Exploring data, designing studies, and using probability to draw statistical conclusions.
Study guide last reviewed September 25, 2026 · see what changed
Format reflects the current published structure — always confirm against AP Central before test day, since formats occasionally change.
| Section | Details | Time | Weight |
|---|---|---|---|
| Section I – Multiple Choice | 40 questions | 1 hr 30 min | 50% |
| Section II – Free Response | 5 short + 1 investigative task | 1 hr 30 min | 50% |
Straight answers about the AP Statistics exam — format, credit, dates, and this site.
Section I – Multiple Choice: 40 questions (1 hr 30 min, 50% of the total score); Section II – Free Response: 5 short + 1 investigative task (1 hr 30 min, 50% of the total score). See the Exam Format tab above for the full breakdown, and always double-check the current year's format on AP Central before test day.
It's multiple-choice questions and written free-response or essay questions.
It depends entirely on the college — most award credit for a 3 or higher, but many selective schools require a 4 or a 5, and some don't grant credit at all for certain scores. Always check your target college's own AP credit policy page rather than assuming a 3 will count everywhere.
AP exams are administered during the College Board's annual testing window in the first two full weeks of May, with the exact date and time for each subject set by the College Board and announced the preceding fall. Check AP Central or your school's AP coordinator for this year's exact date.
Yes. The unit-by-unit study guide, worked examples, practice quiz, and full-length timed practice exam on this page are all free, with no login, signup, or paywall.
Unit-by-unit breakdown of what the course actually covers, in teaching order.
Describing distribution shape (symmetric, skewed left/right), center (mean vs. median, and when each is more appropriate), and spread (range, IQR, standard deviation), plus the 1.5×IQR rule for identifying outliers and constructing/interpreting boxplots, histograms, and dotplots for a single quantitative variable.
Scatterplots for visualizing relationships between two quantitative variables, correlation coefficient (r) measuring strength and direction of linear association, least-squares regression line calculation and interpretation (slope and y-intercept in context), residual plots for checking whether a linear model is appropriate, and the distinction between correlation and causation.
Sampling methods (simple random, stratified, cluster, systematic) and their tradeoffs, sources of bias (undercoverage, nonresponse, response bias), and experimental design principles: control groups, randomization, replication, and blocking, plus the crucial distinction that only well-designed experiments (not observational studies) support causal claims.
Basic probability rules (addition rule, multiplication rule, conditional probability, independence), discrete random variables and their probability distributions, expected value and variance calculations, and named distributions including the binomial (fixed trials, two outcomes) and geometric (trials until first success) distributions with their specific formulas.
The sampling distribution of a sample statistic (like the sample mean or proportion) as the theoretical distribution of that statistic across all possible samples, the Central Limit Theorem guaranteeing approximate normality for large samples regardless of the population's shape, and how standard error (not standard deviation) quantifies sample-to-sample variability.
One-sample and two-sample z-procedures for proportions: confidence intervals (estimate ± margin of error) and hypothesis tests (null/alternative hypotheses, p-values, and conclusions in context), including checking the required conditions (random, independent/10% condition, large counts) before running any procedure.
One-sample and two-sample t-procedures for means (using the t-distribution since population standard deviation is typically unknown), paired versus unpaired (independent samples) t-tests and when each applies, and the same four-step inference process (hypotheses, conditions, calculations, conclusion) applied to means instead of proportions.
Chi-square goodness-of-fit tests (comparing one categorical variable's observed distribution to an expected/claimed distribution), and chi-square tests of independence/homogeneity (examining association between two categorical variables), with the chi-square statistic formula and degrees of freedom calculated differently for each test type.
Inference for the slope of a least-squares regression line, testing whether the true population slope differs significantly from zero (i.e., whether a linear relationship genuinely exists) using a t-test, and constructing confidence intervals for the slope — the course's capstone unit connecting regression from earlier back to formal inference.
Concrete, fully worked examples — this is what separates a page reviewers see as genuinely useful from one that just lists topics.
Dataset: 2, 4, 4, 4, 5, 5, 7, 9. Find the median, IQR, and check for outliers.
Median = average of the 4th and 5th values = (4+5)/2 = 4.5. Q1 = 4 (median of lower half), Q3 = 6 (median of upper half), so IQR = 2. Outlier fences: 4 − 1.5(2) = 1 and 6 + 1.5(2) = 9 — no value falls outside [1, 9], so there are no outliers (9 sits exactly at the fence, not beyond it).
Regression: given ŷ = 2 + 3x and r = 0.85, find the residual for the point (4, 20), and interpret r.
Predicted: ŷ = 2 + 3(4) = 14. Residual = observed − predicted = 20 − 14 = 6 (the model underpredicts this point). r = 0.85 shows a strong, positive linear association — but correlation alone never justifies a causal claim.
Design flaw: a study lets volunteers choose whether they receive a new tutoring program, then compares their test scores to those who opted out.
Since assignment isn't random, this is an observational study, not an experiment — motivated students may have self-selected into tutoring, so any score difference could reflect that pre-existing motivation (a confounding variable) rather than the program itself. Only randomized assignment supports a causal conclusion.
Binomial: X = number of successes in 5 independent trials with p = 0.3. Find P(X = 2) and E(X).
P(X=2) = C(5,2)(0.3)²(0.7)³ = 10(0.09)(0.343) = 0.3087. E(X) = np = 5(0.3) = 1.5.
One-sample z-test: H₀: p = 0.5, Hᴥ: p ≠ 0.5, sample p̂ = 0.58, n = 150.
SE = √[0.5(0.5)/150] ≈ 0.0408. z = (0.58−0.5)/0.0408 ≈ 1.96, giving a two-sided p-value right around 0.05 — a genuinely borderline result worth flagging explicitly in a conclusion rather than rounding to a confident yes/no.
Paired t-test: before/after weights on 15 subjects give mean difference d̄ = 3.2, sd = 4.1.
SE = 4.1/√15 ≈ 1.059. t = 3.2/1.059 ≈ 3.02, df = 14 — compare against the critical t-value for the stated significance level to decide.
Goodness-of-fit: observed counts 30, 25, 45 against an expected equal split (33.33 each).
χ² = ∑(O−E)²/E = (30−33.33)²/33.33 + (25−33.33)²/33.33 + (45−33.33)²/33.33 ≈ 6.5, with df = 2.
Slope inference: a regression gives b₁ = 2.5, SE(b₁) = 0.8, n = 20.
t = 2.5/0.8 = 3.125, df = 18 — testing H₀: β₁ = 0 this way is how you formally decide whether a linear relationship genuinely exists, rather than just eyeballing a scatterplot.
The errors that cost the most points across the whole exam, not just one unit.
Unofficial practice questions written for this study guide — not official AP content and not from College Board. Use them to self-check, then use the real past free-response questions below for authentic exam practice.
Unofficial practice, timed to feel like the real thing. Section I pulls from this page's full multiple-choice bank; Section II gives you free-response prompts to write out and self-score against a rubric. Bank size is still growing toward full exam length — timing is paced per-question, not pulled from this subject's official duration.