IB Mathematics AI

Higher Level

Topic 4: Statistics & Probability

Complete HL Study Guide

Section Topics Covered Level

Statistics (4.1–4.3) Sampling, data presentation, mean, SD, IQR SL

Correlation (4.4, 4.10) Pearson r, regression line, Spearman r■ SL

Probability (4.5–4.7) Combined events, conditional, discrete RVs SL

Distributions (4.8–4.9) Binomial, Normal distributions SL

Hypothesis Testing (4.11) Chi-square, t-test, p-values SL

HL: Reg. & Transforms (4.13–4.14) Non-linear regression, R², linear combos HL

HL: CLT & CI (4.15–4.16) Central Limit Theorem, confidence intervals HL

HL: Poisson (4.17) Poisson distribution, mean = variance = m HL

HL: Critical Regions (4.18) Type I/II errors, critical values HL

HL: Markov Chains (4.19) Transition matrices, steady state HL

 1. Statistics

SL 4.1 — Populations, Samples & Outliers

Population = the entire group being studied. Sample = a smaller subset used for analysis. A random
sample gives every member an equal chance of selection, reducing bias.

Discrete data: countable whole values (e.g. number of students). Continuous data: any value in a range
(e.g. height, temperature).

Outlier definition (must memorise!):

Outlier if: x < Q1 - 1.5 x IQR OR x > Q3 + 1.5 x IQR

Outliers may be valid data points or recording errors — context decides which. Mark outliers with a cross
(x) on box-and-whisker diagrams.

Sampling methods:
Method How it works Key advantage

Simple random Every member equally likely — e.g. lottery draw Eliminates bias

Systematic Every nth member from a list Easy to implement

Stratified Population split into groups; sample proportionally from each

Represents all groups

Quota Like stratified but non-random within groups Faster than stratified

Convenience Whoever is easiest to reach Quick but most biased

SL 4.2 — Data Presentation

Frequency histograms: use equal class intervals; no gaps for continuous data. Class intervals given as
inequalities, e.g. 10 ≤ x < 20. Frequency density histograms are NOT required.
Cumulative frequency graphs: plot running totals against upper class boundaries. Read off the median
at 50%, Q1 at 25%, Q3 at 75%, and any percentile.

Box-and-whisker diagrams: display minimum, Q1, median, Q3, maximum. Use two box plots
side-by-side to compare distributions — compare symmetry, median, and IQR.

Tip: If a box plot is roughly symmetric with the median near the centre, the data may be normally distributed.

SL 4.3 — Measures of Central Tendency & Dispersion

Mean from grouped data — use mid-interval values (midpoint of each class):

x-bar = (Sum of f_i * x_i) / n, where n = Sum of f_i

Modal class = the class with the highest frequency (for equal class intervals only).

IQR = Q3 - Q1. Use your GDC for standard deviation and variance. Variance = (standard deviation)2.

Effect of constant changes on data:

Subtract constant c from all values Add constant c to all values
Mean decreases by c; SD unchanged Mean increases by c; SD unchanged

Multiply all values by k Divide all values by k

Mean x k; SD x k Mean / k; SD / k

Tip: Adding or subtracting a constant shifts the mean but does NOT change spread. Multiplying/dividing scales
BOTH mean and standard deviation.
 2. Correlation & Regression

SL 4.4 — Linear Correlation & Regression Line

Pearson's product-moment correlation coefficient r measures the strength and direction of a linear
relationship between two variables. Always use technology to calculate r.

Value of r Interpretation

r=1 Perfect positive linear correlation

r = -1 Perfect negative linear correlation

r=0 No linear correlation

0.7 to 1 Strong positive correlation

-1 to -0.7 Strong negative correlation

0.3 to 0.7 Moderate positive correlation

0 to 0.3 Weak correlation

Regression line y = ax + b: find using your GDC. The line always passes through the mean point (x-bar,
y-bar). Interpret: a = rate of change (slope), b = y-intercept.

Tip: Correlation does NOT imply causation. Extrapolation (predicting outside the data range) is unreliable — avoid it.

SL 4.10 — Spearman's Rank Correlation

Spearman's rank correlation coefficient r_s measures ANY monotonic relationship (not just linear). Find
using technology. If data items are tied, average their ranks.

Pearson r is best when... Spearman r_s is best when...

Testing only for linear relationships Testing any monotonic relationship

Outliers affect... Neither requires...

Pearson r more than Spearman r_s Derivation or proof in exams
 3. Probability

SL 4.5 — Basic Probability

P(A) = n(A) / n(U)

n(A) = number of outcomes in event A, n(U) = total outcomes in sample space U.

Complementary events: P(A) + P(A') = 1

Expected number of occurrences = P(event) x number of trials. Example: 128 students, P(absent) = 0.1,
expected absences = 12.8.

SL 4.6 — Combined & Conditional Probability

Combined events (general) Mutually exclusive events
P(A u B) = P(A) + P(B) - P(A n B) P(A n B) = 0

Conditional probability Independent events

P(A|B) = P(A n B) / P(B) P(A n B) = P(A) * P(B)

Use Venn diagrams, tree diagrams, sample space diagrams, or tables to organise problems. "Or" in
probability is non-exclusive — A or B or both.

Tip: For conditional probability, P(A|B) means "given B has happened, what is the probability of A?" The denominator
P(B) restricts the sample space to only outcomes where B occurred.

SL 4.7 — Discrete Random Variables

A discrete random variable X has a probability distribution — a table listing each value of X and its
probability P(X = x). All probabilities must sum to 1.

E(X) = Sum of x_i * P(X = x_i)

E(X) is the expected value (mean). If E(X) = 0, the game is fair. Applications include gambling, insurance
pricing, and decision theory.
 4. Probability Distributions

SL 4.8 — Binomial Distribution

Use the binomial distribution when: (1) fixed number of trials n, (2) only two outcomes (success/failure), (3)
constant probability p, (4) trials are independent.

X ~ B(n, p)

Mean Variance
E(X) = np Var(X) = np(1-p)

Use your GDC for all binomial probability calculations. Formal proof of mean and variance is NOT
required.

Tip: Binomial links to expected number of occurrences from SL 4.5 — if X ~ B(n,p), the expected number of
successes is np.

SL 4.9 — Normal Distribution

The normal distribution is bell-shaped and symmetric about the mean mu. It occurs naturally for many
real-world measurements.

X ~ N(mu, sigma^2)

Range Approximate % of data

mu ± sigma ~68%

mu ± 2*sigma ~95%

mu ± 3*sigma ~99.7%

Use technology for all normal probability calculations and inverse normal calculations. For inverse normal:
mean and standard deviation will be given. Transformation to the standardised z-variable is NOT required.
 5. Hypothesis Testing

SL 4.11 — Chi-Square Tests & t-test

Step 1 — State hypotheses: H_0 (null) and H_1 (alternative), expressed as equations, inequalities, or in
words.

Step 2 — Choose significance level: typically 1%, 5%, or 10%.

Step 3 — Run the test on GDC: obtain the p-value and chi-square statistic.

Step 4 — Decision: if p-value < significance level, reject H_0. Otherwise, do not reject H_0.

Chi-Square Test for Independence:

Used to test whether two categorical variables are independent. Set up a contingency table. Maximum 4
rows or 4 columns in exams. Expected frequencies must be greater than 5.

Degrees of freedom = (rows - 1)(columns - 1)

Chi-Square Goodness of Fit Test:

Tests whether observed data fits a specific distribution. Degrees of freedom = n - 1 at SL (subtract 1 extra
for each estimated parameter at HL). Expected frequencies must be > 5 — combine categories if needed.

The t-test:
Compares means of two populations. Use pooled two-sample t-test (assumes equal variance). The
underlying distribution must be normal. Samples will be unpaired; variance unknown at SL.

Tip: Always interpret the result in context: e.g. "There is sufficient evidence at the 5% significance level to conclude
that the variables are not independent."
 6. HL Only Content (4.12–4.19)

AHL 4.12 — Designing Data Collection Methods

Design valid questionnaires: questions should be unbiased, structured with consistent answer choices,
and use precise wording. Select relevant variables from many options.

For chi-square goodness of fit: choose categories appropriately so expected frequencies > 5. Degrees of
freedom: df = (number of categories - 1) - (number of estimated parameters).

Reliability tests Validity tests

Test-retest, parallel forms Content validity, criterion-related validity

AHL 4.13 — Regression of Non-Linear Functions & R²

Fit curves (quadratic, cubic, exponential, power, sine) using least squares regression on your GDC. The
sum of square residuals SS_res measures how well the model fits — smaller SS_res = better fit.

R^2 = 1 - SS_res / SS_tot

R² = 1 means a perfect fit; R² = 0 means the model explains no variability. For linear models, R² = r² (the
square of Pearson's r). R² alone should NOT be the only criterion for choosing between models.

AHL 4.14 — Linear Transformations & Linear Combinations

Single random variable X:

E(aX + b) = aE(X) + b

Var(aX + b) = a^2 * Var(X)

Linear combinations of n independent random variables:

E(a1*X1 + a2*X2 + ... + an*Xn) = a1*E(X1) + a2*E(X2) + ... + an*E(Xn)

Var(a1*X1 + a2*X2 + ... + an*Xn) = a1^2*Var(X1) + a2^2*Var(X2) + ... + an^2*Var(Xn)

(Only valid when variables are INDEPENDENT — variance formula does not hold otherwise.)

Unbiased estimates:

x-bar = (1/n) * Sum of x_i [unbiased estimate of mu]

s^2_(n-1) = (n / (n-1)) * s^2_n [unbiased estimate of sigma^2]

AHL 4.15 — Central Limit Theorem & Normal Combinations

If X ~ N(mu, sigma²), then the sample mean X-bar based on a sample of size n satisfies:

X-bar ~ N(mu, sigma^2 / n)

Central Limit Theorem: for large n (n > 30 is sufficient in exams), X-bar approaches a normal distribution
regardless of the original distribution.

Tip: The standard error of the mean is sigma / sqrt(n) — larger samples give more precise estimates of the
population mean.

AHL 4.16 — Confidence Intervals for the Mean

A confidence interval gives a range of plausible values for the true population mean mu. Interpret as: "We
are X% confident that the true mean lies in this interval."

sigma known sigma unknown

Use normal distribution (z-interval) Use t-distribution (t-interval), regardless of
sample size

AHL 4.17 — Poisson Distribution

Use Poisson when: (1) events are independent, (2) events occur at a uniform average rate.

X ~ Po(m)

Mean Variance
E(X) = m Var(X) = m (same as mean!)

The sum of two independent Poisson distributions is also Poisson: if X ~ Po(a) and Y ~ Po(b), then X + Y ~
Po(a + b).

Tip: Key identifier: mean = variance is the hallmark of a Poisson distribution. Choose between Normal, Binomial, and
Poisson based on the context given.

AHL 4.18 — Critical Values, Critical Regions & Error Types

The critical region is the set of values that lead to rejection of H_0. The critical value is the boundary of
this region.

Type I error Type II error

Reject H_0 when H_0 is actually true (false Fail to reject H_0 when H_0 is actually false
positive) (false negative)

P(Type I) P(Type II)

= significance level alpha Calculated from the actual distribution

Test type Notes

Normal population mean (sigma known) Use z-distribution; matched pairs = single sample

Normal population mean (sigma unknown) Use t-distribution, any sample size

Poisson population mean One-tailed only

Binomial proportion One-tailed only

pmcc = 0 (bivariate normal) Data given in exam; use technology

 Tip: For discrete distributions, the critical region maximises the probability of a Type I error while keeping it at or
below the stated significance level.

AHL 4.19 — Transition Matrices & Markov Chains

A transition matrix T gives probabilities of moving between states. Entry T_ij = P(moving from state j to
state i). Each column sums to 1.

State after n steps: s_n = T^n * s_0

s_0 is the initial state probability vector. Use transition diagrams to represent discrete dynamical systems.

Steady state: as n → infinity, the state vector converges to a fixed vector s_inf. Find it by solving T * s_inf
= s_inf with all entries summing to 1, OR by multiplying T by itself many times.

Tip: The steady state is the eigenvector of T corresponding to eigenvalue 1, scaled so its entries sum to 1. Links to
eigenvalues (AHL 1.15) and adjacency matrices (AHL 3.15).
 7. Formula Reference Sheet

Reference Formula / Rule

4.2 IQR IQR = Q3 - Q1

4.3 Grouped mean x-bar = Sum(f_i * x_i) / n

4.5 Probability P(A) = n(A) / n(U)

4.5 Complement P(A) + P(A') = 1

4.6 Combined events P(A u B) = P(A) + P(B) - P(A n B)

4.6 Mutually exclusive P(A n B) = 0

4.6 Conditional P(A|B) = P(A n B) / P(B)

4.6 Independent P(A n B) = P(A) * P(B)

4.7 Expected value E(X) = Sum of x_i * P(X = x_i)

4.8 Binomial mean E(X) = np

4.8 Binomial variance Var(X) = np(1-p)

4.13 R-squared R^2 = 1 - SS_res / SS_tot

4.14 Linear transform E E(aX + b) = aE(X) + b

4.14 Linear transform Var Var(aX + b) = a^2 * Var(X)

4.14 Unbiased variance s^2_(n-1) = (n/(n-1)) * s^2_n

4.15 Sample mean dist. X-bar ~ N(mu, sigma^2 / n)

4.17 Poisson mean E(X) = m

4.17 Poisson variance Var(X) = m

4.19 Markov chains s_n = T^n * s_0

Key things NOT required in exams:

• Frequency density histograms (SL 4.2)
• Formal proof of mean and variance for Binomial or Poisson distributions
• Transformation to standardised z-variable for Normal distribution
• Derivation of Pearson's r or Spearman's r_s
• Calculating critical regions for t-tests
• Derivation of R² formula

Good luck on your test! Remember: always show your GDC method, interpret results in context, and state
conclusions clearly.