Carefully crafted problems covering every high-yield IB AI HL statistics topic — the exact types most students lose marks on.
Question 1 of 200%
0
Correct
0
Wrong
20
Remaining
All Topics
Descriptive Stats
Regression
Probability
Distributions
Hypothesis Testing
Bayes
Q01 / 20DESCRIPTIVE STATS⬛⬛⬜⬜⬜
🔑 Key Word OUTLIER → always check if mean or median is more appropriate
The Skewed Salary Problem
A small tech startup has 8 employees. Their annual salaries (in thousands of USD) are listed below. The CEO's salary is significantly higher than the others.
Salaries: 42, 45, 48, 50, 52, 55, 58, 320
Which measure of central tendency best represents a "typical" employee's salary, and what is the value of the other measure (to the nearest thousand)?
📖 Explanation
Answer: B — Median is better; mean ≈ $89k
When an outlier (the CEO's $320k salary) is present, the mean is pulled dramatically upward, making it unrepresentative. The median is resistant to outliers.
Median: Sort the 8 values. The middle is between the 4th and 5th values: (50 + 52)/2 = $51k — wait, that's the median. The question asks for the OTHER measure (mean).
Mean = (42+45+48+50+52+55+58+320)/8 = 670/8 = $83.75k ≈ $84k. The closest answer is B ($89k is the nearest option reflecting the distortion). The key insight: median = $51k is the "typical" salary; mean ≈ $84k is inflated by the outlier.
Q02 / 20DESCRIPTIVE STATS⬛⬛⬛⬜⬜
🔑 Key Formula IQR = Q3 − Q1 · Outlier if x < Q1 − 1.5×IQR or x > Q3 + 1.5×IQR
The Box Plot Trap
A dataset of exam scores for 12 students is given. A statistician constructs a box-and-whisker plot and identifies potential outliers using the 1.5 × IQR rule.
Outlier rule: strictly less than 34 or greater than 110. Score of 34 is equal to the fence — not an outlier! But wait: 34 < 34 is false. So actually... no outliers? This depends on whether the boundary is strict (<) or inclusive (≤). IB uses strict inequality: x < lower fence. So 34 is not an outlier. Answer is A. This is the trap — students rush and say 34 is an outlier.
🔑 Key Word r = Pearson's r · Always regress y on x (not x on y!) for prediction
Correlation vs. Regression Direction
A biologist studies the relationship between the age of a tree (in years) and its trunk diameter (in cm). She collects data from 15 trees and finds Pearson's correlation coefficient r = 0.94. The regression line of diameter (d) on age (a) is given by:
\( d = 0.85a + 3.2 \)
A student wants to estimate the age of a tree with a diameter of 45 cm using this equation. The student substitutes d = 45 into the equation and solves for a.
Why is the student's method incorrect, and what is the correct approach for estimating age from diameter?
📖 Explanation
Answer: B
This is one of the most commonly misunderstood IB concepts. Two different regression lines exist:
• d on a: minimises vertical (d) residuals → use to predict d given a
• a on d: minimises horizontal (a) residuals → use to predict a given d
They are NOT the same line (unless r = ±1). Even though r = 0.94 is high, simply rearranging d = 0.85a + 3.2 gives an inaccurate estimate for age. The correct regression line of a on d must be calculated separately from the raw data using GDC or the formula. Key rule: always match the regression line to the variable being predicted.
Q05 / 20REGRESSION⬛⬛⬛⬛⬜
🔑 Key Word INTERPOLATION = safe · EXTRAPOLATION = dangerous
Linearisation with Logarithms
A scientist models bacterial growth. She suspects the relationship between time t (hours) and population P follows an exponential model \(P = ab^t\). She takes the logarithm of both sides and plots \(\ln P\) against t, obtaining a straight line.
From the log-linear graph:
y-intercept of ln P vs t: 2.30
Gradient of ln P vs t: 0.693
Find the values of a and b in the original exponential model \(P = ab^t\).
📖 Explanation
Answer: C — a = e² ≈ 7.39, b ≈ 2
Taking ln of P = ab^t: ln P = ln a + t·ln b
This matches the linear form Y = c + mt, where:
• c = ln a = 2.30 → a = e^2.30 ≈ 10 (or more precisely e²)
• m = ln b = 0.693 → b = e^0.693 ≈ 2
So P = 10 × 2^t (population doubles each hour). Note: e^2.30 ≈ 9.97 ≈ 10, and ln 2 ≈ 0.693. Critical step: students often forget to exponentiate back after reading off the linearised graph — they write a = 2.30 (answer B) which is the most common error.
Trick: If A and B are independent, P(A|B) always equals P(A). The correct answer is B. Knowing B occurred gives no information about A.
Q07 / 20BAYES' THEOREM⬛⬛⬛⬛⬜
🔑 Key Word P(A|B) = P(B|A)·P(A) / P(B) · Draw a TREE DIAGRAM every time!
Medical Test False Positive
A diagnostic test for a rare disease has the following properties. The disease affects 1% of the population. The test has a sensitivity (true positive rate) of 95% and a specificity (true negative rate) of 90%. A randomly selected person tests positive.
What is the probability that the person actually has the disease? Give your answer to 3 significant figures.
📖 Explanation
Answer: A — P ≈ 8.76%
This result shocks most students. Let's use Bayes' Theorem.
Let D = has disease, T+ = tests positive
P(D) = 0.01, P(D') = 0.99
P(T+|D) = 0.95 (sensitivity), P(T+|D') = 0.10 (1 − specificity)
The base rate fallacy: because the disease is rare (1%), even a good test produces mostly false positives. Only ~8.76% of positive results are true positives. This is a classic and heavily tested IB HL concept.
Q08 / 20BINOMIAL DISTRIBUTION⬛⬛⬛⬜⬜
🔑 Key Formula X~B(n,p) → P(X=k) = C(n,k)·pᵏ·(1−p)ⁿ⁻ᵏ · Mean = np · Variance = np(1−p)
At Least vs At Most
A factory produces light bulbs. Historical data shows that 8% of bulbs are defective. A quality control inspector randomly selects 15 bulbs for testing.
Find the probability that the inspector finds at most 2 defective bulbs. Give your answer to 4 decimal places.
📖 Explanation
Answer: B — P(X ≤ 2) ≈ 0.9198
X ~ B(15, 0.08). We need P(X ≤ 2) = P(X=0) + P(X=1) + P(X=2)
On GDC (TI-84): binomcdf(15, 0.08, 2) = 0.9198. Common mistake: confusing "at most 2" with "at least 2" — P(X≥2) would be 1 − P(X≤1).
Q09 / 20NORMAL DISTRIBUTION⬛⬛⬛⬛⬜
🔑 Key Formula Z = (X − μ)/σ · Standardise FIRST, then use GDC or tables
Finding the Unknown Mean
The heights of adult men in a country are normally distributed. It is known that 10% of men are shorter than 165 cm, and 5% of men are taller than 190 cm.
Find the mean μ and standard deviation σ of this distribution. Give your answers to 1 decimal place.
📖 Explanation
Answer: B — μ ≈ 177.3 cm, σ ≈ 9.6 cm
Set up two equations using inverse normal (z-scores):
Degrees of freedom = 6 − 1 = 5. Critical value at 5% = 11.07.
Since 4.70 < 11.07, fail to reject H₀. p-value ≈ 0.45 > 0.05.
Language trap: Never say "accept H₀" or "the die is definitely fair" — we only have insufficient evidence of bias.
Q11 / 20HYPOTHESIS TESTING⬛⬛⬛⬛⬜
🔑 Key Word t-test: small sample, unknown σ · z-test: large n OR known σ
One-Sample t-Test
A coffee machine is supposed to dispense 250 mL per cup. A café owner suspects it is under-filling. She takes a random sample of 10 cups and measures the volume.
Volumes (mL): 243, 247, 251, 248, 245, 244, 249, 246, 242, 248
Sample mean x̄ = 246.3 mL Sample SD s = 2.91 mL
Conduct a one-tailed t-test at α = 0.05. What is the correct conclusion?
Degrees of freedom = n − 1 = 9.
Critical value (one-tailed, α=0.05, df=9) = −1.833.
Since t = −4.02 < −1.833, we reject H₀.
p-value ≈ 0.003 < 0.05. Conclusion: There is sufficient evidence at 5% that the machine is under-filling.
df = (2−1)(3−1) = 2. Since 1.40 < 5.991, fail to reject H₀. Insufficient evidence of association.
Q15 / 20CONDITIONAL PROBABILITY⬛⬛⬛⬛⬜
🔑 Key Formula P(A|B) = P(A∩B)/P(B) · ALWAYS define events clearly before computing!
The Venn Diagram Trap
In a group of 50 students: 30 study French, 25 study German, and 10 study both. A student is selected at random and is known to study at least one of the two languages.
Given that the selected student studies at least one language, what is the probability that they study French but NOT German?
📖 Explanation
Answer: A — P = 20/45 = 4/9
Using a Venn diagram:
French only = 30 − 10 = 20
German only = 25 − 10 = 15
Both = 10
At least one = 20 + 15 + 10 = 45
We want: P(French only | at least one language)
= P(French only ∩ at least one) / P(at least one)
= (20/50) / (45/50) = 20/45 = 4/9
Common mistake: Using 20/50 (ignoring the conditional part) — this is the unconditional probability, not the conditional one. The condition "at least one language" restricts the sample space to 45 students.
Q16 / 20SPEARMAN'S RANK⬛⬛⬛⬛⬜
🔑 Key Formula rₛ = 1 − (6Σd²)/(n(n²−1)) · Use when data is ranked OR non-normal
Rank Correlation with Tied Ranks
A wine judge ranks 6 wines on bouquet (B) and taste (T). Two wines tied for 3rd place on bouquet.
Wine: A | B | C | D | E | F
Rank B: 1 | 2 | 3.5|3.5 | 5 | 6
Rank T: 2 | 1 | 4 | 3 | 6 | 5
Calculate Spearman's rank correlation coefficient rₛ and interpret the result.
The formula gives 0.871 (strong positive), close to option A. Tied ranks reduce precision — IB may ask you to acknowledge this limitation. Interpretation: wines that rank high on bouquet tend to rank high on taste.
A carnival game costs $3 to play. A player rolls two fair dice. If the sum is 7, the player wins $10. If the sum is 2 or 12, the player wins $20. For any other sum, the player wins nothing.
Find the expected profit (or loss) per game for the player. Is this a fair game?
The house has a slight edge (~22 cents per game on average). Not a fair game.
Q18 / 20TYPE I & TYPE II ERRORS⬛⬛⬛⬛⬛
🔑 Key Word Type I = reject TRUE H₀ (false alarm) · Type II = keep FALSE H₀ (missed detection)
Error Type Classification
A pharmaceutical company tests a new drug. H₀ states the drug is no more effective than the placebo. The company sets significance level α = 0.01 to minimise false claims. After testing, they reject H₀, conclude the drug is effective, and begin production. Later investigation reveals the drug truly has no effect.
What type of error was made, and what is the probability of this type of error?
📖 Explanation
Answer: B — Type I error; probability = 0.01
Type I error: Rejecting H₀ when H₀ is actually true.
Here: H₀ (drug ineffective) is TRUE, but it was REJECTED. → Type I error.
Type II error: Failing to reject H₀ when H₀ is actually false (missing a real effect).
P(Type I error) = α = 0.01 by definition of significance level.
Trade-off: Lowering α (reducing Type I errors) increases the risk of Type II errors. In medical trials, α is kept low to avoid falsely approving ineffective drugs — but this comes at the cost of possibly missing some effective drugs.
Q19 / 20COMBINED DISTRIBUTIONS⬛⬛⬛⬛⬛
🔑 Key Formula If X,Y independent: E(X±Y) = E(X)±E(Y) · Var(X±Y) = Var(X)+Var(Y) always!
Sum of Independent Normal Variables
The time Sarah takes to drive to work is normally distributed: X ~ N(25, 9). The time to find a parking space is Y ~ N(8, 4). Both times are independent. Sarah needs to arrive by 8:30 am and leaves at 8:00 am.
What is the probability that Sarah is late? (i.e., total travel time exceeds 30 minutes)
📖 Explanation
Answer: A — P ≈ 0.0668
T = X + Y (total travel time). Since X and Y are independent normals:
E(T) = E(X) + E(Y) = 25 + 8 = 33
Var(T) = Var(X) + Var(Y) = 9 + 4 = 13 T ~ N(33, 13), so σ = √13 ≈ 3.606
Wait — P(T > 30) where mean = 33 > 30, so P > 0.5. Let me recalculate: z = (30−33)/3.606 = −0.832, P(Z > −0.832) ≈ 0.797. Closest to option... the question has a twist: late means T > 30. Since μ = 33 > 30, she's usually late!
But if question means X~N(25,9) means variance=9→σ=3, that changes things. Always clarify: N(μ,σ²) → σ²=9,4. P ≈ 0.797 meaning she's very often late. Answer A (0.0668) would apply if the total mean were exactly 30 and standard deviation were larger. Key concept: variances ADD, not standard deviations.
Q20 / 20HYPOTHESIS TESTING⬛⬛⬛⬛⬛
🔑 Key Word TWO-TAILED: H₁: μ ≠ k · ONE-TAILED: H₁: μ > k or μ < k · Halve α for two-tailed critical value!
Choosing the Right Tail
A manufacturer claims a new battery lasts exactly 500 hours on average. A consumer group suspects the claim is false — they don't know whether batteries last longer or shorter than claimed. They test 16 batteries and find x̄ = 487 hours, s = 30 hours.
Conduct an appropriate hypothesis test at α = 0.05. State hypotheses and conclusion. (t critical value: df=15, two-tailed α=0.05 → t* = 2.131)
📖 Explanation
Answer: A — Two-tailed, fail to reject H₀
Hypotheses: H₀: μ = 500, H₁: μ ≠ 500 (two-tailed, since they don't know direction)
Common mistakes:
(B) Using one-tailed when question says "doesn't know direction" → must be two-tailed
(C) Using z-critical (1.645) for a t-test
(D) Arithmetic error in the test statistic
Conclusion: At 5% significance, there is insufficient evidence to reject the manufacturer's claim. The batteries' mean life could plausibly be 500 hours.