AP · Analysis & Prediction

2023 AP AP Statistics Past Paper: Analysis & Answers

The 2023 AP Statistics Free-Response section consisted of six multi-part questions totaling 24 raw points (scaled to 50% of the exam composite). It evaluated data exploration, experimental design, discrete probability, paired inference for means, bivariate regression inference, and an investigative task integrating sampling distributions of non-standard statistics (the sample range). Mean performance was highest on regression and discrete probability, while the matched pairs t-test proved the most challenging.

Updated: Aug 18, 2026

Analysis Sample paper

Difficulty 3.4/5

Total marks

Duration

90 mins

Most tested

Probability Distributions and Inferential Reasoning

Paper breakdown

Section II, Part A (Questions 1-5): 20 marks · 65 minsSection II, Part B (Investigative Task - Question 6): 4 marks · 25 mins

Top chapters

Unit 4: Probability, Random Variables, and Probability Distributions · 5 marksUnit 1: Exploring One-Variable Data · 5 marksUnit 3: Collecting Data · 4 marksUnit 7: Inference for Quantitative Data: Means · 4 marks

Executive Summary & Difficulty Verdict

The 2023 AP Statistics Free-Response section presented a balanced yet technically rigorous assessment across standard core competencies and non-routine problem solving. With a global mean score of 2.89 and an average score of 0.88/4 on Question 4 (Paired Means Inference), the FRQ section exhibited a moderate-to-challenging difficulty profile. Students demonstrated strong procedural fluency in standard computations such as discrete expected value and least-squares regression arithmetic, but encountered notable friction on experimental precision, paired inference conceptualization, and simulation-based sampling distributions in the Investigative Task.

Where the Marks Were Won and Lost

Unit 1 (One-Variable Data): High accessibility on boxplot construction (Q1b), but marks were dropped in Q1a and Q1c when students failed to systematically address all four distributional characteristics (C-U-S-S: Center, Unusual features, Shape, Spread) with comparative language and units.
Unit 3 (Collecting Data): Q2 saw frequent drops in communication points when students omitted essential randomization mechanics (e.g., failing to specify sampling without replacement or neglecting to define the integer range for random number generators).
Unit 4 (Probability): Discrete calculations in Q3a and Q3c were well handled, but conditional probability in Q3b suffered from the common misstep of assuming independence and multiplying joint events.
Unit 7 (Inference for Means): The steepest mark drop in the exam occurred on Q4, where students persistently misidentified a paired \(t\)-test as a two-sample \(t\)-test and struggled to assess the normality condition using differences in boxplots.
Unit 2 & Unit 9 (Regression & Slope Inference): Q5 was among the highest-scoring questions; however, students regularly forfeited points by interpreting regression slope deterministically rather than stating the predicted or estimated average change.
Investigative Task (Q6): Successfully navigated by students who applied the Central Limit Theorem and sampling distribution standard error \(\sigma_{\bar{x}} = \frac{\sigma}{\sqrt{n}}\), but struggled when interpreting simulated empirical distributions of non-standard statistics (the sample range).

Examiner Pitfalls & Strategic Advice

Definitive Language Trap: Always use non-deterministic phrasing. Never write that a hypothesis test "proves" a claim or that a slope "will increase by exactly \(x\)". Always state there is convincing statistical evidence or that the predicted value changes.
Inference Procedure Identification: Carefully distinguish between the mean of differences (paired data, one-sample \(t\)-procedure) and the difference of two independent means (two-sample \(t\)-procedure).
Direct Comparative Statements: When asked to compare distributions, avoid listing summaries in isolation. Explicitly use comparative conjunctions (e.g., "the median of cold streams is greater than the median of warm streams").

Ready to test yourself?

Turn these notes into exam-style practice. Get unlimited AI questions on this topic with instant marking and explanations.

Practice This Topic

Charts

Marks by chapter

See where the marks were concentrated so revision time goes to the highest-value topics.

Question type mix

Compare the mark share of each paper section and question type.

Constructed Response (Standard FRQ)20 marks · 5 questions
Investigative Task (Extended FRQ)4 marks · 1 questions

Mark accessibility

Estimate which marks were basic, mid-level, or high-difficulty.

75%within easy or medium reach

easy: 7 marksmedium: 11 markshard: 6 marks

Command word frequency

Spot common command words so answers match the expected response style.

Skill weighting

Shows the skill mix this paper tested most heavily.

Study ROI

Bigger bubbles recur more often; higher bubbles carry more marks, helping you rank revision priorities.

Describing and Comparing Distributions (Shape, Center, Spread, Outliers)

5 marks · Difficulty 2 · recurrence 100%

Discrete Probability & Expected Value Calculations

4 marks · Difficulty 2.2 · recurrence 95%

Linear Regression (Slope Interpretation, Residuals, Predictions)

2.5 marks · Difficulty 2.3 · recurrence 90%

Experimental Design & Random Assignment Mechanics

4 marks · Difficulty 3.2 · recurrence 85%

Hypothesis Testing for Means (Paired t-test)

4 marks · Difficulty 4.2 · recurrence 90%

Sampling Distributions of Non-Standard Statistics (Simulation Analysis)

4 marks · Difficulty 4.5 · recurrence 80%

Time vs marks

Compare marks with suggested time allocation to plan exam pacing.

marksmins

Next-year prediction

Topics worth watching next year, with the reason shown directly below each bar.

Inference for Categorical Data: Chi-Square Tests (Goodness of Fit / Association)90%

Unit 8 (Chi-Square) was entirely absent from the 2023 Free-Response operational form and is highly likely to appear in the next test cycle as a primary inference question.

Inference for Categorical Data: Two-Sample Proportions (z-test / z-interval)85%

Unit 6 proportions inference was omitted in favor of paired means and slope inference, making a 2-proportion z-interval or test strongly anticipated.

Sampling Methods & Sources of Bias75%

While Unit 3 was tested via experimental design, observational sampling techniques (stratified, cluster, nonresponse bias) alternate frequently in Question 2.

Cumulative marks ladder

The line is your running mark total question by question; dashed lines are the estimated grade cut-offs. See which question the line crosses your target grade at, so you know how far you must answer cleanly and which questions decide a band.

Examiner insights

Official report

Common pitfalls

Describing a random assignment process without explicitly specifying that slips are drawn 'without replacement' or failing to state the numeric integer range for a random number generator.
Using deterministic language when interpreting regression slopes or stating inference conclusions (e.g., writing 'proves that concrete with fibers prevents cracks' or 'the weight increases by 3.75 kg' without using 'predicted' or 'estimated').
Misapplying conditional probability formulas by multiplying P(A) and P(B) under the false assumption of independence rather than reading joint probabilities directly from the provided discrete distribution table.
Confusing a paired t-test (one-sample mean difference) with a two-sample t-test (difference of two independent means) when analyzing matched-pair pre/post experimental designs.
Forgetting to multiply the one-sided tail probability by 2 when determining the p-value from a t-table for a two-sided test of regression slope.

Misconceptions

Believing that random assignment allows generalization of results to a wider population, rather than establishing causal inference between treatments and response.
Treating expected value as a guarantee for a single trial or rounding it to a whole number rather than recognizing it as a theoretical long-run average.
Believing that the Central Limit Theorem applies to single observations or that a sample size of n = 2 cannot have a normal sampling distribution even when the underlying population is explicitly defined as normal.

Where marks are lost

Question 4, Section 1 & 2: Misidentifying the paired t-test as a two-sample t-test, resulting in incorrect degrees of freedom and test statistics, and failing to verify normality using the sample differences boxplot.
Question 1, Parts (a) & (c): Omitting unusual features (such as gaps or potential outliers) when describing the histogram, and failing to provide direct numerical comparisons between the two distributions.
Question 6, Part (b) & (c): Failing to divide standard deviation by sqrt(n) when calculating probabilities for sample means, and giving an overly wide interval or mode when describing the center of a simulated sampling distribution.

Know what the examiners want. Now go get those marks

thinka is an AI practice app used by 100,000+ students to drill questions and get instant feedback on every answer. You already know the high-frequency topics and where marks are lost. Now practice those specifically on thinka and turn this analysis into actual marks.

Start Practicing Free See pricing

Know what's tested? Drill those topics on thinka. AI questions, instant marking.

Start Practicing Free