AP Statistics Exam Guide: Format, Units & Scoring

AP Statistics is the least mathematical and most linguistic of the AP maths courses. The arithmetic rarely goes beyond a mean and a square root; the difficulty is in choosing the right procedure, checking that its conditions hold, and writing a conclusion in context that a reader can follow. That mismatch — easy computation, hard communication — is why students who breeze through algebra sometimes score a 2, and why the exam rewards a specific kind of preparation.

(AI-Math is an independent study tool and is not affiliated with the College Board. Unit weightings are summarised from the published Course and Exam Description, which is revised periodically — check the current version before building a revision plan around exact percentages.)

The nine units

Unit	Topic	Approx. exam weight
1	Exploring one-variable data	15–23%
2	Exploring two-variable data	5–7%
3	Collecting data	12–15%
4	Probability, random variables, probability distributions	10–20%
5	Sampling distributions	7–12%
6	Inference for categorical data: proportions	12–15%
7	Inference for quantitative data: means	10–18%
8	Inference for categorical data: chi-square	2–5%
9	Inference for quantitative data: slopes	2–5%

Add units 6 through 9 together and inference is roughly 30–40% of the exam. Add unit 5, which exists to justify inference, and half the course is really one idea: what happens to a statistic when you repeat the sampling process. Everything before it is scaffolding.

Exam format

The exam runs 3 hours and splits evenly.

Section I — 40 multiple-choice questions, 90 minutes, 50%. About 2 minutes 15 seconds per question. Many are conceptual and can be answered in 30 seconds; a few require a full calculation.

Section II — 6 free-response questions, 90 minutes, 50%. Five shorter questions of roughly 12–13 minutes each, worth about 75% of the section, and one Investigative Task of roughly 25 minutes worth about 25% of the section. The Investigative Task deliberately extends beyond taught material: it gives you a novel statistic or a modified procedure and asks you to reason with it. You are not expected to have seen it before; you are expected to apply what you know to something unfamiliar.

A graphing calculator is permitted on the whole exam, and a formula sheet plus tables for the normal, t and chi-square distributions are provided. Knowing the formulas is not the constraint — knowing which one belongs is.

How the free-response questions are scored

Each FRQ is scored 0 to 4 on a holistic scale: 4 complete response, 3 substantial, 2 developing, 1 minimal. Within a question, each part is first judged essentially correct (E), partially correct (P) or incorrect (I), and those component labels combine into the 0–4 score. Two consequences follow, and they are the single most useful thing to know about this exam:

1. A part that is right but not communicated is a P, not an E. A correct interval with no interpretation, a correct test statistic with no conclusion in context, a correct condition check that names the condition but never verifies it — all lose the same way.
2. Errors do not necessarily cascade. If you use a wrong value from an earlier part correctly in a later part, the later part can still earn full credit. So never abandon a question because you think part (a) went wrong.

The four-step template that earns the points

Almost every inference question on the exam can be answered with the same skeleton. Memorise it and use it every single time:

1. State — name the procedure and the parameter in context. "A one-sample t-interval for the mean commuting time of all employees at this firm."
2. Plan — check the conditions, with evidence. Random: how were the data collected? Independence: is the sample less than 10% of the population, or was there random assignment? Normality: for proportions, are $np \geq 10$ and $n(1-p) \geq 10$; for means, is $n \geq 30$ or is the plotted sample roughly symmetric with no outliers?
3. Do — compute the statistic and the p-value or the interval. Show the values that go in, not just the answer.
4. Conclude — state the decision, link it to the significance level, and put it in context. "Because the p-value of 0.018 is less than 0.05, we reject the null hypothesis. We have convincing evidence that the mean commuting time exceeds 30 minutes for employees at this firm."

Note what step 4 does not say: it never says the null hypothesis is true, and it never says a result is "proven." A failure to reject is a failure to find evidence, not evidence of no effect. Readers deduct for this every year.

Practise the machinery on our hypothesis test calculator, the confidence interval calculator and the p-value calculator — then write the four steps out by hand, because writing is the graded skill.

A worked practice question

A researcher samples 200 randomly selected commuters in a city and finds 46 of them cycle to work. Construct and interpret a 95% confidence interval for the proportion of all commuters in that city who cycle.

State. A one-sample z-interval for $p$, the proportion of all commuters in this city who cycle to work.

Plan. Random: the sample was randomly selected. Independence: 200 is plausibly less than 10% of all commuters in a city. Normality: $n\hat{p} = 46 \geq 10$ and $n(1-\hat{p}) = 154 \geq 10$. Conditions met.

Do. $\hat{p} = 46/200 = 0.23$. The standard error is

$SE = \sqrt{\frac{0.23 \times 0.77}{200}} = \sqrt{0.000885} \approx 0.02976.$

With $z^{*} = 1.96$, the margin of error is $1.96 \times 0.02976 \approx 0.0583$, so the interval is $0.23 \pm 0.058 = (0.172, 0.288)$.

Conclude. We are 95% confident that the true proportion of commuters in this city who cycle to work is between 0.172 and 0.288.

The interpretation sentence is a scored component in its own right. "There is a 95% chance the true proportion is in this interval" would lose it — the confidence applies to the method across repeated samples, not to this one interval.

Where students lose the most points

• Conditions stated but not verified. Writing "10% condition" without saying what the sample and population are earns nothing.
• Conclusions without context. Every conclusion needs the actual variable and population named.
• Confusing a statistic with a parameter. Hypotheses are about parameters, always: $H_0: p = 0.2$, never $H_0: \hat{p} = 0.2$.
• Describing a distribution incompletely. Shape, centre, spread, and unusual features — all four, every time.
• Reversing correlation and causation. Only a randomised experiment supports a causal claim; an observational study never does, however strong the association.

A realistic study plan

Six weeks out, work backwards from the free-response section rather than forwards from the textbook.

Weeks 1–2. Do one released FRQ set per week, then grade yourself against the published scoring guidelines. Your errors will cluster, and they will almost certainly cluster in communication, not computation.

Weeks 3–4. Drill the inference decision tree until choosing the procedure is automatic: one proportion or two? one mean or two? paired or independent? categorical with more than two categories, so chi-square? a slope, so a t-test for the regression line? Getting to the right procedure in 10 seconds is worth more than any formula.

Week 5. Units 1 through 4, which are the cheapest points on the exam and the easiest to forget after months of inference. Descriptive statistics, study design, and probability rules.

Week 6. Two full timed exams, and rewrite three of your weakest FRQ answers from scratch.

Related: Is AP Statistics hard? and AP Statistics FRQ guide