H2 Maths Sampling Formula Sheet | CLT & Estimators

Study guideUpdated 06 Nov 2026
Marcus Pang
Reviewed by
Marcus Pang·Managing Director (Maths)

H2 Maths sampling formula sheet: sample mean distribution, Central Limit Theorem, unbiased estimators of population mean and variance, and standard error - every key result on o...

Q: What does H2 Maths Notes (JC 1-2): 6.4) Sampling cover?
A: Sample mean distributions, Central Limit Theorem (CLT), and unbiased-estimation workflows aligned with the H2 Maths 2026 syllabus.
Download: Get the H2 Maths Sampling formula sheet (PDF) for quick revision, or the complete notes (PDF) for the full walkthrough.
Before you revise
SEAB labels sub-topic 6.4 Sampling as "for teaching and learning only", but the core ideas (sample mean, E(Xˉ) E(\bar{X}) , Var⁡(Xˉ) \operatorname{Var}(\bar{X}) , CLT) are exactly what you use inside 6.5 Hypothesis Testing. Recap mean/variance notation and get comfortable computing xˉ \bar{x} and s2 s^2 from summarised data.
  • A sample is a small window into a population: Identify the population parameter.
  • The sample mean varies less when sample size is larger: Divide the variance by n for the sample mean.
  • CLT lets large samples behave approximately normal: State whether normality is exact or approximate.

Concrete example: A class sample of 36 students gives a more stable average than a sample of 4 because random highs and lows cancel more strongly.

Status: SEAB's current H2 Mathematics (9758) syllabus PDF is labelled for 2026. Topic 6.4 is marked "for teaching and learning only" in the official syllabus; the sample-mean ideas feed Paper 2 Section B (Probability and Statistics, 60 marks) through Topic 6.5 Hypothesis Testing.


Formulas at a glance

For the sampling formulas below, assume independent observations from the same distribution, with finite variance. The Central Limit Theorem uses n≥30n\geq30 as a rule of thumb; a very skewed population may need a larger sample.

Every result the 9758 syllabus expects you to use, on one screen. The unbiased variance estimator s2s^2 is provided in MF27; the sample mean distribution, CLT statement, and standard error formula are not, so commit them to memory. Worked examples for each appear in the sections below.

Sample mean distribution

QuantityFormula
Expected value of Xˉ\bar{X}E(Xˉ)=μE(\bar{X}) = \mu
Variance of Xˉ\bar{X}Var⁡(Xˉ)=σ2n\operatorname{Var}(\bar{X}) = \dfrac{\sigma^2}{n}
Standard errorσn\dfrac{\sigma}{\sqrt{n}}
Exact distribution (normal population)Xˉ∼N!(μ,σ2n)\bar{X} \sim \mathcal{N}!\left(\mu, \dfrac{\sigma^2}{n}\right)

Central Limit Theorem

ConditionResult
Large nn (typically n≥30n \geq 30), any populationXˉ≈N!(μ,σ2n)\bar{X} \approx \mathcal{N}!\left(\mu, \dfrac{\sigma^2}{n}\right) approximately

Unbiased estimators

QuantityFormula
Sample mean (unbiased estimator of μ\mu)xˉ=∑xn\bar{x} = \dfrac{\sum x}{n}
Unbiased sample variance (estimator of σ2\sigma^2)s2=1n−1[∑x2−(∑x)2n]s^2 = \dfrac{1}{n-1}\left[\sum x^2 - \dfrac{(\sum x)^2}{n}\right]
Mean from shifted sumsxˉ=a+∑(x−a)n\bar{x} = a + \dfrac{\sum (x - a)}{n}
Variance from shifted sumss2=1n−1[∑(x−a)2−(∑(x−a))2n]s^2 = \dfrac{1}{n-1}\left[\sum (x-a)^2 - \dfrac{(\sum (x-a))^2}{n}\right]

Sampling Language

  • A population has true mean μ \mu and variance σ2 \sigma^2 .
  • A simple random sample of size n n is one where every size-n n subset is equally likely.
  • From a sample x1,…,xn x_1, \dots, x_n , the key statistics are:
    • Sample mean xˉ=1n∑xi \bar{x} = \dfrac{1}{n} \sum x_i .
    • Unbiased sample variance s2=1n−1∑(xi−xˉ)2 s^2 = \dfrac{1}{n - 1} \sum (x_i - \bar{x})^2 .
  • Common data issues (good to mention even when not explicitly tested): undercoverage, non-response, and measurement bias.

Statistic-choice checkpoint

Before substituting a formula, decide whether the question is about a random sample mean or about estimating spread from the observed data. The symbols look similar, but they answer different questions.

Question cueQuantity to writeWhat it meansCommon trap
"A random sample of size nn is taken. Find the distribution of Xˉ\bar{X}."E(Xˉ)=μE(\bar{X}) = \mu, Var⁡(Xˉ)=σ2n\operatorname{Var}(\bar{X}) = \dfrac{\sigma^2}{n}The sample mean is a random variable before the sample is observed.Using s2s^2 when the population variance σ2\sigma^2 is supplied.
"The sample has ∑x\sum x and ∑x2\sum x^2. Find an unbiased estimate of the variance."s2=1n−1[∑x2−(∑x)2n]s^2 = \dfrac{1}{n - 1}\left[\sum x^2 - \dfrac{(\sum x)^2}{n}\right]You are estimating the population variance from observed data.Dividing by nn instead of n−1n - 1.
"Use the sample to test a claim about μ\mu."Start with xˉ\bar{x}, then use the sampling model requested by the question.The observed mean becomes evidence for a population claim.Treating xˉ\bar{x} as the true population mean.

Worked check: if σ2=25\sigma^2 = 25 and n=100n = 100, then Var⁡(Xˉ)=25/100=0.25\operatorname{Var}(\bar{X}) = 25/100 = 0.25. Do not divide by 9999, because this is not an unbiased-variance estimate from raw sample data.

Misconception check: s2s^2 estimates population variance after data are collected; σ2/n\sigma^2/n is the variance of the sample mean before observing the sample.


Distribution of the Sample Mean

  • If X1,…,Xn X_1, \dots, X_n are independent with mean μ \mu and variance σ2 \sigma^2 , then Xˉ=1n∑i=1nXi \bar{X} = \dfrac{1}{n} \sum_{i=1}^n X_i has
    • E(Xˉ)=μ E(\bar{X}) = \mu
    • Var⁡(Xˉ)=σ2n \operatorname{Var}(\bar{X}) = \dfrac{\sigma^2}{n} .
  • If the parent population is normal, then Xˉ \bar{X} is exactly normal for any n n .
  • By the Central Limit Theorem (CLT), for sufficiently large n n (often n≥30 n \geq 30 ), Xˉ \bar{X} is approximately N(μ,σ2/n) \mathcal{N}(\mu, \sigma^2 / n) even when the parent population is not normal.

Sample-Mean Model Checkpoint

Before using normalcdf or a z-score, decide why Xˉ \bar{X} can be treated as normal. This prevents the common error of using the right formula with an unsupported model.

Given in the questionModel for Xˉ \bar{X} What to write before calculating
Population is normalExactly normal for any n n Xˉ∼N(μ,σ2/n) \bar{X} \sim \mathcal{N}(\mu, \sigma^2/n) .
Population is not stated normal, but n n is largeApproximately normal by CLTBy CLT, Xˉ≈N(μ,σ2/n) \bar{X} \approx \mathcal{N}(\mu, \sigma^2/n) .
Population shape is unknown and n n is smallNormal model is not justified from the given informationDo not force a normal approximation unless the question supplies another reason.

Common trap: σ/n \sigma/\sqrt{n} is the standard deviation of the sample mean, not the variance. The variance is σ2/n \sigma^2/n .

Example -- Average revision hours

JC students have revision hours with μ=6.4 \mu = 6.4 , σ=1.8 \sigma = 1.8 . For a random sample of n=36 n = 36 , find P(Xˉ>7) P(\bar{X} > 7) .

  1. By the Central Limit Theorem, Xˉ≈N(6.4,1.82/36)=N(6.4,0.09) \bar{X} \approx \mathcal{N}(6.4, 1.8^2 / 36) = \mathcal{N}(6.4, 0.09) .
  2. Standardise: z=7−6.40.3=2.0 z = \frac{7 - 6.4}{0.3} = 2.0 .
  3. Probability P(Xˉ>7)=1−Φ(2.0)≈0.0228 P(\bar{X} > 7) = 1 - \Phi(2.0) \approx 0.0228 .

Example -- Sample size for desired precision (probability form)

Using a normal approximation, how large should n n be so that P(∣Xˉ−μ∣<0.5)≥0.95 P(\lvert \bar{X} - \mu \rvert \lt 0.5) \geq 0.95 when σ=2.4 \sigma = 2.4 and sampling is i.i.d.?

  1. Require P(∣Z∣<0.5σ/n)≥0.95⇒0.52.4/n≥z0.975=1.96 P\left( \lvert Z \rvert \lt \dfrac{0.5}{\sigma / \sqrt{n}} \right) \geq 0.95 \Rightarrow \dfrac{0.5}{2.4 / \sqrt{n}} \geq z_{0.975} = 1.96 .
  2. Solve for n n : n≥1.96×2.40.5=9.408⇒n≥89. \sqrt{n} \geq \dfrac{1.96 \times 2.4}{0.5} = 9.408 \Rightarrow n \geq 89.
  3. Round up to the next whole number: take n=89 n = 89 observations.

Probability-tail checkpoint

When a sampling question gives a probability statement and asks for a cut-off, mark the tail before using invNorm or a z-value. Most wrong answers use the correct standard error but choose the wrong side of the distribution.

Probability statementTail to markCut-off formCommon trap
P(Xˉ>a)=0.05P(\bar{X} \gt a)=0.05Right tail is 0.05, so left area is 0.95.a=μ+z0.95σna=\mu+z_{0.95}\dfrac{\sigma}{\sqrt{n}}Using z0.05z_{0.05}, which gives the lower 5 percent point.
P(Xˉ<a)=0.05P(\bar{X} \lt a)=0.05Left tail is 0.05.a=μ+z0.05σna=\mu+z_{0.05}\dfrac{\sigma}{\sqrt{n}}Making aa above the mean for a lower-tail probability.
P(∣Xˉ−μ∣<d)=0.95P(\lvert\bar{X}-\mu\rvert\lt d)=0.95Middle 0.95 leaves 0.025 in each tail.d=z0.975σnd=z_{0.975}\dfrac{\sigma}{\sqrt{n}}Using z0.95z_{0.95} for a two-sided interval.

Worked check: if Xˉ∼N(50,4/25)\bar{X}\sim\mathcal{N}(50, 4/25) and P(Xˉ>a)=0.05P(\bar{X}\gt a)=0.05, the standard error is 2/5=0.42/5=0.4. The right-tail probability 0.05 means z=1.645z=1.645, so a=50+1.645(0.4)≈50.7a=50+1.645(0.4)\approx50.7.

Misconception check: a small probability does not always mean a negative z-value. First decide whether the small area is on the left tail, right tail, or split across two tails.


Unbiased Estimates from Summarised Data

SEAB explicitly allows questions where the data are summarised as ∑x \sum x and ∑x2 \sum x^2 , or as ∑(x−a) \sum (x - a) and ∑(x−a)2 \sum (x - a)^2 (to reduce calculator rounding).

  • xˉ=∑xn \bar{x} = \dfrac{\sum x}{n} .
  • s2=1n−1[∑x2−(∑x)2n] s^2 = \dfrac{1}{n - 1} \left[\sum x^2 - \dfrac{(\sum x)^2}{n}\right] .
  • Shifted form (given a a ): xˉ=a+∑(x−a)n \bar{x} = a + \dfrac{\sum (x - a)}{n} and s2=1n−1[∑(x−a)2−(∑(x−a))2n] s^2 = \dfrac{1}{n - 1} \left[\sum (x - a)^2 - \dfrac{(\sum (x - a))^2}{n}\right] .

Shifted-sum checkpoint

When a question gives ∑(x−a)\sum (x-a) and ∑(x−a)2\sum (x-a)^2, do not try to reconstruct every raw xx value. Treat the shifted values as the working data, then add the shift back only for the mean.

StepWhat to doWhy it worksCommon trap
1Identify the shift aa.The question has centred the data around aa to keep sums smaller.Treating aa as the sample mean.
2Find the shifted mean: ∑(x−a)n\dfrac{\sum (x-a)}{n}.This is the average distance from aa.Reporting this shifted mean as xˉ\bar{x}.
3Add back the shift: xˉ=a+∑(x−a)n\bar{x}=a+\dfrac{\sum (x-a)}{n}.Every observation was written as aa plus a deviation.Adding aa again inside the variance formula.
4Use the shifted variance formula with divisor n−1n-1.Variance is unchanged by shifting all values by the same constant.Dividing by nn, or expanding to raw ∑x2\sum x^2 unnecessarily.

Worked check: if a=50a=50, n=20n=20, and ∑(x−a)=12\sum(x-a)=12, then the average shifted value is 12/20=0.612/20=0.6, so xˉ=50.6\bar{x}=50.6. The shift changes the mean calculation, but the spread calculation still uses deviations and the divisor n−1n-1.

Misconception check: shifting is a calculation shortcut, not a new data set with a different spread. Add aa back for the mean; do not add aa to s2s^2.

Example -- Using a shift

In a sample of n=40 n = 40 students, the data are summarised as ∑(x−8)=16 \sum (x - 8) = 16 and ∑(x−8)2=92 \sum (x - 8)^2 = 92 . Find xˉ \bar{x} and s2 s^2 .

  1. xˉ=8+1640=8.4 \bar{x} = 8 + \dfrac{16}{40} = 8.4 .
  2. s2=139[92−16240]=85.639=2.19 s^2 = \dfrac{1}{39} \left[92 - \dfrac{16^2}{40}\right] = \dfrac{85.6}{39} = 2.19 (3 s.f.).

Calculator Workflows

  • Use GC statistics mode (1-Var Stats) to obtain xˉ \bar{x} , s s , ∑x \sum x , and ∑x2 \sum x^2 (or the shifted sums if provided).
  • When Xˉ \bar{X} is modelled as normal, compute probabilities using normalcdf after converting to z z -scores.
  • If the question gives a probability and asks for a bound (e.g. find a a such that P(Xˉ>a)=0.05 P(\bar{X} > a) = 0.05 ), use invNorm on the Xˉ \bar{X} distribution (or solve via a=μ+zpσ/n a = \mu + z_p \sigma / \sqrt{n} ).

Exam Watch Points

  • Even though 6.4 is labelled “for teaching and learning only”, you are still expected to use E(Xˉ) E(\bar{X}) and Var⁡(Xˉ) \operatorname{Var}(\bar{X}) when doing 6.5 Hypothesis Testing.
  • Write the model line explicitly: “Xˉ∼N(μ,σ2/n) \bar{X} \sim \mathcal{N}(\mu, \sigma^2/n) ” (normal population) or “approximately normal by CLT” (large n n ).
  • For summarised data, use the unbiased variance with divisor n−1 n - 1 and show the substitution cleanly.
  • Keep rounding until the end; avoid premature rounding of s s when it appears inside s/n s/\sqrt{n} .

Practice Quiz

Apply sample-mean modelling, CLT standardisation, and unbiased-estimation workflows under exam pacing.


Quick Revision Checklist

  • Distinguish population vs sample parameters quickly (μ \mu vs Xˉ \bar{X} ).
  • Compute Var⁡(Xˉ) \operatorname{Var}(\bar{X}) and standard errors without mixing up S S and σ \sigma .
  • Convert probability statements about Xˉ \bar{X} into z z -scores (and back) cleanly.
  • Compute xˉ \bar{x} and s2 s^2 from ∑x \sum x , ∑x2 \sum x^2 or shifted sums without algebra slips.

Want weekly guided practice on Sampling? Our H2 Maths tuition programme builds fluency in this topic through structured problem sets and exam-style drills.


Common exam mistakes

  • Mixing up σ\sigma and ss: Using the population standard deviation σ\sigma in place of the sample standard deviation ss (or vice versa) is a common error. If σ\sigma is unknown, use s2s^2 with divisor n−1n - 1.
  • Forgetting to divide variance by nn: Writing Var⁡(Xˉ)=σ2\operatorname{Var}(\bar{X}) = \sigma^2 instead of Var⁡(Xˉ)=σ2/n\operatorname{Var}(\bar{X}) = \sigma^2/n is one of the most frequent algebraic slips in sampling questions.
  • Invoking CLT without stating “large nn”: The CLT approximation requires nn to be sufficiently large (typically n≥30n \geq 30); always write “by CLT, since nn is large” before using the approximate normal distribution.
  • Premature rounding of ss: Rounding the sample standard deviation ss too early before substituting into s/ns/\sqrt{n} introduces compounding errors. Keep full precision until the final answer.
  • Using the biased formula ∑(xi−xˉ)2/n\sum(x_i - \bar{x})^2 / n: SEAB requires the unbiased estimate with divisor n−1n - 1. Using nn in the denominator is wrong and will not earn the variance mark.

Frequently asked questions

Is there a formula sheet for H2 Maths sampling?
Yes - the "Formulas at a glance" section near the top of this page collects every result you need: the sample mean distribution, the Central Limit Theorem statement, the standard error formula, and both forms of the unbiased estimators (raw sums and shifted sums). Note that MF27 provides the unbiased variance estimator formula s2=1n−1[∑x2−(∑x)2n]s^2 = \dfrac{1}{n-1}\left[\sum x^2 - \dfrac{(\sum x)^2}{n}\right]; however, the sample mean distribution Xˉ∼N(μ,σ2/n)\bar{X} \sim \mathcal{N}(\mu, \sigma^2/n), the CLT approximation, and the standard error σ/n\sigma/\sqrt{n} are not listed in MF27 and must be recalled from memory.

Is Topic 6.4 directly tested in the exam?
Topic 6.4 is labelled “for teaching and learning only” in the official syllabus, so standalone questions specifically on sampling theory are unlikely. However, the key ideas - E(Xˉ)=μE(\bar{X}) = \mu, Var⁡(Xˉ)=σ2/n\operatorname{Var}(\bar{X}) = \sigma^2/n, and unbiased estimation - appear inside Topic 6.5 Hypothesis Testing questions, so you must know them well.

Can I use the GC to compute xˉ\bar{x} and s2s^2 from raw data?
Yes. Use 1-Var Stats on a TI or STAT mode on a Casio to obtain xˉ\bar{x} and sxs_x directly. For summarised data (given ∑x\sum x and ∑x2\sum x^2), show the manual substitution formula in working alongside the GC result.

When do I need to use the CLT versus the exact normal distribution?
If the original population is stated to be normal, Xˉ\bar{X} is exactly normal for any sample size nn. Use the CLT only when the population distribution is unknown or non-normal, and only when nn is large. Always state which case applies.


Other H2 Maths formula sheets

Revising more than one topic? Grab the matching one-page formula sheet:

For the official SEAB reference booklet, see the H2 Maths MF27 formula list.


Sources

Next steps: follow the H2 Maths notes hub into Topic 6.5 - Hypothesis testing and 6.6 - Correlation & regression for full Paper 2 practice.