurement unit.
As a miniature check, take observations 2 and 3. Σx² is 2² + 3² = 13. But (Σx)² is (2+3)² = 25. Confusing the two symbols will usually ruin the compact variance formula before the student reaches the denominator.
The official H1 8865 learning requirement
The 2027 Singapore–Cambridge H1 Mathematics 8865 syllabus explicitly includes unbiased estimates of the population mean and variance based on a sample. It includes problems where information is provided as Σx and Σx², or in shifted form as Σ(x−a) and Σ(x−a)².
The same sampling topic addresses the sample mean as a random variable and sampling distributions. For the 2027 paper, Probability and Statistics carries 60 of the 100 marks, while Pure Mathematics carries 40. Students should check the syllabus for their own eventual A-Level year and follow their JC topic sequence.
Our sampling and Central Limit Theorem guide concentrates on how sample averages vary. This article has a narrower job: recovering and checking the unbiased sample variance from summaries, shifted sums and merged samples.
Worked example one: from five raw observations
Consider the fictional data 8, 10, 12, 14 and 16. These are deliberately chosen teaching numbers, not results from actual eduKateSG learners. There are n = 5 observations.
Their sum is Σx = 8 + 10 + 12 + 14 + 16 = 60, so the sample mean is x̄ = 60/5 = 12. The deviations from the mean are −4, −2, 0, 2 and 4.
Square the deviations to get 16, 4, 0, 4 and 16. Their sum is 40. For an unbiased sample-variance estimate, divide by n−1 = 4, giving s² = 10 and sample standard deviation s = √10 ≈ 3.162.
The point of this first example is to see where the formula comes from: each observed value is compared with the sample mean, and the resulting squared distances are added. The compact formulas introduced next are shorter ways to calculate the same quantity.
If the five values describe the whole population
For the same five values viewed as the complete population, the population descriptive variance is 40/5 = 8, not 10. That difference should be explained by the statistical job, not hidden as a rounding discrepancy.
The two calculations are easy to verify manually because the deviations are small and symmetric. This makes the dataset an effective first teaching example before introducing sums of squares from longer surveys.
Why the unbiased formula uses n minus one
When deviations are measured from the sample mean, they must add to zero. Once n−1 deviations are known, the final deviation is determined by the condition that their sum is zero. One degree of freedom has effectively been used to estimate the centre.
Dividing the sum of squared deviations by n produces a descriptive sample measure, but under the usual independent random-sampling assumptions it tends to underestimate the unknown population variance on average. Dividing by n−1 corrects that downward bias.
Unbiased does not mean perfectly accurate in every sample. It means that across repeated samples satisfying the assumptions, the estimator’s expected value equals the population variance. Individual estimates may be above or below the truth.
A parent-friendly explanation is to imagine repeatedly drawing new groups of five observations. The unbiased estimator is designed to be correct on average across those draws, not to guarantee that every group gives the same estimate.
Worked example two: from the two sums
Now hide the five raw observations and give only n = 5, Σx = 60 and Σx² = 760. The sample mean is still x̄ = 60/5 = 12.
Use the identity Σ(x−x̄)² = Σx² − (Σx)²/n. Here the squared-deviation sum is 760 − 60²/5 = 760 − 720 = 40.
Divide by n−1 = 4 to obtain s² = 40/4 = 10. The standard deviation is √10, just as in the direct calculation. We did not need the individual observations to recover these two summaries.
- Write n, Σx and Σx² separately so their roles remain clear.
- Calculate x̄ = Σx/n.
- Calculate the corrected sum of squares Σx² − (Σx)²/n.
- Divide by n−1 if the question requests the unbiased variance estimator.
- Take the positive square root only if standard deviation is required.
- Check that the variance is non-negative and the units are appropriate.
This process avoids a common trap: subtracting 60² from 760 without dividing by five. That would produce a negative intermediate result, which contradicts the non-negativity of squared deviations.
A second summarised-data problem
Suppose n = 8, Σx = 200 and Σx² = 5200. Find the sample mean and unbiased sample variance. These are again fictional summary numbers for teaching.
The sample mean is 200/8 = 25. The sum of squared deviations is 5200 − 200²/8 = 5200 − 5000 = 200.
The unbiased sample variance is 200/(8−1) = 200/7 ≈ 28.5714. The sample standard deviation is √(200/7) ≈ 5.345.
If the standard deviation feeds another calculation, preserve sufficient precision instead of replacing 200/7 with 28.6 too early. In particular, a normal-distribution function that expects a standard deviation should not be given the variance 28.5714.
Worked example three: using a shifted origin
Questions sometimes provide Σ(x−a) and Σ(x−a)² for a convenient fixed value a. This is a change in the arithmetic reference point, not a different population or a new model.
For the values 8, 10, 12, 14 and 16, choose a = 10. Then d = x−10 gives deviations −2, 0, 2, 4 and 6. Their sum is Σd = 10 and their squared sum is Σd² = 4+0+4+16+36 = 60.
Given only n = 5, a = 10, Σd = 10 and Σd² = 60, calculate the mean as x̄ = a + Σd/n = 10 + 10/5 = 12.
Calculate the unbiased variance as [Σd² − (Σd)²/n]/(n−1) = [60 − 100/5]/4 = (60−20)/4 = 10. Shifting the origin has not changed the spread.
Why shifting by a constant preserves variance
Subtracting the same 10 from every observed value slides the entire set of points to the left on a number line. The distances between observations, and the distances from their new mean, are unchanged.
That is why adding or subtracting a constant changes the mean by that constant but does not change sample variance. A learner who gets a different variance after a pure shift has probably omitted the correction term or used the wrong reference mean.
Recover the raw sums from shifted information
Because x = d+a, we have Σx = Σd + na. In the example, Σx = 10 + 5(10) = 60.
Squaring gives x² = d² + 2ad + a², so Σx² = Σd² + 2aΣd + na². Substitute: 60 + 2(10)(10) + 5(100) = 760.
The raw sums are exactly the same as before. This proves the shifted and unshifted forms are consistent without needing the individual observations to appear on the exam paper.
When the question supplies only shifted totals, work directly from those totals whenever possible. Converting everything back is useful for checking, but it should not become an unnecessary source of arithmetic error.
The shift a need not be the exact sample mean
Our reference value a was 10, while the sample mean was 12. That is acceptable. The purpose of a is often computational convenience: observations near 100 might be easier to work with as deviations from 100.
If a happens to equal x̄, then the shifted deviations sum to zero and the correction term disappears. But if a is simply nearby, the correction term remains necessary to obtain the correct spread.
This is an excellent teaching moment: students should not assume that a familiar-looking reference value printed in the question is automatically the mean. The mean must still be found from the supplied information.
What happens when every value is scaled?
Suppose the original sample is transformed by y = 2x + 3. The new sample mean is 2x̄ + 3 = 2(12) + 3 = 27.
The new sample variance is 2²s² = 4(10) = 40. The factor of two multiplies every deviation by two, and squaring those deviations multiplies their contributions by four. The added constant three merely shifts the entire dataset.
If the transformation is y = −3x + 4, the new sample variance is (−3)²(10) = 90, not −30. The direction of the data reverses, but squared spread cannot become negative.
The expected value and independent variance tutorial explains related probability-model transformations. Here the same scaling logic applies to the numerical spread of the sample observations.
Work backwards from mean and variance
Suppose n = 6, sample mean x̄ = 15 and unbiased sample variance s² = 12. We are asked to recover Σx and Σx².
The sum is Σx = nx̄ = 6(15) = 90. The squared-deviation sum is (n−1)s² = 5(12) = 60.
From Σx² − (Σx)²/n = 60, solve Σx² = 60 + 90²/6 = 60 + 1350 = 1410.
These two summaries do not uniquely determine every raw observation. Many different sets can share the same count, sum and sum of squares. The mathematical task is to recover the summaries necessary for centre and spread, not to invent an unobservable dataset.
Combining two datasets without a wrong average
Suppose Group A contains the fictional observations 8, 10, 12 and 10, while Group B contains 14, 16, 18 and 20. We want the unbiased sample variance of the combined eight observations.
Group A has sum 40 and sum of squares 408. Group B has sum 68 and sum of squares 1176. The combined set therefore has n = 8, Σx = 108 and Σx² = 1584.
Its mean is 108/8 = 13.5. Its squared-deviation sum is 1584 − 108²/8 = 1584 − 1458 = 126. Divide by seven to obtain combined unbiased variance 18.
A tempting but incorrect shortcut is to average the two separate sample variances. Group A’s variance is 8/3; Group B’s is 20/3; their ordinary average is 14/3, about 4.67, not 18.
Why is the combined spread larger? The group means differ: A is centred at 10 and B at 17. When the groups are merged, variation between those centres contributes to the overall dispersion.
This illustrates a wider statistical lesson: groups with small internal variability can form a combined dataset with much greater variability. A correct calculation must incorporate both group sizes and their relative locations.
The sample mean is a random variable, not a guaranteed value
When samples are drawn randomly, the sample mean changes between samples. Under the standard independent-observation assumptions, its expected value is the population mean and its variance is σ²/n.
That formula describes the distribution of sample means, not the distribution of individual observations. If the population standard deviation is σ, the sample-mean standard deviation is σ/√n.
As a simple illustration, suppose a fictional normal population has mean 20 and standard deviation 6, and n = 36 independent observations are sampled. The sample mean has mean 20 and standard deviation 1.
Therefore P(X̄ > 21.5) is the standard normal right tail beyond z = 1.5, approximately 0.0668. This is not the probability that one individual observation exceeds 21.5; a single observation has a different standard deviation.
Our sampling and CLT guide explores that distinction in detail. In the current article, the priority is learning to calculate the sample’s centre and spread accurately before using them in later inference.
A negative variance is a useful warning
Because variance is based on squared deviations, it cannot be negative for ordinary real data. A negative result from a valid sample calculation suggests a formula, input or precision problem.
Check whether Σx² was confused with (Σx)², whether division by n was omitted, and whether the sums were copied accurately. For example, 760 − 3600 is not the correct corrected squared-deviation sum; the needed expression is 760 − 3600/5.
A tiny negative value may sometimes be caused by heavily rounded or inconsistent summary data. Investigate the supplied precision and the question’s instructions rather than silently taking an absolute value and continuing.
This is an example of mathematical sense-checking that no calculator can substitute for. A screen can report an arithmetic output, but the learner must decide whether the output is meaningful.
Ten common mistakes and what to repair
- Treating Σx² and (Σx)² as the same number.
- Using n rather than n−1 for a requested unbiased sample variance.
- Using n−1 blindly when the data explicitly form a full descriptive population.
- Omitting the correction term (Σx)²/n.
- Treating the shifted reference a as the unknown sample mean.
- Forgetting Σx = Σ(x−a) + na.
- Forgetting the 2aΣ(x−a) term when recovering Σx².
- Rounding the mean so early that it distorts the squared deviations.
- Using a variance value when a probability calculation needs standard deviation.
- Reporting squared units for standard deviation or ordinary units for variance.
A student might describe all ten as ‘careless mistakes’, but the corrections differ. The next practice question should target the earliest incorrect step: symbol interpretation, denominator, shift reconstruction, arithmetic or final unit.
A twenty-minute independent revision cycle
- Minutes 1–4: compute the mean and direct squared deviations of 8, 10, 12, 14, 16.
- Minutes 5–8: compute the unbiased variance using only n = 5, Σx = 60 and Σx² = 760.
- Minutes 9–12: use a = 10 and the shifted summaries to recover the same results.
- Minutes 13–16: work backwards from n, mean and variance to obtain Σx and Σx².
- Minutes 17–20: explain one difference between population variance and unbiased sample variance, then log the first error.
Revisit one changed problem two or three days later. A learner who can transfer the formula to shifted or merged data without opening the example is making more meaningful progress than one who only repeats the same numbers.
What the tutor should diagnose in a small group
One learner may calculate direct deviations correctly but mishandle the sum-of-squares notation. A second may use the compact formula reliably yet forget that ‘unbiased’ requires n−1. A third may manage both but fail a shifted-data question because a was mistaken for the mean.
In a three-pupil tutorial, the same dataset can be approached by raw observations, compact summaries and shifted values. Then everyone must independently solve a changed problem, so no student relies on another’s method.
The immutable eduKateSG small-group reference describes the established three-pupil teaching approach near Sixth Avenue MRT at 8 Fourth Avenue, Singapore 268674, commonly in 90-minute weekly sessions.
Families should confirm the current H1 Mathematics tutor, class timetable, fees and availability before planning attendance. The teaching format does not guarantee a particular JC1 place. Bring a recent marked school statistics worksheet for an accurate first diagnosis. See consultation details.
Seven-day review progression
- Day 1: raw observations and the meaning of squared deviations.
- Day 2: compare sample unbiased and population descriptive variance.
- Day 3: solve questions using n, Σx and Σx² only.
- Day 4: work with Σ(x−a) and Σ(x−a)².
- Day 5: reverse the formulas to recover missing sums.
- Day 6: combine two datasets and explain why their centres matter.
- Day 7: complete an unseen mixed problem and retry the first incorrect decision.
If a student is already strong in raw calculations, shorten the early steps and spend more time on shifted summaries, inferential interpretation and choosing the correct statistic from problem wording.
Independent checks with answers
A mean from a sum
If n = 10 and Σx = 150, then x̄ = 150/10 = 15.
A corrected sum of squares
If n = 4, Σx = 20 and Σx² = 120, then Σ(x−x̄)² = 120 − 400/4 = 20.
An unbiased variance
For that same sample, s² = 20/(4−1) = 20/3.
A shifted sum
If n = 5, a = 10 and Σ(x−10) = 10, then Σx = 10 + 5(10) = 60.
A scaled sample
If s² = 7 and all observations are transformed to y = −3x + 4, the new sample variance is 9(7) = 63.
Frequently asked questions
Why use n−1 for the unbiased sample variance?
The sample mean has been estimated from the same observations, using one degree of freedom. The n−1 denominator gives the usual unbiased estimate under standard random-sampling assumptions.
What does Σx² mean?
Square each observation and then add. It is not the same as adding the observations first and then squaring.
Does shifting every value change variance?
No. Adding or subtracting the same constant shifts the centre but preserves distances between observations and hence variance.
Why might a question give Σ(x−a) instead of Σx?
A convenient shift can make arithmetic easier. The sample mean and variance can be recovered without the original full data list.
Can an unbiased sample estimate be wrong for one sample?
Yes. Unbiasedness concerns average behaviour across repeated samples, not perfect accuracy in every realised sample.
Why can two low-variance groups form a high-variance combined sample?
The groups may have quite different means. The combined variance reflects both within-group spread and separation between those centres.
Is this within H1 Mathematics scope?
The 2027 H1 8865 syllabus includes unbiased estimates of mean and variance from samples, including summarised and shifted data. Match the version to the student’s actual examination.
What is the quickest parent check?
Ask the student to solve the five-value example directly and again from summaries, then explain why both give the same variance. A delayed changed example is even better evidence of learning.
Additional worked example: a shifted reference of fifty
Suppose a fictional random sample of six measurements is described by a convenient reference value a = 50, the sum Σ(x−50) = 18, and the squared sum Σ(x−50)² = 102. We are not given the six original values. The useful question is whether these summaries are sufficient to recover the sample mean and unbiased sample variance. They are, because a common shift preserves the distances between the observations.
Define d = x−50. The average deviation is 18/6 = 3, so the original sample mean is x̄ = 50 + 3 = 53. For the corrected squared-deviation sum, calculate Σd² − (Σd)²/n = 102 − 18²/6 = 102 − 54 = 48. The unbiased variance is 48/(6−1) = 9.6, and the sample standard deviation is √9.6 ≈ 3.098.
If a learner insists on recovering each original observation first, ask why that is necessary. Different sets of six observations can share these same summary statistics, so uniquely reconstructing all of them is generally impossible. The point of a summarised-data question is to use information that is already sufficient, rather than invent missing values. This is an important distinction between estimating a characteristic of a dataset and recovering every element of the dataset.
Verify the shifted answer using raw totals
Because x = d + 50, the original raw sum must be Σx = Σd + 6(50) = 18 + 300 = 318. The original sum of squares is Σx² = Σd² + 2(50)Σd + 6(50²) = 102 + 1800 + 15,000 = 16,902. These values are consistent with a mean of 318/6 = 53.
Now calculate the corrected squared-deviation sum from the raw form: 16,902 − 318²/6. Since 318² = 101,124 and 101,124/6 = 16,854, the result is 48. Dividing by five again gives s² = 9.6. This double-check is useful because a student can validate the shifted formula without seeing or needing the original observations.
A strong tutor can turn this into an error-detection exercise by deliberately writing 2aΣd as aΣd, or by forgetting the na² term. The student’s task is to identify which part of the expansion of (d+a)² has disappeared. Explaining that missing term helps repair the algebraic mechanism rather than simply instructing the student to memorise another formula.
Change the reference value, keep the variance
Imagine the same six measurements are now expressed relative to a = 52. Since x−52 = (x−50)−2, the new shifted total is Σ(x−52) = 18−2(6) = 6. The new shifted squared total is Σ(x−52)² = 102−4(18)+4(6) = 54.
The mean is 52 + 6/6 = 53. The corrected sum of squared deviations is 54−6²/6 = 48, leading to the same unbiased variance 9.6. The numerical values of the shifted summaries depend on the chosen reference; the mean and variance of the actual observations do not change.
This is a powerful independent check of the two shifted formulas. If a learner obtains different variances after changing only the reference number, they have almost certainly omitted the correction term or introduced an algebraic sign error. Rather than redo every multiplication without direction, focus on how the shifted deviations relate to the earlier deviations.
A combined-sample calculation with different group sizes
Suppose a hypothetical Group A has 12 observations with mean 30 and unbiased sample variance 25, while Group B has eight observations with mean 40 and unbiased sample variance 16. The groups have different sizes and different centres, so their combined variance should not be found by a casual average of 25 and 16.
For Group A, Σx = 12(30) = 360. Its squared-deviation sum is (12−1)(25) = 275, and therefore Σx² = 275 + 360²/12 = 11,075. For Group B, Σx = 8(40) = 320. Its squared-deviation sum is (8−1)(16) = 112, giving Σx² = 112 + 320²/8 = 12,912.
Combine the summaries: n = 20, Σx = 680 and Σx² = 23,987. The merged mean is 680/20 = 34. The combined squared-deviation sum is 23,987 − 680²/20 = 23,987 − 23,120 = 867. The merged unbiased variance is 867/19 ≈ 45.63.
The merged variance is larger than either separate variance. That is not a mistake: much of the new spread comes from the difference between the group means, 30 and 40. An answer obtained by averaging their variances ignores this extra separation. This example trains students to distinguish within-group variation from variation in the full set of observations.
An outlier changes squared spread much more than a quick glance suggests
Return to the raw fictional sample 8, 10, 12, 14 and 16, whose unbiased sample variance was 10. Now replace the final observation of 16 with 30 while leaving the other four unchanged. The new observations are 8, 10, 12, 14 and 30.
Their sum is 74, so the new mean is 14.8. Their sum of squares is 64 + 100 + 144 + 196 + 900 = 1404. The corrected squared-deviation sum is 1404 − 74²/5 = 1404 − 1095.2 = 308.8. The unbiased sample variance becomes 308.8/4 = 77.2.
Only one of the five values changed, yet the variance increased from 10 to 77.2. Squared deviations give substantial weight to observations far from the mean. A good statistical discussion does not automatically delete an unusual value: the student should first consider whether it belongs to the intended data, whether it was measured correctly and what question the statistic is meant to answer.
The sample-size-one trap
For a sample containing only one observed value, the standard unbiased sample-variance formula is not defined because n−1 equals zero. That does not prove the population has no variation. One observation simply supplies no observed comparison with another observation from which the usual variance estimator can be constructed.
The lone observation equals its own sample mean, so the within-sample squared deviation is zero. Confusing that zero deviation with certain knowledge of the unknown population variance is a conceptual error. This small example helps students understand why the denominator is not a decorative part of the formula.
Check variance and standard deviation when units change
Suppose the observations in the original five-value example are measured in minutes, with mean 12 minutes and unbiased variance 10 minutes squared. If every measurement is converted into seconds using y = 60x, the new mean is 60(12) = 720 seconds.
The new unbiased variance is 60²(10) = 36,000 seconds squared. The standard deviation is 60√10 ≈ 189.7 seconds. Variance receives the square of the scaling factor because each deviation is multiplied by 60 before it is squared.
A student who multiplies the variance by only 60 has treated squared spread like an ordinary measurement. Ask the learner to write the units on each line: the mean remains an ordinary time, standard deviation remains an ordinary time, and variance has squared time units. Unit analysis makes the necessary scaling easier to remember.
A more demanding exam-style final question
A fictional organisation supplies sample summaries relative to a = 52: n = 6, Σ(x−52) = 6 and Σ(x−52)² = 54. A question asks for the unbiased sample variance, the original Σx and the original Σx². This is the same model family as above but uses a different starting representation.
The sample mean is 52 + 6/6 = 53. The unbiased sample variance is [54 − 6²/6]/5 = (54−6)/5 = 9.6. The original sum is 6 + 6(52) = 318. The original squared sum is 54 + 2(52)(6) + 6(52²) = 54 + 624 + 16,224 = 16,902.
This question can be solved entirely without a probability distribution command. The issue is understanding symbols, expanding squared expressions and choosing the unbiased denominator. The result provides three related answers that can check one another rather than three isolated numbers copied from a screen.
For homework, change only n and the two shifted totals, then ask the student to work unaided. Encourage them to reconstruct the raw summaries at the end as a verification, rather than performing the same copied pattern with no concept of what each term measures.
The main idea
A variance formula is useful only when we understand what is being estimated. Begin with the sample or population question, keep the sum notation precise, select the correct denominator and interpret the result in the right units.
For the wider topic, read the sampling and CLT guide, the expected value and variance tutorial, and the Bukit Timah H1 Mathematics overview. Use the official 2027 SEAB syllabus for the stated examination-year scope.