Mean uses all data; median resists outliers; mode for categories
Spread
Range is crude; IQR resists outliers; SD uses every value
Probability rules
AND multiply (independent), OR add (mutually exclusive), P(not) = 1 − P
Permutations vs combinations
Order matters → permute; order irrelevant → combine
Hypothesis testing
Null until evidence: test statistic vs critical value — then conclude in CONTEXT
Correlation
Association is NOT causation — the third variable lurks
1. Averages — which one, and why
Key idea
The three averages
Mean = sum ÷ count (every value matters — outliers DRAG it). Median = the middle value (ORDER the data first — the step candidates skip). Mode = the most frequent value. The exam's question is never 'calculate all three' — it is 'WHICH average best represents this data, and why'.
Example
Worked: wages 2, 3, 3, 5, 27 (thousand) → mean 8 (misleading — the outlier drags it), median 3 (the middle), mode 3. Skewed data with outliers → quote the MEDIAN and say why: the mean is pulled by the extreme value. Ungrouped median with an even count: average the middle two.
2. Spread — range, IQR, standard deviation
Key idea
Measures of spread
Range = max − min (one outlier distorts it). Interquartile range IQR = Q3 − Q1 — the spread of the middle 50%, robust to outliers. Standard deviation = the average distance from the mean (small σ = clustered tight). Box plots display all five numbers: min, Q1, median, Q3, max.
Example
Worked: 1, 3, 5, 7, 9 → median 5, Q1 = 2, Q3 = 8 → IQR = 6. Two classes, same mean 60: one clustered (σ = 4), one spread (σ = 15) — same average, completely different consistency. Comparing CONSISTENCY is an IQR or σ question, never a range question.
3. Averages from tables — grouped data
Grouped data hides exact values — use the MIDPOINT of each class × its frequency. Estimated mean = Σfx ÷ Σf; the modal class is the one with the highest frequency. The exam word is ESTIMATED: grouped data can only approximate — call it that for the precision mark.
Example
Worked: classes 0–10 (f = 4, midpoint 5), 10–20 (f = 6, midpoint 15) → estimated mean = (4×5 + 6×15) ÷ 10 = 110 ÷ 10 = 11. Every class contributes midpoint × frequency — add the column, divide by Σf. Writing the fx column IS the method mark.
4. Probability — the scale and the complement
Key idea
Probability
P(event) = favourable outcomes ÷ total outcomes — always between 0 (impossible) and 1 (certain). P(not A) = 1 − P(A): the complement rule — often the fastest route to an answer. Probabilities of all possible outcomes sum to exactly 1. Experimental probability (relative frequency) ESTIMATES the theoretical value and converges with more trials.
Example
Worked: a fair die: P(prime) = 3/6 = 1/2 (2, 3, 5); P(not even) = 1 − 1/2 = 1/2. A die rolls a six 87 times in 600 throws → 87/600 = 0.145 vs theoretical 1/6 ≈ 0.167 — close, converging. 'Estimate the probability' means divide the counts — 'theoretical' means use the fractions.
5. AND and OR — the two laws
Key idea
The laws
INDEPENDENT events joined by AND: MULTIPLY — P(A and B) = P(A) × P(B). MUTUALLY EXCLUSIVE events joined by OR: ADD — P(A or B) = P(A) + P(B) (they cannot both happen). The general OR with overlap: P(A ∪ B) = P(A) + P(B) − P(A ∩ B) — SUBTRACT the overlap once (the double-count is the classic trap).
Example
Worked: two coins: P(two heads) = 1/2 × 1/2 = 1/4. Die and coin: P(six AND heads) = 1/6 × 1/2 = 1/12. A card: P(even OR prime) on a die = 1/2 + 1/2 − 1/6(only 2 is both... 2 is even and prime — P = 1/6) = 5/6? Check: even = {2,4,6}, prime = {2,3,5}, union = {2,3,4,5,6} → 5/6 ✓.
6. Tree diagrams — multiply along, add across
A tree diagram maps every outcome: probabilities MULTIPLY along a single path (the branches are sequential ANDs) and ADD across different paths leading to the same event. First-branch probabilities always sum to 1; SECOND-branch probabilities CHANGE when events are not independent — without replacement, the bag shrinks (the trap: forgetting the denominator changed).
Example
Worked: bag with 5 red, 3 blue; two drawn WITHOUT replacement: P(RR) = 5/8 × 4/7 = 20/56 = 5/14. P(one of each) = (5/8 × 3/7) + (3/8 × 5/7) = 15/56 + 15/56 = 30/56 = 15/28 — two paths, added. The 4/7 on the second branch is the whole question: the bag has one fewer red.
7. Conditional probability — given that
Key idea
P(B | A)
The probability of B GIVEN A has already happened: P(B|A) = P(A and B) ÷ P(A) — the sample space has SHRUNK to A. Read it straight off a tree's second branch, or off a Venn's overlap ÷ the A circle. Independence test: P(B|A) = P(B) means A tells you nothing about B.
Example
Worked: from the bag above: P(blue | first was red) = 3/7 — direct from the reduced bag (8 − 1 = 7 left, 3 blue unchanged). Venn: P(both) ÷ P(first) = (5/8 × 3/7) ÷ (5/8) = 3/7 ✓ — the formula and the tree agree, as they must.
8. Combinations and permutations — order or not
Tool
Order?
Formula
Worked
Permutation nPr
YES — arrangements
n!/(n − r)!
Arrange 4 of 7 books: 7P4 = 840
Combination nCr
NO — selections
n!/(r!(n − r)!)
Choose 2 of 5: 5C2 = 10
Committee
NO
nCr
Committee of 3 from 8: 8C3 = 56
The exam's one-line discriminator: does REARRANGING the chosen items make a different outcome? Arranging books on a shelf — yes (permutation). Choosing a committee — no (combination). nCr = nPr ÷ r!: every combination is a permutation divided by its own internal orderings.
9. The binomial distribution — fixed trials, two outcomes
Key idea
Binomial
n independent trials, constant success probability p, two outcomes (success/failure): P(X = r) = nCr × p^r × (1 − p)^(n−r). The expected value is simply E(X) = np — the average you would see over many repetitions.
Example
Worked: 10 coin flips, p = 1/2: P(exactly 6 heads) = 10C6 × (1/2)⁶ × (1/2)⁴ = 210 × (1/2)¹⁰ = 210/1024 ≈ 0.205. Expected heads in 10 flips: E(X) = 10 × 1/2 = 5. The three conditions (fixed n, constant p, two outcomes) must be stated — they are the mark before the formula.
10. The normal distribution — the bell curve
Key idea
68–95–99.7
The normal distribution is symmetric about the mean μ, bell-shaped, defined by μ and σ. The rule that answers most questions without tables: 68% of data within ±1σ, 95% within ±2σ, 99.7% within ±3σ. Standardise with z = (x − μ)/σ to use tables or a calculator.
Example
Worked: heights with μ = 60 kg, σ = 8: P(heavier than 76) = P(z > 2) ≈ 2.5% (the tail beyond 95%). P(between 52 and 68) = 68%. The symmetry trick: P(X > μ) = 0.5 always — half the bell, no calculation needed.
11. Sampling — estimates and bias
A RANDOM sample (every member equally likely) estimates the population fairly; a biased sample does not — the exam wants the bias NAMED and its effect explained: 'only morning shoppers' misses evening habits. Capture-recapture: marked/total = recaptured marked/second sample — proportional reasoning in one line.
Example
Worked: 40 fish tagged and released; a second catch of 60 contains 8 tagged → population ≈ 40 × 60 ÷ 8 = 300. The assumption: mixing is complete and the proportion holds — state it for the mark. Biased sampling question: 'a survey of a smartphone app' misses offline users — name who is excluded and why the estimate skews.
12. Correlation and regression — the eternal trap
Key idea
Correlation ≠ causation
Scatter graphs show correlation: positive (both rise), negative (one falls), none. A line of best fit y = mx + c predicts — INTERPOLATION (inside the data) is reliable, EXTRAPOLATION (outside) is not. And correlation never PROVES causation: ice-cream sales correlate with drownings — summer causes both, ice cream causes neither.
Example
Worked: regression y = 2x + 10 from data spanning x = 1 to 10: predict at x = 5 → 20 (interpolating — reliable); at x = 50 → 110 (extrapolating — the model is untested there, say so). The confounder explanation is the essay mark: name the hidden third variable.
13. Summary — the statistics spine
Question type
Method beat
Trap
Averages
Order data first; which one and why
Outliers drag the mean
Spread
IQR for consistency, σ for clustering
Range distorts with outliers
Probability
P = favourable/total; complement shortcut
Probabilities never exceed 1
AND / OR
Multiply independent ANDs; add exclusive ORs
Subtract the overlap once
Trees
Multiply along, add across
Without replacement changes denominators
Counting
Rearranging matters? nPr : nCr
Committees are combinations
Normal
68–95–99.7; z = (x − μ)/σ
P(X > μ) = 0.5 by symmetry
Correlation
Interpolate yes, extrapolate no
Correlation is not causation
Note
Command discipline: 'calculate' = formula, substitution, sensible rounding; 'estimate' = counts divided or midpoints × frequency; 'comment on' = name the outlier/bias/confounder and its effect — the comment IS the mark. Statistics papers reward the reasoning sentence as much as the number.
Spotted a problem with this content? Sign in to send it to our editors.